3D Multi-Class Segmentation of Aortic Branches and Zones in CTA Using Convolutional Neural Networks

Aorta segmentation is a critical task in medical image analysis, crucial for diagnosing and treating cardiovascular diseases. This study presents an automated approach for aorta segmentation using the U-Net convolutional neural network (CNN) architecture. To improve segmentation accuracy, the method integrates Dice Cross-Entropy (DiceCE) Loss with Skeleton Recall Loss, a technique previously introduced to better preserve the structural integrity of segmented objects, particularly thin structures like the aorta's centerline. By reusing this loss function in our framework, we enhance the model's ability to maintain structural continuity. The proposed approach demonstrates high accuracy and robustness. Data augmentation and parameter optimization further contribute to the method's superior performance over traditional techniques, positioning it as a promising tool for clinical application by improving diagnostic precision and reducing manual intervention.