Inside the "black box": Embedding clinical knowledge in data-driven machine learning for heart disease diagnosis

Area of Science:

• Cardiology
• Artificial Intelligence
• Machine Learning

Background:

• Ischemic heart disease (IHD) is a leading cause of death globally.
• Current diagnostic methods for IHD are complex, expensive, and invasive.

Purpose of the Study:

• To develop a novel clinical knowledge-enhanced machine learning (ML) pipeline for predicting IHD.
• To improve the timeliness and cost-effectiveness of IHD prediction.

Main Methods:

• Developed a machine learning pipeline integrating clinical expertise into data analysis, feature selection, and model evaluation.
• Employed one-hot feature encoding to reveal data bias and identify key features.
• Utilized a support vector machine (SVM) classifier.

Main Results:

• The clinical knowledge-enhanced ML pipeline outperformed state-of-the-art data-driven ML models on the Cleveland IHD dataset.
• Achieved 94.4% accuracy and 95% sensitivity using only 7 features.
• Demonstrated the effectiveness of clinical input in enhancing ML model performance.

Conclusions:

• Incorporating clinical expertise into ML models improves performance, interpretability, and trustworthiness for AI in medicine.
• The proposed pipeline offers a more transparent and effective approach to IHD prediction.
• Addresses practical challenges like data bias and model interpretability in clinical AI.