HPRUNET: A High-Precision Network for ALCAPA Segmentation with Residual Learning and Conditional Refinement

Anomalous Left Coronary Artery from the Pulmonary Artery (ALCAPA) is a series of congenital heart defects requiring precise image segmentation for accurate diagnosis. Existing research faced limitations such as loss of image detail, poor contrast handling, and vulnerability to noise leading to decreased accuracy. To address these challenges, this research introduces the High-Pitched Residual U-Network designed for automatic segmentation of ALCAPA from 3D Computed Tomography Angiography images. The proposed model integrates U-Net with residual blocks to extract low and mid-level features, mitigating information degradation. The encoder compresses the image using convolution, max-pooling, and dropout layers, while the decoder reconstructs the segmented image with Conv2DTranspose layers and residual block concatenation. Segmentation is performed using swish activation and sigmoid activation, with segmentation masks refined using Conditional Random Fields. The proposed method achieved an accuracy of 98.65% demonstrating a significant improvement over traditional method. This study highlights the potential of HPRUNET in transforming clinical diagnosis and treatment planning by automating the segmentation process, reducing manual intervention, and improving the detection of ALCAPA.