Key result
Tailored UNet models outperform foundation models for LGE MR segmentation despite lagging in infarct delineation.
Why the study?
Existing deep learning models for myocardial and infarct segmentation rely on small, pre-processed datasets and specific post-reperfusion phases, limiting generalizability for routine clinical practice.
Do well-designed UNet-based deep learning techniques outperform fully automatic foundation models for LGE MR segmentation of myocardial infarcts?
Do well-designed UNet-based deep learning techniques outperform fully automatic foundation models for LGE MR segmentation of myocardial infarcts?
Well-designed UNet-based deep learning models outperform foundation models for LGE MR segmentation, providing stable delineations of the left ventricle and myocardium but requiring further improvement for infarct regions.
Answers are grounded in this paper and cite their sources.
Favors custom UNets over foundation models for LV and myocardial LGE segmentation now; leaves open reliable infarct quantification for clinical use.
Late gadolinium enhancement (LGE) cardiac magnetic resonance (MR) imaging is the modality of choice to assess myocardial infarction (MI) lesions. Nowadays MI volume quantification is not performed routinely in clinical practice. Numerous deep learning (DL) methods have been developed to automate the segmentation of the myocardium and infarct regions. However, most studies rely on relatively small datasets which typically undergo pre-processing steps to standardize images and focus on a specific phase of myocardial infarction following reperfusion therapy. These limitations have impeded the development of models that are generalizable across diverse conditions and thus suitable for routine clinical use. To advance research and establish benchmarks in generalizable learning for myocardial infarct quantification, this paper presents findings from the Myocardial Segmentation with Automated Infarct Quantification (MYOSAIQ) challenge. The dataset set up for the challenge combines 439 CMR volumes from two multicenter clinical trials, with representative data acquired in acute and chronic phases after acute MI. Data were acquired in 16 centers using MRI scanners from three different vendors. Six teams participated until the end of the challenge, employing various baseline models, data augmentation techniques, and confidence strategies. To enhance the significance of this study, we compare the challengers’ results with those of fine-tuned foundation models. Our results indicate that well-designed UNet-based techniques outperform fully automatic foundation models for LGE MR segmentation. While the best methods achieve high-quality and stable delineations of the left ventricle and myocardium under various conditions, they remain improvable in accurately segmenting infarct regions.
No takes yet. Share an insight, caveat, or question.
Bernard et al. (2026) studied Myocardial infarction (n=439). UNet-based deep learning segmentation methods vs. Fully automatic foundation models was evaluated on LGE MR segmentation accuracy (geometric and clinical metrics). Well-designed UNet-based techniques outperformed fully automatic foundation models for LGE MR segmentation, achieving high-quality delineations of the left ventricle and myocardium but remaining improvable for infarct regions.
Synapse has enriched 10 related papers. Consider them for comparative context: