摘要
为了弥补传统的SSD算法在小目标检测中的不足,提出一种结合深度残差网络的SSD目标检测算法,用于医学影像诊断中肺结节小目标的检测识别。具体操作中,首先对肺部CT图像的切片进行预处理操作得到肺实质,通过得到的大肺实质样本对提出的方法进行训练。实验结果表明,与传统的SSD算法相比,提出的方法模型检测的敏感度为84.25%,假阳性率为10.55%,分别比传统的SSD算法在敏感度上提高了6.9%,假阳性率降低了2.7%。
In order to make up for the shortcomings of traditional SSD algorithm in small target detection,this paper combines SSD target detection algorithm with deep residual network to form a new model for detecting lung nodules of small targets.First,pre-processing the slices of lung CT images to obtain the lung parenchyma,and then input a large number of lung parenchyma samples into the model for training.Finally,the sensitivity of the model test was 84.25%,and the false positive rate was 10.55%.Compared with the traditional SSD algorithm,the sensitivity is increased by 6.9%,and the false positive rate is reduced by 2.7%.
作者
汪洋
李建锋
WANG Yang;LI Jian-feng(College of Information Science and Engineering , Jishou University, Jishou Hunan 416000, China)
出处
《佳木斯大学学报(自然科学版)》
CAS
2020年第6期96-100,117,共6页
Journal of Jiamusi University:Natural Science Edition
基金
国家自然科学基金(61962023,61562029)
湘西土家族苗族自治州科研项目(2018sf5013)。
关键词
卷积神经网络
迁移学习
肺结节
残差网络
SSD算法
convolutional neural network
transfer learning
lung nodules
residual network
SSD algorithm