Abstract Details

Title A Machine Learning Approach for Identifying Factors that Contribute to Seizure Freedom Following Temporal Lobectomy for Mesial Temporal Lobe Epilepsy

Topic Epilepsy/Clinical Neurophysiology (EEG)

Presentation(s) S7 - Epilepsy and Clinical Neurophysiology (EEG) 1 (4:30 PM-4:42 PM)

Poster/Presentation
Number 006

Objective This study utilizes a machine learning (ML) approach to analyze which clinical features, preoperative evaluations and pathologic outcomes are predictive of seizure freedom in patients undergoing temporal lobectomy for mesial temporal lobe epilepsy (MTLE).

Background Temporal lobectomy is a highly effective treatment for drug-resistant MTLE. However, studies have found that approximately 30% of patients who undergo this operation experience seizure recurrence within 2 years. While many preoperative features have been correlated with outcomes, no single feature has demonstrated a strong predictive value for achieving seizure freedom.

Design/Methods We retrospectively reviewed the medical records and surgical pathology of 43 patients with a postoperative follow-up of 2 years. Data extracted included preoperative MRI, PET and EEG results, seizure semiology, aura type, side of operation and surgical pathology results. Seizure outcomes for each patient were defined using Engel classification with seizure freedom defined as Engel Class I (n=19). Our ML approach was based on a support vector machine classifier.

Results The combination of various preoperative features and surgical pathology yielded a good prediction performance with area under the curve [AUC] = 0.75, sensitivity = 0.91 and specificity = 0.33. Features most predictive of seizure freedom at 2 years included right-sided operations (100.0), generalized tonic clonic seizure semiology (52.0), auras characterized by fear (42.2), deja vu (39.4) or gustatory sensations (36.0), mesial temporal sclerosis (MTS) pathology ILAE type 1 (33.9) or patients ultimately found to have no hippocampal sclerosis (34.6), and preoperative MRI (31.6) and PET (24.4) scans localizing to the temporal region.

Conclusions Our study uses a ML approach to determine which features are predictive of seizure freedom in patients undergoing temporal lobectomy for MTS. Future work with new patient datasets could be used to predict the preoperative likelihood of seizure freedom.

Authors/Disclosures
Sara E. Ratican, MD (UCSF) PRESENTER	Miss Ratican has received personal compensation in the range of $50,000-$99,999 for serving as a Consultant for Doximity, Inc. . Miss Ratican has stock in Doximity, Inc. .
Seo Ho Song (Geisel School of Medicine at Dartmouth)	Mr. Song has nothing to disclose.
George Zanazzi	No disclosure on file
	No disclosure on file

��ɫ��

��ɫ��