摘要
为了提升医学影像检测的智能化水平,文中对基于深度学习技术的相关图像处理、重构算法展开了研究。以肺部结节的自动检测为应用场景,对X光胸片的纹理特征提取方法进行研究。从灰度统计特征、灰度差异特征及多尺度高斯微分滤波器纹理特征等多个角度,提取了X光胸片的74个纹理特征作为支持向量机算法模型的输入。同时为了防止训练过程中产生的过拟合现象,解决深度学习算法对于训练样本容量的需求,提高样本数量与特征数量的比例,文中还引入了卷积稀疏编码算法对JSRT数据集进行重构,并按照1∶5的比例对算法仿真所需的数据集进行扩充。在分类器选择上,考虑到数据集中正负样本失衡对于分类器训练造成的不利影响,引入了代价敏感支持向量机算法(CS-SVM)。在公开医学影像数据集上进行的仿真结果表明,采用卷积稀疏编码进行数据集扩充后,算法的灵敏度与特异度指标可达到0.788和0.769,分别提升了2.8%和3.8%。
In order to improve the intelligent level of medical image detection,the image processing and reconstruction algorithm based on deep learning technology are studied.Taking the automatic detection of pulmonary nodules as the application scene,the texture feature extraction method of X-ray chest film is studied,and 74 texture features of X-ray chest film from gray statistical features,gray difference features and multi-scale Gauss differential filter texture features are extracted as the input of support vector machine algorithm model.At the same time,in order to prevent the over fitting phenomenon in the training process,solve the demand of deep learning algorithm for the training sample size,and improve the proportion of the number of samples and the number of features,the convolutional sparse coding algorithm to reconstruct the JSRT data set is introduced,and the data set required by the algorithm simulation according to the ratio of 1∶5 is expanded.In the selection of classifiers,considering the negative and positive sample imbalance in the data set,a Cost Sensitive Support Vector Machine(CSSVM)algorithm is introduced.The simulation results on public medical image data set show that the sensitivity and specificity of the algorithm can reach 0.788 and 0.769,which are improved by 2.8%and 3.8%respectively.
作者
王进
冯友红
WANG Jin;FENG Youhong(Medical Equipment Department,Hai’an People’s Hospital Affiliated to Nantong University,Nantong 226600,China;Imaging Department,Hai’an People’s Hospital Affiliated to Nantong University,Nantong 226600,China)
出处
《电子设计工程》
2021年第20期169-173,共5页
Electronic Design Engineering
基金
江苏省科技计划项目(20181561)。
关键词
深度学习
图像重构
特征提取
医学影像
支持向量机
卷积稀疏编码
deep learning
image reconstruction
feature extraction
medical image
support vector machine
convolutional sparse coding
