Towards knowledge-infused automated disease diagnosis assistant.

Area of Science:

• Artificial Intelligence in Medicine
• Computational Linguistics
• Medical Informatics

Background:

• The increasing reliance on digital health platforms necessitates advanced tools for accurate patient diagnosis.
• Traditional diagnostic methods face challenges with the growing complexity of diseases and symptoms.
• Integrating medical knowledge and conversational context is crucial for effective automated diagnosis.

Purpose of the Study:

• To develop a novel disease diagnosis model that leverages patient-doctor interactions and medical knowledge.
• To investigate the impact of incorporating medical knowledge graphs and conversational data in diagnostic systems.
• To create and utilize an empathetic conversational medical corpus for training and evaluation.

Main Methods:

• Proposed a two-channel, knowledge-infused, discourse-aware disease diagnosis model (KI-DDI).
• Employed a transformer-based encoder for patient-doctor communication and a graph attention network (GAT) for symptom-disease embeddings.
• Integrated conversational and knowledge graph embeddings into a deep neural network for disease identification.

Main Results:

• The KI-DDI model significantly outperformed existing state-of-the-art models in disease identification.
• Demonstrated the critical role of doctors' additional symptom extraction beyond patient self-reporting.
• Validated the effectiveness of infusing medical knowledge into the diagnostic process.

Conclusions:

• Automated disease diagnosis can be significantly enhanced by incorporating medical knowledge and analyzing doctor-patient discourse.
• Future work should explore integrating visual sensory information for more comprehensive diagnostic assistance.
• The developed model and corpus provide a strong foundation for advancing AI-driven medical diagnosis.