基于高级语义及注意力的肺结节分割模型  

Pulmonary nodule segmentation model based on advanced semantics and attention

作　　者：丰晓钰 王明泉[1] 李磊磊 朱焕宇 李文波 谢绍鹏 FENG Xiaoyu;WANG Mingquan;LI Leilei;ZHU Huanyu;LI Wenbo;XIE Shaopeng(School of Information and Communication Engineering,North University of China,Taiyuan 030051,China)

机构地区：[1]中北大学信息与通信工程学院,山西太原030051

出　　处：《现代电子技术》2024年第5期60-64,共5页Modern Electronics Technique

摘　　要：为了能够早些发现肺结节患者,进行有效的预防和治疗,便能够大大提升肺癌患者的生存率,针对医学CT图像肺结节分割时存在异质性,会导致分割精度降低,提出一种基于高级语义及注意力的肺结节分割模型。该模型使用VGG16作为主干网络搭建U-net模型;采用金字塔池化模块(PPM),在尽可能保留原信息的情况下,将深层信息进行加强提取,得到更加丰富的高级语义信息;同时利用CA注意力机制强化重要的特征,实现空间和通道方向上的信息整合;使用Focal Loss和Dice Loss函数解决肺结节分割中前背景不均衡和难区分的问题。实验结果显示,所提出的方法在IoU、F1分数指标上较U-net分割算法分别提高了1.33%、0.95%,有效地提升了分割精度,解决了与其他组织对比度低的问题。Detecting lung nodule patients early and carrying out effective prevention and treatment can greatly improve the survival rate of lung cancer patients.However,there is heterogeneity in the segmentation of lung nodules in medical CT(computed tomography)images,which will reduce the segmentation accuracy.In view of this,a lung nodule segmentation model based on advanced semantics and attention is proposed.VGG16 is used as the backbone network to build the model U⁃net.The pyramid pooling module(PPM)is adopted to strengthen the extraction of deep information while retaining the original information as much as possible,so as to obtain more abundant high⁃level semantic information.The CA(coordinate attention)mechanism is used to strengthen important features and realize spatial and channel direction information integration.Focal loss and dice loss functions are used to solve the problem of unbalanced background and difficult distinction in pulmonary nodule segmentation.The experimental results show that the proposed method improves 1.33%and 0.95%in IoU(intersection over union)and F1⁃score metrics in comparison with the U⁃net segmentation algorithm,so it can effectively improve the segmentation accuracy and solves the problem of low contrast with other tissues.

关 键 词：深度学习 医学CT图像 肺结节分割 U-net 注意力机制 金字塔池化 损失函数 分割精度 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP391.41[电子电信—信息与通信工程]