RPS 411 - Neuroanatomy and epilepsy imaging

Author Block: T. Mishael, A. Jasper, H. A. Vanjare, B. T S, J. H. Prasad D; Vellore/IN
Purpose: To determine the role of pre- and post-operative MRI, including advanced imaging techniques, in predicting seizure recurrence after epilepsy surgery, and to identify imaging biomarkers that correlate with long-term outcomes.
Methods or Background: We retrospectively reviewed 102 patients who underwent temporal or extra-temporal epilepsy surgery with ≥12 months follow-up. Pre-operative imaging included 3T MRI (3D T1, FLAIR, T2, DTI), arterial spin labelling (ASL) perfusion, MR spectroscopy (MRS), and fMRI in selected cases. Post-operative MRI was obtained at early (≤72 h) and/or delayed (3–12 mo) intervals. Imaging features—extent of resection (EOR), residual lesions, gliosis, hippocampal remnants, tract integrity, perfusion asymmetry, and functional connectivity—were correlated with Engel/ILAE outcomes using Cox regression and Kaplan–Meier analysis.
Results or Findings: At 24 months, 38% (39/102) experienced seizure recurrence.
Post-operative predictors: Incomplete resection (24%) predicted recurrence (HR 2.4, p=0.006). Residual hippocampal tail remnants (31%) increased risk (HR 1.9, p=0.02). Gliosis was frequent (63%) but non-specific.
Pre-operative predictors: ASL hypoperfusion (HR 2.1, p=0.01), reduced NAA/Cr on MRS (OR 2.7, p=0.03), and reduced hippocampal–default mode connectivity on fMRI (p=0.04) were associated with recurrence.
A Composite Imaging Score (CIS) stratified patients into risk groups with distinct seizure freedom rates: 78% (low risk), 54% (intermediate), 29% (high) (log-rank p<0.001).
Among patients undergoing re-surgery, 58% achieved Engel I–II outcomes.
Conclusion: Integrating pre- and post-operative MRI with advanced imaging improves prediction of seizure recurrence. A composite imaging framework offers a practical tool for risk stratification and personalised follow-up.
Limitations: Single-center design, incomplete availability of advanced imaging for all patients, and variable surgical techniques may limit generalizability. Prospective multi-center validation is warranted.
Funding for this study: This study did not receive any grant or external funding.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was approved by the institutional ethics committee (IRB No. 2025/463)

Author Block: X. Xia¹, Q. Ren¹, X. Meng¹, K. Xue², F. Long²; ¹Qingdao/CN, ²Shanghai/CN
Purpose: To evaluate glymphatic system (GS) functional alterations using global and temporal BOLD-CSF couplings, and to assess microstructural alterations via NODDI parameters, in patients with temporal lobe epilepsy (TLE). We further evaluated their associations with cognition and predictive value for mild cognitive impairment (MCI).
Methods or Background: From January 2024 to May 2025, 71 TLE patients (19 Drug-resistant epilepsy (DRE) ,52 non-DRE) and 49 age- and sex-matched healthy controls (HCs) underwent 5T MRI, including resting-state fMRI and NODDI. BOLD-CSF couplings (gBOLD-CSF and tBOLD-CSF) were computed to assess GS dynamics, while NODDI metrics (global/temporal isotropic volume fraction [isovf], intracellular volume fraction [icvf], and orientation dispersion [od]) quantified microstructural changes. Montreal Cognitive Assessment (MoCA) scores assessed cognition. Partial Spearman's correlations examined associations between metrics and MoCA, adjusting for confounders. Forward stepwise logistic regression developed a model to predict MCI, with receiver operating characteristic (ROC) analysis evaluating performance.
Results or Findings: TLE patients showed lower MoCA scores (p=0.014), reduced global isovf (p<0.001), and elevated global icvf (p=0.016) and od (p<0.001) versus HCs. Post-hoc analyses revealed DRE subgroup exhibited decreased tBOLD-CSF coupling and T_icvf, with increased T_isovf, compared to non-DRE subgroup, (all adjusted p<0.05)(Fig.1). MoCA positively correlated with gBOLD-CSF (r=0.474) and tBOLD-CSF (r=0.436; both FDR-p<0.001) in TLE patients (Fig.2). The combined model (tBOLD-CSF, T_isovf, epilepsy duration) predicted MCI with AUC=0.891 (sensitivity=95.0%, specificity=72.0%), outperformingtBOLD -CSF coupling (p<0.05) (Fig.3).
Conclusion: BOLD-CSF decoupling and NODDI alterations reflect GS dysfunction and microstructural damage in TLE, serving as potential biomarkers for early cognitive decline detection and MCI prediction.
Limitations: The single-center design and potential influence of antiepileptic drug may bias the GS assessments.
Funding for this study: Qingdao Clinical Research Center for Rare Diseases of Nervous System (22-3-7-lczx-3-nsh), and Qingdao Key Health Discipline Development Fund (QDZDZK-2025067).
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The study involving human subjects received approval from the Ethics Committee of Qilu Hospital of Shandong University (KYLL-qdql2020070).

Author Block: Z. Huang, Y. Li, N. Zhu, L. Mazu, J. Chu; Guangzhou/CN
Purpose: This study aims to use the recently proposed sub-voxel QSM, namely APART-QSM, to simultaneously quantify the paramagnetic and diamagnetic susceptibility sources in temporal lobe epilepsy (TLE).
Methods or Background: In this retrospective analysis, a total of 52 healthy subjects (HCs) and 68 patients with TLE were recruited. According to the ILAE 2020 definition, the patients were divided into 30 non-refractory, 31 refractory TLE and 7 patients were lost to follow-up. APART-QSM could distinguish the paramagnetic susceptibility map (Xpara) and diamagnetic susceptibility map (Xdia).
Results or Findings: In the voxel-wise analysis, compared with HCs, TLE patients exhibited deceased Xpara and increased Xdia in a large scale of brain regions. Remarkably, refractory TLE demonstrated significantly decreased Xpara and increased Xdia in both thalamus relative to non-refractory TLE. Decreased Xdia was found in the left supramarginal gyrus in refractory TLE patients compared to non-refractory cohorts (all FDR-q < 0.05 & cluster extent > 100 voxels). No significant differences were observed in disease duration between patient groups. The log-transformed seizure frequency is higher in refractory TLE than that in non-refractory patients (ln [seizure frequency] 0.31 ± 0.34 vs 0.80 ± 0.67, P = 0.001). In the patient cohorts, the disease duration was negatively correlated with Xpara in the left thalamus (r = -0.284, P = 0.026), right thalamus (r = -0.367, P = 0.004) and positive with Xdia in the right thalamus (r = 0.395, P = 0.002). The log-transformed seizure frequency is negatively correlated with Xdia in the left supramarginal gyrus (r = -0.258, P = 0.045).
Conclusion: The sub-voxel QSM offers significant insights into the magnetic susceptibility perturbation of the whole brain in TLE patients.
Limitations: This work is lacking of pathological evidence.
Funding for this study: None.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The institutional ethical committee of the First Affiliated Hospital of Sun Yat-Sen University.

Author Block: O. A. Kargin, R. Ozun Kargin, A. Ceyhan Dirican, H. D. Atakli; Istanbul/TR
Purpose: Juvenile myoclonic epilepsy (JME) is the most prevalent idiopathic generalized epilepsy syndrome, with a typical onset at adolescence. Beyond seizures, JME has increasingly been recognized as a condition involving widespread cognitive dysfunction, particularly affecting memory, executive functions, and attention. This study aimed to investigate cortical structural alterations in JME using surface-based morphometry and to explore their relationship with cognitive performance, with the goal of identifying potential imaging biomarkers.
Methods or Background: Fifty-five patients with JME and 65 healthy controls matched for age, sex, and educational status underwent high-resolution 3D T1-weighted brain MRI. Cortical thickness, sulcal depth, gyrification index, and fractal dimension were extracted using the CAT12/SPM pipeline. Cognitive functions were assessed with a comprehensive neuropsychological battery covering multiple domains including memory, verbal fluency, attention, executive functions, and psychomotor speed. Vertex-wise analyses were conducted using threshold-free cluster enhancement (TFCE). All models were adjusted for age, educational status, depression/anxiety scores, and Euler number.
Results or Findings: Compared to controls, JME patients exhibited significant cortical thinning and increased gyrification, most prominently in perirolandic and medial prefrontal cortices, extending to frontal, parietal, and occipital regions (pFWE<0.001). These effects remained robust after controlling for global mean cortical thickness. Correlation analyses demonstrated moderate positive associations between cortical thickness in frontal regions and cognitive performance, particularly in memory, verbal fluency, and executive domains (pFWE<0.05, r≤0.57).
Conclusion: Our findings highlight region-specific cortical alterations in JME that extend beyond global effects, linking frontal structural abnormalities to clinically relevant cognitive deficits. These results support the role of surface-based morphometric parameters as promising neuroimaging biomarkers of cognitive impairment in JME.
Limitations: This is a cross-sectional study without longitudinal follow-up.
Funding for this study: No funding was received for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The study was approved by the ethics committee of University of Health Sciences, Bakirkoy Prof. Dr. Mazhar Osman Training and Research Hospital for Psychiatric, Neurologic, and Neurosurgical Diseases.

Author Block: X. Chen, L. Song, A. j. Zhang, X. Yang, L. Jiang; Zunyi City/CN
Purpose: To reveal network-level functional reconfiguration in children with self-limited epilepsy with centrotemporal spikes (SeLECTS) by integrating static functional connectivity (sFNC), dynamic functional connectivity (dFNC), and hidden Markov model (HMM) analysis.
Methods or Background: Sixty-one children with SeLECTS and sixty-nine healthy controls (HCs) underwent resting-state fMRI. Independent component analysis identified 34 components, 15 of which were matched to six canonical networks. sFNC was computed from full time series. dFNC was assessed using sliding-window and k-means clustering. And the HMM was applied to resolve dynamic brain activation states.
Results or Findings: The sFNC analysis showed significantly reduced connectivity within the DAN (IC_31–IC_32 and IC_32–IC_33) in SeLECTS compared to HCs. The dFNC analysis identified four recurrent states. In state 1 (IC_32–IC_33) and state 3 (IC_31–IC_32), both reflecting intra-DAN connectivity, SeLECTS showed significantly reduced connectivity compared to HCs. No group differences were found in states 2 and 4. State 2 was characterized by more balanced connectivity and showed fewer fractional windows and shorter mean dwell time in SeLECTS. In contrast, States 1 and 3 represented two extreme connectivity patterns, respectively reflecting low integration and over-integration. The HMM analysis also identified four dynamic states, including distinct extremes of brain network activation—one with markedly increased activation and another with globally reduced activation. Additionally, SeLECTS exhibited prolonged dwell time in a state characterized by suppressed DAN activity.
Conclusion: This study indicates that persistent disruption of DAN connectivity may underlie resting-state network abnormalities in SeLECTS. Imbalanced connectivity states revealed by dFNC and HMM from distinct dynamic perspectives further support network reorganization in SeLECTS.. This static-to-dynamic integrated analysis offers new perspectives on its neuropathological mechanisms.
Limitations: Limitations include cross-sectional design, lack of EEG-fMRI, and uncontrolled medication effects potentially confounding connectivity findings.
Funding for this study: This work was supported by the National Natural Science Foundation of China (Grants No. 82160328), Natural Science Foundation of Guizhou Province (Project No. Qiankehejichu-ZK [2021] yiban 479, Qiankehejichu-ZK [2022] yiban 582), Scientific and Technological Innovation Talent Team for Functional Imaging and Artificial Intelligence Applications in Guizhou Province (Project: QKHRC- CXD(2025)047).
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: A. V. Smirnova, D. Rudenko; Saint Petersburg/RU
Purpose: To assess the concordance between typical MRI patterns and histologically confirmed FCD IIb, and to determine the diagnostic value of these imaging features.
Methods or Background: A total of 93 patients (minimum age in both groups — 3 years) who underwent surgical treatment for focal cortical dysplasia were included in the study. Histopathological examination revealed: FCD type I — 37 cases, FCD type IIb — 56 cases. All patients underwent preoperative MRI on 3T scanners with HARNESS protocols and high‑resolution T2‑weighted sequences (slice thickness 0.6–1.0 mm), with a focus on detecting the transmantle sign.
Results or Findings: The transmantle sign was detected on MRI in 36/56 patients with FCD IIb (64.3%; 95% CI: 50.9–76.0%) and in none of the patients with FCD type I. Demographic data: FCD IIb group — mean age: 17.2 ± 6.4 years (range: 3–32 years), females: 57.1%; FCD I group — mean age: 19.5 ± 7.2 years (range: 3–35 years), females: 51.4%. The difference between groups was statistically significant (Fisher’s exact test, p < 0.0001). Sensitivity — 64.3%, specificity — 100%, positive predictive value (PPV) — 100%, negative predictive value (NPV) — 65.5%.
Conclusion: The transmantle sign demonstrates high specificity (100%) for FCD IIb, with statistically significant differences compared to FCD type I (p < 0.0001). However, its moderate sensitivity (64.3%) means that absence of the sign on MRI, even with 3T scanners and HARNESS protocols, does not rule out the diagnosis. Comprehensive preoperative evaluation integrating clinical and electrophysiological data remains essential.
Limitations: N/A
Funding for this study: N/A
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: H. Anand¹, A. Kapoor², L. H P¹, H. G. K B¹, P. B P¹, A. Kapoor², R. Kanotra³, D. Singh², D. Kumar²; ¹Bengaluru/IN, ²New Delhi/IN, ³Gilbert, AZ/US
Purpose: To evaluate the diagnostic accuracy of a deep learning-based volumetric analysis tool in epilepsy patients with hippocampal sclerosis (HS) across two centers, and to determine which brain regions exhibit the most significant volumetric changes compared to healthy controls (HC).
Methods or Background: This two-center study included 40 subjects:20 patients with MRI-confirmed HS and 20 HC. MRI data were acquired on 3T systems (uMR 780, United Imaging Healthcare,Center-1; and MAGNETOM Skyra, Siemens Healthineers,Center-2). Volumetric analysis was performed using the cascaded weakly supervised confidence integration network (CINet), which applies weak supervision and confidence integration for accurate brain region segmentation. Regions of interest included the hippocampus, amygdala, entorhinal cortex, parahippocampal gyrus, thalamus, and temporal neocortex. A recursive feature elimination pipeline identified discriminative volumetric markers. Classification of HS versus HC was performed using Support-Vector Machine (SVM) and Random-Forest (RF) with 5-fold cross-validation.
Results or Findings: HS patients demonstrated volumetric reductions relative to HC, reflecting both focal hippocampal pathology and broader network involvement. The hippocampus showed the most pronounced atrophy (mean-reduction 27.4%, p < 0.001), followed by the amygdala (21.6%, p < 0.01). Moderate but significant reductions were observed in the entorhinal-cortex (15%) and parahippocampal gyrus (18%) (both p < 0.05). Extra-temporal regions including the thalamus (10%) and temporal neocortex (14%) also showed significant volume loss (p < 0.05). RF classifier achieved an accuracy of 93.3%, while SVM reached 91.3%. Feature selection consistently identified hippocampal and amygdala volumes as the strongest predictors of HS.
Conclusion: This two-center pilot study demonstrates that CINet-based volumetric analysis reliably detects HS-associated brain changes, with RF providing superior classification performance. The integration of deep learning, feature selection, and supervised classifiers offers a robust framework for objective and generalizable epilepsy diagnostics.
Limitations: Small cohort.
Funding for this study: Not Applicable
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: T. M. C. Nguyen; Hanoi/VN
Purpose: Hippocampal sclerosis (HS) is the most frequent pathological finding in temporal lobe epilepsy (TLE) but may be subtle or missed on conventional MRI, limiting pre-surgical decision-making. HipUnfold, an automated surface-based morphometry tool, has shown promise for HS detection and lateralisation. However, prior work focused mainly on volumetry and asymmetry indices. We extended this approach using unsupervised clustering of shape features and compare imaging-derived clusters with histopathological severity.
Methods or Background: We retrospectively included 30 unilateral HS patients who underwent surgical resection and 30 age- and sex-matched controls. Preoperative T1-weighted MRI was processed using HipUnfold to reconstruct hippocampal surfaces and extract structural features (volume, thickness, gyrification, curvature). Morphological features were compared across sclerotic, contralateral, and control hippocampi. Principal component analysis (PCA) was applied to shape-normalized features, followed by unsupervised k-mean clustering.
Results or Findings: HS patients showed significant reductions in hippocampal volume (−24.6%, p < 0.001) with decreased gyrification (−25.6%, p <0,0001) and increased mean curvature (+9.55%, p <0.05). PCA explained 84.4% of variance and identified three clusters: Cluster 1 (n = 17) with severe atrophy and curvature change, and Cluster 2 (n = 29) with milder alterations and Cluster 3 (n=10) with preserved morphology, likely corresponding to non-sclerotic hippocampi. Cluster membership corresponded closely with histopathological grading of sclerosis, providing greater diagnostic granularity than volumetry or asymmetry alone.
Conclusion: Though HipUnfold is validated for HS detection, our study demonstrates that unsupervised clustering of surface features captures HS phenotypes aligned with histopathological sclerotic severity, offering added interpretability for presurgical planning in TLE
Limitations: This pilot study had a modest sample size, which limits statistical power and cluster stability. Most surgical cases showed clear HS on imaging, introducing selection bias.. Larger prospective studies with broader anatomical coverage are needed for validation.
Funding for this study: This study received no external funding.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The study were approved by the ethics committee of the Hanoi Medical University (no 4339/QĐ-ĐHYHN)

Author Block: J. Yoo, R. Stepponat, F. Fischmeister, G. Kasprian; Vienna/AT
Purpose: Understanding the organization of language functions is one of the most rigorously researched neuroscientific topics, as it carries significant importance in clinical settings, such as preoperative planning in neurosurgery, and disease comprehension in neurological disorders.
In our combined DTI tractography and fMRI study, we aim to determine whether experts who work closely with the above-mentioned imaging methods can identify the correct language lateralization solely with DTI tractography in TLE patients as well as healthy controls.
Methods or Background: We performed a questionnaire study with experts who routinely work with DTI tractography as well as fMRI in patients. The experts were shown randomized, partially mirrored images of DTI tractographies of the arcuate and uncinate fasciculi and then attempted to determine the language lateralization based on the images. The results were further validated by calculating the inter-rater variability. Furthermore, fMRI- and DTI-laterality indices (LI) were correlated in order to validate our data.
Results or Findings: The exploratory preliminary results show a correlation of the arcuate fascicle structure to the fMRI LI as proven by previous studies. Furthermore, certain visual qualities of the arcuate fascicle such as a dense, highly organized fiber structure spanning between the temporal and frontal lobe showed to be a visual marker for language lateralization to the same side.
Conclusion: The preliminary results suggest that an evaluation by experts of language lateralization solely by DTI tractography is possible. This finding can help clinicians determine language lateralization or increase their confidence in identifying the dominant hemisphere when fMRI is unavailable or the patient is not able to perform fMRI.
Limitations: Due to the distribution of language laterality by fMRI, atypically lateralized patients are under-represented and further investigation with a larger pool of atypically lateralized patients may be needed.
Funding for this study: Medical University of Vienna
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The study has been performed under amendment of ethics application 1141/2023

Author Block: H. M. H. Sbaihat, B. Bellenberg, A. K. Roenneke, B. Krieger, D. Müller, C. Lukas; Bochum/DE
Purpose: Reliable biomarkers are needed in psychiatry and neurology to improve diagnosis and treatment. While the reliability of resting-state (RS) fMRI has been studied at standard field strengths, it remains underexplored at ultrahigh field (UHF) strengths such as 7 Tesla. This study evaluated the stability of spontaneous brain activity (RSA) and functional connectivity in subcortical gray matter (SCGM), regions central to cognition, emotion, and memory. We assessed RSA with the amplitude of low-frequency fluctuations (ALFF), local connectivity with regional homogeneity (ReHo), and global connectivity with degree centrality (DC).
Methods or Background: Sixteen healthy participants (mean age = 25.3 ± 2.0 years) were selected from a publicly available dataset (Gorgolewski et al., 2015). fMRI and T1-weighted images were acquired on a Siemens MAGNETOM 7T scanner in two sessions one week apart. Preprocessing was performed, and then ALFF, ReHo, and DC were computed in native space for each session. Test–retest reliability was assessed using Lin’s concordance correlation coefficient across SCGM regions, including the amygdala, basal ganglia, caudate, hippocampus, insula, pallidum, putamen, and thalamus.
Results or Findings: Our results indicate moderate-to-strong stability across all selected subcortical gray matter regions, except for the RSA in pallidum. The amygdala, insular cortex, and thalamus demonstrated strong reliability across ALFF, DC, and ReHo. The basal ganglia and caudate were moderately stable for ALFF and ReHo but strong for DC. The hippocampus and putamen showed moderate stability across all measures, while the pallidum exhibited weak ALFF stability but moderate DC and ReHo.
Conclusion: RS measures of SCGM exhibit generally moderate-to-strong stability. These findings support the potential of ultrahigh-field 7T fMRI for reliably assessing spontaneous brain activity and connectivity in subcortical regions, providing a foundation for future biomarker development in psychiatry and neurological disorders.
Limitations: Not applicable.
Funding for this study: No funding was provided for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: publicly available dataset (Gorgolewski and colleagues, 2015)

Author Block: S. Chambers, S. Glatter, L. Krepler, T. Dorittke, J. Binder, M. Weber, D. Prayer, G. Kasprian; Vienna/AT
Purpose: Predictions of epileptogenicity are often biased by inferring risk retrospectively from symptom onset. Fetal MRI allows identification of malformations of cortical development (MCD) before clinical manifestation. We present a preliminary analysis of the epileptogenicity of suspected MCD in prenatal MRI, validated through postnatal follow-up. Our aim is to stratify epilepsy risk by MCD subtype and location, providing less biased estimates to support the transition from reactive to preemptive, disease-modifying interventions.
Methods or Background: We retrospectively reviewed fetal MRIs performed at the Medical University of Vienna (2005–2024) with suspected MCD. Scans included T2-weighted sequences in three planes at 1.5/3T, with ≥1 year postnatal follow-up. Two neuroradiologists independently assessed primary MCD features (e.g., schizencephaly, polymicrogyria, dysplasia) and lesion location. Clinical outcomes included epilepsy and/or neurodevelopmental delay. Group comparisons used chi-squared/Fisher’s exact tests; stepwise logistic regression assessed multivariable effects.
Results or Findings: Forty-four cases (18 female) were included with mean follow-up of 6.3 years. Sixteen patients (36.4%) developed epilepsy and 19 (43.2%) neurodevelopmental delay. Microlissencephaly and dysgyria were strongly associated with epilepsy (both p=0.013), with frontal lobe involvement being the strongest predictor (OR=13, p<0.001), remaining significant in multivariable regression. Combining location to pathology, schizencephaly also reached significance (p=0.013), whereas periventricular nodular heterotopia and hippocampal malrotation showed nonsignificant associations.
Conclusion: Our exploratory, preliminary analysis highlights the epileptogenicity of microlissencephaly, dysgyria, and frontal lobe involvement, while schizencephaly and polymicrogyria may be less predictive in a univariate analysis than assumed. Our findings support the questionable epileptogenicity of periventricular nodular heterotopia and suggest possible two-hit mechanisms.
Limitations: The limitations are the small sample size and consequent difficulty of disentangling co-occurring features, a multi-center study is currently ongoing. Furthermore, our assessment is limited to early-onset epilepsy due to the mean follow-up period of 6 years.
Funding for this study: No funding was received for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The IRB number is EK 2167/2023

Author Block: A. Chauhan; Bangalore/IN
Purpose: Accurate localization of epileptogenic zones is critical for surgical success in drug-resistant epilepsy, yet conventional MRI and EEG often provide incomplete information. This study investigates an artificial intelligence (AI) framework that integrates MRI-based structural and functional connectomics with EEG data to enhance presurgical mapping.
Methods or Background: A cohort of 72 patients with drug-resistant focal epilepsy underwent 3T multiparametric MRI (structural MRI, diffusion tensor imaging, resting-state fMRI) and scalp EEG. Radiomics and graph-theoretical features were extracted from MRI, while functional connectivity metrics were derived from EEG. These multimodal datasets were integrated using a supervised machine learning pipeline (random forest and support vector classifier), trained on postsurgical outcome as ground truth.
Results or Findings: The AI-integrated model achieved an accuracy of 87% in localizing epileptogenic networks, outperforming MRI-only (72%) and EEG-only (68%) models. Notably, connectomic features capturing disrupted hub regions in the temporal and frontal lobes were strongly predictive of seizure recurrence risk. Surgical resections guided by AI-predicted zones correlated with improved seizure freedom rates (Engel Class I outcomes in 79% vs. 61% with standard planning).
Conclusion: AI-driven integration of MRI and EEG connectomics provides a robust tool for identifying epileptogenic zones, enhancing presurgical planning, and improving clinical outcomes in drug-resistant epilepsy.
Limitations: Single-center design and modest sample size may limit generalizability. Future work should validate the framework in multicenter datasets and explore real-time clinical deployment.
Funding for this study: Not Applicable
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: M. Stuempflen, P. Kienast, V. Schmidbauer, M. Weber, J. Tischer, T. Dorittke, J. Binder, D. Prayer, G. Kasprian; Vienna/AT
Purpose: Failure of fetal interneuron migration arising from the ganglionic eminence (GE) may lead to a wide spectrum of neuropsychiatric and neurodevelopmental disorders. Early detection of anomalies may improve the MRI phenotyping of neurodevelopmental diseases. This atlas-based fetal MRI study aimed to quantitatively assess development of and alterations in GE volume in fetuses with previously diagnosed abnormal GE volumes based on subjective assessment only.
Methods or Background: This retrospective study investigated 17 fetuses (20 fetal MRIs, mean gestational age 26.3 weeks, SD 3.3) with subjectively enlarged ganglionic eminence with concurrent structural central nervous system anomalies based on assessment of experienced fetal neuroimaging experts. Three-dimensional volumetry was performed based on super-resolution fetal MRI and compared to age-matched neurotypical controls (94 fetuses, 100 MRIs, mean gestational age 27.2 weeks, SD 3.6).
Results or Findings: Only 25% were found to include patients with GE hyperplasia, while 60% were found to show normal and 15% smaller GE volumes than healthy references. Most patients (80%) were found to have increased total brain volume - mostly due to high ventriculomegaly (75%). Brain parenchyma volume was enlarged in only 20%. No correlation was found between GE volumes and volumes of ten other substructures of the fetal head.
Conclusion: The study highlights the unreliability of expert visual assessment of GE size, despite excellent examination conditions and emphasizes the necessity of three-dimensional volumetry. Specifically in patients with structural CNS anomalies quantitative fetal neuroimaging will serve as an emerging tool for deep phenotyping, which identifies abnormalities within the GE prior to the structural emergence of cortical malformations.
Limitations: The findings of this study should be confirmed in larger patients cohorts and histological studies, which were unavailable due to the in vivo nature of this study.
Funding for this study: None.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Approval of ethics committee obtained (number EK 1585/2021).