微创医学

目的探讨MaZda纹理分析技术预测甲状腺癌颈部淋巴结转移的临床应用价值。方法选取广西医科大学第一附属医院经病理学检查确诊的63例甲状腺癌患者为研究对象，其中46例发生颈部淋巴结转移，17例未发生颈部淋巴结转移。应用MaZda软件纹理分析技术分析甲状腺CT扫描静脉期图像。为确保分析结果的准确性，分别使用Fisher系数、分类错误率+平均相关系数（POE+ACC）算法及互信息（MI）三种不同降维方式处理图像；选用MaZda自带的分析软件B11判别图像的纹理特征，对三种降维方式分别进行原始数据分析（RDA）、主成分分析（PCA）、线性判别分析（LDA）和非线性判别分析（NDA）4种分析，获得甲状腺癌颈部淋巴结转移的误判率并进行ROC检验，以及SPSS统计分析。结果 POE+ACC+NDA组合对甲状腺癌颈部淋巴结转移的误判率最低，为5/63（7.94%）。利用MaZda软件筛选出WavEnLL_S-1、Perc.10%、Teta2和S（0，4）SumAve、S（0，5）SumAve、S（3，-3）SumAve、S（4，4）SumAve、S（4，-4）SumAve、S（5，5）SumAve、S（5，-5）SumAve共10个最优纹理特征参数。WavEnLL_S-1、Perc.10%、S（0，4）SumAve、S（0，5）SumAve、S（3，-3）SumAve、S（4， 4）SumAve、S（4，-4）SumAve、S（5，，5）SumAve、S（5，-5）SumAve纹理参数的曲线下面积（AUC）分别为0.738、0.793、0.748、0.751、0.745、0.750、0.747、0.753、0.747，Teta2的AUC为0.421。10个最优纹理参数中，除Teta2外，其他9个最优纹理参数的AUC均>0.5，差异均有统计学意义（均P<0.05）。结论利用MaZda纹理分析技术可以预测CT扫描中甲状腺癌是否发生颈部淋巴结转移，为临床的进一步诊断和治疗提供帮助。

Objective To investigate the clinical application value of MaZda texture analysis in predicting cervical lymph node metastasis of thyroid carcinoma. Methods A total of 63 patients with thyroid carcinoma diagnosed by pathological examination at the First Affiliated Hospital of Guangxi Medical University were enrolled as research subjects, among whom, 46 cases had cervical lymph node metastasis and 17 cases had no cervical lymph node metastasis. The texture analysis technology of MaZda software was applied to objectively analyze the venous phase images of thyroid CT scans. To guarantee the accuracy of the analysis results, dimensionality reduction methods such as Fisher coefficient, classification error rate + average correlation coefficient (POE+ACC) algorithm, and mutual information (MI) were utilized to process the images. The B11 analysis software built into MaZda was utilized to distinguish image texture features. For each of the three dimensionality reduction methods, four types of analyses were performed: Raw Data Analysis (RDA), Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), and Nonlinear Discriminant Analysis (NDA). And the misdiagnosis rate of cervical lymph node metastasis of thyroid carcinoma was obtained, followed by ROC curve analysis and SPSS statistical analysis. Results The combination of POE+ACC+NDA had the lowest misdiagnosis rate for cervical lymph node metastasis in thyroid carcinoma, which was 5/63 (7.94%). Ten optimal texture parameters were screened out by MaZda software, including WavEnLL_S-1, Perc.10%, Teta2, and S(0, 4)SumAve, S(0, 5)SumAve, S(3, -3)SumAve, S(4, 4)SumAve, S(4, -4)SumAve, S(5, 5)SumAve, S(5, -5)SumAve. The area under the curve (AUC) values for the texture parameters WavEnLL_S-1, Perc.10%, S(0, 4)SumAve, S(0, 5)SumAve, S(3, -3)SumAve, S(4, 4)SumAve, S(4, -4)SumAve, S(5, 5)SumAve, and S(5, -5)SumAve were 0.738, 0.793, 0.748, 0.751, 0.745, 0.750, 0.747, 0.753, and 0.747, respectively, while the AUC for Teta2 was 0.421. Among the 10 optimal texture parameters, the AUC of 9 parameters (excluding Teta2) was exceeded 0.5, and all differences were statistically significant (all P<0.05). Conclusion MaZda texture analysis can be used to predict whether cervical lymph node metastasis occurs in thyroid carcinoma on CT scans, providing assistance for further clinical diagnosis and treatment.