基于增强MRI的亚区域影像组学模型可预测乳腺癌患者新辅助化疗后的病理完全反应  

Enhanced MRI-based subregional radiomics model can predict pathological complete response after neoadjuvant chemotherapy in breast cancer patients

作　　者：李永[1] 崔书君[1] 杨飞[1] 张凡[2] 殷晓霞 LI Yong;CUI Shujun;YANG Fei;ZHANG Fan;YIN Xiaoxia(Department of Medical Imaging,The First Affiliated Hospital of Hebei North University,Zhangjiakou 075000,Hebei,China;Department of Pathology,The First Affiliated Hospital of Hebei North University,Zhangjiakou 075000,Hebei,China)

机构地区：[1]河北北方学院附属第一医院医学影像部,河北张家口075000 [2]河北北方学院附属第一医院病理科,河北张家口075000

出　　处：《山东大学学报(医学版)》2025年第1期81-89,共9页Journal of Shandong University:Health Sciences

基　　金：河北省医学重点课题(20210091)。

摘　　要：目的基于对比增强MRI开发并验证预测乳腺癌患者新辅助化疗(neoadjuvant chemotherapy,NAC)后病理完全缓解(pathological complete response,pCR)的亚区域影像组学模型。方法回顾性纳入155例接受NAC的乳腺癌患者(训练组108名,验证组47名),采用K-means算法对每位患者的感兴趣区域进行亚区聚类,然后对每个亚区使用Pyradiomics软件包,提取患者对比增强MRI中的高维影像组学特征。利用Mann-Whitney U检验、斯皮尔曼相关系数和最小绝对值收敛和选择算子(least absolute shrinkage and selection operator,LASSO)算法进行特征筛选。筛选后的最优特征通过逻辑回归算法在训练集中建立各个亚区单模型,并采用相同的逻辑回归算法融合亚区单模型建立亚区总模型,使用列线图进行可视化,在验证集中对所有模型的效能进行验证。结果患者的临床病理因素在训练集和验证集中差异无统计学意义,而基于K-means算法患者的最佳聚类结果为5类,由5个亚区单模型融合建立的亚区总模型在训练集和验证集的受试者工作特征(receiver operating characteristic,ROC)曲线下面积(area under curve,AUC)分别为0.898(0.839~0.957)和0.828(0.695~0.961)。在验证集中,ROC曲线、临床决策曲线与校准曲线提示亚区总模型具有比亚区单模型更好的判别能力、校准能力与临床实用性。结论亚区总模型可揭示肿瘤异质性并帮助临床医生在治疗前准确预测NAC后的pCR,具有一定预测乳腺癌患者疗后预后的能力,表明人工智能在改善乳腺癌患者个体化治疗方面的潜力。Objective To develop and validate a subregional radiomics model based on contrast-enhanced MRI to predict pathological complete response(pCR)after neoadjuvant chemotherapy(NAC)in breast cancer patients.Methods A total of 155 breast cancer patients who underwent NAC were retrospectively included,with 108 in the training set and 47 in the validation set.Subregion clustering was performed for each patient's region of interest by using the K-means algorithm,and then high-dimensional radiomics features in the patients'contrast-enhanced MRI were extracted using the Pyradiomics software package for each subregion.Mann-Whitney U test,Spearman correlation coefficient and least absolute shrinkage and selection operator(LASSO)algorithm were utilized for feature screening.The filtered optimal features were used to build each subregion single model in the training set by Logistic regression algorithm,and the same algorithm was used to fuse the subregion single models to build the subregion total model,which was visualized by using the nomogram,and the effectiveness of all models was verified in the validation set.Results The clinicopathological factors of the patients were not statistically different between the training and validation sets.The best clustering results for patients based on the K-means algorithm were 5 classes,and the area under curve(AUC)of the receiver operating characteristic(ROC)of the total subarea model built from the fusion of the 5 subarea single models in the training set and validation set was 0.898(0.839-0.957)and 0.828(0.695-0.961),respectively.In the validation set,the ROC curves,clinical decision curves and calibration curves suggested that the subarea total model had better discriminative ability,calibration ability and clinical utility than the subarea single model.Conclusion The subregion total model can reveal tumor heterogeneity and help clinicians accurately predict pCR after NAC before treatment and has some ability to predict the post-treatment prognosis,highlighting the potential of arti

关 键 词：影像组学 新辅助化疗 乳腺癌 预测模型 亚区分析 

分 类 号：R737.9[医药卫生—肿瘤]