KDA-UNet: A Deep Learning Framework for Segmentation in Medical Imaging

Medical imaging plays a crucial role in the diagnosis, treatment, and assessment of diseases. Accurate segmentation of bones and muscles in medical images is of significant practical importance for assisting physicians in diagnosis. However, the low contrast between bones and muscles in medical images, coupled with substantial variations in the edges and shapes of different bones, presents considerable challenges. To address these issues, this paper proposes a segmentation network based on the U-Net architecture, incorporating KAN convolution and a dual attention mechanism, referred to as KDA-UNet. The KAN convolution effectively captures local spatial information in images, compensating for the limitations of standard convolutional layers in modeling spatial features. The dual attention mechanism helps the network focus on the most critical regions and channels within the image, significantly enhancing the feature extraction performance and capability of the segmentation network.