Superpixel Combining U-NET for Pancreas Segmentation

In order to improve the performance of pancreas segmentation, this paper proposes a pancreas segmentation method combined superpixel and U-NET. Firstly, we propose a medical superpixel segmentation method. Then we map and reduce dimensionality to obtain visual summary image according to the result of superpixel segmentation. Finally, we use the visual summary image and superpixel position information as the input of U-NET to obtain the pancreas segmentation result. The experimental results on the NIH pancreas public dataset show that the DSC of this method is 87.9%, which is higher than all current pancreas segmentation methods;and the method is faster than U-NET.