基于CT影像组学预测肾透明细胞癌与集合管闰细胞起源肿瘤的研究  

Radiomics Model for Predicting the Clear Cell Renal Cell Carcinoma and Eosinophilic Tumor of Intercalated Cell Origin Based on CT Imaging

作　　者：王煦[1] 洪楠[1] 钟珺文 刘雨璐 孙超[1] 刘涛[1] 曾亚奇 尹平 WANG Xu;HONG Nan;ZHONG Junwen(Department of Radiology,Peking University People's Hospital,Beijing 100044,P.R.China)

机构地区：[1]北京大学人民医院放射科,100044

出　　处：《临床放射学杂志》2023年第7期1164-1168,共5页Journal of Clinical Radiology

基　　金：国家重点研发计划(子课题)(编号:2017YFC0109003)。

摘　　要：目的探讨基于CT影像组学方法于术前预测肾透明细胞癌和嗜酸细胞腺瘤/嫌色癌的可行性。方法回顾性搜集经病理确诊的208例肾透明细胞癌和集合管闰细胞起源肿瘤灶(嗜酸细胞腺瘤/嫌色癌瘤)。依据世界卫生组织第五版(2022)泌尿和男性生殖系统肿瘤分类,选取透明细胞癌111例,嗜酸细胞腺瘤29例,嫌色癌68例,将嗜酸细胞腺瘤和嫌色癌合并为一组共97例。之后随机按照8∶2的比例分入训练组和测试组,选取出最具预测潜能的特征,采用13类建模法建立影像组学模型,最终使用AUC曲线、敏感度和准确性等评估模型的诊断效能。结果共19个特征被筛选入鉴别透明细胞癌和闰细胞起源肿瘤的模型中。得到104套模型,其中例如Z分数归一化>>高斯过程模型表现良好,在训练组AUC值0.985,F1值0.93,召回率和精确度分别为0.842和0.985,敏感度和特异度分别为0.88、0.989,准确性为0.939。在测试组,AUC和F1分值分别为0.984和0.889,召回率和精确度分别为0.88和0.941,敏感度和特异度分别为0.842、0.957,准确性为0.905。此外,另得到22套近似的高效准确预测模型。结论基于CT的影像组学模型对术前预测肾透明细胞癌和嗜酸细胞腺瘤/嫌色癌瘤有较好的效果。Objective To explore the feasibility of preoperatively predicting the clear cell Renal Cell Carcinoma(ccRCC)and Renal Oncocytoma(RO)/Chromophobe Renal Carcinoma(ChRCC)using radiomics models based on CT-enhanced imaging.Methods 208 cases of pathologically confirmed clear cell Renal Cell Carcinoma and Tumors of intercalated cell origin(RO/ChRCC)patients admitted were collected retrospectively.According to the fifth edition of the WHO classification of tumors of the urinary and male reproductive system,111 cases of ccRCC and 97 cases that Tumors of intercalated cell origin(RO:29;ChRCC:68)were selected,RO and ChRCC Combined into a Group of 97 Cases.Then they were randomly divided into the training group and the test group according to the ratio of 8:2,and Texture Features with the most characteristics of predictive potential were selected.13 types of modeling methods were used to establish radiomics models,and finally the diagnostic performance of the model was evaluated using the AUC curve,sensitivity and accuracy,and some other indicators.Results A total of 19 features were screened into the model to differentiate ccRCC from Eosinophilic tumors.A total of 104 sets of models were obtained.For example,in Z_score_scalerGaussianProcess model,the training group AUC was 0.985,and F1-Score was 0.93.The recall and precision were 0.842 and 0.985,respectively.The sensitivity and specificity were 0.88 and 0.989,respectively,and the accuracy was 0.939.In the testing group,The AUC and F1-Score were 0.984 and 0.889,respectively.The recall and precision were 0.88 and 0.941,respectively.The sensitivity and specificity were 0.842 and 0.957,respectively,and the accuracy was 0.905.In addition,we get another 22 sets of approximate efficient and accurate prediction models.Conclusion CT-based radiomics model has good effect on preoperative prediction of renal clear cell carcinoma and oncocytoma/chromophobe carcinoma.

关 键 词：肾 透明细胞癌 嗜酸细胞腺瘤 嫌色细胞癌 影像组学 预测 评估模型 

分 类 号：R737.11[医药卫生—肿瘤] R730.44[医药卫生—临床医学]