Intelligent Questionnaire for Seizures: Developing an AI-Based Diagnostic Tool for Epilepsy in Low-Resource Areas

BACKGROUND

Epilepsy affects 50,000 children in Texas alone and is one of the most common neurologic disorders in childhood. Health outcomes in epilepsy depend on multiple factors, including access to subspecialty care. The diagnosis of epilepsy requires interpretation and integration of complex health care data. However, in rural areas, epilepsy specialists are scarce, leaving pediatricians and community physicians to diagnose and manage epilepsy. To mitigate delayed access and aid in diagnosing seizures in low-resource settings, several seizure questionnaires have been published. However, no questionnaire has a satisfactory sensitivity and specificity for different seizure types, likely because standard questionnaires require standardized answers and are unable to adapt questions based on the history like a physician would do in a clinical encounter.

PROJECT DESCRIPTION

Machine learning models can support many applications that require mapping raw data into applicable and interpretable insights that can assist in clinical decision-making.

We propose to build an interactive language interface based on a large-scale language model (LLM) that can support physicians in diagnosing epilepsy. We will start with small seed data annotated by experts and generate data using LLMs to train the system. Our system will converse with patients, asking clarification and follow-up questions based on patient history and their initial description of the seizure events.

GOAL

The goal of our project is an AI-based “intelligent” questionnaire as a seizure diagnostic tool by approximating the diagnostic algorithm of a trained epilepsy expert clinician. We plan to use this tool in the future as part of the screening process for our existing epilepsy outreach project to improve access to epilepsy care in rural areas of Texas.