Machine learning for cardiology.

Area of Science:

• Cardiology
• Artificial Intelligence
• Machine Learning

Background:

• Physicians are increasingly using artificial intelligence (AI) tools, specifically machine learning (ML) techniques, in cardiology.
• This review focuses on ML-based risk scores commonly used in cardiovascular research, examining their assumptions, biases, merits, and shortcomings.

Purpose of the Study:

• To review current applications of AI and ML in cardiology.
• To analyze the adoption and effectiveness of ML-based risk scores in cardiovascular research.
• To identify underutilized ML techniques and future research directions.

Main Methods:

• Review of recent cardiology literature on AI and ML applications.
• Detailed introduction to principal ML-based risk scores, explaining their function without requiring deep technical expertise.
• Comparative analysis of ML approaches with traditional statistical methods.

Main Results:

• ML-based risk scores are increasingly adopted in cardiology, though their frequency of use varies.
• Neural networks are gradually being integrated, but techniques like semi-supervised and federated learning are underutilized.
• Underutilization may stem from a lack of awareness or perceived complexity, despite their potential benefits.

Conclusions:

• Significant potential exists for advanced ML techniques in cardiology, particularly semi-supervised and federated learning.
• Semi-supervised learning can leverage partially labeled datasets, while federated learning enables collaborative model building across institutions.
• Future research should explore and promote the adoption of these underutilized ML methods to advance cardiovascular research and practice.