Artificial Intelligence Applied to Electrocardiogram-Based Cardiovascular Risk Assessment: A Systematic Review

Area of Science:

• Cardiology
• Medical Informatics
• Artificial Intelligence

Background:

• Cardiovascular diseases (CVDs) are a leading global cause of death.
• Electrocardiograms (ECGs) are common but have subjective interpretation limitations.
• Artificial intelligence (AI) offers potential for enhanced prognostic information extraction from ECGs.

Purpose of the Study:

• To systematically review studies using AI techniques on ECGs for cardiovascular risk prediction and mortality.
• To assess the performance of AI models solely using ECG signals for cardiovascular outcomes.
• To identify current trends and challenges in AI-ECG research for risk stratification.

Main Methods:

• Systematic literature search across multiple databases.
• Inclusion of original studies using ECGs as the sole input for AI models.
• Narrative synthesis of data from eleven included retrospective cohort studies.

Main Results:

• Eleven studies, mostly retrospective cohorts from high-income countries, utilized Convolutional Neural Networks (CNNs).
• AI-ECG models predicted all-cause mortality, cardiovascular death, and major adverse cardiovascular events (MACE).
• Reported Area Under the Receiver Operating Characteristic (AUROC) values ranged from 0.63 to 0.961, with some models exceeding traditional risk scores.

Conclusions:

• AI-ECG models demonstrate potential for early detection of subclinical cardiovascular disease and improved risk stratification.
• AI can identify prognostic information even from normal ECGs, advancing personalized cardiovascular risk assessment.
• Further research is crucial for diverse populations, model interpretability, prospective validation, and equitable clinical integration.