High Throughput Deep Learning Detection of Mitral Regurgitation

Abstract

Background: Diagnosis of mitral regurgitation (MR) requires careful evaluation of echocardiography with Doppler imaging. This study presents the development and validation of a fully automated deep learning pipeline for identifying apical-4-chamber view videos with color Doppler and detection of clinically significant (moderate or severe) mitral regurgitation from transthoracic echocardiography studies. Methods: A total of 58,614 studies (2,587,538 videos) from Cedars-Sinai Medical Center (CSMC) were used to develop and test an automated pipeline to identify apical-4-chamber view videos with color Doppler across the mitral valve and then assess mitral valve regurgitation severity. The model was tested on an internal test set of 1,800 studies (80,833 videos) from CSMC and externally evaluated in a geographically distinct cohort of 915 studies (46,890 videos) from Stanford Healthcare (SHC). Results: In the held-out CSMC test set, the view classifier demonstrated an AUC of 0.998 (0.998 - 0.999) and correctly identified 3,452 of 3,539 MR color Doppler videos (sensitivity of 0.975 (0.968-0.982) and specificity of 0.999 (0.999-0.999) compared with manually curated videos). In the external test cohort from SHC, the view classifier correctly identified 1,051 of 1,055 MR color Doppler videos (sensitivity of 0.996 (0.990 ? 1.000) and specificity of 0.999 (0.999 ? 0.999) compared with manually curated videos). For evaluating clinically significant MR, in the CSMC test cohort, moderate-or-severe MR was detected with AUC of 0.916 (0.899 - 0.932) and severe MR was detected with an AUC of 0.934 (0.913 - 0.953). In the SHC test cohort, the model detected moderate-or-severe MR with an AUC of 0.951 (0.924 - 0.973) and severe MR with an AUC of 0.969 (0.946 - 0.987). Conclusions: In this study, we developed and validated an automated pipeline for identifying clinically significant MR from transthoracic echocardiography studies. Such an approach has potential for automated screening of MR and precision evaluation for surveillance.

Opportunistic Screening of Chronic Liver Disease with Deep Learning Enhanced Echocardiography

“…Comparison between disparate imaging modalities such as ultrasound and MRI will have inherent limitations due to different modality-specific accuracy. However, in totality this study suggests that the clinical utility of high throughput disease screening using AI is promising, particularly for early disease, and enhances the utility of pre-existing imaging data 11,12,30 . Further studies are warranted to establish the optimal clinical workflow for opportunistic liver disease screening among CVD patients and downstream treatment.…”

“…Artificial intelligence (AI) can identify diseases and characteristics that may not be readily observable by the human eye [9][10][11][12][13] , predict disease progression 14 , mortality 15 , and improve measurement accuracy of cardiac parameters [16][17][18][19] . Our study aims to develop and validate an AI computer vision approach to leverage echocardiographic images and videos to detect CLD.…”