Multi-Source and Multi-Sequence Myocardial Pathology Segmentation Using a Cascading Refinement CNN

TL;DR

The paper addresses the challenge of assessing myocardial tissue viability after infarction by leveraging multi-sequence MR data (LGE, T2, and cine) and proposes MS-CaRe-CNN, a two-stage cascading refinement CNN that first segmentates anatomical cardiac structures and then discriminates tissue viability (healthy, scar, edema). The approach handles missing sequences via dataset grouping and employs strong 3D data augmentation to mitigate domain shift, using a 5-fold ensemble to improve generalization. Quantitatively, the method achieves DSCs of 62.31% for scar (with 82.65% precision) and 63.78% for scar+edema (with 87.69% precision), demonstrating strong performance on challenging small structures and enabling downstream tasks such as personalized therapy planning. Overall, MS-CaRe-CNN provides a robust, end-to-end framework for multi-sequence myocardial pathology segmentation and viability assessment in clinical imaging workflows.

Abstract

Myocardial infarction (MI) is one of the most prevalent cardiovascular diseases and consequently, a major cause for mortality and morbidity worldwide. Accurate assessment of myocardial tissue viability for post-MI patients is critical for diagnosis and treatment planning, e.g. allowing surgical revascularization, or to determine the risk of adverse cardiovascular events in the future. Fine-grained analysis of the myocardium and its surrounding anatomical structures can be performed by combining the information obtained from complementary medical imaging techniques. In this work, we use late gadolinium enhanced (LGE) magnetic resonance (MR), T2-weighted (T2) MR and balanced steady-state free precession (bSSFP) cine MR in order to semantically segment the left and right ventricle, healthy and scarred myocardial tissue, as well as edema. To this end, we propose the Multi-Sequence Cascading Refinement CNN (MS-CaRe-CNN), a 2-stage CNN cascade that receives multi-sequence data and generates predictions of the anatomical structures of interest without considering tissue viability at Stage 1. The prediction of Stage 1 is then further refined in Stage 2, where the model additionally distinguishes myocardial tissue based on viability, i.e. healthy, scarred and edema regions. Our proposed method is set up as a 5-fold ensemble and semantically segments scar tissue achieving 62.31% DSC and 82.65% precision, as well as 63.78% DSC and 87.69% precision for the combined scar and edema region. These promising results for such small and challenging structures confirm that MS-CaRe-CNN is well-suited to generate semantic segmentations to assess the viability of myocardial tissue, enabling downstream tasks like personalized therapy planning.

Figures (2)

• Figure 1: Overview of the proposed method, MS-CaRe-CNN, which is a 2-stage CNN cascade that semantically segments cardiac structures in 3D from multi-sequence data, namely, LGE MR, T2 MR and the end-diastolic phase of bSSFP cine MR. At Stage 1, given the three MR scans concatenated in channel dimension, MS-CaRe-CNN predicts the left and right ventricle as well as the myocardium which is then concatenated with the original image information in channel dimension. This initial prediction is then further processed in Stage 2, where the model is trained to additionally distinguish myocardial tissues by separating it into healthy tissue, scar and edema.
• Figure 2: Qualitative results on the validation set of the CARE2024 Challenge MyoPS++ track. We show corresponding examples of LGE MR (col. 1), T2 MR (col. 2) and the end-diastolic phase of bSSFP cine MR images (col. 3), as well as predictions of Stage 1 (col. 4) and Stage 2 (col. 5). In Stage 1, colors represent the left ventricle (red), right ventricle (blue) and myocardium as one label (green). Colors of Stage 2 refer to the left ventricle (red), right ventricle (blue), healthy myocardial tissue (green), scarred myocardial tissue (yellow) and edema (cyan).