腮腺多形性腺瘤与腺淋巴瘤的放射组学研究

Radiomic study of pleomorphic adenoma and adenolymphoma in parotid gland

目的研究放射组学在腮腺多形性腺瘤与腺淋巴瘤中的应用价值。方法回顾性分析44例腮腺多形性腺瘤与40例腮腺腺淋巴瘤增强CT图像,采用纹理分析软件MaZda对肿瘤最大层面感兴趣区(ROI)进行分析,用费希尔系数(Fisher)、分类误差概率与平均相关系数(POE+ACC)、互信息(MI)分别筛选10个最佳纹理参数,再结合B11模块中的原始数据分析(RDA)、主成分分析(PCA)、线性判别分析(LDA)、非线性判别分析(NDA)对肿瘤进行判定,以最小错误判别法作为基准,计算错误率、灵敏度、特异度、准确度。结果纹理参数均值(Mean)在3种降维方法中均被选出;在纹理判别分析中,Fisher/RDA和Fisher/PCA的错误率最低(3.57%);在预测腮腺肿瘤类别时,灵敏度和准确度最高的是Fisher/RDA和Fisher/PCA,灵敏度均为97.73%,准确度均为96.43%,特异性最高为MI/NDA(97.5%)。结论放射组学能够用于区分腮腺多形性腺瘤与腺淋巴瘤。

Objective To study the application value of radiomics in the pleomorphic adenoma and adenolymphoma in the parotid gland.Methods The contrast-enhanced CT images of 44 cases with pleomorphic adenoma in parotid gland and 40 cases with adenolymphoma in parotid gland were retrospectively analyzed.The region of interest(ROI)at the maximum level of the tumor was analyzed using MaZda,a texture analysis software,and 10 optimal texture parameters were screened out using Fisher’s coefficient(Fisher),probability of classification and average correlation coefficient(POE+ACC),and mutual information(MI).Then the tumor was determined through original data analysis(RDA),principal component analysis(PCA),linear discriminant analysis(LDA),and non-linear discriminant analysis(NDA)in the B11 module.The minimum error entropy was used as the criterion to calculate the error rate,sensitivity,specificity,and accuracy.Results The mean value of the texture parameter(Mean)was selected in three methods of dimensionality reduction.In texture discriminant analysis,Fisher/RDA and Fisher/PCA generated the lowest error rate(3.57%).When predicting the type of parotid tumor,Fisher/RDA and Fisher/PCA presented the highest sensitivity(97.73%)and the highest accuracy(96.43%),and MI/NDA presented the highest specificity(97.5%).Conclusion Radioomics can be used to differentiate pleomorphic adenoma from adenolymphoma in the parotid gland.