Medical Dialogue Generation via Intuitive-then-Analytical Differential Diagnosis (2401.06541v1)

Abstract: Medical dialogue systems have attracted growing research attention as they have the potential to provide rapid diagnoses, treatment plans, and health consultations. In medical dialogues, a proper diagnosis is crucial as it establishes the foundation for future consultations. Clinicians typically employ both intuitive and analytic reasoning to formulate a differential diagnosis. This reasoning process hypothesizes and verifies a variety of possible diseases and strives to generate a comprehensive and rigorous diagnosis. However, recent studies on medical dialogue generation have overlooked the significance of modeling a differential diagnosis, which hinders the practical application of these systems. To address the above issue, we propose a medical dialogue generation framework with the Intuitive-then-Analytic Differential Diagnosis (IADDx). Our method starts with a differential diagnosis via retrieval-based intuitive association and subsequently refines it through a graph-enhanced analytic procedure. The resulting differential diagnosis is then used to retrieve medical knowledge and guide response generation. Experimental results on two datasets validate the efficacy of our method. Besides, we demonstrate how our framework assists both clinicians and patients in understanding the diagnostic process, for instance, by producing intermediate results and graph-based diagnosis paths.

Introduction to Medical Dialogue Systems

Medical dialogue systems (MDS) are a rapidly growing field of research, aiming to enhance the capabilities of healthcare services. These systems are designed to provide diagnosis, treatment plans, and health advice through automated conversations. Accurate diagnosis plays a critical role in medical dialogues and lays the groundwork for future consultations and patient care.

The Challenge in Modeling Differential Diagnosis

In a clinical setting, physicians use a combination of intuition and analytical reasoning to formulate a differential diagnosis—a list of possible conditions to guide further investigation. Intuitive reasoning is fast and experiential, while analytic reasoning entails a systematic and methodical approach to refining diagnoses.

Past approaches in MDS have primarily leveraged pre-trained LLMs for dialogue generation, but these often lack a meticulously grounded diagnostic process. As a result, while responses may appear coherent, they often fail to offer an interpretable, diagnostic-based rationale, which is key for both clinician and patient acceptance.

Proposing an Intuitive-then-Analytic Framework

The solution presented in the discussed paper is a framework called Intuitive-then-Analytic Differential Diagnosis (IADDx), which aims to mimic the clinician's reasoning process. IADDx consists of two principal stages:

1. Intuitive Association: By examining patient conditions within the dialogue, a preliminary list of diseases is generated via similarity-based retrieval of past cases and disease documentation.
2. Analytic Refinement: A diagnosis-oriented graph that includes body systems, organs, diseases, and symptoms is created. Through enhanced entity embeddings and multi-disease classification, IADDx refines the initial list of diseases, yielding a more accurate and interpretable diagnosis.

This method translates to more precise response generation, as medical knowledge is retrieved based on a differential diagnosis, and guides the conversation flow.

Verified through Experimental Validation

Upon testing the proposed framework on two medical datasets, IADDx demonstrates a notable improvement over baseline models, especially in generating responses that are coherent and medically accurate. The paper observed marked enhancements in automatic evaluation metrics like B-1, B-2, and B-4, which are indicators of the response's quality and the model's language understanding.

Additionally, IADDx's potential is reaffirmed through a human evaluation where medical professionals rated generated responses higher in fluency, knowledge accuracy, and overall quality compared to established models.

Conclusion

IADDx stands as an innovative response to the challenge of effectively modeling differential diagnosis in MDS. It not only solidifies the importance of a structured diagnosis process in conversation generation but also underscores the usefulness of combining intuitive and analytical reasoning for more informative and reliable medical dialogue systems. Given its performance in experiments, IADDx is poised to make significant contributions to improving MDS and thereby enhancing virtual healthcare services.