The use of Artificial Intelligence in Magnetic Resonance Imaging of Epilepsy: A Systematic Review and Meta-Analysis

Chen, J.; Sahlas, E.; Zhou, Y.; Chen, N.; Xie, J.; Wadia, F.; Armstrong, S.; Caciagli, L.; Weil, A.; Dudley, R.; Schrader, D. V.; Bernasconi, A.; Bernasconi, N.; Bernhardt, B.

doi:10.1101/2025.09.19.677393

bioRxiv · 10.1101/2025.09.19.677393

The use of Artificial Intelligence in Magnetic Resonance Imaging of Epilepsy: A Systematic Review and Meta-Analysis

Chen, J.·Sahlas, E.·Zhou, Y.·Chen, N.·Xie, J.·Wadia, F.·Armstrong, S.·Caciagli, L.·Weil, A.·Dudley, R.·Schrader, D. V.·Bernasconi, A.·Bernasconi, N.·Bernhardt, B.

View original source ↗Export metadata ↓

Abstract

BackgroundThe application of artificial intelligence (AI)/machine learning (ML) to MRI can be a powerful tool to streamline clinical decision-making, yet variability amongst MRI sequences and algorithms have hindered appropriate assessment of reliability and generalizability. MethodsWe conducted a systematic review and meta-analysis of the ability of current AI/ML models operating on MRI data for: 1) epilepsy diagnosis, 2) temporal lobe epilepsy lateralization, 3) lesion localization, and 4) post-surgical outcome prediction. Searches were conducted across PubMed, Medline, and Embase databases from inception until January 1, 2025. We selected studies that employed AI/ML models trained on any MRI modality to classify at least ten patients across the four main objectives for qualitative assessment, and further included in the meta-analysis if they reported an accuracy rate. The primary outcome of the meta-analysis was the overall accuracy of AI/ML models trained on MRI data. The secondary outcome was the concomitant risk of bias evaluation using PROBAST. FindingsWe identified 158 studies for qualitative evaluation and 127 studies for inclusion in the meta-analysis. AI/ML on multimodal MRI could accurately distinguish epilepsy patients from healthy controls (overall accuracy: 88% [85-90]), lateralize temporal lobe epilepsy (90% [87-93]), localize epileptogenic lesions (82% [74-88]), and predict post-surgical seizure-freedom (83% [78-87]). Overall, a high risk of bias remains in the literature; participant bias remained high across all outcomes (64-87%), as well as predictor (88-100%) and analysis (83-100%) bias. Outcome bias was low only for AI/ML studies predicting post-surgical outcomes. InterpretationOur results support promising accuracy of AI/ML models in epilepsy diagnostics and prognostics but remain highly susceptible to bias in participants, predictors, outcome, and analysis domains which limits current translation to routine clinical practice. We encourage closer interdisciplinary collaboration between clinical and scientific groups to improve validation studies based on thorough study design, analysis, and reporting.

Explore related subjects

neuroscience Preprint

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Chen, J., Sahlas, E., Zhou, Y., Chen, N., Xie, J., Wadia, F., Armstrong, S., Caciagli, L., Weil, A., Dudley, R., Schrader, D. V., Bernasconi, A., Bernasconi, N., Bernhardt, B.. 2025-09-20. The use of Artificial Intelligence in Magnetic Resonance Imaging of Epilepsy: A Systematic Review and Meta-Analysis. https://doi.org/10.1101/2025.09.19.677393

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

Isogenic forebrain organoids uncover early neurodevelopmental alterations and imbalances in neuronal function leading to hyperexcitation in Gaucher disease

Gaucher disease is a rare lysosomal storage disorder caused by autosomal recessive mutations in the GBA1 gene, encoding the lysosomal enzyme glucocerebrosidase. Gaucher disease is classified in 3 different subtypes depending on the presence and severity of neurological involvement, with type 2 resulting in fatal early-onset neuropathology and patients exhibiting developmental delays, seizures and early death. Studies investigating disease mechanisms of neuronopathic Gaucher disease are mainly based on animal models and focus predominantly on late neuronal phenotypes. Here, we established healthy control and Gaucher disease patient-derived iPSC lines and engineered them to obtain isogenic control and disease lines. Using these lines, we generated cortical and subpallial brain organoids in which we identified early-onset lipid dysregulation in form of glucosylceramide accumulation, highly elevated glucosylsphingosine, and a later increase in ganglioside levels, recapitulating clinical findings. Furthermore, single-cell transcriptomic profiling uncovered novel phenotypes in both cortical and subpallial forebrain organoids. Subpallial alterations consisted of an early increase in migrating interneurons in subpallial organoids, which upregulated cholesterol metabolism. Cortical alterations showed early upregulation of mitochondrial genes and a downregulation of proliferation, with a subsequent switch from GABAergic to glutamatergic neuron fate with a striking increase in gene expression related to the synaptic assembly. Functional assays demonstrated a marked hyperexcitability of cortical organoids and reduced response to GABA-A receptor blockage in Gaucher disease. Additional 2D neuronal network models confirmed the organoid data and showed that both glutamatergic and GABAergic neurons contribute to the phenotype, with hyperexcitability of Gaucher glutamatergic neurons and incapacity of Gaucher GABAergic neurons to balance the excessive excitation. This alteration represents a clinically significant phenotype as many patients exhibit an excitation/inhibition imbalance leading to treatment-resistant seizures, hastening their decline. In conclusion, our defined human models of Gaucher disease identify novel and clear phenotypes that can be used for drug screening or aid in development of new therapeutic strategies to ameliorate Gaucher disease.

neuroscience↗

bioRxiv2026

Oxytocin and Vasopressin Immunoreactivity Differs Across Auditory Brainstem Nuclei in Rodents with Distinct Social Systems

Oxytocin (OT) and vasopressin (AVP) are neuropeptide hormones involved in regulating animal social behavior and a broad spectrum of physiological processes. Although their distributions are well documented in neuroendocrine regions of the forebrain and midbrain, their expression in the hindbrain remains poorly understood. Here, we used immunohistochemistry to quantify OT and AVP immunoreactive puncta within three auditory brainstem nuclei, the lateral superior olive (LSO), the medial superior olive (MSO), and the medial nucleus of the trapezoid body (MNTB) in six wild-caught rodent species differing in sociality. We also quantified the volume of these nuclei and examined variation in total brain volume across species and sociality. OT and AVP puncta count differed among species and social groups. Group-living species exhibited higher OT and AVP puncta counts than monogamous and solitary species in the LSO and MNTB. In the MSO, OT puncta counts did not differ among social groups, whereas AVP puncta counts were higher in group-living than in monogamous and solitary species. Total brain volume and the volumes of the MNTB and MSO differed among species, but not across social groups, whereas LSO volume did not differ among species or sociality. These findings revealed sociality-related variation in OT and AVP immunoreactive puncta within auditory brainstem circuits and suggest that neuropeptide signaling within early auditory brainstem pathways may contribute to the neural integration of social and auditory information.

neuroscience↗

bioRxiv2026

Connexin 40 deficiency alters the temporal profile of postictal oxygen dynamics following focal seizures.

Epilepsy is increasingly recognized as a disorder involving both neuronal and vascular dysfunction. While connexin signaling has been implicated in epileptogenesis, the contribution of vascular connexins to seizure associated cerebrovascular pathology remains poorly understood. Connexin40 (Cx40) is an endothelial gap junction protein that plays a crucial role in vascular communication and blood-flow regulation. Seizures induce dynamic changes in cerebral perfusion and oxygenation, including prolonged postictal hypoperfusion/hypoxia. To determine whether Cx40 influences postictal hypoxia following focal seizures, we examined seizure characteristics and postictal oxygen dynamics in Cx40 knockout (Cx40-/-) mice using an established focal hippocampal seizure model. Electrically kindled seizures were elicited in wild-type and Cx40-/- mice, and local hippocampal tissue oxygenation was continuously monitored before and after seizure induction. Seizure duration did not differ between genotypes, indicating comparable seizure severity. Interestingly, Cx40 deletion altered the temporal pattern of postictal oxygen recovery, producing greater early hypoxia and a delayed secondary rebound in pO2 despite similar peak oxygen levels and overall hypoxic burden. These findings demonstrate that loss of Cx40 selectively alters the temporal profile of postictal oxygen dynamics without affecting seizure duration. Taken together, the results suggest that endothelial gap junctional communication contributes to postictal vascular recovery and identify Cx40 as a potential modulator of seizure associated neurovascular dysfunction.

neuroscience↗