Research on Improved Abdominal Multi-Organ Segmentation Method Based on TransUNet and Multi-Domain Attention Fusion

Abstract

Abdominal multi-organ segmentation holds significant value in intelligent medical diagnosis. However, due to factors such as large variations in organ morphology, blurred boundaries, and significant scale differences, existing segmentation models still face challenges of insufficient representation and inadequate feature fusion. To address these issues, this paper introduces a Multi-Domain Attention Fusion (MDAF) module into the TransUNet framework to achieve effective alignment and deep fusion of spatial-domain and long-range features. Experimental results indicate that the proposed method can effectively mitigate semantic biases in cross-domain feature fusion, providing an efficient and scalable solution for medical image segmentation.