基于改进YOLOv8s的肺结节检测算法  

Pulmonary Nodule Detection Algorithm Based on Improved YOLOv8s

作　　者：郭相均 蒋朝根[1] GUO Xiangjun;JIANG Chaogen(School of Computing and Artificial Intelligence,Southwest Jiaotong University,Chengdu 611756,China)

机构地区：[1]西南交通大学计算机与人工智能学院,四川成都611756

出　　处：《现代信息科技》2025年第7期87-92,共6页Modern Information Technology

摘　　要：肺癌已成为世界范围内死亡率较高的常见癌症之一,肺结节是肺癌的早期表现。针对CT影像中肺结节检测困难的问题,提出一种基于改进YOLOv8s的肺结节检测算法。主干网络采用改进的YOLOv8s,使用空间深度转换卷积替换传统的步长卷积和池化层,避免步长卷积和池化层在处理低分辨率图像或小物体时导致细粒度信息的丢失。加入坐标注意力模块,考虑了肺结节影像的通道间关系和位置信息,使模型能更准确地定位并识别目标区域。使用Focal Loss替换交叉熵损失函数,以解决数据集样本标签数量较少的问题。采用自适应激活函数,不仅能提升网络的稳定性,还能提高网络精度。使用LUNA16数据集验证算法性能,改进后的肺结节检测算法检测精度达到了77.8%,较Faster R-CNN和YOLOv8s检测算法分别提升了3.7%和8.6%。Lung cancer has become one of the common cancers with high mortality worldwide.Pulmonary nodules are the early manifestations of lung cancer.Aiming at the difficulty of pulmonary nodule detection in CT images,a pulmonary nodule detection algorithm based on improved YOLOv8s is proposed.The backbone network uses improved YOLOv8s,and uses Space-to-Depth Convolution to replace the traditional step-size convolution and pooling layer to avoid the loss of fine-grained information caused by step-size convolution and pooling layer when processing low-resolution images or small objects.The coordinate attention module is added to consider the inter-channel relationship and position information of the pulmonary nodule image,so that the model can locate and identify the target area more accurately.Focal Loss is used to replace the cross-entropy loss function to solve the problem of small number of sample labels in the dataset.The adaptive activation function can not only improve the stability of the network,but also improve the accuracy of the network.The LUNA16 dataset is used to verify the performance of the algorithm.The detection accuracy of the improved pulmonary nodule detection algorithm reaches 77.8%,which is 3.7%and 8.6%higher than that of Faster R-CNN and YOLOv8s detection algorithms,respectively.

关 键 词：肺结节检测 YOLOv8s 无步长卷积 自适应激活函数 空间注意力机制 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]