一种基于改进YOLOv5s网络的结直肠腺瘤实时检测方法  

A real-time method for colorectal adenoma detection based on an improved YOLOv5s network

作　　者：刘爽[1] 田兆星 李浩然 常颖[2] 吴思蓓 薛林雁[1] LIU Shuang;TIAN Zhaoxing;LI Haoran;CHANG Ying;WU Sibei;XUE Linyan(National & Local Joint Engineering Research Center of Metrology Instrument and System, College of Quality and Technical Supervision, Hebei University, Baoding 071002, China;Department of Gastroenterology, Affiliated Hospital of Hebei University, Baoding 071000, China)

机构地区：[1]河北大学质量技术监督学院,计量仪器与系统国家地方联合工程研究中心,河北保定071002 [2]河北大学附属医院消化内科,河北保定071000

出　　处：《河北大学学报（自然科学版）》2022年第3期327-336,共10页Journal of Hebei University(Natural Science Edition)

基　　金：河北省自然科学基金资助项目(H2019201378);河北大学自然科学多学科交叉研究计划资助项目(DXK201914);河北大学校长科研基金资助项目(XZJJ201914;XZJJ201917;XZJJ201918);大中学生科技创新能力培育专项项目(22E50041D)。

摘　　要：为解决现有的基于深度学习的结直肠息肉检测算法计算复杂或检测精度较低,不能在速度和精度方面同时满足实时检测的问题,提出了一种基于单阶段目标检测网络YOLO(you only look once)v5s的结直肠腺瘤实时检测方法.在YOLOv5s的主干网中融入通道注意力机制,并以BCEWithLogitsLoss代替其原有的交叉熵损失函数BCELogits,以此提升网络性能.选取2074张腺瘤图片和包含19700帧的20段腺瘤视频,按照3∶1的比例构建结直肠腺瘤训练集和测试集.测试结果表明,结直肠腺瘤检测的平均精度为93.6%,检测速度为93帧/s,验证了该系统可以在肠镜的退镜过程中实时检测腺瘤性息肉,且具有较好的检测性能.There are two major issues in the existing deep learning-based algorithms for colorectal polyp detection which are leading to the failure of real-time detection:one is the complicate calculation and the other is poor detection precision.In this paper,we proposed a real-time method for colorectal adenoma detection based on a single-stage target detection network YOLOv5s,in which the channel attention mechanism was added to the backbone and the loss function of BCEWithLogitsLoss was adopted instead of the original one of BCELogits.We collected 2074 endoscopic adenoma images and 20 adenoma videos with 19700 frames,and then assigned them as training dataset and testing dataset according to the ratio of 3∶1.

关 键 词：计算机辅助诊断 腺瘤 卷积神经网络 YOLOv5s 实时检测 

分 类 号：TP391.7[自动化与计算机技术—计算机应用技术]