Objectives: This study aims to review state of the art (SOTA) deep learning (DL) methods that segment cardiac images from magnetic resonance imaging (MRI), computed tomography (CT), and ultrasound imaging modalities, and to discuss the strengths and weaknesses of these approaches as well as current research challenges. Material and Methods: Recent advances in DL architecture include convolutional neural networks (CNN), fully convolutional networks (FCN), U-Nets, V-Nets, and recurrent neural networks (RNN). Studied and reviewed the application of these architectures to cardiac segmentation tasks and considered preprocessing, loss functions, and hybrid approaches (i.e., DL combined with traditional segmentation methods). Results: DL methods consistently outperform conventional segmentation techniques in accuracy and efficiency. MRI segmentation is most widely studied due to the availability of public datasets, while CT and ultrasound present challenges such as motion artifacts, speckle noise, and annotation sparsity. Advanced models (e.g., Dense U-Net, OmegaNet, multi-view CNN) achieve high Dice scores; however, 3D models require greater computational resources and risk overfitting. Conclusion: Deep learning has transformed cardiac imaging segmentation by improving its predictive accuracy and clinical utility. However, challenges remain in data availability, computational complexity, and interpretability. Future work should focus on lightweight architectures, implementing semi-supervised learning, and utilizing explainable AI to facilitate broader clinical implementation.
Nandhagopal et al. (Sat,) studied this question.