Colorectal cancer glandular cell segmentation algorithm based on improved UNet
Segmentation of medical images plays a crucial role in computer-aided diagnosis.UNet is a primary method for image segmentation,but the encoder in the UNet network suffers from insufficient extraction of colorectal cancer glandular cell tissue features and inadequate semantic fusion,leading to suboptimal segmentation results.To address this issue,an algorithm named RCG-UNet is proposed.This algorithm improves the UNet encoder through a residual structure,Ghost convolution,and a channel attention mechanism,and replaces the ReLU activation function with the Mish activation function.This enhances the feature information extraction and fusion of images,thereby improving the precision of colorectal cancer glandular cell segmentation.The network was tested on the GlaS and CRAG datasets,showing an increase in the Dice similarity coefficient by 1.4%and 1.5%,and an improvement in the mean Intersection over Union(mIoU)by 1.2%and 1.5%,respectively,compared to the traditional UNet network.
medical image segmentationUNetfeature fusioncolorectal cancer
万方数据
维普
