ECG-based data-driven solutions for diagnosis and prognosis of cardiovascular diseases: A systematic review

Area of Science:

• Cardiology and Artificial Intelligence (AI)

Background:

• Cardiovascular diseases (CVD) are a major global health concern, with electrocardiograms (ECG) vital for diagnosis, yet interpretation faces challenges due to a shortage of skilled cardiologists.
• Machine learning (ML) and deep learning (DL) offer advanced computer-assisted solutions for ECG interpretation, but often lack explainability and may exhibit bias.
• Existing literature reviews inadequately address the crucial Trustworthy AI aspects (explainability, bias, ethical, legal, and societal implications - ELSI) in ML/DL models for ECG-based CVD diagnosis and prognosis.

Approach:

• This systematic review provides a holistic analysis of data-driven models for ECG-based CVD detection.
• The review examines various dimensions including CVD types, dataset characteristics, input modalities, ML/DL algorithms (with a focus on DL), and Trustworthy AI elements.
• Challenges within these dimensions are identified, and concrete recommendations are provided for researchers.

Key Points:

• Identifies trends in ML/DL applications for ECG analysis in CVD.
• Highlights the critical need for explainability, bias mitigation, and ethical considerations in AI for cardiology.
• Addresses the gap in comprehensive reviews focusing on Trustworthy AI in ECG-based CVD diagnostics.

Conclusions:

• This review offers a comprehensive understanding of the current landscape of ML/DL in ECG-based CVD diagnosis and prognosis.
• It emphasizes the importance of integrating Trustworthy AI principles to ensure reliable and ethical AI solutions in cardiology.
• Provides actionable insights and recommendations to guide future research and development in this critical area.