Enhanced Automatic Coronary Sinus Veins Segmentation Model Based on Hausdorff Distance Loss Function and Multi-Architecture Neural Network Ensemble

The key point of the study consists of performing automatic coronary sinus (CS) vein segmentation based on Hausdorff distance loss function and the multi-architecture neural network ensemble.A proprietary cardiac computed tomography protocol is implemented to provide adequate data quality.The dataset consists of two subsets: 133 patients selected for cardiac resynchronization therapy (CRT subset) and 127 candidates recruited for surgical interventions due to ventricular arrhythmias (Extra subset).To test the method, 39 of the 133 cases in the CRT subset were randomly selected.The remaining patients of the CRT subset and all cases of the extra subset are used to train neural networks.The ensemble neural network achieves a Dice score of 0.77, 30.8 mm in terms of Hausdorff distance, and 1.04 mm for mean surface distance in the test subset.In the study, an improvement in automatic CS anatomic reconstruction is presented, which could be helpful for pacemaker implantation and follow-up procedures.