Diagnostic Segmentation Based on Kidney Medical Image

Lesion segmentation of medical images is an important component smart medicine. The development deep learning technology followed by rapid advancement in lesion images. Though the present can retain spatial features, insufficient features are retained with low accuracy. Our proposed PST-UNet model combines transformer U-shaped structure and better infuses encoder's multi-scale using convolution fusion module. adopts two types block Swin transform at encoder decoder ends respectively. Renal data tends to a normal distribution. Therefore, preserve more enhance precision renal segmentation, full GELU (Gaussian Error Linear Unit) activation function introduced end. Similarly, end, block, function, up-sampling jumper wires from module also introduced.