基于U-Net的病理图像细胞核分割

Medical image segmentation is one of the important research directions of image processing, and the segmentation of cell nuclei in medical images is an important basis for doctors and pathologists to diagnose and classify cancer grades. However, how to efficiently and accurately segment cell nuclei in pathological images has always been the current research hotspot. In cell nucleus segmentation experiments, the segmentation effect is often poor or even difficult to segment due to the blurred edge contour information of the nucleus, the low contrast between the nucleus and the background, and the adhesion of cells in various organs. In order to solve the above problemsPathological image segmentation of breast cancer cell nucleus based on U-Net, this paper uses U-Net to realize the breast cancer cell segmentation experiment. U-Net uses downsampling to extract the shallow information of the image, extracts the deep information of the image through upsampling, and uses the skip connection structure to make the shallow information and the image. The deep information is effectively combined so that accurate nuclear segmentation can be achieved. In this paper, the U-Net is tested on the MoNuSeg public data set, the Dice coefficient is 73.66%, and the Miou intersection ratio reaches 73.78%. The experiments show that the U-Net network is an effective method for cell nucleus segmentation.