改进U-Net网络的结直肠息肉分割模型

Improved Colorectal Polyp Segmentation Model of U-Net Network

针对结直肠息肉边界模糊性和形状、大小多样性导致的分割不精确的问题,基于U-Net网络构建级联卷积注意力模块,以保留更多的有用特征;采用多尺度特征聚合模块来捕捉更多的特征与细节;将空间注意力嵌入到感受野模块,扩大网络感受野的同时提供更丰富的空间信息表示。实验结果表明,该模型在Kvasir数据集上展现出了良好的分割效果。相比传统U-Net模型,其在IoU,Dice上分别提高了3.14%,5.31%,在医学影像计算机辅助诊断中具有很大的应用潜力。

To address the issue of imprecise segmentation caused by blurry boundaries and diverse shapes and sizes of colorectal polyps,a cascaded convolutional attention module is constructed based on the U-Net network to retain more useful features.Adopting a multi-scale feature aggregation module to capture more features and details.Embedding spatial attention into the receptive field module expands the network receptive field while providing richer spatial information representation.The experimental results show that the model exhibits good segmentation performance on the Kvasir dataset.Compared with the traditional U-Net model,it has improved by 3.14%and 5.31%on IoU and Dice,respectively,and has great potential for application in computer-aided diagnosis of medical imaging.