FIT-graph: A multi-grained evolutionary graph based framework for disease diagnosis.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Decision Support Systems

Background:

• Machine learning, particularly knowledge graphs, aids early disease diagnosis and treatment optimization.
• Current knowledge graph methods struggle to fully utilize multi-granularity and temporal information in medical records.
• This limitation restricts the quality and comprehensiveness of machine learning-driven diagnoses.

Purpose of the Study:

• To propose a novel disease diagnosis framework, FIT-Graph, that addresses limitations in current knowledge graph approaches.
• To enhance the organization and utilization of multi-grained and temporal information from medical records.
• To improve the accuracy and comprehensiveness of disease inference for clinical applications.

Main Methods:

• Developed FIT-Graph, a novel disease diagnosis framework utilizing medical multi-grained evolutionary graphs.
• Efficiently organized extracted information across various granularities and time stages.
• Maximized retention of valuable information for disease inference and ensured comprehensiveness and validity.

Main Results:

• FIT-Graph demonstrated superior performance compared to baseline models on two real-world clinical datasets (cardiology and respiratory).
• The framework improved baseline performance by approximately 5% across multiple evaluation indices.
• Experimental results validate the effectiveness of FIT-Graph in disease diagnosis applications.

Conclusions:

• FIT-Graph effectively organizes and leverages multi-grained and temporal information from medical records for enhanced disease diagnosis.
• The proposed framework offers a significant advancement over existing knowledge graph-based methods in clinical settings.
• FIT-Graph has the potential to optimize the diagnosis and treatment process, leading to improved patient outcomes.