Predicting Epileptogenic Tubers in Patients with Tuberous Sclerosis Complex Using a Fusion Model Integrating Lesion Network Mapping and Machine Learning

Accurate localization of epileptogenic tubers (ETs) in patients with tuberous sclerosis complex (TSC) is essential but challenging because ETs lack distinct pathological or genetic markers that differentiate them from other cortical tubers. Approximately 60% of patients fail to have their ETs identified through noninvasive preoperative evaluations, highlighting the clinical need for an efficient, noninvasive ET localization method.

Using MRI data from training datasets, we developed a novel noninvasive fusion model that combines a risk model based on lesion network mapping with a prediction model using brain functional connectivity and random forest algorithms. A retrospective analysis was conducted on TSC patients with epilepsy who underwent resective surgery. Tubers were classified as true-ETs, false-ETs, or true non-ETs based on resection locations and postoperative seizure-freedom. The model calculated and ranked the ET probability for each tuber for every patient, and its accuracy was assessed based on postoperative seizure outcomes.