Predicting antiseizure medication treatment outcome using AI and Structural MRI

Motivation: Implementation of AI-driven precision medicine and finding the most effective Antiseizure Medications. Goal(s): To predict outcomes of drug interventions in epilepsy patients and categorize them into distinct seizure outcome groups. Approach: The research employs both patient characteristic, clinical, and MRI features going throguh a feature selection step followed by binary classification using Support Vector Machine, Naïve Bayes, Decision Tree, and Ridge Regression. Results: Ridge regression combined with genetic algorithm outperformed the others, achieving an accuracy of 0.77 and AUC (Area Under the Curve) of 0.80 in predicting seizure outcome. This success was attained using a total of 18 MRI features and 10 ASMs. Impact: Our model may help selection of the most effective ASM for individual patients. This may reduce the need for consecutive drug trials involving ineffective medications, thereby alleviating associated burdens.