Applications of Large Language Models (LLMs) in Healthcare Focus on Stroke Prediction and Management

Abstract

Stroke is a major cause of death and long-term disability in the world, with severe impacts in low- and middle-income countries like Algeria where quick diagnosis and treatment are often unavailable. Artificial Intelligence, especially Large Language Models (LLMs), offers new hope for improving stroke care. This thesis introduces AVCIntelliCare, a smart system to predict, detect, and respond to strokes quickly. First, we collected stroke-related data and transformed it into text for analysis. Next, we trained and fine-tuned a biomedical LLM (BioMedNLP-BiomedBERT , biobert ) to better understand stroke symptoms, Reaching optimal performance . We chose theFAST-ED test over other methods for its fast and reliable stroke detection. Then, we built AVCIntelliCare, combining a mobile app (AVCGuard) for stroke prediction using our fine-tuned LLM model( BioMedNLPBiomedBERT) and stroke detection with FAST-ED, and a central platform (AVCAlert) for emergency alerts. For high-risk cases , the system asks for a phone number and automatically sends alerts with the user’s GPS location and time to medical services. Designed with Arabic and English support, AVCIntelliCare is accessible to Algerian users. Despite challenges like data privacy, model overfitting, and limited resources, this system provides a scalable, easy-to-use solution to reduce stroke delays and improve outcomes. Future work includes connecting to hospitals, adding local dialects, and testing with doctors