摘 要：结直肠息肉准确分割，可以辅助医生诊断肠胃疾病，有效降低结直肠癌的发病风险。为解决息肉准确分割的问题，在 Transformer 模型和 U-Net 模型相融合的基础上提出了 UT-Former 模型。首先，采用一系列预处理技术对原始图像进行处理。其次，借助结直肠的图像，基于UT-Former网络结构设计结直肠息肉分割模型。再次，对UT-Former模型进行训练得到最佳模型，并将 U-Net 模型作为对比实验。最后，通过 Dice 指数评价 UT-Former 模型的有效性，并与 U-Net 模型进行对比。实验结果表明，UT-Former 模型可以准确地预测结直肠息肉，为患者提供早期预后信息。
关键词：结直肠；Transformer；息肉分割；U-Net；卷积神经网络
DOI:10.19850/j.cnki.2096-4706.2022.22.019
中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2022）22-0078-04
Colorectal Polyp Segmentation Model Based on UT-Former
YANG Ying, HAN Jincang, ZHANG Yangjie
(Lanzhou University of Finance and Economics, Lanzhou 730020, China)
Abstract: Accurate segmentation of colorectal polyps can help doctors diagnose gastrointestinal diseases and effectively reduce the risk of colorectal cancer. In order to solve the problem of accurate polyp segmentation, UT-Former model is proposed based on the fusion of Transformer model and U-Net model. First, a series of preprocessing techniques are used to process the original image. Secondly, with the help of colorectal images, a colorectal polyp segmentation module is designed based on the UT-Former network structure. Thirdly, the UTFormer model is trained to get the best model, and U-Net model is used for a contrast experiment. Finally, the validity of UT-Former model is evaluated by Dice index and compared with U-Net model. The experimental results show that UT-Former model can accurately predict colorectal polyps and provide early prognosis information for patients.
Keywords: colorectal; Transformer; polyp segmentation; U-Net; convolutional neural network
参考文献：
[1] 周雄，胡明，李子帅，等 .2020 年全球及中国结直肠癌流行状况分析 [J/OL]. 海军军医大学学报，2022：1-9[2022-06-12].https://kns.cnki.net/kcms/detail/detail.aspx?dbcode=CAPJ& dbname=CAPJLAST&filename=DEJD20220917001&uniplatform =NZKPT&v=dcBgsJbVkeiKaQiXwzTpEvXWVS8_qQ08m0LTBlp sPanDbPggSP2jjS2eYfbq7fo0.
[2] 沈志强，林超男，潘林，等 . 基于同构化改进的 U-Net 结直肠息肉分割方法 [J]. 中国生物医学工程学报，2022，41（1）：48-56.
[3] 张庆辉.卷积神经网络综述 [J].中原工学院学报，2017（3）：82-86+90.
[4] RONNEBERGER O，FISCHER P，BROXT. U-Net: Convolutional Networks for Biomedical Image Segmentation [EB/OL].[ 2022 - 0 6 - 2 4 ].https://link.springer.com/chapt er/10.1007/978-3-319-24574-4_28.
[5] 武芳，王鸿雁 .U-Net 及其改进算法在医学图像分割中的研究进展 [J]. 电子世界，2021（20）：36-37.
[6] DOSOVITSKIY A，BEYERL，KOLESNIKOVA，et al. An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale [EB/ OL].[2022-07-01].https://blog.csdn.net/qq_42014059/article/ details/122098994.
[7] 梁礼明，周珑颂，尹江，等 . 融合多尺度 Transformer 的皮肤病变分割算法 [J/OL]. 吉林大学学报（工学版），2022：1-13[2022-06-29].
[8] JHA D，SMEDSRUD P H，RIEGLER M A，et al. KvasirSEG: A Segmented Polyp Dataset [EB/OL].[2022-06-23].https://link. springer.com/chapter/10.1007/978-3-030-37734-2_37.
[9] ABADI M. TensorFlow: learning functions at scale. In Proceedings of the 21st ACM SIGPLAN International Conference on Functional Programming (ICFP) [EB/OL].[2022-06-23].https://dl.acm. org/doi/abs/10.1145/2951913.2976746.
[10] BOLEI Z，HANG Z，XAVIER P, et al. Scene Parsing through ADE20K Dataset.Computer Vision and Pattern Recognition (CVPR) [EB/OL].[2022-06-23].http://groups.csail.mit.edu/vision/
datasets/ADE20K/.
作者简介：杨颖（1997—），女，汉族，湖北黄冈人，硕士研究生在读，研究方向：数据分析与信息处理。