Prospective Human Validation of Artificial Intelligence Interventions in Cardiology: A Scoping Review

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Informatics

Background:

• Integration of artificial intelligence (AI) into clinical practice is hindered by limited evidence of its effectiveness compared to human experts.
• Real-world validation of AI in cardiovascular care against gold standards is scarce.

Purpose of the Study:

• Identify AI interventions in cardiology prospectively validated against human experts or gold standards.
• Assess the effectiveness of these AI interventions.
• Determine future research directions for AI in cardiology.

Main Methods:

• Systematic review of Scopus and MEDLINE databases.
• Inclusion of peer-reviewed publications from January 2015 to December 2023.
• Focus on studies with prospective human validation of AI in cardiology.

Main Results:

• 64 studies were included, with 92.2% published after 2020.
• AI interventions reported improvements over human experts in 68.75% of studies.
• Common applications included imaging, ejection fraction, arrhythmia, and coronary artery disease, primarily using convolutional neural networks and image data.

Conclusions:

• AI demonstrates potential in cardiology, frequently matching human expert performance for defined tasks.
• A limited number of randomized controlled trials highlight the need for continued validation.
• Further research should focus on real-world settings and joint human-AI decision-making.