Diagnostic Accuracy of AI Algorithms in Aortic Stenosis Screening: A Systematic Review and Meta-Analysis

Area of Science:

• Cardiology
• Medical Artificial Intelligence
• Diagnostic Accuracy

Background:

• Aortic stenosis (AS) is often diagnosed late, after symptoms manifest.
• Early detection of AS is crucial for timely intervention and improved patient outcomes.
• The study addresses the need for accurate and accessible AS screening methods.

Purpose of the Study:

• To systematically review and meta-analyze the diagnostic accuracy of artificial intelligence (AI) algorithms for screening aortic stenosis (AS).
• To evaluate the performance of AI in detecting AS across diverse data modalities.
• To assess the feasibility and scalability of AI-driven AS screening.

Main Methods:

• Conducted a systematic search across six databases to identify relevant studies.
• Performed a diagnostic meta-analysis using parameters like sensitivity, specificity, and AUC for AI algorithms.
• Included AI algorithms utilizing electrocardiograms (ECG), chest radiographs, auscultation audio, electronic stethoscopes, and wearable sensor data.

Main Results:

• Pooled estimates for AI algorithms showed high sensitivity (0.83) and specificity (0.81) for AS diagnosis.
• The area under the curve (AUC) was 0.909, indicating outstanding diagnostic accuracy.
• Heterogeneity was observed across subgroups based on continent, AS type, data source, and AI method; publication bias was noted.

Conclusions:

• Deep learning approaches demonstrate high sensitivity, feasibility, and scalability for identifying patients with moderate or severe AS.
• AI-powered screening holds significant potential for early and accurate detection of aortic stenosis.
• Further research may refine AI algorithms and address identified sources of heterogeneity and bias.