Logo image
Back
A Survey on Cardiac MRI Segmentation: from Classical Methods to State-of-the-Art Deep Learning
Conference paper

A Survey on Cardiac MRI Segmentation: from Classical Methods to State-of-the-Art Deep Learning

Hamed Aghapanah, Reza Saboori Amleshi, Ali Saeedi Rad and Masood Noruzi
2025 32nd National and 10th International Iranian Conference on Biomedical Engineering (ICBME), pp 369-375
19 Nov 2025
DOI: https://doi.org/10.1109/ICBME68496.2025.11392392

Abstract

Accuracy Benchmark testing Cardiac MRI Segmentation Deep learning Explainable AI Federated learning Hybrid Methods Image segmentation Machine Learning Methods Myocardium Robustness Survey Surveys Magnetic Resonance Imaging
Accurate and timely diagnosis of cardiac pathologies relies heavily on Cardiovascular Magnetic Resonance (CMR) imaging, the gold standard for assessing myocardial structure, function, and tissue characteristics. A critical step in CMR analysis is the segmentation of heart chambers-particularly the left ventricle, right ventricle, and myocardium-to derive essential clinical parameters such as ejection fraction, ventricular volumes, and myocardial mass. Manual segmentation, while accurate, is labor-intensive and subject to inter-observer variability, limiting its scalability in clinical practice. This has driven the need for automated, reliable, and reproducible segmentation methods. Classical approaches, including active contours and level sets, struggle with noise and low contrast. In contrast, deep learning models-especially U-Net variants, transformers, and hybrid architectures-have achieved expert-level accuracy, enabling fully automated quantification. However, challenges remain in generalizability across scanners and centers, robustness to artifacts, model interpretability, and integration into clinical workflows. This review addresses these gaps by systematically evaluating contemporary techniques, highlighting advances in deep and hybrid models, public benchmarks, and emerging solutions such as explainable AI and federated learning. This work emphasizes the importance of translating methodological advances into clinically viable solutions, fostering the adoption of secure, interpretable, and scalable AI-based CMR segmentation in routine practice.

Metrics

1 Record Views

Details

Logo image