This project explores the use of a U-Net based 2D convolutional neural network (CNN) for the fully automatic segmentation of the right ventricle (RV) in cardiac magnetic resonance (CMR) images. The study utilizes a dataset of 5,729 short-axis cine CMR slices from 100 individuals, with training performed on 50 patients and testing on 40.
The CNN model, trained for 18 epochs, achieved a median Dice similarity coefficient (DSC) of 0.898 and a median 95th percentile Hausdorff distance (HD) of 5.1 mm on the test dataset, demonstrating robust performance. Despite challenges with specific heart regions, particularly at end-systole and the heart apex, the model effectively segmented the RV across a diverse patient group.
This work shows promising results for automating RV segmentation, potentially reducing the time-intensive manual delineation process currently used in clinical practice.