Mutual enhancing learning-based automatic segmentation of CT cardiac substructure.

Researchers

Journal

Modalities

Models

Abstract

Current segmentation practice for thoracic cancer RT considers the whole heart as a single organ despite increased risks of cardiac toxicities from irradiation of specific cardiac substructures. This may be due to time consuming process of manually segmenting up to 15 cardiac substructures that can have large anatomic variations from one patient to the other. In this work, a new deep learning (DL) -based mutual enhancing strategy is introduced for an accurate and automatic segmentation, especially of smaller substructures. Our proposed method is consisted of three subnetworks: retina U-net, classification module and segmentation module. Retina U-Net is used as a backbone network architecture that aims to learn deep features from the whole heart. Whole heart feature maps from retina U-net are then transferred to four different sets of classification modules to generate classification localization maps of coronary arteries, great vessels, chambers of the heart, and valves of the heart. Each classification module is in sync with its corresponding subsequent segmentation module in a bootstrapping manner, which allows them to share their encoding paths to generate mutual enhancing strategy. Segmentation accuracies were statistically compared through dice similarity coefficient (DSC), Jaccard, 95% Hausdorff distance (HD95), mean surface distance (MSD), root mean square distance (RMSD), center of mass distance (CMD), and volume difference (VD). The proposed method yielded good spatial consistency to the ground truth segmentations of cardiac substructures. Our method generated substructure segmentations with significantly (P < 0.05) better accuracy for small substructures, especially coronary arteries, compared to three competing methods, 3D U-Net, mask R-CNN and mask scoring R-CNN. Promising results of this work demonstrate potential of this framework to be used as a tool to rapidly generate substructure segmentations followed by physician’s review to improve clinical efficiency and treatment outcomes.© 2022 Institute of Physics and Engineering in Medicine.