Abstracts

Improving Surgical Decision-Making in Drug-Resistant Mesial Temporal Lobe Epilepsy with Multimodal Phenotyping

Abstract number : 3.187
Submission category : 2. Translational Research / 2D. Models
Year : 2025
Submission ID : 907
Source : www.aesnet.org
Presentation date : 12/8/2025 12:00:00 AM
Published date :

Authors :
Presenting Author: John Thomas, PhD – Duke University

Chifaou Abdallah, MD – McGill University
Thandar Aung, MD,MS – University of Pittsburgh Medical School
Irena Doležalová, MD, PhD – Brno Epilepsy Center, First Department of Neurology, St. Anne’s University Hospital, Faculty of Medicine, Masaryk University, Brno, Czech Republic
Alyssa Ho, M.Sc. – Duke University
Kassem Jaber, M.Sc. – Duke University
Erica Minato, MS – McGill University
Olivier Aron, MD – Department of Neurology, University Hospital of Nancy, Lorraine University, F-54000 Nancy, France
Stephan Chabardes, MD, PhD – CHU Grenoble-Alpes, Université Grenoble Alpes, Inserm, U1216, Grenoble Institute Neurosciences, Grenoble, France
Sophie Colnat-Coulbois, MD, PhD – CHU Grenoble-Alpes, Université Grenoble Alpes, Inserm, U1216, Grenoble Institute Neurosciences, Grenoble, France
Jeffery Hall, MD – McGill University
Petr Klimes, PhD – Brno Epilepsy Center, First Department of Neurology, St. Anne’s University Hospital, Faculty of Medicine, Masaryk University, Brno, Czech Republic
Lorella Minotti, MD – CHU Grenoble-Alpes, Université Grenoble Alpes, Inserm, U1216, Grenoble Institute Neurosciences, Grenoble, France
Milan Brazdil, MD, PhD – Brno Epilepsy Center, First Department of Neurology, St. Anne’s University Hospital, Faculty of Medicine, Masaryk University, Brno, Czech Republic
François Dubeau, MD – McGill University
Jorge Gonzalez-Martinez, MD, PhD – Department of Neurosurgery, University of Pittsburgh Medical Center, Pittsburgh, PA, USA
Philippe Kahane, MD, PhD – CHU Grenoble-Alpes, Université Grenoble Alpes, Inserm, U1216, Grenoble Institute Neurosciences, Grenoble, France
Louis Maillard, MD, PhD – Department of Neurology, University Hospital of Nancy, Lorraine University, F-54000 Nancy, France
Prachi Parikh, MD – Duke University
Derek Southwell, MD PhD – Duke University
David Carlson, PhD – Duke University
Jean Gotman, PhD – McGill University
Birgit Frauscher, M.D. Ph.D. – Duke University

Rationale: Mesial temporal lobe epilepsy (MTLE) is often regarded as the most uniform subtype of focal epilepsy, with favorable surgical outcomes in approximately 63% of patients (range: 59–67% [1]). However, this success rate has plateaued, possibly due to the limited number of stereo-electroencephalography (SEEG) cases and the lack of variability in patient characteristics, which impede the ability of individual centers to learn from experience during pre-surgical evaluations [2]. Multimodal data phenotyping presents a promising approach to identify patient subgroups more likely to benefit from surgery, enabling the discovery of clinical and electrophysiological phenotypes associated with favorable outcomes.

Methods: We analyzed 48 patients from five centers with unilateral MTLE who underwent standard temporal lobe resection; 21 (43.8%) of them did not achieve seizure freedom. Ten features were extracted, five qualitative (from clinical text data) and five quantitative (from raw SEEG signals), from three modalities, namely, clinical meta-data, seizure semiology, and electrophysiology, all routinely used in pre-surgical evaluations. We employed a two-step evaluation using leave-one-patient-out cross-validation to assess whether a model combining these features could distinguish seizure-free (Engel Ia) from non-seizure-free outcomes (Engel Ib-IV). A decision tree-based classifier (XGBoost) was used as the classification model.

Results: The model significantly distinguished between seizure-free and non-seizure-free outcomes (Wilcoxon rank-sum test, p = 0.02; Cliff’s delta = 0.38; Figure 1a), suggesting improved predictive capability in comparison with current clinical practice. Surgical decisions were made under the assumption that all patients will achieve seizure-freedom, resulting in a balanced accuracy of 50% (100% sensitivity, 0% specificity). Models incorporating different feature combinations achieved stepwise improvements in balanced accuracy: 1.7% with clinical meta-data alone, 11.2% with meta-data and semiology, and 16% with meta-data, semiology, and electrophysiology. The performance comparison between the current clinical standard and the models was also analyzed in terms of specificity, i.e., how many non-seizure-free outcomes can be predicted a priori, for a fixed sensitivity of 70%. The optimal model was able to recommend against surgery in an additional 32% of patients (Figure 1b).

Conclusions: This study demonstrates that combining multimodal clinical and electrophysiological data using a decision tree model significantly improves the ability to predict surgical outcomes compared to the current clinical standard. Phenotyping may therefore serve as a valuable tool to identify patients most likely to benefit from surgery and to avoid ineffective interventions, ultimately supporting more-informed, personalized, and data-driven clinical decision-making.

Funding:

References: [1] Avigdor et al. medRxiv, 2025 [2] Thomas et al. Journal of neural engineering, 2025.

Funding: CIHR (PJT-175056), AES Infrastructure grant (#1319635, 2024-2025)



Translational Research