Comprehensive aortic stenosis characterization using multi-view deep learning

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate aortic stenosis (AS) assessment requires integrating structural and functional information.
• Current artificial intelligence (AI) models often use only structural or functional data.

Purpose of the Study:

• To develop EchoNet-AS, an integrated AI approach for comprehensive AS severity assessment.
• To combine convolutional neural networks for valve motion analysis and segmentation models for Doppler measurements.

Main Methods:

• Developed EchoNet-AS, an open-source, end-to-end AI model.
• Utilized convolutional neural networks for video analysis and segmentation models for Doppler measurements.
• Trained on over 210,000 images and validated on multiple large, diverse cohorts.

Main Results:

• EchoNet-AS achieved high accuracy in classifying AS severity, with AUCs up to 0.989 across internal and external validation cohorts.
• The integrated approach outperformed models using single data types (videos or Doppler measurements).
• Demonstrated robust performance and generalization across multiple healthcare systems.

Conclusions:

• EchoNet-AS effectively synthesizes B-mode and Doppler echocardiographic data for accurate AS assessment.
• The AI model shows significant potential as an automated clinical decision support tool for AS.
• The approach demonstrated robust generalization to external validation cohorts.