基于改进PraNet的结直肠息肉图像分割算法  

Colorectal Polyp Image Segmentation Algorithm Based on Improved PraNet

作　　者：陈政豪 常君明[1] CHEN Zhenghao;CHANG Junming(School of Artificial Intelligence,Jianghan University,Wuhan 430056,China)

机构地区：[1]江汉大学人工智能学院,湖北武汉430056

出　　处：《现代信息科技》2025年第1期60-65,70,共7页Modern Information Technology

摘　　要：结直肠息肉的早期精确分割对于预防结直肠癌至关重要。尽管PraNet模型在图像分割领域取得了显著成效,但在处理边界模糊和尺寸多变的息肉时仍面临挑战。为了提升PraNet在结直肠息肉图像分割中的性能,提出了一种融合图神经网络(GNN)的改进算法。将图像中的像素或超像素视为图结构中的节点,利用GNN学习图结构特征捕捉了图像中的局部上下文信息,结合特征融合机制将GNN学习特征与PraNet原始特征结合。实验结果表明,此PraNet改进算法在mDice和mIoU指标上分别达到了94.1%和88.3%,显著优于U-Net和FCN等对比模型。因此,该算法能够显著提高结直肠息肉图像分割的精度,为结直肠癌的预防提供了更加可靠的技术支持。Early and precise segmentation of colorectal polyps is essential for the prevention of CRC.Although the PraNet model has achieved remarkable results in thefield of image segmentation,it still faces challenges in dealing with polyps with blurred boundaries and variable sizes.In order to improve the performance of PraNet in the image segmentation of colorectal polyps,an improved fusion Graph Neural Network(GNN)algorithm is proposed.The pixels or superpixels in the images are regarded as nodes in the graph structure.By using GNN to learn the graph structure features,the local context information in the images is captured.Combined with the feature fusion mechanism,the GNN learning features are integrated with the PraNet original features.The experimental results show that this improved PraNet algorithm achieves 94.1%and 88.3%in mDice and mIoU indexes,respectively,significantly better than the comparison models such as U-Net,FCN and so on.Therefore,this algorithm can significantly improve the image segmentation accuracy of colorectal polyps,and provide more reliable technical support for the prevention of CRC.

关 键 词：PraNet算法 结直肠息肉 GNN 图像分割 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP391.4[自动化与计算机技术—控制科学与工程]