An Interpretable Data-Driven Medical Knowledge Discovery Pipeline Based on Artificial Intelligence.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Data Mining

Background:

• Knowledge discovery from electronic medical records (EMRs) is crucial in medical research.
• Automatic validation of discovered medical knowledge remains a significant challenge.
• Expert validation is often required, limiting scalability and efficiency.

Purpose of the Study:

• To develop a data-driven, closed-loop pipeline for medical knowledge discovery and automatic validation.
• To integrate interpretable machine learning and deep learning for enhanced medical knowledge extraction.
• To improve the performance of prognostic predictive models using discovered medical knowledge.

Main Methods:

• Proposed a pipeline comprising Data Generator, Medical Knowledge Mining, Medical Knowledge Evaluation, and Medical Knowledge Application.
• Utilized interpretable machine learning and deep learning techniques for knowledge discovery and validation.
• Incorporated medical expert participation throughout the pipeline to ensure medical effectiveness.

Main Results:

• The pipeline successfully discovered and automatically validated medical knowledge from EMRs.
• Incorporating discovered medical knowledge significantly improved a traditional heart failure prognostic model.
• A scale model developed from the discovered knowledge demonstrated good predictive performance.

Conclusions:

• The proposed closed-loop pipeline enables efficient and automated medical knowledge discovery and validation.
• This approach enhances the performance of clinical predictive models by integrating data-driven insights.
• The pipeline's design, with expert involvement, ensures the clinical relevance and effectiveness of discovered knowledge.