FURI | Summer 2025

Agentic AI Framework for Pediatric Seizure Onset Zone Detection

Identifying seizure onset zones (SOZ) in pediatric epilepsy is the key to planning surgeries, but it’s often time-consuming and limited by the need for an expert’s input or large datasets. This project aims to build an agentic AI system that bridges expert knowledge with advanced language models to classify SOZs using very little data. It will specifically use expert-informed deep learning (e.g., DeepXSOZ) alongside zero-shot LLM methods (e.g., CuKPL), where the AI generates and uses smart text-based prompts to reason about medical images. The system will also include a feedback loop from doctors to help it improve over time. This approach offers an efficient way to support clinical decisions while making the system explainable and adaptable across patients and hospitals.