Automatic Deep Learning-based Myocardial Contours Segmentation from Cine MRI Images

Ejection fraction (EF) represents important predictor of adverse cardiovascular events in patients with coronary heart diseases (CHD). In Addition, Regional Wall Motion Abnormalities (RWMA) have greater prognostic values in discriminating between stunned or hibernating myocardial segments that largely help in the therapeutic decision. Therefore, it is important to accurately compute this parameter to ensure a good support for clinical left ventricle (LV) diagnosis. In this work, we propose a new method based on ResNet-UNet architecture to detect cardiac myocardial contours from MRI images. The proposed algorithm is trained using two datasets. A total of 240 patients were included in this study with 6000 MRI images. The proposed framework showed a Dice index of Dice Similarity Coefficient (DSC) of 0.97, 0.94, 0.92, and 0.94 for LVED, LVES, Myocardium ED, and Myocardium ES, respectively. The Hausdorff index was 4.8 mm and 7.9 mm, respectively for end diastolic LV and myocardium. The results showed improved performance compared to SOTA over the same public dataset.