Interactive model enhances pancreas segmentation accuracy in CT images

A novel deep-learning network, RRM-TransUNet, significantly improves pancreas segmentation accuracy in CT images. Achieving a dice similarity coefficient of 93.82% and an average symmetric surface distance error of 1.12 mm on multiple datasets, this model integrates clinical expertise with user interaction. By using advanced techniques like rotary position embedding and a mixture of experts mechanism, RRM-TransUNet outperforms previous methods, providing clinicians with a more reliable and intuitive segmentation tool.