Automated Detection of Aortic Stenosis Using Machine Learning.

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Aortic stenosis (AS) is a prevalent yet often underdiagnosed valvular heart disease.
• Limited two-dimensional echocardiography is frequently used, but automated detection methods are needed for improved screening.
• Early detection and treatment of AS are crucial for better patient outcomes.

Purpose of the Study:

• To develop and validate methods for automated detection of aortic stenosis (AS) using limited two-dimensional echocardiographic data.
• To assess the performance of convolutional neural networks in identifying and grading AS severity.
• To establish a foundation for a novel, AI-driven screening tool for AS.

Main Methods:

• Convolutional neural networks were trained, validated, and tested on 2D transthoracic echocardiographic datasets.
• The networks performed two sequential tasks: echocardiographic view identification and study-level AS grading.
• Performance was evaluated using balanced accuracy and area under the receiver operator curve (AUROC).

Main Results:

• Fully automated AS screening achieved an AUROC of 0.96.
• The AI distinguished significant AS from no/mild AS with an AUROC of 0.86.
• External validation on over 8,500 echocardiograms confirmed high performance (AUROC 0.91) using limited views.

Conclusions:

• Modern neural networks enable fully automated detection of AS from limited 2D echocardiographic data.
• These findings support the development of a novel, AI-powered screening method for aortic stenosis.
• Automated detection has the potential to improve AS diagnosis rates and timely treatment initiation.