目的 结直肠息肉检测可以有效预防癌变，然而人工诊断往往存在较高漏检率，使用深度学习技术可以提供有助于诊断的细粒度信息，辅助医生进行筛查。实际场景中，息肉形态各异和息肉边缘模糊的特点会严重影响算法的准确性。针对这一问题，本文提出了一种边缘概率分布模型引导的结直肠息肉分割网络(edge distribution guided high-resolution network, HRNetED)。方法 本文所提的HRNetED网络使用HRNet结构作为网络主干，并设计了一种堆叠残差卷积模块，显著降低模型参数量的同时提高模型性能；此外，本文使用边缘概率分布模型来描述息肉边缘，提高模型对边缘检测的稳定性；最后，本文在多尺度解码器中引入边缘检测任务，以加强模型对息肉边缘的感知；结果 本文在Kvasir-Seg、ETIS、CVC-ColonDB、CVC-ClinicDB和CVC-300五个数据集上进行测试。最终，HRNetED在CVC-ClinicDB和CVC-300数据集上的Dice系数和mIoU指标均优于现有其他算法，且在CVC-ClinicDB数据集上相较于先前最优模型分别获得了1.25%和1.37%的提升；在ETIS数据集上，Dice系数表现优于现有最优算法；在CVC-ColonDB数据集上，Dice和mIoU处于较优水平。此外，HRNetED在Kvasir-Seg、ETIS、CVC-ColonDB数据集上的HD95距离相较于现有最优算法分别降低了0.315%、29.19%和2.95%，在CVC-ClinicDB和CVC-300数据集上表现排在次优处，同样具有良好的性能。结论 本文提出的HRNetED网络在多个数据集中表现稳定，对于小目标、模糊息肉有较好的感知能力，对息肉轮廓检测能力更强。

Objective As a disease with high prevalence and high harm, colorectal cancer is seriously threatening human life and health. According to statistics, nearly 95% of colorectal cancer cases are caused by the development of early colon polyps, so if colorectal polyps can be detected in time and closely observed by medicine, the incidence of colorectal cancer can be effectively reduced. Artificial diagnosis often has a high rate of missing polyps. The use of deep learning technology can provide fine-grained information that is helpful for diagnosis, such as the location and shape of polyps, and assist doctors in screening, which is of great value for the prevention and treatment of colorectal cancer. Therefore, the rapid development of deep learning in recent times has brought great breakthroughs to computer-aided diagnosis technology in the medical field, models such as convolutional neural networks and Vision Transformer have shown good medical task processing capabilities. The use of computer technology for auxiliary diagnosis has gradually become a trend. In view of the characteristics of colorectal polyp images such as excessive morphological differences and unclear edges, we propose a edge probability distribution guided high-resolution network for colorectal polyp segmentation called HRNetED, which performs well in multiple colorectal polyp datasets and has good clinical application significance. Method The HRNetED network proposed in this paper takes the HRNet structure as the network backbone, which can ensure the full exchange of multi-scale features and ensure the accuracy of the model output by maintaining a high-resolution convolutional branch. In addition, a stack residual convolution module (SRC) is designed. The SRC module extracts the output of each convolution kernel by splitting a single convolution into four subconvolutions and connecting them serially, so as to obtain the characteristics of multi-receptive fields. Finally, pointwise convolution is used for feature fusion, and residual connection is introduced to avoid model performance degradation. To a certain extent, SRC solves the limitation of insufficient receptive fields in a single convolution operation, and significantly reduces the number of model parameters and improves model performance through convolution splitting. Considering the different morphological sizes, large color differences and inconsistent imaging quality of colorectal polyp images, we designed a multi-scale decoder to simultaneously supervise and learn the output results of different scales, and introduces edge detection tasks into the structure to strengthen the perception of polyp edges. For the problem that the edge of polyp is not clear, we used the edge probability distribution model based on Gaussian distribution to describe the polyp edge, so that the model does not need to return the accurate edge position information, but only needs to predict the heat map of the edge distribution, which effectively reduces the difficulty of model convergence and improves the perception ability and robustness of the model in the edge semantic ambiguous region. In the dataset configuration part, we follow the experimental steps of mainstream networks such as PraNet: 900 images from the Kvasir-Seg dataset and 550 images in CVC-ClinicDB were used as the training set, for a total of 1450 images; All images from ETIS, CVC-ColonDB, and CVC-300, as well as the remaining images from Kvasir-Seg and CVC-ClinicDB, were combined as test sets, and all images were scaled to 256×256 dimensions simultaneously. In the model training part, FocalLoss and BCELoss were used for supervised training of edge detection and polyp segmentation tasks, respectively. Iterative using cosine annealing learning rate adjustment strategy and Adam optimizer. In the model testing phase, the model is evaluated using the Dice coefficient and the mIoU metric. Result We tested on 5 publicly available colorectal polyp datasets, including Kvasir-Seg, ETIS, CVC-ColonDB, CVC-ClinicDB, and CVC-300, and compared with existing colorectal polyp segmentation algorithms, including HRNetv2, PraNet, UACANet, MSRF-Net, BDG-Net, SSFormer and ESFPNet. From the comparison results, it can be seen that the Dice coefficient and mIoU metric of HRNetED on CVC-ClinicDB and CVC-300 data sets are better than other existing algorithms, and compared with the previous optimal model on CVC-ClinicDB data set, HRNetED achieved 1.25% and 1.37% improvement respectively. In ETIS dataset, the Dice and mIoU of HRNetED were 82.41% and 71.21%, respectively, and the Dice coefficient was better than the existing optimal algorithm. On the CVC-ColonDB dataset, the Dice and mIoU of the proposed algorithm are 80.55% and 71.56%, respectively. In addition, the HD95 distance of HRNetED on the Kvasir-Seg, ETIS, and CVC-ColonDB datasets is 0.315%, 29.19%, and 2.95% lower than the existing optimal algorithms,and on the CVC-ClinicDB and CVC-300 datasets Performance is second best, also has good performance. Conclusion The HRNetED network we proposed in this paper performs well in colorectal polyp segmentation tasks, and from the subjective segmentation results, HRNetED performs stably in multiple datasets, and has good perception of small targets and fuzzy polyps, also has stronger ability to detect polyp contours; Through ablation experiments, it can be seen that the stacked residual convolution module proposed in this paper can greatly reduce the number of model parameters and improve the performance of the model, and the edge probability distribution model proposed for the problem of edge fuzzy region can effectively improve the performance of the network.