基于CT纹理分析在鉴别结直肠癌旁肿瘤沉积与转移淋巴结中的价值

The Value of CT Texture Anglysis in Differentiating Tumor Deposition and Metastatic Lymph Nodes Adjacent to Colorectal Cancer

目的探讨基于CT图像的3D纹理分析在鉴别结直肠癌旁肿瘤沉积与转移淋巴结中的诊断价值。方法回顾分析经手术病理证实且CT图像上可辨识的结直肠癌旁肿瘤沉积灶45枚,转移淋巴结灶45枚,全腹部CT扫描均在术前1周内完成。通过MaZda软件对静脉期增强轴位图像进行3D纹理分析,提取各个病灶的3D纹理参数共794个特征值,联合软件内置费尔系数(Fisher)、分类错误概率联合平均相关系数(POE+ACC)和交互信息(MI)3种方法的(FPM)进行特征降维得到30个最优纹理参数,并由内置的B11模块的原始数据分析(RDA)、主成分分析(PCA)、线性分类分析(LDA)和非线性分类分析(NDA)对这些纹理参数进行分类分析,各纹理特征参数的诊断结果以误判率形式显示;采用SPSS 19.0软件比较TD与MLN的30个最优纹理参数间的统计学差异,对差异有统计学意义的纹理参数进行ROC曲线分析并计算AUC并比较其诊断效能。结果肿瘤沉积组及转移淋巴结组纹理特征分类中,NDA对两种病变的误判率(4.44%~10.00%)较RDA(15.5%~26.67%)、PCA(16.67%~31.11%)和LDA(10.00~30.00%)均低。基于螺旋CT静脉期增强图像的三维纹理特征,采用NDA的误判率最低,其中FPM联合NDA的误判率最低,误判率为4.44%。在肿瘤沉积组与转移淋巴结组30个最优纹理参数中,30个最优纹理参数进行ROC曲线分析中,有14个特征AUC>0.70,其中>0.9有1个。结论基于螺旋CT静脉期增强图像的3D纹理分析对结直肠癌旁肿瘤沉积及转移淋巴结的鉴别诊断具有很好的客观的量化证据。

Objective To explore the diagnostic value of 3 D texture analysis based on CT images in the diagnosis of tumor deposits and metastatic lymph nodes adjacent to colorectal cancer.Methods A retrospective analysis of 45 colorectal tumor deposits and 45 metastatic lymph node foci that were confirmed by surgery and pathology and identifiable on CT images were performed.CT scans of the entire abdomen were completed within 1 week before surgery.Used MaZda software to perform 3 D texture analysis on the venous phase enhanced axial image,extract 3 D texture parameters of each lesion,a total of 794 eigenvalues,combined with the software built-in Fair coefficient(Fisher),classification error probability and combined average correlation coefficient(POE+ACC).The 30 optimal texture parameters were obtained by feature reduction and the three methods of interactive information(MI)(FPM),the original data analysis(RDA),principal component analysis(PCA),linear classification analysis(LDA)and non-linear classification analysis(NDA)wereclassifed and analyzed in these texture parameters,and the diagnosis results of each texture feature parameter were displayed in the form of misjudgment rate.SPSS 19.0 software was used to compare the statistics between the 30 optimal texture parameters of TD and MLN.We analyzed the differences between the two groups and the ROC curve of the texture parameters with statistical significance,then calculated the AUC,and compared the diagnostic power.Results compared with RDA(15.5%-26.67%),PCA(16.67%-31.11%)and LDA(10.00%-30.00%),the false rate of NDA was lower in TD group and mln group.Based on the three-dimensional texture features of spiral CT veno phase enhanced images,the rate of misjudgment using non-linear classification analysis(NDA)is the lowest,and the rate of misjudgment using FPM combined with non-linear classification analysis(NDA)is 4.44%.Among the 30 optimal texture parameters of TD group and mlngroup,there are 14 features AUC>0.70 in the ROC curve analysis of 30 optimal texture parameters,and 1 feature AUC>0.9.Conclusion 3 D texture analysis based on the enhanced image of spiral CT in venous phase has a good objective quantitative evidence for the differential diagnosis of tumor deposition and metastasis in colorectal cancer.