基于U-net的结直肠息肉诊断方法的探索与发现  

Exploration and Discovery of Colorectal Polyp Diagnosis Methods Based on U-Net

作　　者：陆万里 陈欣 张琴 Lu Wan-li;Chen Xin;Zhang Qin(Chongqing University of Engineering,Chongqing 400056,China)

机构地区：[1]重庆工程学院,重庆400056

出　　处：《科学与信息化》2025年第4期160-162,共3页Technology and Information

基　　金：大学生创新创业训练计划项目,项目名称:基于磁共振成像的膀胱肿瘤分级和分期预测诊断,项目编号:S202412608012。

摘　　要：旨在提高早期结直肠癌的诊断准确性,采用的技术路线清晰划分为数据获取、清洗、建模和测试4个关键步骤;创新性地应用了数据增强和图像分割技术,结合迁移学习模型U-net,显著提升了检测模型的性能。测试结果表明,模型训练达到了极低的损失值(趋近于0.045)和高准确度(0.9990)。此外,还有效解决了数据标签不一致、数据集不平衡和模型选择等关键问题,最终实现了对CT影像中息肉的精确识别和定位。This study aims to improve the diagnostic accuracy of early-stage colorectal cancer.The technical approach is clearly divided into four key steps:data acquisition,data cleaning,modeling,and testing.Innovatively,data augmentation and image segmentation techniques are applied in combination with the transfer learning model U-Net,which significantly enhances the performance of the detection model.The test results show that the model training achieved an extremely low loss value(approaching 0.045)and high accuracy(0.9990).Additionally,the study effectively addresses key issues such as inconsistent data labeling,imbalanced datasets,and model selection,ultimately enabling precise identification and localization of polyps in CT images.

关 键 词：结直肠息肉检测 图像分割算法 U-net模型 

分 类 号：R73[医药卫生—肿瘤]