AI-aided general clinical diagnoses verified by third-parties with dynamic uncertain causality graph extended to also...

Area of Science:

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

Background:

• Artificial intelligence (AI)-aided general clinical diagnosis shows promise for primary care physicians.
• Existing machine learning methods face challenges with generalization and interpretability in clinical settings.
• Dynamic Uncertain Causality Graphs (DUCGs) leverage expert knowledge to overcome these limitations.

Purpose of the Study:

• To extend the Dynamic Uncertain Causality Graph (DUCG) framework.
• To incorporate representation and inference algorithms for non-causal classification relationships.
• To develop and validate DUCG-based knowledge bases for common chief complaints in general clinical diagnosis.

Main Methods:

• Constructed six distinct knowledge bases for specific chief complaints (arthralgia, dyspnea, cough, epistaxis, fever with rash, abdominal pain).
• Developed subgraphs representing diseases and causalities related to each chief complaint, irrespective of hospital department.
• Synthesized subgraphs into comprehensive chief complaint knowledge bases and validated using independent third-party hospital data.

Main Results:

• Achieved high total diagnostic precisions for the six knowledge bases, ranging from 96.5% to 100%.
• Ensured a minimum diagnostic precision of 80% for every individual disease within the knowledge bases.
• Demonstrated the effectiveness of the extended DUCG model in real-world clinical diagnostic scenarios.

Conclusions:

• The extended DUCG framework effectively addresses limitations of traditional machine learning in clinical diagnosis.
• DUCGs provide a robust and interpretable approach for AI-aided general clinical diagnosis.
• This methodology offers a reliable tool for enhancing diagnostic accuracy in primary care settings.