WLKA-RVS: a retinal vessel segmentation method using weighted large kernel attention

Retinal vessel segmentation is an important task in medical image analysis and has a wide range of applications in the diagnosis and treatment of retinal diseases. However, existing segmentation methods still have some shortcomings in accurately segmenting thin vessels. Based on this observation, we propose a Retinal Vessel Segmentation method based on Weighted Large Kernel Attention (WLKA-RVS), which aims to improve the accuracy of retinal vessel segmentation to better assist physicians in clinical diagnosis and treatment. Our method consists of an encoder and a decoder. In the encoder, a convolution stem first reduces the dimension of the input image. Then, feature extraction is performed by four stages of Swin Transformer modules, each stage with a downsampling layer. In the decoder, there are four different stages of Weighted Large Kernel Attention Block (WLKAB) corresponding to the Swin Transformer modules in the encoder. Then WLKA-RVS applies the Patch Expanding module to achieve upsampling. Finally, a linear layer outputs the final results. We have performed extensive experiments comparing several recent advanced models on three public datasets. WLKA-RVS led by 0.32%, 1.24%, and 0.71% in the mAcc metric, respectively. At the same time, the inference speed of WLKA-RVS met the real-time requirements for medical diagnosis. A series of experiments demonstrated the efficiency, robustness, and applicability of WLKA-RVS.