Stacked BCDU-Net with Semantic CMR Synthesis: Application to Myocardial Pathology Segmentation Challenge

Abstract

Accurate segmentation of pathological tissue, such as scar tissue and edema, from cardiac magnetic resonance images (CMR) is fundamental to the assessment of the severity of myocardial infarction and myocardial viability. There are many accurate solutions for automatic segmentation of cardiac structures from CMR. On the contrary, a solution has not as yet been found for the automatic segmentation of myocardial pathological regions due to their challenging nature. As part of the Myocardial Pathology Segmentation combining multi-sequence CMR (MyoPS) challenge, we propose a fully automatic pipeline for segmenting pathological tissue using registered multi-sequence CMR images sequences (LGE, bSSFP and T2). The proposed approach involves a two-staged process. First, in order to reduce task complexity, a two-stacked BCDU-net is proposed to a) detect a small ROI based on accurate myocardium segmentation and b) perform inside-ROI multi-modal pathological region segmentation. Second, in order to regularize the proposed stacked architecture and deal with the under-represented data problem, we propose a synthetic data augmentation pipeline that generates anatomically meaningful samples. The outputs of the proposed stacked BCDU-NET with semantic CMR synthesis are post-processed based on anatomical constrains to refine output segmentation masks. Results from 25 different patients demonstrate that the proposed model improves 1-stage equivalent architectures and benefits from the addition of synthetic anatomically meaningful samples. A final ensemble of 15 trained models show a challenge Dice test score of 0.665 ± 0.143 and 0.698 ± 0.128 for scar and scar + edema, respectively.

C. Martín-Isla and M. Asadi-Aghbolaghi—contributed equally to this work.