• Sidrah Laldin, Cedrique Shum-Tim, Satya Prakash, and Dominique Shum-Tim.
• Section of Cardiac Surgery, Department of Surgery, University of Manitoba, 409 Tache Avenue, St. Boniface Hospital, Winnipeg, MB, R2H 2A6.
• J. Thorac. Cardiovasc. Surg. 2025 Jan 20.

Objective(S)We conduct a comparative study that employs the use of multiple dynamic deep learning algorithms to develop predictive models with video-based echocardiographic images using sample size determination as a key variable to assess optimal performance metrics.MethodsOur study compares performance of 3D convolutional neural networks, video vision transformers, and hybrid convolutional neural networks and Long Short-Term Memory models within both supervised and semi-supervised domains using variable sample sizes.ResultsFor supervised learning, the ResNet3D model achieved the lowest Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) across all training set sizes (200, 400, and 800-video datasets), with the best performance observed on the 800-video training set (MAE: 7.409, RMSE: 10.216). In the semi-supervised setting, both ResNet3D and ResNet+LSTM models benefited from the inclusion of unlabelled data, particularly at larger dataset sizes.ConclusionsBecause semi-supervised models use both labelled and unlabelled data sets, our findings are significant in showing that performance of certain predictive models using mixtures of unlabelled and labelled data is comparable to those using only labelled data with similar sample sizes, thus obviating the need for large sample sizes of labelled data.Copyright © 2025. Published by Elsevier Inc.

3 keywords...

hide…