Proposing Multiple Methods for End-Systole and End-Diastole Detection from Echocardiography Videos
Year: 2026
Type: conference-paper
Abstract: Accurate identification of end-systolic (ES) and end-diastolic (ED) frames is a crucial step in assessing cardiac function from 2D echocardiography, including ventricular volume and ejection fraction. In recent years, numerous studies have used deep neural networks (DNNs) to identify ES and ED frames from 2D echocardiography videos. The main limitations of using DNNs in this context are their limited generalizability, the requirement for a large amount of training data, and their weakness in modeling temporal dependencies. Therefore, developing methods that can detect ES and ED frames with low computational cost and high generalizability is still important. Thus, this paper proposes three different methods to identify ES and ED frames from echocardiography videos. These methods are based on the Laplacian eigenmap (LE), the maximum Laplacian eigenvalue (MLE), and the non-negative tensor factorization (NTF) method. To evaluate the performance of these proposed methods, they were applied to twenty-two videos from the EchoNet-Dynamic dataset. The proposed methods were also compared to the locally linear embedding (LLE) based method and the non-negative matrix factorization (NMF) based approach from previous studies. The results showed that the LE, MLE, and NTF methods can predict ES from a 2D echocardiography video with respective mean errors of 2, 2.5, and 1.6 frames; while they can predict ED with respective mean errors of 3.4, 3.3, and 2.8 frames. Overall, these methods can effectively be used to detect ES and ED frames from 2D echocardiography videos.
Language: English
Subfield: Cardiology and Cardiovascular Medicine
Field: Medicine
Domain: Health Sciences
Open Access status: gold
OpenAlex ID: W7220958740