Adjacency-Aware Co-Occurrence Loss

Adding co-occurrence loss improved overall consistency and reduced misclassification. The adjacency-aware co-occurrence loss uses a binary anatomical adjacency matrix to encode connected artery pairs. The co-occurrence loss penalizes predicted co-occurrence for non-adjacent artery pairs. For missing expected connections, the method compares predicted and ground-truth co-occurrence features in keypoint regions with Sobel-derived edge maps.