Medical-informed machine learning: integrating prior knowledge into medical decision systems.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Clinical Data Science

Background:

• Machine Learning (ML) shows promise in clinical medicine, but few models impact patient care.
• ML alone often fails to capture the complexity of clinical data.
• Integrating medical domain knowledge is crucial for effective ML in healthcare.

Purpose of the Study:

• To review strategies for integrating medical knowledge into ML pipelines.
• To analyze the impact of this integration on ML model performance.
• To provide guidance for future integration efforts in clinical ML.

Main Methods:

• Comprehensive review of existing medical knowledge integration techniques in ML.
• Mapping integration strategies across the ML pipeline (pre-processing to evaluation).
• Case study on diabetes prediction demonstrating knowledge integration at each ML stage.

Main Results:

• Integration enhances ML model accuracy, interpretability, data efficiency, and clinical guideline adherence.
• Integrated models outperform purely data-driven approaches, improving generalization.
• Knowledge integration is effective in limited data scenarios and ensures model conformity with clinical standards.

Conclusions:

• Integrating medical domain knowledge significantly benefits ML applications in clinical settings.
• Challenges include refining domain knowledge representation and its contribution to ML models.
• Further research is needed to optimize knowledge integration for enhanced clinical ML.