Robust Deep Learning for Myocardial Scar Segmentation in Cardiac MRI with Noisy Labels

2025-06-26

AI Summary

Automated precise segmentation of myocardial scar in cardiac MRI faces critical challenges including label noise, data heterogeneity, and severe class imbalance. To address these, we propose a robust segmentation framework that (i) incorporates a KL-divergence-based loss to mitigate the impact of noise inherent in semi-automatically generated annotations; (ii) integrates multi-scale and large-scale data augmentation to enhance out-of-distribution generalization and model robustness; and (iii) leverages state-of-the-art deep learning fine-tuning strategies for fine-grained differentiation between acute and chronic scars. Evaluated on public and multi-center datasets, our method surpasses SOTA approaches—including nnU-Net—across key metrics such as Dice score and Hausdorff distance. Qualitatively, the segmentations exhibit improved smoothness and stronger anatomical consistency. These advances significantly enhance reliability for clinical scar assessment and personalized treatment planning.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Segmentation
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

The accurate segmentation of myocardial scars from cardiac MRI is essential for clinical assessment and treatment planning. In this study, we propose a robust deep-learning pipeline for fully automated myocardial scar detection and segmentation by fine-tuning state-of-the-art models. The method explicitly addresses challenges of label noise from semi-automatic annotations, data heterogeneity, and class imbalance through the use of Kullback-Leibler loss and extensive data augmentation. We evaluate the model's performance on both acute and chronic cases and demonstrate its ability to produce accurate and smooth segmentations despite noisy labels. In particular, our approach outperforms state-of-the-art models like nnU-Net and shows strong generalizability in an out-of-distribution test set, highlighting its robustness across various imaging conditions and clinical tasks. These results establish a reliable foundation for automated myocardial scar quantification and support the broader clinical adoption of deep learning in cardiac imaging.

Problem

• Automated myocardial scar segmentation in cardiac MRI
• Addressing label noise from semi-automatic annotations
• Handling data heterogeneity and class imbalance

Innovation

• Fine-tuning state-of-the-art deep learning models
• Using Kullback-Leibler loss for noisy labels
• Extensive data augmentation for heterogeneity