Deep learning-assisted aortic stenosis detection and grading based on multiview versus single-view echocardiography.

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Deep learning (DL) shows promise for automating echocardiogram interpretation, enhancing clinical accuracy and efficiency.
• A fully automated pipeline for aortic stenosis (AS) analysis using DL is largely unexplored.
• This study addresses the need for streamlined clinical assessment of AS.

Purpose of the Study:

• To develop a deep learning (DL) framework for automated aortic stenosis (AS) analysis in echocardiography.
• To streamline clinical AS assessment through an automated pipeline.
• To evaluate the performance of the DL framework in classifying views, detecting AS, and assessing its severity.

Main Methods:

• Utilized 499 AS studies (1,996 views) from 17,436 VHD cases for training, validation, and internal testing.
• Employed a prospectively collected dataset of 3,278 echocardiograms as a real-world test set.
• Developed a DL framework for automated view classification, AS detection, and severity assessment using multiview and single-view algorithms.

Main Results:

• The DL model achieved high performance in AS detection (AUC=0.942) in the prospective test dataset.
• Excellent correlation was observed between DL-graded metrics and manual measurements for key AS parameters (e.g., AV peak velocity r=0.94, mean peak gradient r=0.91).
• The DL model demonstrated superior performance in identifying severe AS (AUC=0.976) compared to moderate (AUC=0.907) and mild AS (AUC=0.874).

Conclusions:

• The proposed DL algorithm demonstrates significant potential for automating and enhancing the efficiency of clinical workflows for AS screening and grading.
• The framework offers a promising tool for objective and reproducible AS assessment in echocardiography.
• This automated approach can aid clinicians in faster and more accurate diagnosis and management of AS.