Combining Multi-Sequence and Synthetic Images for Improved Segmentation of Late Gadolinium Enhancement Cardiac MRI

Accurate segmentation of the cardiac boundaries in late gadolinium\nenhancement magnetic resonance images (LGE-MRI) is a fundamental step for\naccurate quantification of scar tissue. However, while there are many solutions\nfor automatic cardiac segmentation of cine images, the presence of scar tissue\ncan make the correct delineation of the myocardium in LGE-MRI challenging even\nfor human experts. As part of the Multi-Sequence Cardiac MR Segmentation\nChallenge, we propose a solution for LGE-MRI segmentation based on two\ncomponents. First, a generative adversarial network is trained for the task of\nmodality-to-modality translation between cine and LGE-MRI sequences to obtain\nextra synthetic images for both modalities. Second, a deep learning model is\ntrained for segmentation with different combinations of original, augmented and\nsynthetic sequences. Our results based on three magnetic resonance sequences\n(LGE, bSSFP and T2) from 45 different patients show that the multi-sequence\nmodel training integrating synthetic images and data augmentation improves in\nthe segmentation over conventional training with real datasets. In conclusion,\nthe accuracy of the segmentation of LGE-MRI images can be improved by using\ncomplementary information provided by non-contrast MRI sequences.\n