Publication Details

Abstract

Despite the growth of digital health, existing symptom checkers often fail to provide high diagnostic accuracy or actionable pathways for patient care. This thesis introduces DocAI, an integrated platform designed to bridge the "reliability gap" and the "navigation gap" in preliminary medical diagnostics. The framework employs a dual-engine inference pipeline that integrates a heuristic symptom-matching engine with a Random Forest machine learning classifier. By grounding these engines in a medical knowledge graph and utilizing advanced Document AI models (e.g., LayoutLMv3 and DONUT) for medical record extraction, DocAI provides a robust diagnostic tool that connects patients directly to clinical specialists. Evaluations using 45 standardized clinical vignettes demonstrate that the combined Dual-Engine approach achieved approximately 75% accuracy, significantly outperforming the Heuristic Engine (60%) or ML Engine (68%) when used in isolation.