CT纹理分析在鉴别腰椎多发性骨髓瘤与骨质疏松中的价值
The value of CT texture analysis in differentiating lumbar multiple myeloma from osteoporosis
朱心雨 1郭立 1张雨柔 1黄鹏1
作者信息
- 1. 昆明医科大学第二附属医院放射科 云南 昆明 650000
- 折叠
摘要
目的 探讨CT纹理分析在鉴别腰椎多发性骨髓瘤(MM)与骨质疏松中的价值.方法 选取 40 例腰椎多发性MM(MM组)与腰椎骨质疏松 40 例(骨质疏松组)患者的资料进行分析,基于术前CT轴位图像,分别选择一个MM椎体和骨质疏松椎体勾画感兴趣区(ROI),利用MaZda软件提取图像纹理特征,用费希尔算法(Fisher)、分类错误率+平均相关系数算法(POE+ACC)、交互信息算法(MI)三种方法对提取的特征参数进行降维筛选,其每种方法筛选出 10 个最优纹理特征参数,然后利用原始数据分析法(RDA)、主要成分分析法(PCA)、线性判别分析法(LDA)、非线性判别分析法(NDA)四种分析方法分别结合降维筛选出的特征参数,计算各组合的分类误判率.对 30 个最优纹理特征参数进行统计学意义分析,获取具有统计学意义参数的受试者工作(ROC)曲线.对有意义的参数行Spearman相关分析做进一步筛选,筛选出的参数构建Logistic回归模型.结果 三种降维方法中,POE+ACC结合NDA组合对两种疾病的误判率最低(2.50%),15 个纹理特征具有统计学意义.经筛选后获得 135dr_GlevNonU、Perc.99%、S(0,4)Correlat和WavEnLH_s-5的 4 个参数构建模型,ROC曲线分析表明该模型诊断效果较好,曲线下面积(area under the curve,AUC)为 0.968.结论 基于 CT 图像的纹理分析有助于鉴别 MM 与骨质疏松,且 135dr_GlevNonU,Perc.99%,S(0,4)Correlat 和WavEnLH_s-5 构建的联合模型诊断效果较好.
Abstract
Objective To investigate the value of CT texture analysis in differentiating lumbar multiple myeloma(MM)and lumbar spine osteoporosis.Methods We retrospectively analyzed the data of 40 patients with MM and 40 patients with lumbar spine osteoporosis,selected one myeloma vertebra and osteoporotic vertebra respectively to outline the region of interest based on preoperative CT axial images,and extracted the image texture features using MaZda software.Fisher coefficient,classification error probability and average correlation coefficients,as well as mutual information coefficient were used to reduce the dimension-ality of the extracted feature parameters.10 optimal texture parameters were selected by each method.Then,the four methods of raw data analysis,principal component analysis,linear discriminant analysis,and nonlinear discriminant analysis were used to combine the extracted feature parameters with the dimensionality reduction screening and the classification misclassification rate of each combination was calculated.Statistical significance analysis was performed on the 30 optimal texture feature parameters to obtain subject operator characteristic curves with statistically significant parameters.Spearman correlation analysis was per-formed on the significant parameters for further screening,and logistic regression models were constructed for the screened pa-rameters.Results Among the three downscaling methods,the combination of POE plusACC combined with NDA had the low-est false positive rate(2.50%)for both diseases,and 15 texture features were statistically significant.The four parameters,i.e.,135dr_GlevNonU,Perc.99%,S(0,4)Correlat,and WavEnLH_s-5 were obtained after screening to construct the model,and ROC curve analysis showed that the model was more effective for diagnosis,with an area under the curve of 0.968.Conclusion Texture analysis based on CT images can help to identify MM and osteoporosis,the combined model constructed by 135dr_GlevNonU,Perc.99%,S(0,4)Correlat and WavEnLH_s-5 has a better diagnostic efficacy.
关键词
纹理特征/骨髓瘤/骨质疏松/体层摄影术,X线计算机Key words
Textural features/Myeloma/Osteoporosis/Tomography,X-ray computed引用本文复制引用
基金项目
云南省卫生健康委医学学科带头人培养计划(D-2019024)
出版年
2024
NETL
NSTL
万方数据