基于L1范数主成分分析网络的肝功能分级方法  

Liver function classification based on L1-norm principalcomponent analysis network

作　　者：张正言[1] 黄炜嘉[1] 奚彩萍[2] 杨魏 张惠惠 ZHANG Zhengyan;HUANG Weijia;XI Caiping;YANG Wei;ZHANG Huihui(Ocean College,Jiangsu University of Science and Technology,Zhenjiang 212100,China;School of Automation,Jiangsu University of Science and Technology,Zhenjiang 212100,China;.Department of Intervention,Jiangsu Province Hospital,Nanjing 210029,China)

机构地区：[1]江苏科技大学海洋学院,镇江212100 [2]江苏科技大学自动化学院,镇江212100 [3]江苏省人民医院介入科,南京210029

出　　处：《江苏科技大学学报（自然科学版）》2023年第5期65-71,共7页Journal of Jiangsu University of Science and Technology:Natural Science Edition

基　　金：国家自然科学基金资助项目(61901195);江苏科技大学博士科研启动基金(1032931604)。

摘　　要：针对肝功能分级方法主要基于患者血液学检查的生物化学指标,具有创伤性,且存在时效性不足等问题.考虑到肝脏CT图像与其病理组织具有一定的关联性,从肝脏CT图像入手,提出一种基于L1范数主成分分析网络(L1-PCANet)的肝功能分级方法.首先,利用L1范数主成分分析(L1-PCA)算法改进主成分分析网络(PCANet)模型中卷积核参数的学习方法,增强对离群数据和噪声的鲁棒性,进而提取出CT图像中肝脏感兴趣区域的深度层级特征,然后,在网络输出层引入等距特征映射(Isomap)算法对特征进行非线性降维,进一步去除冗余信息,最后,利用支持向量机对模型进行优化训练,实现肝功能分级.结果表明:改进模型的分级准确率、查准率、查全率、F1值分别为78.67%,78.10%,87.33%和0.8246,相比原始PCANet模型分别提高了5.81%、2.73%、9.33%和5.8%,有效地提高了肝功能分级的准确率.At present,the method of liver function classification is mainly based on the biochemical indexes of patients′hematological examinations,which is traumatic and time-sensitive.Considering the correlation between liver CT image and liver pathological tissue,a liver function classification approach based on the L1-norm principal component analysis network(L1-PCANet)is proposed starting from liver CT image.To improve the learning method of the convolution kernel parameters in principal component analysis network(PCANet),the L1-norm principal component analysis(L1-PCA)methodology with robustness to noise and outliers is employed first,and then the depth hierarchical features of the liver region of interest in the CT image are extracted.Furthermore,the isometric feature mapping approach is used in the output layer of PCANet to conduct nonlinear dimensionality reduction of the features,removing redundant information from the deep hierarchical features.Finally,the support vector machine is employed to classify liver function.The experimental findings show that the proposed L1-PCANet model′s classification accuracy,precision,recall,and F1 values are 78.67%,78.10%,87.33%,and 0.8246,respectively,which are 5.81%,2.73%,9.33%,and 5.8%higher than those of the PCANet model.The proposed method is efficient,non-invasive,and repeatable,and it can provide clinicians with rapid and accurate supplemental diagnosis.

关 键 词：肝功能分级 主成分分析网络 L1范数 等距特征映射 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]