Enhanced Liver Segmentation Using Hybrid U-Net-Transformer Architecture

Segmentation of the liver plays a crucial role in the detection and diagnosis of liver disorders. Conventional manual segmentation methods are time-intensive, susceptible to operator-dependent inconsistencies, and ineffective for extensive clinical applications. To overcome these obstacles, this study presents a hybrid U-Net model with a Transformer bottleneck, which enhances accuracy in segmentation by capturing both local and global contextual information. The model extracts spatial characteristics by utilizing the U-Net encoder, which are then handled by the Transformer encoder to refine global representations before passing through the U-Net decoder. Through the combination of transformer architecture and convolutional neural networks (CNNs), this integration ensures reliable liver segmentation. The proposed model obtained Dice Similarity Coefficients (DSC) of 98.09% and 98.12%, respectively, when evaluated using two publicly available benchmark datasets, CHAOS and 3D-IRCADb. These findings show that the hybrid U-Net-Transformer model enhances segmentation accuracy while maintaining computing efficiency.