3DCNN Model for Left Ventricular Ejection Fraction Evaluation in Echocardiography

Echocardiography is a popular medical imaging technique for identifying cardiovascular issues. One of the most critical indicators of heart function is the left ventricular ejection fraction (LVEF). However, there is inter-observer variability in the evaluation of the ejection fraction. This study aims to calculate the EF value that can be used to identify cardiomyopathy (abnormalities in the heart muscle that can lead to heart failure), evaluate whether a patient is eligible for specific chemotherapies, and establish the need for specific medical devices. This study presents a new 3DCNN model for video datasets containing echocardiogram sequences capturing heart motion throughout time. The proposed 3DCNN architecture efficiently captures the spatiotemporal patterns in echocardiograms using temporal information. The model can extract spatial and temporal information from the input data by including three-dimensional convolutions. The network architecture has multiple 3D convolutional layers, max pooling, and fully-connected layers for regression tasks. This model assessed our methodology using the EchoNet-Dynamic dataset to confirm its efficacy. Also, this model outperformed the most recent approach with outstanding results regarding performance metrics. The proposed model made quantitative errors with an MAE of 4.71 (RMSE) of 1.30 and an (R2) value of 0.68. These measurements show this 3DCNN model’s outstanding ability to predict LVEF from echocardiography video sequences with high levels of accuracy.