CoD, Towards an Interpretable Medical Agent using Chain of Diagnosis
Junying Chen, Chi Gui, Anningzhe Gao, Ke Ji, Xidong Wang, Xiang Wan, Benyou Wang
2024-07-24

Summary
This paper introduces Chain-of-Diagnosis (CoD), a new method designed to improve how AI models diagnose medical conditions. It focuses on making the decision-making process of these models more understandable and transparent, similar to how a doctor thinks.
What's the problem?
While large language models (LLMs) have improved medical diagnosis, they often lack clarity in their reasoning. This makes it hard for doctors and patients to trust the AI's decisions because they can't see how the model arrived at its conclusions. Without clear explanations, it's difficult to ensure that the AI is making accurate and safe recommendations.
What's the solution?
The authors developed CoD, which organizes the diagnostic process into a series of steps that reflect a doctor's thought process. This method outputs a confidence level for each diagnosis, showing how sure the model is about its conclusions. By using this structured approach, CoD helps identify important symptoms and provides a clearer understanding of the diagnosis process. They created a model called DiagnosisGPT that can diagnose over 9,600 diseases and has been shown to perform better than other existing models on diagnostic tasks.
Why it matters?
This research is important because it enhances the reliability and trustworthiness of AI in healthcare. By making AI diagnostics more interpretable, doctors can better understand and verify the AI's recommendations, leading to safer and more effective patient care. This could significantly improve how medical professionals use technology in their practices.
Abstract
The field of medical diagnosis has undergone a significant transformation with the advent of large language models (LLMs), yet the challenges of interpretability within these models remain largely unaddressed. This study introduces Chain-of-Diagnosis (CoD) to enhance the interpretability of LLM-based medical diagnostics. CoD transforms the diagnostic process into a diagnostic chain that mirrors a physician's thought process, providing a transparent reasoning pathway. Additionally, CoD outputs the disease confidence distribution to ensure transparency in decision-making. This interpretability makes model diagnostics controllable and aids in identifying critical symptoms for inquiry through the entropy reduction of confidences. With CoD, we developed DiagnosisGPT, capable of diagnosing 9604 diseases. Experimental results demonstrate that DiagnosisGPT outperforms other LLMs on diagnostic benchmarks. Moreover, DiagnosisGPT provides interpretability while ensuring controllability in diagnostic rigor.