Aims
This study was conducted as part of the CRT-DRIVE project aimed at improving the efficiency of cardiac resynchronization therapy (CRT). The study sought to develop an approach for reconstructing anatomy of coronary sinus veins using the proprietary cardiac computer tomography (CT) protocol. These cardiac structures are crucial for pre-procedural assessment and successful CRT but are often poorly visible. The goal of the study was to accurately identify these target structures and improve the efficiency of CRT.
Methods
The method used in this study is based on the CRT-DRIVE clinical study CT scan protocol, which is specifically designed to examine the heart anatomy and the cardiac venous system in patients with heart failure (HF) and a reduced LV ejection fraction (LV EF <50%). The reconstruction method relies on a combination of a state-of-the-art neural network (NN) for medical image segmentation called SWIN-UNET and the k-fold cross-validation ensemble. Our researchers collected a high-quality dataset, consisting of 101 cardiac CT studies. Each image was marked by radiologist and further independently validated by two cardiologists. To test the method, 30 cases were randomly selected, while the other 71 cases were used to train five neural networks during a 5-fold cross-validation procedure. The trained NNs were ensembled, and the final quality was estimated.
Results
The results of the study indicate that the basic NN achieved a quality of 0.74 in terms of the dice score during the 5-fold cross-validation procedure. The final quality of the ensemble estimated using the test subset was 0.76 in terms of the dice score.
Conclusion
The study achieved a satisfactory quality of CS anatomy reconstruction, and a high-quality dataset was collected. Furthermore, the study confirmed the efficiency of the proposed cardiac CT protocol for the reconstruction of CS veins in patients with chronic HF and reduced LV EF.