Enhancing the Interpretability of Malaria and Typhoid Diagnosis with Explainable AI and Large Language Models

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Diagnostic Systems

Background:

• Malaria and Typhoid fever are significant health challenges in tropical regions, often complicated by drug resistance and environmental factors.
• Accurate diagnosis is critical for effective treatment and reducing mortality, but traditional methods struggle with symptom overlap.
• Existing machine learning (ML) models provide accurate predictions but lack transparency, hindering clinical adoption due to their "black box" nature.

Purpose of the Study:

• To enhance the interpretability and transparency of AI-driven diagnostic systems for malaria and typhoid fever.
• To build trust among healthcare providers by clarifying the decision-making process of diagnostic ML models.
• To integrate explainable AI (XAI) techniques with ML models for improved medical diagnostics.

Main Methods:

• Implementation of explainable AI (XAI) models, including Local Interpretable Model-agnostic Explanations (LIME) and Large Language Models (LLMs) like GPT.
• Utilized Random Forest (RF) as the primary predictive model, with LIME for feature importance and ChatGPT 3.5 for explanation generation.
• Developed a mobile application integrating RF, LIME, and GPT for a transparent malaria and typhoid diagnosis system.

Main Results:

• The Random Forest model demonstrated superior performance compared to other tested models.
• LIME effectively identified key diagnostic features, enhancing model interpretability.
• ChatGPT 3.5 showed a comparative advantage over other LLMs in providing clear explanations.

Conclusions:

• The integration of RF, LIME, and GPT creates a more transparent and interpretable AI diagnostic system for malaria and typhoid.
• Dataset quality and computational demands for real-time deployment are key limitations.
• AI-driven diagnostics, enhanced by XAI, hold significant potential for improving healthcare in resource-limited settings.