Quantitative, Functional and Connectomic Analysis of Brain MRI

Number: 0739

Table Of Contents

Policy

Scope of Policy

This Clinical Policy Bulletin addresses automated quantification, functional magnetic resonance imaging (fMRI) and connectomic analysis of multi-modal brain MRI.

1. Medical Necessity

   Aetna considers functional magnetic resonance imaging (fMRI) medically necessary to identify the eloquent cortex in pre-surgical evaluation of persons with brain tumors (except temporal tumors), epilepsy (except temporal neocortical epilepsy), or vascular malformations.

2. Experimental, Investigational, or Unproven

   Aetna considers the following experimental, investigational or unproven

   1. Connectomic analysis of multi-modal brain MRI;
   2. Automated quantification of brain MRI;
   3. Functional MRI (fMRI) for the following indications: 
       
      1. Use to identify the eloquent cortex in pre-surgical evaluation of persons with temporal neocortical epilepsy or temporal tumors; 
      2. For the diagnosis, monitoring, prognosis, or surgical management of all other indications, including any of the following conditions/diseases (not an all-inclusive list):
         
         
         1. Alzheimer's disease
         2. Anger and aggressive behaviors
         3. Anxiety disorder
         4. Anoxic-ischemic brain injury
         5. Attention-deficit hyperactivity disorder
         6. Autism spectrum disorder
         7. Bipolar disorder
         8. Childhood mal-treatment
         9. Chronic pain (including fibromyalgia)
         10. Disorders of consciousness (e.g., locked-in syndrome, minimally conscious state (subacute/chronic; traumatic/non-traumatic), and coma/vegatative state)
         11. Emotion-expressive suppression
         12. Migraines
         13. Multiple sclerosis
         14. Obsessive-compulsive disorder
         15. Panic disorder
         16. Parkinson's disease
         17. Psychosis
         18. Psychotic depression
         19. Schizophrenia
         20. Sleep behavior disorder
         21. Stroke/stroke rehabilitation
         22. Trauma (e.g., head injury).

3. Related Policies

   • CPB 0279 - Magnetic Source Imaging / Magnetoencephalography

Background

Functional magnetic resonance imaging (fMRI) is a type of functional brain imaging technology. It localizes regions of activity in the brain by measuring blood flow and/or metabolism following task activation, and is generally used to identify the eloquent cortex in the brain. Eloquent cortex refers to specific cortical areas in the brain that directly controls function, such as language (e.g., Broca's area, Wernicke's area) and sensorimotor function (e.g., sensorimotor cortex). Injury, dissection or removal of that area can result in major focal neurological deficits (Byrne, 2016). Functional MRI has been used to map these areas when planning a tumor resection. Furthermore, fMRI has been employed for the diagnosis, monitoring, prognosis, or surgical management of many diseases/conditions (e.g., Alzheimer's disease, brain tumors, epilepsy, multiple sclerosis (MS), Parkinson's disease, stroke, trauma, vascular malformations, and vegetative state/coma).

The bulk of published evidence concerning the clinical applications of fMRI centers on its use in pre-surgical planning. In particular, studies involving language fMRI mainly address its use in pre-surgical planning for epilepsy, arterio-venous malformations (AVMs), and brain tumors (Bookheimer, 2007). It has been suggested that fMRI of the brain reduces the need for invasive testing of seizure disorder patients being considered for surgical treatment. Woermann et al (2003) compared the determination of language dominance using fMRI with results of the Wada test in 100 patients with different localization-related epilepsies. These investigators found 91 % concordance between both tests. The overall rate of false categorization by fMRI was 9 %, ranging from 3 % in left-sided temporal lobe epilepsy (TLE) to 25 % in left-sided extra-temporal epilepsy. The authors noted that language fMRI might reduce the necessity of the Wada test for language lateralization, especially in TLE.

Sabsevitz and colleagues (2003) examined whether pre-operative fMRI predicts language deficits in patients with epilepsy undergoing left anterior temporal lobectomy (L-ATL). A total of 24 patients with L-ATL underwent pre-operative language mapping with fMRI, pre-operative intra-carotid sodium amobarbital (Amytal)/Wada testing for language dominance, as well as pre- and post-operative neuropsychological testing. Functional MRI laterality indexes (LIs), reflecting the inter-hemispheric difference between activated volumes in left and right homologous regions of interest, were calculated for each patient. Relationships between the fMRI LI, Wada language dominance, and naming outcome were examined. Both the fMRI LI (p < 0.001) and the Wada test (p < 0.05) were predictive of naming outcome. Functional MRI showed 100 % sensitivity and 73 % specificity in predicting significant naming decline. Both fMRI and the Wada test were more predictive than age at seizure onset or pre-operative naming performance. The authors concluded that pre-operative fMRI predicted naming decline in patients undergoing L-ATL surgery.

Medina et al (2005) prospectively evaluated effect of fMRI on diagnostic work-up and treatment planning in patients with seizure disorders who are candidates for surgical treatment. A total of 60 consecutively enrolled patients (27 females and 33 males; mean age of 15.8 +/- 8.7 years; range of 6.8 to 44.2 years) were examined. Forty-five (75 %) patients were right-handed, 9 (15 %) were left-handed, and 6 (10 %) had indeterminate hand dominance. Prospective questionnaires were used to evaluate diagnostic work-up, counseling, and treatment plans of the seizure team before and after fMRI. Confidence level scales were used to determine effect of fMRI on diagnostic and therapeutic thinking. Paired-t test and 95 % confidence interval analyses were performed. In 53 patients, language mapping was performed; in 33, motor mapping; and in 7, visual mapping. The study revealed change in anatomical location or lateralization of language-receptive area -- (Wernicke's area) (28 % of patients) as well as language-expressive area (Broca's area) (21 % of patients). Statistically significant increases were found in confidence levels after fMRI in regard to motor and visual cortical function evaluation. In 35 (58 %) of 60 patients, the seizure team thought that fMRI results altered patient and family counseling. In 38 (63 %) of 60 patients, fMRI results helped to avoid further studies, including the Wada test. In 31 (52 %) and 25 (42 %) of 60 patients, intra-operative mapping and surgical plans, respectively, were altered because of fMRI results. In 5 (8 %) patients, two-stage surgery with extra-operative direct electrocortical stimulation mapping (ESM) was averted, and resection was accomplished in one-stage. In 4 (7 %) patients, extent of surgical resection was altered because eloquent areas were identified close to seizure focus. The authors concluded that fMRI results influenced diagnostic and therapeutic decision making of the seizure team; results indicated a change in language dominance, an increase in confidence level in identification of critical brain function areas, alterations in patient and family counseling as well as intra-operative mapping and surgical approach.

Functional MRI has been used in pre-surgical planning for patients with brain tumors as well as vascular malformations. Pouratian and colleagues (2002) evaluated the utility of pre-operative fMRI to predict if a given cortical area would be deemed essential for language processing by ESM. These investigators studied patients with vascular malformations, specifically AVMs and cavernous angiomas, in whom blood-flow patterns are abnormal and in whom a perfusion-dependent mapping signal may be questionable. A total of 10 patients were studied (7 with AVMs and 3 with cavernous angiomas). These researchers used a battery of linguistic tasks, including visual object naming, word generation, auditory responsive naming, visual responsive naming, and sentence comprehension, to identify brain regions that were consistently activated across expression and comprehension linguistic tasks. In a comparison of ESM and fMRI activations, the researchers varied the matching criteria (overlapping activations, adjacent activations, and deep activations) and the radii of influence of ESM (2.5, 5, and 10 mm) to determine the effects of these factors on the sensitivity and specificity of fMRI. The sensitivity and specificity of fMRI were dependent on the task, lobe, and matching criterion. For the population studied, the sensitivity and specificity of fMRI activations during expressive linguistic tasks were found to be up to 100 % and 66.7 %, respectively, in the frontal lobe, and during comprehension linguistic tasks up to 96.2 % and 69.8 %, respectively, in the temporal and parietal lobes. The sensitivity and specificity of each disease population (AVMs and cavernous angiomas) and of individuals were consistent with those values reported for the entire population studied. The authors concluded that pre-operative fMRI is a highly sensitive pre-operative planning tool for identifying cortical areas that are essential for language; and that this imaging modality may play a future role in pre-surgical planning for patients with vascular malformations.

Anderson et al (2006) examined the utility of fMRI as a determinant of lateralization of expressive language in children with cerebral lesions. Functional MRI language lateralization was attempted in 35 children (29 with epilepsy) aged 8 to 18 years with frontal or temporal lobe lesions (28 left hemisphere, 5 right hemisphere, and 2 bilateral). Axial and coronal fMRI scans through the frontal and temporal lobes were acquired at 1.5 Tesla (T) by using a block-design, covert word-generation paradigm. Activation maps were lateralized by blinded visual inspection and quantitative asymmetry indices (hemispheric and inferior frontal regions of interest, at p < 0.001 uncorrected and p < 0.05 Bonferroni corrected). A total of 30 children showed significant activation in the inferior frontal gyrus. Lateralization by visual inspection was left in 21, right in 6, and bilateral in 3, and concordant with hemispheric and inferior frontal quantitative lateralization in 93 % of cases. Developmental tumors and dysplasias involving the inferior left frontal lobe had activation overlying or abutting the lesion in 5 of 6 cases. Functional MRI language lateralization was corroborated in 6 children by frontal cortex stimulation or intra-carotid Amytal testing (IAT) and indirectly supported by aphasiology in a further 6 cases. In 2 children, fMRI language lateralization was bilateral, and corroborative methods of language lateralization were left. Neither lesion lateralization, patient handedness, nor developmental versus acquired nature of the lesion was associated with language lateralization. Involvement of the left inferior or middle frontal gyri increased the likelihood of atypical language lateralization. The authors concluded that this study suggests that fMRI lateralizes language in children with cerebral lesions.

Stancanello et al (2007) attempted to validate a method to exploit functional information for the identification of functional organs at risk (fOARs) in CyberKnife radiosurgery treatment planning. Five patients affected by AVMs and scheduled to undergo radiosurgery were scanned prior to treatment using computed tomography (CT), three-dimensional rotational angiography (3D-RA), T2 weighted and blood oxygenation level dependent echo planar imaging MRI. Tasks were chosen on the basis of lesion location by considering those areas which could be potentially close to treatment targets. Functional data were superimposed on 3D-RA and CT used for treatment planning. The procedure for the localization of fMRI areas was validated by direct ESM on 38 AVM and tumor patients undergoing conventional surgery. Treatment plans studied with and without considering fOARs were significantly different, in particular with respect to both maximum dose and dose volume histograms; consideration of the fOARs allowed quality indices of treatment plans to remain almost constant or to improve in 4 out of 5 cases compared to plans with no consideration of fOARs. The authors concluded that the presented method provides an accurate tool for the integration of functional information into AVM radiosurgery, which might help to minimize undesirable side effects and to make radiosurgery less invasive.

There is evidence for the use of fMRI in pre-surgical planning for epilepsy and monitoring of language function during tumor resection.

Roux et al (2003) analyzed the usefulness of pre-operative language fMRI by correlating fMRI data with intra-operative ESM results for patients with brain tumors. Naming and verb generation tasks were used, separately or in combination, for 14 right-handed patients with tumors in the left hemisphere. Acquired fMRI data were analyzed with statistical parametric mapping software, with two standard analysis thresholds (p < 0.005 and then p < 0.05). The fMRI data were then registered in a frameless stereotactic neuro-navigational device and correlated with direct brain mapping results. These researchers used a statistical model with the fMRI information as a predictor, spatially correlating each intra-operatively mapped cortical site with fMRI data integrated in the neuro-navigational system (site-by-site correlation). Eight patients were also studied with language fMRI post-operatively, with the same acquisition protocol. These investigators observed high variability in signal extents and locations among patients with both tasks. The activated areas were located mainly in the left hemisphere in the middle and inferior frontal gyri (F2 and F3), the superior and middle temporal gyri (T1 and T2), and the supra-marginal and angular gyri. A total of 426 cortical sites were tested for each task among the 14 patients. In frontal and temporo-parietal areas, poor sensitivity of the fMRI technique was observed for the naming and verb generation tasks (22 % and 36 %, respectively) with p < 0.005 as the analysis threshold. Although not perfect, the specificity of the fMRI technique was good in all conditions (97 % for the naming task and 98 % for the verb generation task). Better correlation (sensitivity, 59 %; specificity, 97 %) was achieved by combining the two fMRI tasks. Variation of the analysis threshold to p < 0.05 increased the sensitivity to 66 % while decreasing the specificity to 91 %. Post-operative fMRI data (for the cortical brain areas studied intra-operatively) were in accordance with brain mapping results for 6 of 8 patients. Complete agreement between pre- and post-operative fMRI studies and direct brain mapping results was observed for only 3 of 8 patients. The authors concluded that with the paradigms and analysis thresholds used in this study, language fMRI data obtained with naming or verb generation tasks, before and after surgery, were imperfectly correlated with intra-operative brain mapping results. A better correlation could be obtained by combining the fMRI tasks. The overall results of this study showed that language fMRI could not be used to make critical surgical decisions in the absence of direct brain mapping. Other acquisition protocols are needed for evaluation of the potential role of language fMRI in the accurate detection of essential cortical language areas.

Benke and associates (2006) noted that recent studies have claimed that language fMRI can identify language lateralization in patients with TLE and that fMRI-based findings are highly concordant with the conventional assessment procedure of speech dominance, the IAT. These researchers attempted to establish the power of language fMRI to detect language lateralization during pre-surgical assessment and compared the findings of a semantic decision paradigm with the results of a standard IAT in 68 patients with chronic intractable right and left TLE (rTLE, n = 28; lTLE, n = 40) who consecutively underwent a pre-surgical evaluation program. The patient group also included 14 (20.6 %) subjects with atypical (bilateral or right hemisphere) speech. Four raters used a visual analysis procedure to determine the laterality of speech-related activation individually for each patient. Overall congruence between fMRI-based laterality and the laterality quotient of the IAT was 89.3 % in rTLE and 72.5 % in lTLE patients. Concordance was best in rTLE patients with left speech. In lTLE patients, language fMRI identified atypical, right hemisphere speech dominance in every case, but missed left hemisphere speech dominance in 17.2 %. Frontal activations had higher concordance with the IAT than did activations in temporo-parietal or combined regions of interest. Because of methodological problems, recognition of bilateral speech was difficult. The authors concluded that these data provide evidence that language fMRI as used in the present study has limited correlation with the IAT, especially in patients with lTLE and with mixed speech dominance. They noted that further refinements regarding the paradigms and analysis procedures will be needed to improve the contribution of language fMRI for pre-surgical assessment.

Petrella and colleagues (2006) prospectively evaluated the effect of pre-operative fMRI localization of language and motor areas on therapeutic decision making in patients with potentially resectable brain tumors. A total of 39 consecutive patients (19 men, 20 women; mean age of 42.2 years) referred for fMRI for possible tumor resection were evaluated. A pre-operative diagnosis of brain tumor was made in all patients. Sentence completion and bilateral hand squeeze tasks were used to map language and sensorimotor areas. Neurosurgeons completed questionnaires regarding the proposed treatment plan before and after fMRI and after surgery. They also gave confidence ratings for fMRI results and estimated the effect on surgical time, extent of resection, and surgical approach. The effect of fMRI on changes in treatment plan was assessed with the Wilcoxon signed rank test. Differences in confidence ratings between altered and un-altered treatment plans were assessed with the Mann-Whitney U test. The estimated influence of fMRI on surgical time, extent of resection, and surgical approach was denoted with summary statistics. Treatment plans before and after fMRI differed in 19 patients (p < 0.05), with a more aggressive approach recommended after imaging in 18 patients. There were no significant differences in confidence ratings for fMRI between altered and un-altered plans. Functional MRI resulted in reduced surgical time (estimated reduction, 15 to 60 minutes) in 22 patients who underwent surgery, a more aggressive resection in 6, and a smaller craniotomy in 2. The authors concluded that fMRI enables the selection of a more aggressive therapeutic approach than might otherwise be considered because of functional risk. In certain patients, surgical time may be shortened, the extent of resection increased, and craniotomy size decreased. 

Di et al (2007) assessed the differences in brain activation in response to presentation of the patient's own name spoken by a familiar voice (SON-FV) in patients with vegetative state (VS) and minimally conscious state (MCS). By using fMRI, these investigators prospectively studied residual cerebral activation to SON-FV in 7 patients with VS and 4 patients with MCS. Behavioral evaluation was performed by means of standardized testing up to 3 months post-fMRI. Two patients with VS failed to show any significant cerebral activation, while 3 patients with VS showed SON-FV induced activation within the primary auditory cortex. Finally, 2 patients with VS and all 4 patients with MCS not only showed activation in primary auditory cortex but also in hierarchically higher order associative temporal areas. The 2 patients with VS showing the most widespread activation subsequently showed clinical improvement to MCS observed 3 months after their fMRI scan. The authors concluded that cerebral responses to patient's own name spoken by a familiar voice as measured by fMRI might be a useful tool to pre-clinically distinguish MCS-like cognitive processing in some patients behaviorally classified as vegetative. 

The American College of Radiology (ACR)'s guideline on neurological imaging for patients with epilepsy (Karis et al, 2006) noted that the data provided by MRI are essential in the pre-surgical evaluation of patients with medically refractory epilepsy, but noted that structurally detectable abnormalities are absent in many patients. In these patients, functional studies provide useful information on localization of the seizure focus. In this regard, functional imaging techniques, including positron emission tomography, single-photon emission computed tomography, magnetic source imaging, and fMRI, have contributed to the pre-surgical evaluation of patients with epilepsy. The ACR guideline provided appropriateness ratings (1 = least appropriate; 9 = most appropriate) on fMRI for the following indications:

• Chronic epilepsy, poor therapeutic response. Surgery candidate (rating = 5; may be helpful in pre-surgical planning).
• New onset of seizure. Ethyl alcohol, and/or drug-related (rating = 2).
• New onset seizure. Aged 18 to 40 years (rating = 2).
• New onset seizure. Aged greater than 40 years (rating = 2).
• New onset seizure. Focal neurological deficit (rating = 2).

Additionally, the ACR's guideline on neurological imaging for patients with head trauma (Davis et al, 2006) provided an appropriateness rating of 2 for patients with sub-acute or chronic closed head injury with cognitive and/or neurological deficit(s).

The Ontario Ministry of Health and Long-Term Care's review on functioning brain imaging (2006) stated that there may be a role for fMRI in the identification of surgical candidates for tumor resection. The review also stated that there may be some clinical utility for fMRI in pre-surgical functional mapping.

The assessment by the Ontario Ministry of Health and Long-Term Care (2006) stated that there is limited clinical utility of functional brain imaging in the management of patients with MS at this time. This is in agreement with the European Federation of Neurological Societies' guideline on the use of neuro-imaging in the management of MS (Filippi et al, 2006), which stated that the use of non-conventional MRI techniques (e.g., fMRI, diffusion tensor MRI, magnetization transfer MRI, and MR spectroscopy) is not recommended.

Rocca and colleagues (2008) used fMRI to examine the properties of the mirror neuron system (MNS) in patients with MS. Using a 3 tesla scanner, these researchers acquired fMRI in 16 right-handed patients with relapsing-remitting MS and 14 controls. Two motor tasks were studied. The first consisted of repetitive flexion-extension of the last 4 fingers of the right hand (simple task) alternated to epochs of rest; the second (MNS task) consisted of observation of a movie showing the hand of another subject while performing the same task. During the simple task, compared to controls, patients with MS had more significant activations of the contralateral primary sensori-motor cortex and supplementary motor area. During the MNS task, both groups showed the activation of several visual areas, the infra-parietal sulcus, and the inferior frontal gyrus (IFG), bilaterally. The IFG and the visual areas were significantly more active in patients than controls. The between-group interaction analysis showed that in patients with MS, part of the regions of the MNS were more active also during the simple task. The authors concluded that the findings of this study suggested increased activation of the MNS in patients with MS with a normal level of function and widespread damage of the central nervous system. The potentialities of this system in facilitating clinical recovery in patients with MS and other neurological conditions should be investigated.

In an editorial that accompanied the afore-mentioned article, Phillips (2008) stated that fMRI has tremendous potential for assessing and better understanding MS. He noted that it is important to remember that fMRI is an indirect measurement of neuronal activity. Also, it has been reported that there is altered brain perfusion in patients with MS. Changes in perfusion may alter the sensitivity and statistical characteristics of fMRI. Currently, it is unclear to what extent altered tissue perfusion complicates the interpretation of fMRI in MS. Furthermore, the enhanced activation patterns observed in MS have also been shown in other neurological conditions such as Alzheimer's disease, Parkinson's disease, and stroke.

In a randomized, double-blind, placebo-controlled study, Atri et al (2011) examined the feasibility and test-retest reliability of encoding-task fMRI in mild Alzheimer disease (AD). These investigators studied 12 patients with mild AD (mean [SEM] Mini-Mental State Examination score, 24.0 [0.7]; mean Clinical Dementia Rating score, 1.0) who had been taking donepezil hydrochloride for more than 6 months from the placebo-arm of a larger 24-week study (n = 24, 4 scans on weeks 0, 6, 12, and 24, respectively). They performed whole-brain t maps (p < 0.001, 5 contiguous voxels) and hippocampal regions-of-interest analyses of extent (percentage of active voxels) and magnitude (percentage of signal change) for novel-greater-than-repeated face-name contrasts. These researchers also calculated intra-class correlation coefficients and power estimates for hippocampal regions of interest. Task tolerability and data yield were high (95 of 96 scans yielded favorable-quality data). Whole-brain maps were stable. Right and left hippocampal regions-of-interest intra-class correlation coefficients were 0.59 to 0.87 and 0.67 to 0.74, respectively. To detect 25.0 % to 50.0 % changes in week-0 to week-12 hippocampal activity using left-right extent or right magnitude with 80.0 % power (2-sided α = 0.05) requires 14 to 51 patients. Using left magnitude requires 125 patients because of relatively small signal to variance ratios. The authors concluded that encoding-task fMRI was successfully implemented in a single-site, 24-week, AD randomized controlled trial. Week 0 to 12 whole-brain t maps were stable, and test-retest reliability of hippocampal fMRI measures ranged from moderate to substantial. Right hippocampal magnitude may be the most promising of these candidate measures in a leveraged context. These initial estimates of test-retest reliability and power justify evaluation of encoding-task fMRI as a potential biomarker for signal of effect in exploratory and proof-of-concept trials in mild AD. They stated that validation of these results with larger sample sizes and assessment in multi-site studies is warranted.

Burgmer and colleagues (2010) stated that studies with functional neuroimaging support the hypothesis of central pain augmentation in fibromyalgia syndrome (FMS) with functional differences in areas of the medial pain system. These investigators examined if these findings are unique to patients with FMS. BOLD-signal patterns during and before tonic experimental pain were compared to healthy controls and patients with rheumatoid arthritis (RA) as a chronic pain disorder of somatic origin. These researchers expected different BOLD-signal patterns in areas of the medial pain system that were most pronounced in patients with FMS. An fMRI-block design before, during and after an incision was performed in patients with FMS (n = 17), RA (n = 16) and in healthy controls (n = 17). A 2-factorial model of BOLD-signal changes was designed to explore significant differences of brain activation between the groups during the pain stimulus. Additionally, the correlation of brain activity during the anticipation of pain with the amount of the impending pain was determined. These researchers observed a FMS-unique temporal brain activation of the frontal cortex in patients with FMS. Moreover, areas of the motor cortex and the cingulate cortex presented a FMS-specific relation between brain activity during pain anticipation and the magnitude of the subsequent pain experience. The authors concluded that these findings support the hypothesis that central mechanisms of pain processing in the frontal cortex and cingulate cortex may play an important role in patients with FMS.

Tregellas et al (2010) noted that 3-(2,4-Dimethoxybenzylidene)-anabaseine (DMXB-A) is a partial agonist at alpha7-nicotinic acetylcholine receptors and is now in early clinical development for treatment of deficits in neurocognition and sensory gating in schizophrenia. During its initial phase II test, fMRI studies were conducted to determine whether the drug had its intended effect on hippocampal inhibitory interneurons. Increased hemodynamic activity in the hippocampus in schizophrenia is found during many tasks, including smooth pursuit eye movements, and may reflect inhibitory dysfunction. Placebo and 2 doses of drug were administered in a random, double-blind cross-over design. After the morning drug/placebo ingestion, subjects underwent fMRI while performing a smooth pursuit eye movement task. Data were analyzed from 16 non-smoking patients, including 7 women and 9 men. The 150-mg dose of DMXB-A, compared with placebo, diminished the activity of the hippocampus during pursuit eye movements, but the 75-mg dose was ineffective. The effect at the 150-mg dose was negatively correlated with plasma drug levels. The findings are consistent with the previously established function of alpha7-nicotinic receptors on inhibitory interneurons in the hippocampus and with genetic evidence for deficits in these receptors in schizophrenia. Imaging of drug response is useful in planning future clinical tests of this compound and other nicotinic agonists for schizophrenia.

Astrakas et al (2012) stated that the number of individuals suffering from stroke is increasing daily, and its consequences are a major contributor to invalidity in today's society. Stroke rehabilitation is relatively new, having been hampered from the long-standing view that lost functions were not recoverable. Nowadays, robotic devices, which aid by stimulating brain plasticity, can assist in restoring movement compromised by stroke-induced pathological changes in the brain that can be monitored by MRI. Multi-parametric MRI of stroke patients participating in a training program with a novel Magnetic Resonance Compatible Hand-Induced Robotic Device (MR_CHIROD) could yield a promising biomarker that, ultimately, will enhance the ability to advance hand motor recovery following chronic stroke. Using state-of-the art MRI in conjunction with MR_CHIROD-assisted therapy can provide novel biomarkers for stroke patient rehabilitation extracted by a meta-analysis of data. Successful completion of such studies may provide a ground breaking method for the future evaluation of stroke rehabilitation therapies. Their results will attest to the effectiveness of using MR-compatible hand devices with MRI to provide accurate monitoring during rehabilitative therapy. Furthermore, such results may identify biomarkers of brain plasticity that can be monitored during stroke patient rehabilitation. The potential benefit for chronic stroke patients is that rehabilitation may become possible for a longer period of time after stroke than previously thought, unveiling motor skill improvements possible even after 6 months due to retained brain plasticity.

Wager et al (2013) noted that persistent pain is measured by means of self-report, the sole reliance on which hampers diagnosis and treatment. Functional magnetic resonance imaging holds promise for identifying objective measures of pain, but brain measures that are sensitive and specific to physical pain have not yet been identified. In 4 studies involving a total of 114 participants, these researchers developed an fMRI-based measure that predicts pain intensity at the level of the individual person. In study 1, they used machine-learning analyses to identify a pattern of fMRI activity across brain regions -- a neurologic signature -- that was associated with heat-induced pain. The pattern included the thalamus, the posterior and anterior insulae, the secondary somatosensory cortex, the anterior cingulate cortex, the peri-aqueductal gray matter, and other regions. In study 2, these investigators tested the sensitivity and specificity of the signature to pain versus warmth in a new sample. In study 3, they assessed specificity relative to social pain, which activates many of the same brain regions as physical pain. In study 4, these researchers evaluated the responsiveness of the measure to the analgesic agent remifentanil. In study 1, the neurologic signature showed sensitivity and specificity of 94 % or more (95 % confidence interval [CI]: 89 to 98) in discriminating painful heat from non-painful warmth, pain anticipation, and pain recall. In study 2, the signature discriminated between painful heat and non-painful warmth with 93 % sensitivity and specificity (95 % CI: 84 to 100). In study 3, it discriminated between physical pain and social pain with 85 % sensitivity (95 % CI: 76 to 94) and 73 % specificity (95 % CI, 61 to 84) and with 95 % sensitivity and specificity in a forced-choice test of which of 2 conditions was more painful. In study 4, the strength of the signature response was substantially reduced when remifentanil was administered. The authors concluded that it is possible to use fMRI to assess pain elicited by noxious heat in healthy persons. Moreover, they state that future studies are needed to assess whether the signature predicts clinical pain.

Magland and Childress (2014) stated that real-time fMRI is especially vulnerable to task-correlated movement artifacts because statistical methods normally available in conventional analyses to remove such signals cannot be used in the context of real-time fMRI. Multi-voxel classifier-based methods, although advantageous in many respects, are particularly sensitive. These researchers systematically studied various movements of the head and face to determine to what extent these can "masquerade" as signal in multi-voxel classifiers. A total of 10 subjects were instructed to move systematically (12 instructed movements) throughout fMRI exams and data from a previously published real-time study was also analyzed to determine the extent to which non-neural signals contributed to the high reported accuracy in classifier output. Of potential concern, whole-brain classifiers based solely on movements exhibited false positives in all cases (p < 0.05). Artifacts were also observed in the spatial activation maps for 2 of the 12 movement tasks. In the retrospective analysis, it was determined that the relatively high reported classification accuracies were (fortunately) mostly explainable by neural activity, but that in some cases performance was likely dominated by movements. The authors concluded that movement tasks of many types (including movements of the body, eyes, and face) can lead to false positives in classifier-based real-time fMRI paradigms.

The University of Michigan Health System’s clinical guideline on "Attention-deficit hyperactivity disorder" (2013) listed functional magnetic resonance imaging as one of the search terms for the update of a previous version of this guideline. Moreover, the updated guideline stated that "Diagnosis is based on the Diagnostic and Statistical Manual of Mental Disorders, fourth edition (DSM-IV-TR) criteria. The three main types are primary hyperactive, primary inattentive, and combined. No specific test can make the diagnosis".

Furthermore, UpToDate reviews on "Attention deficit hyperactivity disorder in children and adolescents: Clinical features and evaluation" (Krull, 2014) and "Adult attention deficit hyperactivity disorder in adults: Epidemiology, pathogenesis, clinical features, course, assessment, and diagnosis" (Bukstein, 2014) do not mention the use of fMRI as a diagnostic tool.

Wager and colleagues (2013) noted that persistent pain is measured by means of self-report, the sole reliance on which hampers diagnosis and treatment. Functional MRI holds promise for identifying objective measures of pain, but brain measures that are sensitive and specific to physical pain have not yet been identified. In 4 studies involving a total of 114 participants, these researchers developed an fMRI-based measure that predicts pain intensity at the level of the individual person. In study 1, they used machine-learning analyses to identify a pattern of fMRI activity across brain regions -- a neurologic signature -- that was associated with heat-induced pain. The pattern included the thalamus, the posterior and anterior insulae, the secondary somatosensory cortex, the anterior cingulate cortex, the periaqueductal gray matter, and other regions. In study 2, these researchers tested the sensitivity and specificity of the signature to pain versus warmth in a new sample. In study 3, they assessed specificity relative to social pain, which activates many of the same brain regions as physical pain. In study 4, these investigators assessed the responsiveness of the measure to the analgesic agent remifentanil. In study 1, the neurologic signature showed sensitivity and specificity of 94 % or more (95 % CI: 89 to 98) in discriminating painful heat from non-painful warmth, pain anticipation, and pain recall. In study 2, the signature discriminated between painful heat and non-painful warmth with 93 % sensitivity and specificity (95 % CI: 84 to 100). In study 3, it discriminated between physical pain and social pain with 85 % sensitivity (95 % CI: 76 to 94) and 73 % specificity (95 % CI: 61 to 84) and with 95 % sensitivity and specificity in a forced-choice test of which of 2 conditions was more painful. In study 4, the strength of the signature response was substantially reduced when remifentanil was administered. The authors concluded that it is possible to use fMRI to assess pain elicited by noxious heat in healthy persons. Moreover, they stated that future studies are needed to assess whether the signature predicts clinical pain.

Furthermore, the Work Loss Data Institute’s guideline on "Pain (chronic)" (2013) listed fMRI as one of the interventions that were considered, but are not recommended.

UpToDate reviews on "Locked-in syndrome’ (Caplan, 2015) and "Stupor and coma in adults" (Young, 2015) do not mention functional MRI as a management tool.

Furthermore, an UpToDate review on "Treatment and prognosis of coma in children’ (Thompson and Williams, 2015) states that "Other neuroimaging modalities, MR spectroscopy, functional MRI, positron emission tomography are not useful in the evaluation of coma prognosis]. Studies, awaiting validation, suggest that these tools may help discriminate between persistent vegetative state, minimally conscious state and other states of impaired consciousness".

Anoxic-Ischemic Brain Injury

An UpToDate review on "Hypoxic-ischemic brain injury: Evaluation and prognosis" (Weinhouse and Young, 2016) states that "In the future, larger studies may find a role for standard MRI as well as functional neuroimaging, such as positron emission tomography (PET) and functional MRI (fMRI), in the prognostic assessment of adults with anoxic-ischemic brain injury …. fMRI studies have the potential to detect network processing of sensory and motor responses, showing some evidence of awareness in a small proportion of behaviorally unresponsive patients. However, the performance and interpretation of these studies remains complex and is still investigational. There are also ethical issues regarding quality of life in decision-making that need to be resolved, namely whether patients who can generate such binary responses can participate in a decision-making process".

Psychotic Depression

O'Connor and Agius (2015) stated that psychotic depression is widely accepted as a specific subtype of unipolar major depression. Magnetic resonance imaging studies have begun to investigate the neurobiological changes that differentiate this subtype of major depression from non-psychotic depression. Any differences may eventually be useful in aiding diagnosis patients for whom there is diagnostic uncertainty. This review collated the currently available evidence. These investigators performed a systematic search of the Medline, PubMed, Embase & Web of Science databases was used to identify all articles comparing structural grey matter or fMRI differences between adults (18 years or older) with previously diagnosed psychotic and non-psychotic depression in pre-defined regions of interest (hippocampus, amygdala, cingulate, insula and frontal cortices). The results were collated and organized according to brain region. There was a paucity of studies addressing structural and functional changes differentiating these 2 disorders and recommendations regarding use of these modalities in diagnosis cannot be made. From the available studies decreases in frontal cortex grey matter volumes may differentiate psychotic from non-psychotic depression while further studies are needed to confirm decreases in insula cortex volumes. Functional MRI studies showed associations between altered activity in these 2 regions and cognitive impairments in patients with psychotic depression. The volumes of putative emotional processing regions including the amygdala, hippocampus and anterior cingulate showed no difference between psychotic and non-psychotic depression. The authors concluded that structural and functional changes in the higher associative regions of the frontal and insular cortices appeared to differentiate psychotic and non-psychotic depression to a greater degree than changes in putative emotional processing regions. The quality of the evidence both in terms of numbers of studies available and sample sizes involved was very poor; but in regard to directing future study, understanding the neurobiology of psychotic depression may benefit from a more detailed assessment of these 2 regions.

Temporal Neocortical Epilepsy and Temporal Tumors

On behalf of the American Academy of Neurology (AAN), Szaflarski and colleagues (2017) evaluated the diagnostic accuracy and prognostic value of fMRI in determining lateralization and predicting post-surgical language and memory outcomes. An 11-member panel rated available evidence according to the 2004 AAN process. At least 2 panelists reviewed the full text of 172 articles and selected 37 for data extraction. Case reports, reports with less than 15 cases, meta-analyses, and editorials were excluded. The authors concluded that the use of fMRI may be considered an option for lateralizing language functions in place of intracarotid amobarbital procedure (IAP) in patients with medial temporal lobe epilepsy (MTLE; Level C), temporal epilepsy in general (Level C), or extra-temporal epilepsy (Level C). For patients with temporal neocortical epilepsy or temporal tumors, the evidence is insufficient (Level U). They stated that fMRI may be considered to predict post-surgical language deficits after anterior temporal lobe resection (Level C). The use of fMRI may be considered for lateralizing memory functions in place of IAP in patients with MTLE (Level C), but is of unclear utility in other epilepsy types (Level U). Moreover, they stated that fMRI of verbal memory or language encoding should be considered for predicting verbal memory outcome (Level B); and fMRI using non-verbal memory encoding may be considered for predicting visuospatial memory outcomes (Level C). These investigators noted that pre-surgical fMRI could be an adequate alternative to IAP memory testing for predicting verbal memory outcome (Level C).

Anxiety Disorder

Wang and colleagues (2018) noted that impairments in emotion regulation, and more specifically in cognitive re-appraisal, are thought to play a key role in the pathogenesis of anxiety disorders. However, the available evidence on such deficits is inconsistent. To further illustrate the neurobiological underpinnings of anxiety disorder, the present meta-analysis summarized fMRI findings for cognitive re-appraisal tasks and investigated related brain areas. These investigators performed a comprehensive series of meta-analyses of cognitive reappraisal fMRI studies contrasting patients with anxiety disorder with healthy control (HC) subjects, employing an anisotropic effect-size signed differential mapping approach. They also conducted a subgroup analysis of medication status, anxiety disorder subtype, data-processing software, and MRI field strengths. Meta-regression was used to explore the effects of demographics and clinical characteristics. A total of 8 studies, with 11 datasets including 219 patients with anxiety disorder and 227 HC, were identified. Compared with HC, patients with anxiety disorder showed relatively decreased activation of the bilateral dorsomedial prefrontal cortex (dmPFC), bilateral dorsal anterior cingulate cortex (dACC), bilateral supplementary motor area (SMA), left ventromedial prefrontal cortex (vmPFC), bilateral parietal cortex, and left fusiform gyrus during cognitive re-appraisal. The subgroup analysis, jackknife sensitivity analysis, heterogeneity analysis, and Egger's tests further confirmed these findings. The authors concluded that they identified the most robust functional neuroimaging findings on cognitive re-appraisal in anxiety disorder. The results demonstrated that patients with anxiety disorder could not recruit the pre-fronto-parietal network, including the dmPFC, dACC, SMA, vmPFC, and parietal cortex, to down-regulate their emotion response. These findings provided robust evidence that impairment of pre-fronto-parietal neuronal circuits may play an important role in the pathogenesis of anxiety disorder. This finding may provide novel targets for medical or cognitive-behavioral interventions and neuromodulation approaches (e.g., transcranial magnetic stimulation). These researchers stated that with longitudinal data, future investigations should further explore whether these functional abnormities are associated with structural changes or influenced by disease severity and medication status.

The authors stated that this study had several drawbacks. First, the number of fMRI studies included was small; the literature search yielded only 8 studies with 11 relevant patients versus controls comparisons. This could affect the generalizability of these findings, particularly in the subgroup meta-analyses and meta-regressions analyses. Second, this meta-analysis was based on co-ordinates from published studies rather than raw statistical maps, which might reduce its accuracy. Third, the heterogeneity of the data acquisition and analysis techniques, including MRI field strengths, slice thickness, voxel size, and data-processing software, may reduce the accuracy of these results. Fourth, this meta-analysis included studies of medicine-naive patients who had undergone a medication wash-out period before scanning, so that longer-term influences of medication on brain function could not be completely excluded. Although these researchers conducted a subgroup meta-analysis of the medicine-naive, these results should be interpreted with caution. Finally, some patients with anxiety disorder had co-morbid major depression. Anxiety and major depressive disorders may have different disorder-specific deficits in the neural mechanisms of cognitive re-appraisal. Although the patients fulfilled their criteria for co-morbid major depression with anxiety disorder being the primary diagnosis, the influence of major depression could not be completely ruled out.

Childhood Mal-Treatment

Heany and associates (2018) stated that childhood mal-treatment, including abuse and neglect, may have sustained effects on the integrity and functioning of the brain, alter neurophysiological responsivity later in life, and pre-dispose individuals toward psychiatric conditions involving socio-affective disturbances. This meta-analysis quantified associations between self-reported childhood mal-treatment and brain function in response to socio-affective cues in adults. A total of 17 fMRI studies reporting on data from 848 individuals examined with the Childhood Trauma Questionnaire were included in a meta-analysis of whole-brain findings, or a review of region of interest findings. The spatial consistency of peak activations associated with mal-treatment exposure was tested using activation likelihood estimation, using a threshold of p < 0.05 corrected for multiple comparisons. Adults exposed to childhood mal-treatment showed significantly increased activation in the left superior frontal gyrus and left middle temporal gyrus, and decreased activation in the left superior parietal lobule and the left hippocampus. Although hyper-responsivity to socio-affective cues in the amygdala and ventral anterior cingulate cortex in correlation with mal-treatment severity was a replicated finding in region of interest studies, null results were reported as well. The authors concluded that these findings suggested that childhood mal-treatment had sustained effects on brain function into adulthood, and high-lighted potential mechanisms for conveying vulnerability to development of psychopathology. These findings need to be further investigated.

Obsessive-Compulsive Disorder

Lu and co-workers (2020) presented a study protocol for a single-blind, randomized controlled trial (RCT) to examine the feasibility and efficacy of mindfulness-based cognitive therapy. A total of 120 un-medicated Chinese obsessive-compulsive disorder (OCD) patients will be randomized to the mindfulness-based cognitive therapy group, the selective serotonin reuptake inhibitor group or the psycho-education group for 11 sessions in 10 weeks. A range of scales for clinical symptoms and fMRI will be completed at baseline (week 0), mid-intervention (week 4), post-intervention (week 10) and the 6-month follow-up (weeks 14, 22 and 34). The authors stated that this study will have relevance to decisions about therapeutic options for un-medicated OCD patients.

Panic Disorder

Ni and colleagues (2020) stated that panic disorder (PD) is a prevalent anxiety disorder, however, its neurobiology remains poorly understood. It has been proposed that the pathophysiology of PD is related to an abnormality in a particular neural network. However, most studies investigating resting-state functional connectivity (FC) have relied on a priori restrictions of seed regions, which may bias observations. These investigators examined changes in intra- and inter-network FC in the whole brain of patients with PD using resting-state fMRI. A voxel-wise data-driven independent component analysis was performed on 26 PD patients and 27 healthy controls (HCs). They compared the differences in the intra- and inter-network FC between the 2 groups of subjects using statistical parametric mapping with 2-sample t-tests. PD patients exhibited decreased intra-network FC in the right anterior cingulate cortex (ACC) of the anterior default mode network, the left pre-central and post-central gyrus of the sensorimotor network, the right lobule V/VI, the cerebellum vermis, and the left lobule VI of the cerebellum network compared with the HCs. The intra-network FC in the right ACC was negatively correlated with symptom severity. None of the pairs of resting state networks showed significant differences in functional network connectivity between the 2 groups. The authors concluded that these results suggested that the brain networks associated with emotion regulation, interoceptive awareness, and fear and somatosensory processing may play an important role in the pathophysiology of PD. 

Furthermore, an UpToDate review on "Panic disorder in adults: Epidemiology, pathogenesis, clinical manifestations, course, assessment, and diagnosis" (Roy-Byrne, 2020) does not mention functional MRI as a management option.

Psychosis

Gonzalez-Vivas and colleagues (2019) noted that little is known regarding changes in brain functioning after 1st-episode psychosis (FEP). Such knowledge is important for predicting the course of disease and adapting interventions. and fMRI has become a promising tool for examining brain function at the time of symptom onset and at follow-up. These researchers carried out a systematic review of longitudinal fMRI studies with FEP patients according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. Resting-state and task-activated studies were considered together. A total of 11 studies were included; they reported on a total of 236 FEP patients who were evaluated by 2 fMRI scans and clinical assessments; 5 studies found hypo-activation at baseline in prefrontal cortex areas, 2 studies found hypo-activation in the amygdala and hippocampus, and 3 others found hypo-activation in the basal ganglia. Other hypo-activated areas were the anterior cingulate cortex, thalamus and posterior cingulate cortex; 10 out of 11 studies reported (partial) normalization by increased activation after anti-psychotic treatment. A minority of studies observed hyper-activation at baseline. The authors concluded that this review of longitudinal FEP samples studies showed a pattern of predominantly hypo-activation in several brain areas at baseline that may normalize to a certain extent following treatment. These investigators stated that these findings should be interpreted with caution given the small number of studies and their methodological and clinical heterogeneity.

Language and Memory Decline (Evaluation After Epilepsy Surgery)

Schmid and colleagues (2018) noted that the European Union-funded E-PILEPSY project was launched to develop guidelines and recommendations for epilepsy surgery. In a systematic review, these investigators evaluated the diagnostic accuracy of fMRI, Wada test, magnetoencephalography (MEG), and functional transcranial Doppler sonography (fTCD) for language and memory decline after surgery. They carried out a literature search using PubMed, Embase, and CENTRAL. The diagnostic accuracy was expressed in terms of sensitivity and specificity for post-operative language or memory decline, as determined by pre- and post-operative neuropsychological assessments. If 2 or more estimates of sensitivity or specificity were extracted from a study, 2 meta-analyses were conducted, using the maximum ("best case") and the minimum ("worst case") of the extracted estimates, respectively. A total of 28 papers were eligible for data extraction and further analysis. All tests for heterogeneity were highly significant, indicating large between-study variability (p < 0.001). For memory outcomes, meta-analyses were conducted for Wada tests (n = 17) using both memory and language laterality quotients. In the best case, meta-analyses yielded a sensitivity estimate of 0.79 (95 % CI: 0.67 to 0.92) and a specificity estimate of 0.65 (95 % CI: 0.47 to 0.83). For the worst case, meta-analyses yielded a sensitivity estimate of 0.65 (95 % CI: 0.48 to 0.82) and a specificity estimate of 0.46 (95 % CI: 0.28 to 0.65). The overall quality of evidence, which was assessed using Grading of Recommendations Assessment, Development, and Evaluation (GRADE) methodology, was rated as very low. Meta-analyses concerning diagnostic accuracy of fMRI, fTCD, and MEG were not feasible due to small numbers of studies (fMRI, n = 4; fTCD, n = 1; MEG, n = 0). This also applied to studies concerning language outcomes (Wada test, n = 6; fMRI, n = 2; fTCD, n = 1; MEG, n = 0). The authors concluded that meta-analyses could only be conducted in a few subgroups for the Wada test with low-quality evidence. Thus, more evidence from high-quality studies and improved data reporting are needed. Moreover, these researchers stated that the large between-study heterogeneity underlined the necessity for more homogeneous and thus comparable studies in future research.

Sleep Behavior Disorder

Campabadal and colleagues (2021) noted that isolated rapid eye movement sleep behavior disorder (iRBD) is a harbinger for developing clinical synucleinopathies; and MRI has been suggested as a tool for understanding the neural bases of iRBD and its evolution. In a systematic review, these researchers analyzed original full text articles on structural MRI as well as fMRI in patients with video-polysomnography-confirmed iRBD according to systematic procedures suggested by the PRISMA guideline. The literature search was performed via the PubMed database for articles related to structural MRI and fMRI in iRBD from 2000 to 2020. Investigations to-date have been diverse in terms of methodology, but most agree that patients with iRBD have structural changes in deep gray matter nuclei, cortical gray matter atrophy, and disrupted functional connectivity within the basal ganglia, the cortico-striatal and cortico-cortical networks. In addition, there is evidence that MRI detects structural and functional brain alterations associated with the motor and non-motor symptoms of iRBD. The authors concluded that this review highlighted the need for larger, multi-center, longitudinal studies that will aid in identifying structural and functional patterns of brain degeneration. These researchers stated that it is anticipated that MRI may soon be able to help with the monitoring of disease progression, and perhaps even be of use in determining the short-term risk of subsequent phenoconversion. Finally, larger case-controlled further research must study iRBD longitudinally, implementing multi-modal imaging that uses complex approaches based on data-driven and unsupervised machine learning. This should provide greater insights into the natural course of iRBD.

The authors stated that the main drawback of this study was its small sample size of 10 patients with iRBD and the persistent controversy over the use of T1-weighted images to study structural connectivity. Some investigators argued that gray matter structural connectivity could not be considered a direct measure of connectivity, unlike the diffusion tensor imaging (DTI) approach, while others considered that this approach provided additional insights into the brain network topographical organization. Given evidence that white matter changes could be demonstrated with DTI approaches and that structural connectivity research is possible via diffusion MRI tractography, this is a promising field that requires further investigation with larger iRBD cohorts.

Anger and Aggressive Behaviors

Nikolic et al (2022) stated that reactive aggression in response to perceived threat or provocation is part of humans' adaptive behavioral repertoire; however, high levels of aggression can result in the violation of social and legal norms. Understanding brain function in individuals with high levels of aggression as they process anger- and aggression-eliciting stimuli is critical for refining explanatory models of aggression; thus, improving interventions. These researchers noted that 3 neurobiological models of reactive aggression -- the limbic hyperactivity, prefrontal hypoactivity, and dysregulated limbic-prefrontal connectivity models -- have been proposed. However, these models were based on neuroimaging studies involving mainly non-aggressive individuals, leaving it unclear which model best described brain function in those with a history of aggression. These investigators carried out a systematic literature search (PubMed and Psycinfo) and Multilevel Kernel Density meta-analysis (MKDA) of 9 fMRI studies (8 included in the between-group analysis [i.e., aggression versus control groups], 5 in the within-group analysis). Studies examined brain responses to tasks putatively eliciting anger and aggression in individuals with a history of aggression alone and relative to controls. Individuals with a history of aggression exhibited greater activity in the superior temporal gyrus and in regions comprising the cognitive control and default mode networks (right posterior cingulate cortex, precentral gyrus, precuneus, right inferior frontal gyrus) during reactive aggression relative to baseline conditions. Compared to controls, individuals with a history of aggression exhibited increased activity in limbic regions (left hippocampus, left amygdala, left para-hippocampal gyrus) and temporal regions (superior, middle, inferior temporal gyrus), and reduced activity in occipital regions (left occipital cortex, left calcarine cortex). The authors concluded that the findings of this study lend support to the limbic hyperactivity model in individuals with a history of aggression, and further indicated altered temporal and occipital activity in anger- and aggression-eliciting conditions involving face and speech processing. The clinical implications of these findings need to be further investigated in well-designed studies.

Autism Spectrum Disorder

Santana et al (2022) noted that the diagnosis of autism spectrum disorder (ASD) is still based on behavioral criteria via a lengthy and time-consuming process. Much effort is being made to identify brain imaging biomarkers and develop tools that could facilitate its diagnosis. In particular, using Machine Learning classifiers based on resting-state fMRI (rs-fMRI) data is promising; however, there is an ongoing need for further research on their accuracy and reliability. In a systematic review and meta-analysis, these researchers examined the available evidence in the literature so far. They employed a bi-variate random-effects meta-analytic model to examine the sensitivity and specificity across the 55 studies that offered sufficient information for quantitative analysis. The findings indicated overall summary sensitivity and specificity estimates of 73.8 % and 74.8 %, respectively. Support Vector Machine (SVM) stood out as the most used classifier, presenting summary estimates above 76 %. Studies with bigger samples tended to obtain worse accuracies, except in the subgroup analysis for Artificial Neural Network (ANN) classifiers. The use of other brain imaging or phenotypic data to complement rs-fMRI information appeared promising, achieving higher sensitivities when compared to rs-fMRI data alone (84.7 % versus 72.8 %). Finally, this analysis showed area under the curve (AUC) values between acceptable and excellent. The authors concluded that given the many limitations indicated in this study, further well-designed studies are needed to extend the potential use of those classification algorithms to clinical settings, and the quantitative meta-analytical results presented here should be taken with caution.

The authors stated that this study had several drawbacks. The main one was the sample overlap between the studies, especially considering the lack of information on the patient selection process and the large number of studies that used the ABIDE database. Sample overlap induced a correlation structure among empirical outcomes, which, if not accounted for, could harm the statistical properties of meta-analysis methods and resulted in higher rates of false positives. Therefore, it was unclear to which extent this overlap could have biased the results obtained. In addition, due to the tremendous heterogeneity of ASD, this high degree of overlap may limit the interpretability and generalizability of this analysis. Despite that, these researchers highlighted that all the significant results obtained in their analyses were reasonable and in line with the literature. Furthermore, they clearly stated this drawback throughout the study and hoped that it would serve as a guide to future works, eventually reaching a state where more robust analyses could be performed. These investigators also stated that considering the significant heterogeneity within the selected publications, the summary estimates obtained via the meta-analysis had to be interpreted with caution and in light of the methodologic quality of the studies. Most studies provided only limited information regarding the patient samples and their clinical characteristics. However, detailed information regarding the subjects’ disease status, symptoms, current medication, history of interventions, or co-morbidities was crucial for examining the potential of the proposed models to be used in clinical practice. Therefore, the impact of those variables on classification accuracy needs to be better examined. The authors stated that the studies included in this analysis identified ASD-distinctive brain patterns as compared to healthy volunteers. Nevertheless, it is critical to examine the patterns of brain abnormalities that differentiate between different psychiatric disorders. Additionally, the results obtained in this meta-analysis did not apply to individuals under 5 years of age since almost none of the studies included individuals with such low age. Furthermore, some methodological steps were not examined in these analyses, such as the data pre-processing and feature selection procedures; these aspects still need to be evaluated to define their effects on classification accuracy.

Emotion-Expressive Suppression

Sikka et al (2022) stated that expressive suppression refers to the inhibition of emotion-expressive behavior (e.g., facial expressions of emotion). Although it is a commonly used emotion regulation strategy with well-documented consequences for well-being, little is known regarding its underlying mechanisms. In a systematic review, these investigators synthesized functional neuroimaging studies on the neural bases of expressive suppression in non-clinical populations. The 12 studies included in this review contrasted the use of expressive suppression to simply watching emotional stimuli. Results showed that expressive suppression consistently increased activation of frontoparietal regions, especially the dorsolateral and ventrolateral prefrontal cortices and inferior parietal cortex; but decreased activation in temporo-occipital areas. Results regarding the involvement of the insula and amygdala were inconsistent with studies showing increased, decreased, or no changes in activation. The authors concluded that these mixed findings underscored the importance of distinguishing expressive suppression from other forms of suppression and highlighted the need to pay more attention to experimental design and neuroimaging data analysis procedures. They noted that involvement of emotion-generative regions (amygdala and insula) in expressive suppression remains inconclusive; and mixed results stemmed from conceptual and methodological issues that need to be addressed in future research.

Migraines

Schramm et al (2023) noted that migraine is a highly prevalent primary headache disorder. Despite a high burden of disease, key disease mechanisms are not entirely understood. Functional MRI is an imaging method using the blood-oxygen-level-dependent signal, which has been increasingly employed in migraine research over recent years. In a systematic review, these investigators examined recent findings using fMRI for the investigation of migraine. They carried out a systematic search and selection of fMRI uses in migraine from April 2014 to December 2021 (PubMed and references of identified articles according to the PRISMA guidelines). Methodological details and main findings were extracted and synthesized. Out of 224 studies identified, 114 were included after selection. Repeatedly emerging structures of interest included the insula, brainstem, limbic system, hypothalamus, thalamus, and functional networks. Assessment of functional brain changes in response to treatment is emerging, and machine learning (ML) has been used to examine potential fMRI-based markers of migraine. The authors concluded that a wide variety of fMRI-based metrics were found altered across the brain for heterogeneous migraine cohorts, partially correlating with clinical parameters and supporting the concept to conceive migraine as a brain state. However, a majority of findings from previous studies have not been replicated, and studies varied considerably regarding image acquisition and analyses techniques. Therefore, while fMRI appears to have the potential to advance the understanding of migraine pathophysiology, replication of findings in large representative datasets and precise, standardized reporting of clinical data would benefit the field and further increase the value of observations.

Schizophrenia Spectrum Disorders

Voineskos et al (2024) noted that functional neuroimaging emerged with great promise and has provided fundamental insights into the neurobiology of schizophrenia; however, it has faced challenges and criticisms, most notably a lack of clinical translation. In a systematic review, these investigators provided a critical summary of the available evidence on functional neuroimaging, in particular fMRI, in schizophrenia. They began by reviewing research on fMRI biomarkers in schizophrenia and the clinical high-risk phase through a historical lens, moving from case-control regional brain activation to global connectivity and advanced analytical approaches, and more recent machine learning (ML) algorithms to identify predictive neuroimaging features. Findings from fMRI studies of negative symptoms as well as of neuro-cognitive and social cognitive deficits were then reviewed. Functional neural markers of these symptoms and deficits may represent promising treatment targets in schizophrenia. Next, these researchers summarized fMRI research related to anti-psychotic medication, psychotherapy and psychosocial interventions, as well as neurostimulation, including treatment response and resistance, therapeutic mechanisms, and treatment targeting. In addition, these researchers reviewed the use of fMRI and data-driven approaches to dissect the heterogeneity of schizophrenia, moving beyond case-control comparisons, as well as methodological considerations and advances, including consortia and precision fMRI. Finally, limitations and future directions of research in the field were discussed. The authors concluded that the findings of this comprehensive review suggested that, in order for fMRI to be clinically useful in the care of patients with schizophrenia, research should address potentially actionable clinical decisions that are routine in schizophrenia treatment, such as which anti-psychotic should be prescribed or whether a given patient is likely to have persistent functional impairment. These researchers stated that the potential clinical use of fMRI is influenced by and must be weighed against cost and accessibility factors; future investigations on the use of fMRI in prognostic and therapeutic response studies may consider including a health economics analysis to make a tangible clinical impact.

The authors stated that although fMRI has been highly impactful in psychiatry research in the last 30 years, it is associated with several limitations, which have until now hampered its deployment in clinical settings. These researchers stated that if fMRI is to become a useful diagnostic/prognostic tool in the care of patients with schizophrenia (e.g., to predict conversion to psychosis from at‐risk states, to predict response to certain antipsychotic medications, or to guide precision treatment), these limitations will need to be overcome. These investigators divided these limitations into 3 categories: technical, experimental, and conceptual. Technical limitations are those concerning data collection and analysis. Experimental limitations are those that arise in the conduct of clinical fMRI research, such as sample size and power limitations, and sampling biases. Conceptual limitations denote issues in interpretation of fMRI findings in clinical schizophrenia research.

Gallucci et al (2024) stated that depressive symptoms in schizophrenia spectrum disorders (SSDs) negatively impact suicidality, prognosis, and quality of life (QOL). Despite this, effective treatments are limited, mainly because the neural mechanisms underlying depressive symptoms in SSDs remain poorly understood. In a systematic review, these investigators provided an overview of studies that examined the neural correlates of depressive symptoms in SSDs using neuroimaging techniques. They searched Medline, PsycINFO, Embase, Web of Science, and Cochrane Library databases from inception through June 19, 2023. Specifically, these researchers focused on structural and fMRI, entailing T1-weighted imaging measuring brain morphology; diffusion-weighted imaging examining white matter integrity; or T2*-weighted imaging measures of brain function. The search yielded 33 articles; 14 structural MRI studies, 18 fMRI studies, and 1 multi-modal fMRI/MRI study. Reviewed studies showed potential commonalities in the neurobiology of depressive symptoms between SSDs and major depressive disorders, especially in subcortical and frontal brain regions, although confidence in this interpretation was limited. These researchers stated that this review underscored a notable knowledge gap in the understanding of the neurobiology of depression in SSDs, marked by inconsistent approaches and few studies examining imaging metrics of depressive symptoms. Furthermore, inconsistencies across studies' findings emphasized the necessity for more direct and comprehensive research focusing on the neurobiology of depression in SSDs. The authors concluded that future studies should go beyond "total score" depression metrics and adopt more nuanced assessment approaches considering distinct subdomains, which could reveal unique neurobiological profiles and inform investigations of targeted treatments for depression in SSDs; and a potential future direction may be to examine both unique and shared neural correlates across the 2 disorders. Moreover, these researchers noted that the limited availability of studies using diffusion MRI metrics or examining brain function at the network level underscored the need for further investigations.

Connectomic Analysis of Multi-Modal Brain MRI

The clinical evidence supporting connectomic analysis spans two major neurosurgical domains, neuro-oncology and epilepsy, with the strongest and most mature evidence base in pre-surgical brain tumor planning. The most robust clinical application of connectomics is in glioma surgery, where the goal is to maximize extent of resection while preserving neurological function (the "onco-functional balance"). 

A large retrospective study of 192 consecutive brain tumor surgeries using preoperative AI-based connectomic mapping (Quicktome platform) demonstrated clinical feasibility, with median Karnofsky Performance Status remaining stable at 3 months postoperatively. Tractometry analysis in a subgroup showed postoperative shifts toward improved interhemispheric white matter symmetry in multiple tracts. This study evaluated the feasibility and exploratory impact of integrating an artificial intelligence–based connectomics platform into routine brain tumor surgery, with a secondary aim of assessing postoperative white matter changes using tractometry (Sistiaga et al., 2026). The investigators conducted a retrospective review of 192 consecutive tumor resections performed between April 2023 and April 2025 using preoperative connectomic mapping, and measured functional outcomes with the Karnofsky Performance Status at baseline and approximately 3 months after surgery. In a small opportunistic subgroup of 13 patients with paired preoperative and postoperative imaging, tractometry analysis quantified fractional anisotropy across six major white matter tracts, along with derived hemispheric asymmetry indices designed to detect subtle microstructural changes. The results showed that connectomic mapping was successfully incorporated into all surgical planning workflows without apparent delay, and median functional status remained stable at 3 months, suggesting preservation of short term outcomes. In the tractometry subgroup, 12 of 13 patients demonstrated postoperative shifts toward reduced interhemispheric asymmetry in at least one tract, with patterns varying by tumor laterality, and asymmetry metrics appeared more sensitive than raw fractional anisotropy values for detecting postoperative microstructural alterations. However, the study has several important limitations, including its retrospective and descriptive design without a control group, lack of direct evidence that connectomics altered surgical decisions or outcomes, a very small and nonuniform tractometry subgroup, variable timing of postoperative imaging, reliance on fractional anisotropy as an indirect and potentially confounded marker of white matter integrity, absence of detailed neurocognitive assessments beyond Karnofsky Performance Status, and lack of correction for multiple statistical comparisons in exploratory analyses. Overall, the findings support the feasibility of large scale clinical implementation of connectomics in neurosurgical oncology and generate hypotheses regarding postoperative network level changes that require prospective validation.

A multi-institutional survey of 70 patients across four academic centers found that integrating fMRI and DTI into surgical planning resulted in a significant shift from awake to asleep craniotomies (51% overall, p < 0.0001), a "much more aggressive" surgical plan in 39% of cases, larger extent of resection in 61%, and shorter-than-expected surgical durations in 51% of cases. This multi-institutional survey study by Talekar et al. (2026) evaluated how preoperative fMRI and DTI influence neurosurgical decision-making and outcomes in brain tumor resection across four academic centers. Seventy patients from Thomas Jefferson University (TJU, n=51), University Hospital Basel (n=11), University of Pennsylvania (n=4), and Johns Hopkins University (JHU, n=3) underwent preoperative task-based fMRI and DTI. Six neurosurgeons completed structured pre- and post-imaging surveys assessing changes in surgical approach, craniotomy type, extent of resection, operative duration, and diagnostic confidence. Integration of fMRI/DTI resulted in a significant shift toward asleep craniotomies, which increased to 51% overall (Chi-square P < .0001), most pronounced at TJU (P = 0.01). This suggests that preoperative functional mapping provided sufficient confidence in the localization of eloquent cortex to forgo awake mapping in many cases. In addition, fMRI led to a "much more aggressive" resection strategy in 39% of cases globally, most prominently at TJU (74%) and UPenn (50%). JHU was an outlier, reporting decreased aggressiveness in 33.3% of cases. DTI had a similar but slightly attenuated effect, with "much more aggressive" as the top response in 34% of cases. The fMRI was rated as more clinically valuable than DTI in 53.4% of cases overall, with TJU reporting the highest rate (72%). A larger extent of resection was reported in 61% of cases, and surgical durations were shorter than expected in 51% of cases. Combined fMRI/DTI had a "strong positive" influence on surgery in 71% and on clinical care in 68% of cases, with significant inter-institutional variability (P < 0.001). However, the study relied on subjective surgeon surveys rather than objective outcome measures, and the sample was unevenly distributed across institutions (73% from TJU). The small sample sizes at UPenn (n=4) and JHU (n=3) limit generalizability of site-specific findings. The study design was observational without a control group.

Preoperative function-specific connectome analysis has been shown to predict surgery-related aphasia after glioma resection with 73.3% accuracy and 78.3% sensitivity, using machine learning models trained on nTMS-derived connectome network properties. This study aimed to determine whether preoperative function specific connectome analysis could predict the risk of surgery related aphasia in patients undergoing resection of language eloquent gliomas (Ille, et al., 2022). The investigators performed a post hoc analysis of 60 patients with left hemispheric perisylvian gliomas and intact baseline language function, dividing them into those who developed postoperative aphasia and those who did not. Preoperative evaluation combined navigated transcranial magnetic stimulation language mapping with diffusion tensor imaging based tractography to construct individualized connectomes, and graph theoretical metrics such as average degree, global efficiency, local efficiency, and path length were calculated across multiple network configurations. These features were then incorporated into a machine learning model with cross validation to predict aphasia outcomes. The results demonstrated that patients who did not develop postoperative aphasia exhibited higher network connectivity and efficiency across several connectome metrics, while patients with aphasia showed reduced network integration, particularly in language related networks. Several connectome parameters were significantly associated with aphasia risk, and when combined in a machine learning model, they predicted surgery related aphasia with approximately 73 to 77 percent accuracy and sensitivity around 78 percent. These findings suggest that preoperative network level properties may reflect the brain’s capacity for functional compensation and are useful predictors of postoperative language outcomes. However, the study has several limitations, including the absence of a healthy control group, relatively small sample size for machine learning analysis, lack of functional MRI confirmation of language dominance, heterogeneity in tumor characteristics such as grade and edema, and limited neurocognitive assessment beyond selected aphasia subtests. Overall, the study provides evidence that connectome based metrics can help stratify aphasia risk prior to surgery, though further validation in larger and more comprehensive cohorts is needed. This represents a meaningful advance in preoperative risk stratification for language-eloquent tumors.

A systematic review of 41 studies analyzing high-grade gliomas found that tumors have distinct impacts on the functional connectome based on location - frontal gliomas decrease global and local efficiency bilaterally, temporal gliomas alter bilateral functional connectivity with preserved small-world properties, and parietal gliomas primarily affect local connectivity. Connectomic metrics (functional connectivity, local efficiency, global efficiency) improved predictive models for post-resection complications when combined with clinical and structural data. This systematic review aimed to synthesize current evidence on how high grade gliomas affect structural and functional brain networks and to evaluate the role of connectomics based approaches, including machine learning, in predicting postoperative outcomes (Burrington, et al., 2025). The authors conducted a PRISMA guided literature review of studies published up to October 19, 2023, searching multiple databases for articles that used diffusion tensor imaging or resting state functional MRI to assess connectivity changes in patients with high grade gliomas; 41 studies met inclusion criteria after screening. The results demonstrated that high grade gliomas produce location dependent alterations in brain networks, with structural connectivity often increased among highly connected hub regions and decreased among peripheral nodes, while functional connectivity changes varied by tumor location but commonly included reductions in local efficiency and long distance interhemispheric connectivity with relative preservation of global efficiency and small world network organization. Frontal and temporal tumors were associated with bilateral network disruptions, whereas parietal and insular tumors tended to produce more localized effects, and integration of connectomic metrics such as functional connectivity, global efficiency, and local efficiency into predictive models improved the accuracy of forecasting postoperative complications and survival compared with traditional clinical or radiologic models alone. However, the study highlights several limitations, including heterogeneity across included studies, limited data on less common tumor locations such as the occipital lobe, incomplete understanding of how global network metrics relate to clinical outcomes, and lack of standardization in imaging acquisition, preprocessing, and analytic methodologies, all of which restrict comparability and generalizability of findings. Overall, the review supports the growing role of connectomics as a framework for understanding glioma related network disruption and improving prediction of surgical outcomes, while emphasizing the need for standardized and prospective studies.

Case series have demonstrated that multimodal structural and functional connectome imaging can identify language network parcellations preoperatively, correlating with intraoperative awake mapping findings (speech arrest and paraphasic errors at tumor boundaries corresponding to predicted functional regions). Shah et al. (2023) presented a case report demonstrating that preoperative connectomics imaging correlates with intraoperative awake language mapping findings during brain tumor surgery. The authors reported a single instructive case of a patient with a brain tumor impinging on the language area. Preoperative connectomics imaging was performed using machine learning-based software to generate individualized connectome maps, which identified the tumor's proximity to specific parcellations of the language network. The patient then underwent an awake craniotomy with intraoperative direct cortical stimulation for language mapping. During intraoperative awake mapping, stimulation at the tumor boundary elicited speech arrest and paraphasic errors. Critically, the cortical sites where these language disruptions occurred correlated with the functional regions predicted by the preoperative connectomics imaging. The connectomics-derived parcellations provided an anatomical and functional explanation for the specific types of language deficits observed intraoperatively. Subsequently, Shah et al. (2025) reported a two-patient case series investigating the use of multimodal connectome imaging, combining structural and functional connectivity data, to guide resection of left frontal opercular brain lesions while preserving language function. A machine learning–based software was used to create individualized structural and functional connectome maps by reparcellating the Human Connectome Project Multi-Modal Parcellation (HCP-MMP) atlas based on each patient's anatomic MRI, diffusion MRI, and resting-state functional MRI (rs-fMRI). Structural connectivity imaging identified at-risk parcellations near the lesion, and seed-based analysis of regions of interest (ROIs) characterized functional relationships within the language network. Multimodal, connectome-guided resections were performed with concurrent intraoperative neuromonitoring. Two patients with left frontal lesions were included, one patient with a WHO Grade IV gliosarcoma, and another patient with an intracerebral abscess. Preoperative functional neuroimaging revealed individualized, distinct patterns of functional connectivity between language network parcellations for each patient. Both patients underwent connectome-guided resections with intraoperative neuromonitoring and demonstrated intact or improved language function relative to baseline at follow-up. Postoperative imaging showed functional reorganization between Brodmann areas 44 and 45 (corresponding to Broca's area) and other parcellations of the language network, suggesting network-level plasticity after surgery. The authors concluded that preoperative visualization of both structural and functional connectivity of language areas can be meaningfully integrated into a multimodal surgical approach alongside intraoperative neuromonitoring to help preserve language function during intracranial surgery. They also suggest these imaging modalities may serve as tools for monitoring postoperative functional recovery.

Duffau's paradigm-shifting work has advocated for a "connectome-based resection" approach that considers subcortical pathways and network plasticity, enabling resection in areas traditionally deemed "inoperable" by leveraging functional reallocation. This narrative review aimed to examine how advances in brain connectomics have transformed the conceptual framework and clinical practice of glioma surgery, shifting from a traditional localization based paradigm toward a dynamic network oriented approach that optimizes the balance between tumor resection and functional preservation (Duffau, et al., 2021). The author synthesized existing neuroscientific and clinical evidence on brain connectivity, neuroplasticity, and intraoperative mapping, emphasizing the role of distributed cortico subcortical networks rather than isolated cortical regions in mediating cognition, language, and behavior. The review describes methodological approaches such as functional neuroimaging, intraoperative direct electrical stimulation during awake surgery, and connectome informed surgical planning, highlighting how these tools allow real time identification of critical neural pathways and support individualized surgical strategies tailored to each patient’s functional priorities. The results of this synthesis indicate that adopting a connectomic perspective enables more extensive resections, including in regions previously considered inoperable, while preserving neurological and higher order cognitive function through the exploitation of neuroplasticity and dynamic network reorganization. The review further reports that techniques such as awake mapping and network guided "oncological disconnection surgery" are associated with lower rates of permanent deficits, improved quality of life, high rates of return to work, and prolonged survival, reflecting a paradigm in which surgical boundaries are defined by functional network limits rather than anatomic tumor margins alone. However, several limitations are acknowledged, including the inability of current preoperative imaging to reliably predict individualized network organization or functional reserve, variability in neuroplastic potential across patients and tumor types, limited precision in mapping distributed networks preoperatively, and the challenge of standardizing connectome based approaches in clinical practice. Overall, the review concludes that a connectome based, personalized, and dynamic surgical strategy represents a major advancement in neurooncology, though further research is needed to refine predictive tools and better characterize individual network reorganization.

A large single-institution experience (191 consecutive patients) at Washington University demonstrated that rs-fMRI had a significantly lower failure rate than task-based fMRI (13% vs. 38.5%, p < 0.001), making it particularly valuable for patients unable to cooperate with task-based paradigms due to cognitive impairment, young age, or need for sedation. This study aimed to evaluate the feasibility, reliability, and clinical integration of resting state functional MRI as a tool for preoperative functional mapping in neurosurgical patients, particularly compared with traditional task based fMRI approaches (Leuthardt, et al., 2018). The investigators conducted a retrospective analysis of 191 consecutive patients who underwent resting state functional MRI at a single institution, many of whom also received task based fMRI, and implemented a novel automated workflow using a multilayer perceptron algorithm to process imaging data and generate maps of major functional networks that were integrated into the clinical imaging infrastructure and surgical navigation systems. The results demonstrated that resting state functional MRI could successfully map multiple functional networks, including motor and language systems, in a single short imaging session without requiring patient participation, and was broadly applicable across diverse clinical populations such as children, cognitively impaired individuals, and sedated patients. Importantly, resting state functional MRI showed a significantly lower failure rate than task based fMRI, with failure occurring in 13 percent of resting state studies compared with 38.5 percent of task based studies, and it was able to provide functional localization even in cases where task based imaging failed. These findings support the clinical utility of automated resting state mapping as a practical and scalable approach for identifying eloquent cortex and aiding surgical planning, especially in patients unable to perform task based paradigms. However, the study has several limitations, including its retrospective design, lack of prospective outcome data to determine the impact on surgical decision making or patient outcomes, and reliance on a locally developed imaging and processing platform that may limit generalizability to other institutions without similar infrastructure. Overall, the study provides evidence that resting state functional MRI can be effectively integrated into clinical practice and may represent a more reliable alternative to conventional task based functional imaging in neurosurgical planning.

Automated rs-fMRI network extraction tools have shown good spatial agreement with intraoperative direct electrical stimulation, with distances < 1 cm in 78–100% of cases depending on the functional domain. This technical note aimed to introduce and evaluate a novel automated pipeline, ReStNeuMap, for processing resting state functional MRI data to enable reliable presurgical mapping of functional brain networks in individual patients (Zacà, et al., 2019). The authors described the architecture and workflow of the pipeline, which incorporates automated preprocessing, quality control procedures, and independent component analysis with template matching to identify functional networks, and then reported a preliminary validation study in 6 patients undergoing brain tumor or vascular lesion resection with intraoperative direct electrical stimulation as the reference standard. The results demonstrated that ReStNeuMap successfully identified functional networks corresponding to eloquent cortical areas in all patients, with strong spatial agreement between resting state functional MRI–derived networks and intraoperative stimulation mapping, as distances between modalities were less than 1 cm in 78 percent of motor sites, 87.5 percent of language sites, and 100 percent of visual and speech articulation sites. These findings suggest that automated resting state functional MRI analysis can provide accurate, noninvasive functional localization and may complement intraoperative mapping to guide surgical planning and reduce mapping time. However, several limitations were noted, including the small sample size, lack of validation against task based functional MRI or alternative analytic methods, potential susceptibility to neurovascular uncoupling leading to false negative results, absence of certain preprocessing corrections such as geometric distortion adjustment, and restricted capability to assess networks beyond those predefined for presurgical mapping. Overall, the study provides preliminary evidence supporting the feasibility and clinical potential of automated resting state functional MRI pipelines, while emphasizing the need for larger, multicenter studies to confirm reliability and generalizability.

Connectomic analysis in epilepsy is an active area of research with promising but less clinically mature evidence compared to neuro-oncology. Studies using resting-state fMRI have demonstrated that patients who fail epilepsy surgery show increased functional segregation outside the surgical resection zone preoperatively, suggesting that connectomic markers could help identify candidates likely to benefit from surgery. This narrative review aimed to synthesize recent research on how connectomics can improve understanding of network dysfunction in drug resistant epilepsy and enhance prediction of surgical outcomes and treatment strategies (Johnson, et al., 2022). The authors conducted a focused review of studies published within the preceding 18 months, examining evidence from structural and functional connectivity analyses using modalities such as functional MRI, diffusion imaging, and intracranial electrophysiology, with an emphasis on hypothesis driven investigations linking network features to surgical outcomes and cognitive effects. The review found that epilepsy is better conceptualized as a distributed epileptogenic network rather than a single localized epileptogenic zone, and that connectomic measures can help predict surgical success, as patients with widespread functional segregation outside the resection zone or abnormal connectivity patterns are more likely to experience persistent seizures after surgery. Additional findings highlighted that both functional and electrophysiologic connectivity metrics, including network segregation, node strength, and integration, are associated with outcomes, while postoperative network disconnection correlates with seizure freedom and changes in large scale networks can explain cognitive effects after treatment. The review also emphasized the potential of connectomics to guide alternative strategies such as neuromodulation or network targeted therapies and to identify biomarkers of treatment response, including electrophysiologic signatures during deep brain stimulation. However, the authors noted important limitations in the field, including heterogeneity of methodologies, difficulty interpreting graph theory metrics biologically, small sample sizes in many studies, lack of standardized analytic approaches, and the risk of overinterpretation without hypothesis driven frameworks, all of which limit immediate clinical translation and generalizability. Overall, the review concludes that connectomics is a valuable tool for advancing understanding of epilepsy as a network disorder and may improve surgical planning and outcomes if applied rigorously and integrated with clinical decision making. 

In mesial temporal lobe epilepsy, higher preoperative network resilience to targeted attack on topologically central nodes was associated with seizure-free outcomes after temporal lobectomy. This longitudinal study aimed to investigate how functional brain network organization differs before and after anterior temporal lobectomy in patients with mesial temporal lobe epilepsy and whether connectomic features can predict surgical outcomes and recovery mechanisms (Liao, et al., 2016). The authors analyzed resting state functional MRI data from 37 patients preoperatively and 24 patients postoperatively, classifying participants into seizure free and non seizure free groups, and applied graph theoretical network analysis to construct functional connectomes, assess network resilience to simulated node removal, and evaluate dynamic changes in global and nodal network properties over time. The results showed that patients who achieved seizure freedom exhibited greater preoperative network resilience to targeted attacks on highly connected nodes, suggesting more robust and redundant connectivity, and demonstrated distinct patterns of postoperative network reorganization, particularly involving the temporoparietal junction and its connections to the ventral prefrontal cortex. Additionally, the study found that greater surgical damage to functionally important network nodes was associated with reduced postoperative small world network properties, indicating impaired global and local information processing capacity. These findings support the concept of epilepsy as a network disorder and suggest that connectomic metrics may help predict surgical success and characterize mechanisms of functional recovery. However, the study has several limitations, including a relatively small sample size, grouping of left and right temporal lobe epilepsy patients which may obscure lateralized effects, short postoperative follow up duration, lack of detailed cognitive assessments, and reliance on group level analyses that may limit applicability to individual patient prediction. Overall, the study provides evidence that longitudinal connectome analysis can offer clinically relevant insights into surgical outcomes and brain network reorganization in epilepsy. 

A study of 37 patients with temporal lobe epilepsy showed that preoperative connectome changes (increased segregation of the ipsilateral anterior temporal lobe) and postoperative integration of contralateral regions were associated with clinical variables including hippocampal atrophy and seizure frequency, offering potential biomarkers for surgical planning. This longitudinal neuroimaging study aimed to characterize how focal temporal lobe pathology and its surgical resection reshape large scale structural brain networks and how these changes relate to individual clinical features in patients with drug resistant temporal lobe epilepsy (Larivière, et al., 2024). The investigators analyzed diffusion MRI data from 37 patients scanned before and after anterior temporal lobectomy and compared them with 31 matched healthy controls, constructing structural connectomes via tractography and applying dimensionality reduction to derive connectome gradients that capture global axes of brain connectivity, followed by statistical modeling to assess preoperative abnormalities, longitudinal reorganization, and associations with clinical variables using partial least squares analysis. The results demonstrated that prior to surgery, patients exhibited marked connectome alterations characterized by increased segregation of the ipsilateral anterior temporal lobe and widespread abnormalities in temporoparietal and orbitofrontal regions, indicating disrupted network integration beyond the lesion site. Following surgery, reorganization was observed both locally and distally, with continued segregation near the resection area but increased integration of contralateral temporoparietal regions, suggesting compensatory network plasticity. Furthermore, regions showing the greatest preoperative abnormalities tended to undergo the largest postoperative changes, and multivariate analyses identified a clinical imaging signature in which increased postoperative integration was associated with greater hippocampal atrophy, lower seizure frequency, and presence of generalized seizures. These findings support the concept that focal lesions and their surgical removal lead to widespread and coordinated reorganization of brain networks that reflect both disease effects and recovery mechanisms. However, the study has limitations, including reliance on diffusion MRI tractography which can be affected by technical constraints such as crossing fibers and resolution limitations, absence of longitudinal imaging in healthy controls for comparison of temporal changes, and lack of modeling of virtual resections in control subjects to isolate surgery specific effects, as well as the retrospective design and moderate sample size which may limit generalizability and predictive application. Overall, the study provides important evidence that connectome gradient analysis can capture clinically relevant brain network reorganization in epilepsy and may inform individualized understanding of surgical outcomes.

A 2026 Lancet Neurology review highlighted that widespread abnormalities in structural (DTI-based) and functional (rs-fMRI and stereo-EEG-based) networks correlate with postoperative seizure recurrence, especially when abnormal regions are not resected. However, multiple barriers remain for clinical translation, including the need for large validation datasets, user-friendly analytic tools, and demonstration of added value over traditional clinical metrics. This narrative review aimed to evaluate emerging technologies, evolving surgical strategies, and the role of network and molecular approaches in advancing personalized epilepsy surgery for patients with drug resistant focal epilepsy (Yardi, et al., 2026). The authors conducted a structured literature review of studies published between 2019 and 2025 using databases such as MEDLINE and PubMed, synthesizing evidence on resective surgery, minimally invasive approaches such as laser interstitial thermal therapy, neuromodulation techniques, and advances in computational, network, and multiomic analyses to improve treatment selection and outcome prediction. The review found that epilepsy is increasingly understood as a network disorder, leading to expansion of treatment options beyond traditional resection to include ablation and neuromodulation, with resective surgery still offering the highest rates of seizure freedom but newer approaches providing alternatives for patients with distributed or eloquent network involvement. Advances in imaging, intracranial electrophysiology, and machine learning were shown to improve localization of epileptogenic networks and prediction of outcomes, while emerging genetic and molecular biomarkers may further refine individualized treatment strategies. Despite these advances, outcome prediction remains imperfect, and variability in surgical success is influenced by patient specific network characteristics, extent of resection, and underlying pathology, with widespread network abnormalities associated with poorer outcomes. The review also highlighted significant global disparities in access to epilepsy surgery and limitations in translating advanced network analyses into clinical practice due to the need for large datasets, standardized tools, and prospective validation of predictive models. Additional limitations include reliance on heterogeneous observational data, lack of robust comparative effectiveness studies between surgical modalities, limited prospective validation of predictive algorithms, and early stage evidence for molecular biomarkers. Overall, the study concludes that integrating network neuroscience, computational modeling, and multiomic data holds promise for improving precision in epilepsy surgery, but further validation and broader implementation are needed to achieve consistent clinical impact.

Connectome-based approaches in pediatric epilepsy surgery remain limited, though emerging evidence suggests potential for optimizing surgical strategies and predicting outcomes through network analysis. This narrative review aimed to summarize current knowledge on connectome based approaches in pediatric epilepsy surgery and to explore how network level analyses can inform surgical planning, predict outcomes, and guide future personalized strategies (De Benedictis, et al., 2023). The authors conducted a qualitative synthesis of existing literature, focusing on studies of structural and functional brain connectivity derived from modalities such as diffusion tensor imaging, functional MRI, electroencephalography, and intraoperative neurophysiology, alongside emerging computational and machine learning methods, to examine the role of white matter anatomy, intraoperative mapping, postoperative connectome changes, and network based predictors of outcomes in children with drug resistant epilepsy. The review found that epilepsy is increasingly understood as a network disorder rather than a focal lesion alone, with widespread alterations in structural and functional connectivity contributing to seizure generation and propagation, and that surgical interventions can induce significant reorganization of brain networks, including compensatory increases in contralateral connectivity and adaptive plasticity associated with preserved or improved neurocognitive function. Network characteristics were also shown to influence clinical outcomes, with specific patterns of connectivity reorganization associated with seizure freedom or recurrence and with neurocognitive trajectories, while advanced approaches such as multimodal connectomic integration, machine learning models, and in silico simulations demonstrated promising ability to improve prediction of surgical outcomes and optimize individualized treatment planning. However, the study highlights several limitations, including the relatively limited pediatric specific evidence compared with adult populations, heterogeneity in methodologies and imaging protocols, variability in developmental brain maturation that complicates interpretation of connectivity measures, and the need for larger, standardized, multimodal datasets and prospective validation to translate connectome based tools into routine clinical practice. Overall, the review emphasizes that connectome based frameworks offer a promising pathway toward precision epilepsy surgery in children but require further methodological refinement and validation.

Despite promising results, several challenges persist: lack of standardization in data acquisition and processing, issues with reproducibility, susceptibility to artifacts (particularly in the setting of tumor-related edema and mass effect), and the need for prospective validation in larger cohorts. Most connectomic studies remain retrospective and hypothesis-generating, and the field awaits randomized evidence demonstrating improved patient outcomes compared to standard mapping approaches. Wu, et al. (2021) reviewed the clinical applications of MRI-based structural and functional connectomics, aiming to evaluate how noninvasive brain network mapping can inform diagnosis, surgical planning, and treatment of neurological disorders while critically assessing its limitations. The authors employed a narrative review methodology that synthesizes evidence from diffusion MRI tractography and functional MRI approaches, including resting-state and task-based paradigms, and describes standard acquisition and processing pipelines such as motion correction, registration, tractography modeling, and functional connectivity analyses using seed-based correlation or independent component analysis. Across multiple clinical domains, the results highlight that connectomic techniques can improve identification of eloquent cortex, guide neurosurgical planning for tumors and epilepsy, refine targeting for deep brain stimulation in movement and psychiatric disorders, and provide prognostic biomarkers such as predicting seizure outcomes or cognitive decline after radiotherapy. The review also emphasizes emerging approaches including lesion network mapping, effective connectivity modeling, and machine learning, which further enhance the ability to map disease-relevant brain networks and personalize interventions. However, the study underscores significant limitations that constrain clinical translation, including variability in MRI data quality, susceptibility to motion and distortion artifacts, lack of standardization in acquisition and processing methods, limited test–retest reliability especially for functional connectivity, and absence of a true in vivo gold standard for validating tractography results. Additional methodological challenges include variability in software and modeling approaches, potential for false positive or negative connections, biases in region of interest selection, and difficulty interpreting functional significance of identified networks. Overall, the authors conclude that although MRI-based connectomics holds substantial promise for precision neuroscience and individualized care, broader clinical adoption will require greater validation, standardization, and integration with complementary modalities.

A systematic review evaluated noninvasive preoperative brain mapping techniques for glioma surgery, specifically navigated transcranial magnetic stimulation, functional magnetic resonance imaging, and magnetoencephalography, with the objective of assessing their accuracy, reliability, clinical utility, and accessibility in optimizing extent of resection and preserving neurological function (Leone et al., 2025). The authors conducted a PRISMA-guided systematic review of studies published between 1997 and January 2024 using databases including PubMed, EMBASE, PLOS, and the Cochrane Library, selecting human studies with more than 10 patients and excluding case reports and small or heterogeneous cohorts. A total of 128 studies were included in the qualitative synthesis, comprising 48 studies on navigated transcranial magnetic stimulation, 56 on functional magnetic resonance imaging, and 24 on magnetoencephalography. Data extraction focused on study characteristics and key performance metrics such as accuracy relative to direct cortical stimulation, spatial resolution, reliability, and feasibility. Methodological quality was assessed using the Joanna Briggs Institute checklist for case series. Overall, navigated transcranial magnetic stimulation demonstrated the highest accuracy for motor mapping, with close agreement to direct cortical stimulation and meaningful clinical benefits including reduced postoperative deficits, improved extent of resection, and the ability to generate individualized risk stratification models when combined with diffusion tensor imaging tractography. Functional magnetic resonance imaging showed utility in identifying functional regions but was limited by susceptibility to false positives and false negatives related to coactivation and neurovascular uncoupling, which reduced its reliability for surgical planning, especially near tumors. Magnetoencephalography offered high temporal resolution and reasonable accuracy for motor mapping but was constrained by high cost, technical complexity, and spatial discrepancies compared with direct cortical stimulation. Across all modalities, language mapping results were inconsistent, with no standardized approach identified, although repetitive transcranial magnetic stimulation provided high negative predictive value that may inform decisions regarding awake versus asleep surgery. Key limitations of the study include substantial heterogeneity across included studies, particularly in language mapping methodologies and outcomes, limited availability of randomized controlled trials, and variability in operator expertise and institutional practices that affect generalizability. Additionally, many analyses were based on observational data and case series, which carry inherent risks of bias, and the relatively small number of magnetoencephalography studies limits definitive conclusions regarding its comparative effectiveness.

Automated Post-Processing Analysis of Previously Acquired Brain MRI Data

The evidence for automated post-processing analysis of previously acquired brain MRI data, encompassing lesion identification, characterization and quantification, along with brain volume quantification and/or a severity score (CPT 0865T and 0866T) demonstrates emerging clinical utility in multiple sclerosis monitoring and dementia assessment, with validation data showing improved sensitivity for lesion detection compared to standard radiology reports. However, significant limitations persist including low positive predictive values in real-world settings, scanner-dependent variability, and limited clinical validation for individualized patient management decisions.

Sima et al. (2021) conducted a microsimulation health economic analysis to evaluate the impact of software-assisted brain MRI on treatment decisions, outcomes, and costs in relapsing-remitting multiple sclerosis (RRMS) patients. The objective was to determine whether adding quantitative MRI analysis, specifically lesion detection and brain volume loss measurement, improves detection of disease activity and leads to better clinical and economic outcomes compared with clinical monitoring alone or visual MRI assessment. To achieve this, the authors developed a decision-analytic Markov model simulating a cohort of 1,000 RRMS patients over 10- and 15-year horizons from a US healthcare perspective. The model incorporated both clinical (relapses, EDSS progression) and subclinical (new lesions, brain atrophy) disease activity, along with treatment effects from low- and high-efficacy disease-modifying therapies (DMTs). Four decision strategies were compared: clinical monitoring without MRI, visual MRI assessment (NEDA-3), software-assisted lesion detection (NEDA-3 with software), and software-assisted lesion plus brain atrophy detection (NEDA-4). Outcomes included time with undetected disease activity, quality-adjusted life years (QALYs), and health-state costs, while explicitly modeling diagnostic inaccuracies such as missed lesions or measurement error. The results demonstrated that incorporating MRI, and particularly software-assisted MRI, substantially improved early detection of disease activity. The average time patients remained on low-efficacy therapy with undetected progression decreased from approximately 2.8–3.2 years with clinical assessment alone to about 1.0–1.3 years with software-assisted MRI. This earlier detection enabled faster escalation to high-efficacy therapies, which translated into improved long-term outcomes. Over 10–15 years, software-assisted strategies yielded incremental gains of 0.23–0.37 QALYs compared with clinical monitoring alone. Economically, these clinical benefits led to lower downstream costs driven by better health states, with estimated annual savings of approximately $1,500–$2,200 per patient despite no inclusion of DMT or MRI acquisition costs. Overall, more advanced decision strategies (particularly NEDA-4 with software) showed the greatest clinical and economic benefit, largely due to improved sensitivity in detecting subclinical disease activity. The study has several important limitations. First, it is based on a simulated cohort rather than real-world patient data, meaning results depend on assumptions about disease progression, treatment effects, and diagnostic performance. Second, costs were limited to health-state costs and did not include major drivers such as DMT costs, MRI acquisition, or adverse events, which may influence real-world cost-effectiveness. Third, the model simplified clinical decision-making by applying fixed thresholds for treatment switching and did not incorporate patient-specific factors such as disease severity, treatment tolerability, or clinician judgment. Additional limitations include excluding inter-rater variability in clinical assessments (e.g., EDSS), assuming constant treatment effects over time, and modeling measurement error primarily for brain atrophy but not for other clinical variables. Finally, while the model shows improved outcomes with software-assisted MRI, it also identified a small risk of false-positive detection of disease activity due to measurement error, which could influence treatment decisions.

Barnett et al. (2023) aimed to provide real world clinical validation of an artificial intelligence based MRI analysis tool, iQ-MS, for monitoring disease activity in patients with multiple sclerosis. The objective was to determine whether this AI tool could improve the detection of radiologic disease activity, particularly new or enlarging lesions and brain atrophy, compared with conventional qualitative radiology reports and to assess whether its quantitative outputs were comparable to those generated by a specialized core MRI reading laboratory. The study also sought to address a known limitation in routine MS care, namely that standard radiology reports are largely qualitative and may lack sensitivity for subtle but clinically meaningful changes in lesion burden and brain volume. The investigators conducted a retrospective, multicenter analysis of 397 paired MRI scans from 282 patients with multiple sclerosis acquired in routine clinical practice, typically about 12 months apart. All scans included standardized MRI sequences and were analyzed using three approaches: conventional clinical radiology reports, a blinded core MRI reading laboratory using established quantitative methods, and the fully automated AI tool. An expert consensus panel established the reference standard for lesion activity by resolving discrepancies across methods. The AI system used deep learning algorithms to perform lesion detection, segmentation, and volumetric analyses, including calculation of brain volume and percentage brain volume change. Outcomes included sensitivity and specificity for detecting new or enlarging lesions, agreement with core lab volumetrics, and correlations with clinical disability measures. The results demonstrated that the AI tool had substantially higher sensitivity than routine radiology reports for detecting new or enlarging lesions, achieving 93.3 percent sensitivity compared with 58.3 percent for radiology reports, with only a small reduction in specificity. Its performance was comparable to the core MRI laboratory, particularly in cases with consistent imaging protocols. The AI tool also showed strong agreement with the core lab for quantitative measures such as lesion volume and brain volume, including percentage brain volume change, with closely aligned mean values and strong correlations. Importantly, the study found that conventional radiology reports frequently failed to identify brain atrophy and inconsistently described lesion burden, whereas the AI tool provided standardized, quantitative outputs and additional contextual benchmarking relative to reference populations. Overall, the findings support that AI based MRI quantification can improve detection of subclinical disease activity and enhance the clinical utility of imaging in MS management. Several limitations were noted. The majority of scans were obtained from a limited number of MRI platforms and many scan pairs were acquired using consistent protocols, which may limit generalizability to more heterogeneous real world settings. The determination of expert consensus as the reference standard was not fully blinded, which could introduce bias, although it was partially validated by independent review. Some comparisons, such as lesion segmentation, may have been influenced by shared training data between the AI tool and core lab methods. Additionally, volumetric analyses were restricted to scans meeting quality criteria, and variability between methods at the individual patient level highlights ongoing challenges in applying brain volume metrics clinically. Finally, the tool is designed as an adjunct rather than a replacement for radiologist interpretation, and its impact on clinical outcomes was not directly assessed.

A systematic review by Mendelsohn et al. (2023) evaluated the evidence supporting commercially available quantitative volumetric MRI reporting tools (QReports) for multiple sclerosis (MS), focusing on their technical characteristics, validation, and clinical utility. The objective of the study was to provide a comprehensive synthesis of available commercial QReports used in MS, including their technical features and the extent of their validation, in order to support evidence-based clinical decision-making. The authors aimed to categorize validation evidence using the Quantitative Neuroradiology Initiative (QNI) framework into technical validation (e.g., segmentation accuracy), clinical validation (correlation with clinical outcomes or clinician use), and in-use evaluation (real-world or health economic impact). Methods followed PRISMA guidelines and included a systematic search of FDA-cleared and CE-marked products, supplemented by conference searches and vendor outreach, with inclusion criteria requiring tools to provide MRI-based brain and lesion volumetrics with normative comparisons and structured reporting. A parallel literature review identified peer-reviewed studies evaluating these tools, which were classified according to validation type. The results identified 10 commercial QReport vendors and 38 relevant validation studies. All tools provided automated segmentation of brain and MS lesions using standard MRI sequences (primarily T1 and T2-FLAIR), enabled both cross-sectional and longitudinal analyses, and contextualized outputs against normative databases ranging from approximately 620 to 8000 subjects. Most tools incorporated automated quality control, methods to address inter-scanner variability, and visual overlays for interpretation. In terms of evidence, 30 studies addressed technical validation, demonstrating generally good agreement with manual segmentation or established algorithms, while only 7 studies provided clinical validation, typically limited to correlations with clinical metrics such as EDSS or cognitive scores. Critically, only one study evaluated clinician end-user interaction and workflow impact, and a single study examined in-use economic implications through simulation modeling, suggesting potential benefits in disease monitoring and treatment decisions but without robust real-world confirmation. The study highlights several key limitations. First, there is a significant gap in clinical validation, particularly regarding real-world utility, clinician adoption, and impact on decision-making, which limits confidence in translating technical performance into clinical benefit. Second, in-use evaluations are nearly absent, with minimal evidence on cost-effectiveness, workflow integration, or patient outcomes. Third, heterogeneity in validation methodologies and lack of standardized evaluation frameworks hinder comparison across tools. Additional limitations include potential omission of some CE-marked tools due to database constraints, reliance on vendor-reported technical details that could not be independently verified, and the rapidly evolving nature of the field, meaning newer tools or validations may not have been captured. Overall, the review concludes that although commercial QReports for MS demonstrate strong technical performance and increasing regulatory approval, there remains insufficient clinical and real-world evidence, particularly involving clinician use, to support widespread adoption, underscoring the need for more rigorous, standardized validation and implementation studies.

Peters et al. (2025) evaluated the performance of an artificial intelligence based software tool, mdbrain, for detecting new and enlarging lesions on longitudinal MRI in patients with multiple sclerosis, with a particular focus on its clinical utility and the effect of using different MRI scanners. The objective was to determine whether AI assisted analysis could improve lesion detection compared with routine radiology reports and to assess how variability in imaging hardware influences diagnostic accuracy, given that consistent imaging conditions are often not achievable in routine practice. The investigators conducted a retrospective, single center diagnostic study using follow up MRI examinations from patients with multiple sclerosis performed in 2023. Imaging studies were categorized into those acquired on the same MRI scanner and those acquired on different scanners, including differences in field strength and manufacturer. All studies were analyzed using the mdbrain AI software, which employs a deep learning architecture for lesion detection and segmentation. The results of the AI analysis were compared against routine radiologic reports and a gold standard reference established by an experienced neuroradiologist who reviewed all imaging data. Key performance measures included sensitivity, specificity, positive and negative predictive values, and lesion level agreement using the Dice similarity coefficient. The results demonstrated that the AI tool exhibited high sensitivity and very high negative predictive value for detecting new or enlarging lesions, particularly when follow up imaging was performed on the same scanner, where sensitivity reached 1.0 and negative predictive value was 1.0. Performance declined when different scanners were used, though sensitivity remained relatively high at 0.786 with a negative predictive value of 0.954. In contrast, specificity and positive predictive value were lower, especially in cases involving different scanners, reflecting a substantial number of false positive lesion detections. False positives were primarily driven by signal intensity differences, imaging artifacts, or minor changes in lesion size that were not clinically relevant. The findings also showed that AI analysis could identify lesions missed on routine radiology interpretation in a subset of cases, suggesting potential incremental value in clinical workflows. Overall, the study indicates that AI-based tools can be useful for ruling out disease activity but require expert review to confirm suspected new lesions. Several limitations were identified. The retrospective, single center design may limit generalizability. The study only included MRI scans acquired with a standardized in house protocol and comparable image quality, which may not reflect broader real world variability in imaging parameters. The analysis focused exclusively on FLAIR based lesion detection and did not evaluate other MRI sequences that could influence performance. The gold standard evaluation incorporated information from both the AI outputs and radiology reports, which could introduce bias. Additionally, lesion localization differences between AI and human interpretation were not assessed, and baseline segmentation errors were not evaluated. Finally, performance variability across different scanners highlights a key limitation of current AI tools, as differences in signal characteristics can substantially affect accuracy, particularly by increasing false positive detections.

Hindsholm et al. (2025) aimed to prospectively validate a commercial artificial intelligence tool, mdbrain, for automated assessment of MRI scans in patients with multiple sclerosis, with specific objectives to evaluate its impact on radiologist workflow, assessment time, and diagnostic accuracy in detecting new and enlarging lesions in routine clinical practice. The investigators conducted a prospective, real world clinical validation study that included 112 patients with multiple sclerosis undergoing routine follow up MRI between September and December 2024. MRI data from current and prior scans were analyzed independently by four neuroradiologists, both with and without AI assistance, using a structured reporting protocol. The AI tool employed a deep learning convolutional neural network approach for lesion segmentation and longitudinal lesion detection. Assessment times were recorded, and radiologists completed post-assessment questionnaires regarding tool usability. Quantitative comparisons were made between radiologist-only assessments, radiologist with AI assistance, and AI-only outputs, with performance metrics such as sensitivity, specificity, positive predictive value, and negative predictive value calculated at the case level for detection of new and enlarging lesions. The results showed that use of the AI tool led to a modest reduction in average MRI assessment time, decreasing from 5.3 minutes without AI to 4.8 minutes with AI, although this difference of approximately 27 seconds was not statistically significant. Radiologists reported that the AI tool was helpful in 87 percent of cases, particularly for lesion counting and identification of small lesions. In terms of diagnostic performance, the AI demonstrated high negative predictive value, indicating strong ability to rule out disease progression, but low positive predictive value, reflecting frequent false positive lesion detections. Sensitivity for detecting new lesions was moderate to high depending on the comparator, but specificity and accuracy were reduced by false positives, often related to small or ambiguous imaging features. Overall detection rates of new or enlarging lesions did not differ significantly between AI-assisted and unassisted radiologist assessments. The findings suggest that the tool may be useful for triaging or screening purposes rather than replacing expert interpretation. Several limitations were noted. The study cohort included heterogeneous MRI data acquired across multiple scanner types and protocols, which may have influenced performance but could not be fully evaluated due to limited subgroup sizes. Each case was assessed only once per method, preventing direct intra-reader comparisons and limiting assessment of inter-rater variability. The study focused on case-level lesion detection and did not evaluate spatial accuracy of lesion segmentation. Additionally, the absence of a definitive ground truth for lesion progression introduces inherent uncertainty in performance evaluation. Challenges such as false positive detections, missed small lesions, and dependence on image quality further limit reliability. Finally, workflow integration issues and variability in radiologist experience may affect generalizability of the findings.

Hayes et al. (2026) investigated whether automated cross sectional MRI brain volumetric measures have prognostic value for long term disability outcomes in patients with relapsing remitting multiple sclerosis. The primary objective was to evaluate associations between baseline measures of brain volume and lesion burden and subsequent disability outcomes, including confirmed disability worsening, progression independent of relapse activity, conversion to secondary progressive multiple sclerosis, and confirmed disability improvement. The study addressed a key gap in clinical practice, where conventional MRI lesion metrics have limited prognostic utility, by examining whether quantitative volumetric measures could provide more clinically meaningful predictors of disease progression. This was a large, multicenter international cohort study using prospectively collected data from the MSBase Registry, including 1598 patients with relapsing remitting multiple sclerosis. MRI scans were analyzed using an automated software platform (Icobrain) that generated cross sectional volumetric measurements of whole brain volume, gray matter volume, white matter volume, and lateral ventricular volume, as well as lesion metrics such as T1 black hole lesion volume and T2 lesion volume. Associations between these imaging measures and disability outcomes were assessed using multivariable Cox proportional hazards models, adjusting for relevant clinical covariates such as age, disease duration, baseline disability status, and treatment exposure. Disability outcomes were rigorously defined using standardized criteria, and follow up extended for a median of several years, allowing for assessment of sustained clinical changes. The results demonstrated that cross sectional brain volumetric measures were significantly associated with long term disability outcomes. Higher whole brain, gray matter, and white matter volumes were generally associated with a lower hazard of disability worsening and a higher likelihood of disability improvement, while higher lateral ventricular volume was consistently associated with worse outcomes across all endpoints. Lateral ventricular volume emerged as a particularly robust predictor, showing significant associations with confirmed disability worsening, progression independent of relapse activity, secondary progressive conversion, and disability improvement. In contrast, lesion volume measures showed more limited prognostic value, with only T1 black hole lesion volume significantly associated with risk of conversion to secondary progressive disease. Overall, the findings support that measures of brain atrophy, especially ventricular enlargement, reflect underlying neuroaxonal loss and provide clinically meaningful prognostic information beyond conventional lesion metrics. Several limitations were noted. Registry based data may be subject to variability in data quality and imaging acquisition, although the large sample size enhances generalizability. The study relied on cross sectional measurements and could not fully capture dynamic longitudinal changes or distinguish acute from chronic lesion pathology. The lack of contrast enhanced MRI limited assessment of active inflammation, and certain volumetric measures such as regional gray matter structures and spinal cord metrics were not available, which may also have prognostic relevance. Additionally, potential confounding effects related to treatment initiation and transient brain volume changes were addressed through sensitivity analyses but cannot be entirely excluded. These limitations highlight the need for further studies incorporating more detailed imaging and longitudinal assessments to refine prognostic modeling. 

Temmerman et al., (2025) evaluated the clinical relevance of longitudinal brain volume loss in multiple sclerosis within a real world setting while explicitly addressing variability introduced by different MRI scanners. The primary objective was to determine whether annualized brain volume loss differs between patients with multiple sclerosis and healthy controls, while secondary objectives included assessing the relationship between brain volume loss and clinical disability progression and identifying baseline predictors of pathological brain atrophy. The study sought to address the limited adoption of brain volumetry in clinical practice due to methodological variability and to assess whether harmonization techniques could enable meaningful interpretation of real world imaging data. This retrospective longitudinal cohort study included 72 patients with multiple sclerosis who underwent at least two MRI scans approximately 48 to 60 months apart across multiple centers, along with 27 healthy controls. Brain volume changes in whole brain, total gray matter, cortical gray matter, and deep gray matter were quantified using an automated volumetric pipeline (msBrain). To address variability from different MRI scanners, the investigators applied a similarity index as a post acquisition harmonization method. Clinical outcomes were assessed using multiple validated measures including the Expanded Disability Status Scale, Timed 25 Foot Walk Test, 9 Hole Peg Test, Symbol Digit Modalities Test, and a composite functional score. Statistical analyses included group comparisons, correlation analyses, and regression models to evaluate associations between imaging changes and clinical outcomes. The results showed no significant difference in annualized brain volume loss between patients with multiple sclerosis and healthy controls. However, within the multiple sclerosis cohort, greater gray matter volume loss, particularly in total and cortical gray matter, was significantly associated with clinical worsening, as measured by performance on the 9 Hole Peg Test and composite functional scores. Regression analyses demonstrated that changes in whole brain and gray matter volumes independently predicted clinical decline. Despite these associations, baseline demographic and clinical variables did not predict the likelihood of developing pathological levels of brain atrophy. These findings indicate that while overall brain volume loss rates may not differ from controls in this real world cohort, gray matter atrophy remains clinically meaningful and correlates with functional deterioration. The study has several limitations. The relatively small sample size and substantial data loss due to quality control and harmonization thresholds reduced statistical power and may have limited detection of between group differences. Scanner variability, although addressed with a similarity index, may still have influenced results, and the harmonization method does not account for biological factors affecting brain volume. The retrospective design and reliance on two timepoints limited the ability to assess longitudinal patterns of confirmed disability progression. Potential confounding factors such as pseudoatrophy and treatment effects could not be fully excluded. Additionally, the absence of detailed regional volumetric measures, such as specific gray matter structures or spinal cord imaging, restricts more granular assessment of disease mechanisms.

D’hooghe et al., (2019) evaluated whether a single MRI based volumetric assessment obtained during routine clinical practice contributes to the prediction of physical and cognitive disability in patients with multiple sclerosis. The primary objective was to determine the added value of automated brain and lesion volume measurements in explaining disability outcomes, specifically Expanded Disability Status Scale scores for physical disability and Symbol Digit Modalities Test performance for cognitive function, using real world clinical data where the applicability of volumetric tools has remained uncertain. The investigators conducted a retrospective study of 470 adults with multiple sclerosis who underwent MRI as part of standard care. Brain and lesion volumes were quantified using an automated software tool that segmented white matter lesions and different brain compartments from 3D FLAIR and T1 weighted MRI sequences. Clinical variables including disability scores, demographic characteristics, disease features, and treatment status were collected, and statistical modeling was performed using stepwise linear regression to assess predictors of disability at baseline. In addition, longitudinal analyses were conducted using Kaplan–Meier estimates and Cox proportional hazards models to evaluate time to worsening in disability outcomes across quartiles of brain volume. The results showed that both brain and lesion volume measurements significantly improved prediction of disability outcomes compared with models using clinical variables alone. Inclusion of volumetric data increased the explained variance for physical disability from 17 percent to 28 percent and for cognitive disability from 9 percent to 25 percent. Whole brain volume was the most important contributor to physical disability, followed by age and lesion volume, whereas lesion volume and whole brain volume were key contributors to cognitive performance. Lower brain volumes, particularly those in the lowest quartile, were associated with an increased risk of cognitive decline over time, although no clear relationship was observed between baseline brain volume and worsening of physical disability. The findings support a threshold effect in which patients with the lowest brain volumes are at greater risk of cognitive deterioration. Several limitations were acknowledged. The retrospective design and reliance on routinely collected clinical data may limit generalizability. MRI scans were obtained from multiple centers with different scanners and protocols, introducing potential technical variability despite quality control measures. Clinical and imaging assessments were not always performed at the same time, and imputation of missing data was required for some analyses. Longitudinal follow up was incomplete for some patients, potentially affecting estimates of disability progression. In addition, heterogeneity in treatment exposure and the use of real world datasets may introduce confounding factors that are difficult to fully control. 

Fragoso et al., (2017) examined the relationship between brain volumetric measurements and clinical disease activity in multiple sclerosis using real world data, with the objective of determining whether disability and recent relapse activity correlate with MRI derived measures of brain volume obtained from the MSmetrix platform. The study aimed to evaluate the clinical relevance of automated volumetric analysis as a tool for assessing neurodegeneration and disease severity in routine practice. The investigators conducted a cross sectional study using data from 185 patients with multiple sclerosis whose MRI scans had been analyzed with MSmetrix. Brain volumetric measures included total brain volume, gray matter volume, and lesion load volume, which were normalized to account for age, sex, and head size. Clinical data included disability measured by the Expanded Disability Status Scale and the number of relapses in the previous year. Statistical analyses included nonparametric tests and correlation analyses to assess relationships between imaging variables and clinical outcomes. The results demonstrated significant correlations between reduced brain volume and increased disability, with both total brain and gray matter volumes inversely associated with disability scores. Higher lesion load and a greater number of relapses in the preceding year were also significantly associated with lower brain volume and higher disability. Additionally, lesion volume correlated with both increased relapse frequency and greater disability, indicating that inflammatory disease activity and neurodegeneration are closely linked. These findings support the clinical relevance of volumetric MRI measures as indicators of disease burden and progression in multiple sclerosis. The study has several limitations. Its cross sectional design limits the ability to assess causality or longitudinal progression. The analysis relied on a single MRI per patient, which restricts evaluation of temporal changes in brain volume. The use of real world clinical data introduces variability in imaging acquisition and clinical assessment. Additionally, while volumetric measures were normalized, residual confounding factors related to patient characteristics and imaging differences may remain. 

Brewer et al. (2009) evaluated a fully automated magnetic resonance imaging postprocessing system for quantifying regional brain volumes to improve the detection of focal atrophy in Alzheimer disease, with the objective of determining whether automated volumetric analysis can accurately identify neurodegenerative changes that are typically assessed through more labor intensive methods. The authors aimed to address the need for practical and reliable imaging biomarkers that could assist in early diagnosis, risk stratification, and monitoring of disease progression in clinical settings. The investigators conducted a technical validation study using MRI data from 40 elderly subjects, including individuals with mild probable Alzheimer disease and age matched healthy controls. Structural T1 weighted MRI scans were processed using the NeuroQuant software, which performs fully automated segmentation of brain structures through a sequence of preprocessing steps that include quality checks, correction for imaging artifacts, skull stripping, and probabilistic atlas based labeling of brain regions. Automated volumetric results were compared with those obtained through independent computer assisted semimanual segmentation performed by an expert, and statistical measures such as intraclass correlation coefficients were used to assess agreement between methods. Group comparisons were also conducted to evaluate the sensitivity of the automated system in detecting disease related differences in regional brain volumes. The results demonstrated a high level of agreement between automated and semimanual segmentation methods, with strong correlations observed across multiple brain regions, particularly in medial temporal structures such as the hippocampus and lateral ventricles that are known to be affected early in Alzheimer disease. The automated system was sensitive to characteristic patterns of atrophy, including reduced hippocampal and amygdala volumes and increased ventricular volumes in patients with Alzheimer disease compared with controls. No significant differences were observed in brain regions not typically involved in early disease, supporting the specificity of the method. These findings indicate that automated volumetric analysis can provide objective and clinically meaningful measurements of neurodegeneration that align closely with established reference methods.Several limitations were noted or implied. The study used a relatively small sample size and relied on existing imaging datasets rather than prospectively collected data, which may limit generalizability. The validation was conducted under controlled conditions and performance in more heterogeneous real world imaging environments was not fully assessed. The method may be affected by substantial structural abnormalities such as large hemorrhages, which can interfere with accurate segmentation of adjacent regions. Additionally, while the system provides quantitative measures, broader clinical validation is required to establish its role in routine care and in predicting longitudinal outcomes.

Wittens et al. (2021) evaluated the diagnostic performance of automated MRI volumetry using icobrain dm for Alzheimer’s disease (AD) in a real-world clinical setting. The objective was to determine whether icobrain dm improves diagnostic accuracy across the AD continuum (healthy controls, subjective cognitive decline, mild cognitive impairment, and AD dementia) and to compare its performance with the widely used FreeSurfer software, while also assessing whether combining multiple brain volumes enhances diagnostic classification. The study used a large, retrospective, multicenter cohort (REMEMBER) including 820 participants who underwent routine clinical MRI. Automated segmentation of global, cortical, and subcortical brain regions was performed, and diagnostic performance was evaluated using ROC analyses and logistic regression models. Additionally, stepwise backward regression was applied to identify optimal combinations of volumetric measures for distinguishing disease stages. The results demonstrated that icobrain dm provided robust and efficient volumetric analysis, outperforming FreeSurfer in processing time (minutes vs hours) and robustness (no failures vs multiple pipeline failures) while also showing superior diagnostic performance for key structures. Specifically, hippocampal volume, temporal cortex, whole brain volume, and lateral ventricles were the most informative individual biomarkers for distinguishing AD dementia from healthy controls, with hippocampal measures a, chieving the highest accuracy (AUC up to ~0.87). Importantly, combining multiple brain volumes significantly improved diagnostic accuracy beyond any single metric. The best-performing multivariate model achieved an AUC of 0.914 with sensitivity of 86.3% and specificity of 83.0% for distinguishing AD dementia from healthy controls. Across disease stages, patterns of progressive atrophy were observed, particularly in hippocampal and temporal regions supporting known AD pathophysiology. icobrain dm also demonstrated better discrimination than FreeSurfer for several measures, including whole brain, hippocampal volumes, and lateral ventricles, while maintaining stable performance across heterogeneous real-world imaging data. The study highlights several limitations. First, the retrospective multicenter design introduced variability in MRI acquisition (e.g., slice thickness, scanner type, and image quality), which may affect generalizability and accuracy. Second, while age and sex adjustments were applied, there were baseline demographic differences between groups that could introduce residual confounding. Third, automated volumetry does not capture all disease-relevant information and is not specific to AD, requiring integration with other biomarkers and clinical assessment. Fourth, image quality variability, especially lower-resolution scans. reduced sensitivity in some analyses. Finally, the study did not include direct comparison with manual segmentation in this dataset and remains subject to inherent biases of algorithm-based segmentation approaches. Overall, the study concludes that automated MRI volumetry, particularly when combining multiple brain regions, has meaningful diagnostic value in routine clinical practice and can improve accuracy and efficiency in AD diagnosis, supporting broader adoption of such tools in real-world settings.

Sima et al. (2024) evaluated the diagnostic performance of an artificial intelligence (AI)–based assistive software tool (icobrain aria) for detecting and quantifying amyloid-related imaging abnormalities (ARIA) on MRI in patients with Alzheimer disease receiving amyloid-β–targeting therapies. The objective was to determine whether radiologists’ diagnostic accuracy improves when using the software compared with standard unassisted interpretation, given the known difficulty of identifying subtle ARIA findings that are critical for treatment monitoring and dosing decisions. The study used a multiple-reader, multiple-case (MRMC) design in which 16 board-certified radiologists independently reviewed 199 paired MRI scans (baseline and follow-up) from clinical trials (PRIME, EMERGE, ENGAGE), both with and without software assistance. The AI tool analyzes longitudinal MRI data using deep learning models to detect ARIA-E (edema/sulcal effusion) and ARIA-H (microhemorrhage/siderosis), quantify lesion burden, assign severity, and generate annotated outputs for concurrent radiologist review. The results demonstrated that software assistance significantly improved diagnostic accuracy across primary and secondary endpoints. For ARIA-E detection, the mean area under the ROC curve (AUC) increased from 0.82 unassisted to 0.87 assisted (difference +0.05, p=0.001), and for ARIA-H detection from 0.78 to 0.83 (difference +0.04, p=0.001). Sensitivity improved substantially (ARIA-E: 71% to 87%; ARIA-H: 69% to 79%), particularly for mild cases where detection is most challenging, while specificity remained above 80% despite slight decreases. Secondary analyses confirmed improved performance for severity classification and lesion localization, as well as higher interreader agreement when using the software (e.g., Kendall coefficient increased from 0.72 to 0.81 for ARIA-E severity). The software notably reduced missed cases, especially subtle ARIA-E findings, and enhanced consistency among readers without increasing interpretation time. Overall, the findings suggest that AI-assisted interpretation improves both sensitivity and reproducibility of ARIA detection, which is clinically important for safely managing disease-modifying therapies. The study has several limitations. It used retrospective imaging data from clinical trials (primarily aducanumab studies), which may limit generalizability to broader real-world populations and other therapies, although ARIA characteristics are expected to be similar. The reader cohort excluded radiologists with prior ARIA experience and received training before the study, which may overestimate baseline performance compared with routine practice. Sample size, while powered for primary endpoints, was relatively limited for some subgroup analyses. The software was not trained to detect large macrohemorrhages, and occasional discordances between software output and expert consensus highlight reliance on model training data and the need for continued radiologist oversight. Finally, the tool is intended as a decision-support system and requires integration with clinical context and multidisciplinary interpretation rather than standalone use. Overall, this study demonstrates that AI-based assistive software can meaningfully enhance radiologist performance in detecting and characterizing ARIA, particularly improving sensitivity for subtle findings and interreader consistency, though validation in broader clinical settings remains necessary.

References