A Femoral Head Segmentation Algorithm Based on an Improved TransUNet

To assist clinicians in accurately and rapidly determining the classification of femoral head necrosis from CT images, a femoral head region segmentation algorithm based on TransUNet is proposed to automatically segment femoral head regions in CT images. In the skip connections of TransUNet, a DABlock attention module is introduced, and the original Transformer component is replaced with a Swin Transformer Block, which enhances spatial texture representation and channel feature extraction while reducing redundant information, thereby improving the model's capability to segment target regions. Experimental results on a self-constructed CT image dataset of femoral head necrosis show that the improved TransUNet model achieves a precision of 98.94% and a recall of 99.13%, representing improvements of 2.07% and 1.45%, respectively. These results demonstrate the effectiveness of the proposed joint segmentation and classification algorithm for the identification and assessment of femoral head necrosis.