CT形态特征、性别联合放射组学鉴别腮腺多形性腺瘤与腺淋巴瘤

Differentiation of pleomorphic adenoma and adenolymphoma of parotid gland by CT morphological features,gender and radiomics

目的探讨基于CT平扫的形态特征、性别联合放射组学模型对腮腺多形性腺瘤(PA)与腺淋巴瘤(AL)的鉴别应用。方法回顾性分析经病理证实的56例PA与49例AL的形态特征,观察分析其形状、边界、囊变、多发以及部位,提取并分析CT平扫图像中肿瘤的6种放射组学特征,包括灰度直方图(HA)、绝对梯度(AG)、灰度共生矩阵(GLCM)、自回归模型(AR)、灰度游程矩阵(GLRLM)和小波变换(WT),对两组间有统计学意义的放射组学特征参数进行筛选,分别以径向基函数核(RBFK)、多项式核(PK)和线性核(LK)对筛选后的放射组学特征建立支持向量机(SVM)分类模型并联合性别及形态特征建立联合模型,运用受试者工作特征曲线(ROC)评价诊断效能。结果最终从287个放射组学特征参数中筛出12个特征建立分类模型,以RBFK为核的分类模型诊断效能最高,对应的灵敏度、特异度、准确率及曲线下面积(AUC)分别为90.2%、82.5%、89.6%及0.883;PA以女性多见,AL以男性多见;与PA相比,AL更易多发及囊变(P<0.05);而2组间边界是否清楚、形状是否规则以及肿瘤的部位无明显差异(P>0.05)。放射组学特征联合性别及形态特征(多发与囊变)建立以RBFK为核的联合模型的灵敏度、特异度、准确率及AUC分别为95.1%、87.6%、92.8%及0.963。结论基于性别及CT形态特征联合放射组学特征建立的联合模型能够在术前对PA与AL进行有效鉴别。

Objective This study aimed to explore the differential application of pleomorphic adenoma(PA)and adenolymphoma(AL)in parotid gland on the basis of the morphological characteristics of CT plain scan and gender combined with radiomic model.Methods The morphological features of 56 cases of PA and 49 cases of AL confirmed by pathology were analyzed retrospectively.The morphological characteristics of shape,boundary,cystic degeneration,multiple occurrence,and location of the tumors were observed and analyzed.Six kinds of radiologic features of tumors in CT plain scan images were extracted and analyzed,including gray histogram,absolute gradient,gray-level co-occurrence matrix,autoregressive model,gray-level run length matrix,and wavelet transform.They were used to screen the statistically significant radiomic characteristic parameters between groups.The radial basis function kernel(RBFK),polynomial kernel(PK),and linear kernel(LK)of the support vector machine(SVM)classification model were established for the screened radiomic features.A joint model combined with gender and morphological features was also established.The receiver operator characteristic curve was used to evaluate the diagnostic efficiency of classification models and joint model.Results A total of 12 features were screened out from 287 radiomic feature parameters to establish classification models.The classification model with RBFK as core had the highest diagnostic efficiency,and the corresponding sensitivity,specificity,accuracy,and area under the curve(AUC)were 90.2%,82.5%,89.6%,and 0.883,respectively.PA was more common in women,whereas AL was more common in men.AL was more prone to multiple and cystic degeneration than PA(P<0.05).No significant difference was observed between the two groups in terms of boundary,shape,and the location of the tumor(P>0.05).The sensitivity,specificity,accuracy,and AUC of the combined model based on the radiomic characteristics of RBFK,gender,and morphological characteristics(multiple and cystic changes)were 95.1%,87.6%,92.8%,and 0.963,respectively.Con⁃clusion The combination of morphological characteristics based on radiomics characteristics,gender and morphological characteristics could effectively distinguish PA and AL before operation.