Proof-of-Concept Automated Framework for Intraoperative Transesophageal Echocardiography: View Classification and Biventricular Function Assessment

术中经食道超声心动图自动化框架的概念验证:视图分类和双心室功能评估

阅读:1

Abstract

OBJECTIVE: To develop a proof-of-concept automated framework for integrated intraoperative transesophageal echocardiography (TEE) interpretation, encompassing view classification, left ventricular ejection fraction (LVEF) and right ventricular systolic function (RVSF) assessment, and tricuspid regurgitation (TR) grading. DESIGN: This cross-sectional study used TEE video clips and reports from patients undergoing cardiac surgery from 2018 to 2023. SETTING: Two hospitals within the University of Pennsylvania Healthcare System. PARTICIPANTS: Study participants were ≥18 years of age and undergoing any cardiac surgery in which an intraoperative TEE was performed. We used 6,900 intraoperative TEE studies from 6,016 unique patients. INTERVENTIONS: Receipt of intraoperative TEE examination during cardiac surgery. MEASUREMENTS AND MAIN RESULTS: The view classification model was trained on TEE clips from 1,432 studies, and the diagnostic model was trained on clips from 6,400 studies, for a total of more than 700,000 individual TEE clips for training. Performance was evaluated on 945 test clips from 8 studies for view classification and 500 studies for diagnostic prediction. No external testing was performed. The view classifier achieved 86% accuracy compared to the expert agreement top-1 weighted accuracy of 74%. The diagnostic model differentiated LVEF >55% versus ≤55% (area under the receiver operating characteristic curve [AUROC] = 0.86) and LVEF >30% versus ≤30% (AUROC = 0.95). Continuous LVEF prediction was assessed (Pearson r = 0.79, mean absolute error = 7.8). The model differentiated normal versus abnormal RVSF (AUROC = 0.82) and differentiated less-than-moderate versus moderate-or-greater right ventricular dysfunction (AUROC = 0.92). Additionally, the model differentiated no TR versus any TR (AUROC = 0.69) and differentiated less-than-moderate versus moderate-or-greater TR (AUROC = 0.79). CONCLUSIONS: This proof-of-concept automated approach showed encouraging results for view classification and quantification tasks, suggesting potential for further development toward intraoperative utility.

特别声明

1、本页面内容包含部分的内容是基于公开信息的合理引用;引用内容仅为补充信息,不代表本站立场。

2、若认为本页面引用内容涉及侵权,请及时与本站联系,我们将第一时间处理。

3、其他媒体/个人如需使用本页面原创内容,需注明“来源:[生知库]”并获得授权;使用引用内容的,需自行联系原作者获得许可。

4、投稿及合作请联系:info@biocloudy.com。