基于增强CT构建鉴别肾透明细胞癌ISUP分级的神经网络模型

韩冬; 张喜荣; 贾永军; 任革; 吕蕊花; 史琳娜; 贺太平

doi:10.3971/j.issn.1000-8578.2021.20.0440

基于增强CT构建鉴别肾透明细胞癌ISUP分级的神经网络模型

A Neural Network Model Based on Enhanced CT for Distinguishing ISUP Grade of Clear Cell Renal Cell Carcinoma

摘要

摘要:
目的基于增强CT构建鉴别肾透明细胞癌（ccRCC）ISUP分级的神经网络模型。
方法收集本单位病理确诊的ccRCC患者131例，ISUP低级别92例、高级别39例。按5:5分层抽样将患者分为训练集和验证集。由影像科医师对ccRCC增强CT图像进行评价。对患者一般特征及增强CT特征采用递归特征消除（RFE）进行降维，用于神经网络建模及验证。
结果患者一般特征及增强CT特征经RFE后降维为14个特征，重要性排序前5的特征为生长方式、坏死、区域淋巴结肿大、肿瘤大小及假包膜。基于该5个特征构建的神经网格模型在训练集鉴别低、高级别ccRCC的AUC为0.8844（95%CI: 0.8062~0.9626），敏感度为89.47%，特异性为82.61%。验证集中的AUC为0.7924（95%CI: 0.6567~0.9280），敏感度为75.00%，特异性为86.96%。
结论基于增强CT构建ccRCC ISUP分级的神经网络模型有较好的诊断效能。

Abstract:
Objective To establish a neural network model based on enhanced CT for distinguishing ISUP grade of clear cell renal cell carcinoma (ccRCC).
Methods We collected 131 cases of ccRCC, with 92 cases of low ISUP grade and 39 cases of high ISUP grade. Patients were divided into training set and validation set according to 5:5 stratified sampling. The enhanced CT images of each ccRCC patient were evaluated by the radiologist. Recursive feature elimination (RFE) was used to reduce the dimension of patients' general features and enhanced CT features, which was used for neural network modeling and validation.
Results Patients' general features and enhanced CT features were verified by RFE method and then reduced to 14 features. The top 5 features were growth pattern, necrosis, enlargement of lymph nodes, tumor size and capsule. The AUC of the neural network model based on these 5 features in training set was 0.8844 (95%CI: 0.8062-0.9626), sensitivity was 89.47% and specificity was 82.61%; and those in validation set were 0.7924 (95%CI: 0.6567-0.9280), 75.00% and 86.96%, respectively.
Conclusion The neural network model of ccRCC ISUP grade based on enhanced CT has relatively high diagnostic efficiency.

HTML全文

参考文献(20)

施引文献

资源附件(0)