DR-Unet++: An Approach for Left Ventricle Segmentation from Magnetic Resonance Images

Precise and automatic segmentation of the left ventricle (LV) in cardiac magnetic resonance imaging (MRI) can help cardiologists make accurate diagnoses and decisions. In the revolution of Deep Learning, many LV image segmentation models have been developed and shown promising performance. Nevertheless, segmentation of the LV is still a nontrivial task due to challenges in the cardiac MR images such as the presence of intensity inhomogeneity, clutter, and object size variations. In this study, we propose DR- Unet++, an enhanced DR-Unet architecture, for cardiac segmentation tasks. The DR-Unet++ model, which utilizes advanced Deep Learning techniques, provides better feature extraction for the segmentation of desired segmented objects. Our tests on the ACDC dataset and Sunnybrook Cardiac dataset show that DR-Unet++ outperforms other modern models in terms of Dice coefficient and IoU metric and demonstrates its potential in biomedical image segmentation tasks.