OptiHealth: An AI-Driven Health Management System with Multi-Modal Analysis and Personalized Recommendations

Abstract

This thesis presents three interconnected AI systems designed to improve medical image analysis and patient support. First, CXR-Adaptive enhances chest X-ray disease detection using YOLOv12 with adaptive augmentation techniques, achieving a 6.5% improvement over the baseline model. Second, SS-UNETR introduces a transformer-based architecture for brain tumor segmentation, outperforming the Swin UNETR baseline on the BraTS 2023 dataset. Third, OptiHealth combines medical vision models with LLMs and Retrieval-Augmented Generation (RAG) to provide personalized diet recommendations, specialist referrals, and conversational healthcare assistance through multiple integrated clinical tools.

Overall, the study demonstrates how AI can move beyond image interpretation toward actionable healthcare guidance. While the systems achieved strong benchmark performance and positive expert evaluations, future work should focus on cross-institution validation, larger clinical studies, and real-world healthcare integration.