Machine learning with electrocardiograms: A call for guidelines and best practices for 'stress testing' algorithms.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Current hospital ECG interpretation relies on manually coded knowledge-based systems.
• Supervised machine learning (ML) approaches use labeled ECG data to develop automated classification rules.
• Deep neural networks are emerging as powerful ML algorithms for ECG analysis.

Purpose of the Study:

• To describe current computer-based ECG interpretation methods.
• To discuss new opportunities and challenges in ECG ML.
• To explore the potential of hybrid knowledge and data-driven algorithms.

Main Methods:

• Comparison of knowledge-engineered algorithms with supervised machine learning (ML) techniques.
• Discussion of deep neural networks for ECG classification.
• Analysis of the impact of open ECG datasets on research focus.

Main Results:

• Both knowledge engineering and ML have pros and cons for ECG interpretation.
• Hybrid algorithms combining knowledge and data-driven techniques present a promising avenue.
• Open ECG datasets can align ML research with clinical needs.

Conclusions:

• There is a need for standardized guidelines for evaluating ECG ML algorithms, including stress testing.
• Hybrid ECG ML approaches offer potential for enhanced diagnostic accuracy.
• Further research into open datasets and robust evaluation methods is crucial for advancing ECG analysis.