Author + information
- Wenhao Jiang,
- Kit Hang Lee,
- Zhiyu Liu,
- Yiting Fan,
- Ka-Wai Kwok and
- Alex Pui-Wai Lee
Accurate assessment of the left ventricular ejection fraction (LVEF) is crucial for making therapeutic decisions. The echocardiographic method for quantification of LVEF recommended by the American Society of Echocardiography is the biplane Simpson's method. However, this method requires manual tracing of the endocardial border in two apical orthogonal planes, which is time-consuming and subject to measurement variability. In this study, we aim to assess LVEF by utilizing deep learning (DL) algorithms.
Apical 4- and 2-chamber views were acquired from 102 patients. LVEF of these patients were calculated by the biplane Simpson's formula as assessment targets. The samples with LVEF >50% were labelled as abnormal. We partitioned the cohort into 2 sets (79.4% training, 20.6% validation). A DL model was trained to assess LVEF and detect abnormality. Its performance is then evaluated by the reserved validation set.
Disease prevalence of our cohort was 29.4%. The area under the curve for the classification result on validation set was 0.87, and the average analysis time was 10 ± 3 ms per subject. Sensitivity, specificity and accuracy were 88.9%, 73.3% and 77.8%, respectively.
DL model provides accurate and reproducible assessment of LVEF. Integrating this classification model into clinical workflow will permit timely interventions in diagnosis.
Poster Hall, Hall F
Sunday, March 17, 2019, 3:45 p.m.-4:30 p.m.
Session Title: Non Invasive Imaging: Echo 4
Abstract Category: 28. Non Invasive Imaging: Echo
Presentation Number: 1271-327
- 2019 American College of Cardiology Foundation