基于CT影像组学及深度学习模型区分腮腺多形性腺瘤和基底细胞腺瘤  

To Distinguish Parotid Pleomorphic Adenoma from Basal Cell Adenoma Based on CT Radiomics and Deep Learning Model

作　　者：刘奇峰 贺克武 丁治民 姚琪 王亚奇 LIU Qifeng;HE Kewu;DING Zhimin(Department of Medical Imaging,the Third Affiliated Hospital of Anhui Medical University,Hefei,Anhui Province 230061,P.R.China)

机构地区：[1]安徽医科大学第三附属医院影像中心,合肥230061 [2]皖南医学院第一附属医院放射科,芜湖241001

出　　处：《临床放射学杂志》2025年第4期611-617,共7页Journal of Clinical Radiology

基　　金：2023年安徽省高等学校科学研究项目(自然科学类)(编号:2023AH040253)。

摘　　要：目的探讨基于CT影像组学及深度学习模型联合临床影像资料区分腮腺多形性腺瘤(PA)和基底细胞腺瘤(BCA)的价值。方法回顾性搜集经病理证实为PA和BCA的157例患者的临床及影像资料,勾画肿瘤的感兴趣区域并提取影像组学及深度学习特征,使用最小冗余最大相关以及最小绝对收缩和选择算子对特征降维和筛选,并分别计算影像组学评分和深度学习评分。采用单因素及多因素Logistic回归分析筛选独立危险因素,使用Logistic回归分别构建临床、传统组学、深度学习和联合模型,并绘制联合模型列线图。使用受试者工作特征曲线下面积评价模型的效能,DeLong检验比较各模型之间的差异,校准曲线评价模型的拟合度,决策曲线评价模型的临床净收益。结果最终筛选出12个影像组学特征和14个深度学习特征。多因素Logistic回归分析显示肿瘤最大径、影像组学评分和深度学习评分为独立危险因素。训练集和测试集中,联合模型的曲线下面积最高(分别为0.948、0.981),与临床模型(分别为0.675、0.761)、传统组学模型(分别为0.883、0.872)之间差异均有统计学意义(训练集P值分别为<0.001、0.026,测试集P值分别为0.007、0.039)。校正曲线和决策曲线显示联合模型的拟合良好且临床净收益最高。结论基于CT影像组学及深度学习模型联合临床影像资料能较准确地区分PA和BCA,可为临床诊疗提供更多新思路。Objective To investigate the value of CT-based radiomics and deep learning model combining clinical imaging data in differentiating pleomorphic adenoma(PA)from basal cell adenoma(BCA).Methods The clinical and imaging data of 157 patients with PA and BCA proved by pathology were retrospectively collected.The region of interest of tumor were outlined,and then the radiomics and deep learning features were extracted.Feature dimensionality reduction and selection were performed by using minimal-Redundancy-Maximal-Relevance and Least Absolute Shrinkage and Selection Operator.Radiomics and deep learning scores were calculated separately.Univariate and multivariate Logistic regression analyses were used to identify independent risk factors,using Logistic regression to construct clinical,traditional radiomics,deep learning and combined models,respectively,and then plot nomogram for the combined model.The performance of each model was evaluated by using the area under the receiver operating characteristic curve(AUC),using DeLong's test to compare differences between models.Calibration curves and decision curve analysis(DCA)were used to evaluate calibration and clinical net benefit,respectively.Results A total of 12 radiomics features and 14 deep learning features were selected.Multivariate Logistic regression analysis revealed tumor maximum diameter,radiomics score,and deep learning score were independent risk factors.In both the training and testing set,the AUC of the combined model was the highest(0.948 and 0.981,respectively),and has statistical significance(The P values of training set were<0.001 and 0.007,respectively;The P values of testing set were 0.026 and 0.039,respectively)with the clinical model(0.675 and 0.761,respectively)and the traditional radiomic model(0.883 and 0.872,respectively).Calibration curves and DCA showed good calibration and the highest clinical net benefit of the combined model.Conclusion The CT-based radiomics and deep learning model,combined with clinical imaging data can accurately differen

关 键 词：多形性腺瘤 基底细胞腺瘤 深度学习 影像组学 计算机断层成像 

分 类 号：R73[医药卫生—肿瘤]