Evaluating artificial intelligence-enabled medical tests in cardiology: Best practice.

Area of Science:

• Cardiovascular Research
• Biomedical Data Science
• Cardiac Electrophysiology

Background:

• Machine learning (ML) is increasingly applied in cardiovascular research.
• Cardiac electrophysiology presents unique challenges due to large, imbalanced datasets.
• Evaluating ML studies in this field requires specific guidelines.

Purpose of the Study:

• To highlight opportunities and challenges in evaluating ML studies in cardiovascular research.
• To provide guidance on best practices for assessing ML applications in cardiac electrophysiology.
• To establish a framework for the quality evaluation of ML-based medical tests.

Main Methods:

• Utilized examples from cardiac electrophysiology research.
• Focused on supervised machine learning study evaluation.
• Developed recommendations for reporting and presentation of ML studies.

Main Results:

• Recommended proper cohort selection and strict separation of training/testing data.
• Advocated for comparison against non-ML reference models.
• Suggested specific metrics and plots for reporting ML model performance, especially for time series and image data.

Conclusions:

• Adherence to recommended principles ensures the quality of ML studies.
• The proposed best practices serve as a blueprint for evaluating ML in cardiac electrophysiology and related fields.
• Standardized evaluation methods will advance the reliable application of ML in cardiovascular medicine.