集成机器学习构建肾透明细胞癌免疫影像分型及其对生存预测的价值研究
DOI:
作者:
作者单位:

1.南京医科大学第一附属医院江苏省人民医院放射科;2.南京医科大学第一附属医院江苏省人民医院泌尿外科

作者简介:

通讯作者:

中图分类号:

基金项目:

国家自然科学基金项目


Ensemble Machine Learning-based Immune-Radiomic Phenotype in Predicting Survival of Clear Cell Renal Cell Carcinoma
Author:
Affiliation:

Department of Radiology,the First Affiliated Hospital of Nanjing Medical University

Fund Project:

The National Natural Science Foundation Project

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    目的:基于集成机器学习及增强CT构建肾透明细胞癌(clear cell renal cell carcinoma, ccRCC)免疫影像分型(immuno-radiomics phenotype, ImRad)并探讨其对患者预后生存的预测价值。方法:收集癌症基因图谱(The Cancer Genome Atlas,TCGA)数据库113例ccRCC患者的临床、影像及RNAseq表达信息,提取全肿瘤影像组学特征,基于集成机器学习(machine learning, ML)进行特征筛选并构建免疫浸润、肿瘤突变负荷、免疫耗竭相关基因的ImRad。进一步多因素Cox回顾分析ImRad对患者总生存(overall survival, OS)的预测效能。结果:经ML构建30个ImRad分型中,经五折法验证,基于朴素贝叶斯算法对肿瘤免疫微环境预测效能最佳(曲线下面积0.724-0.956)。113例ccRCC患者中位OS为31(10~54)月,与基于临床、病理及ImRad单模态指标对比,融合临床-病理及ImRad对OS预测效能最佳(C指数:融合模型0.938,组学模型 0.756,临床模型0.924)。ImRad特征中,RadMast_cells_activated等8个特征是OS的独立预测因子。结论:基于集成机器学习及CT组学分析可预测ccRCC免疫微环境并提高患者术后生存的预测效能。

    Abstract:

    Objective: To construct hybrid immune-radiomic (ImRad) phenotypes of clear cell renal cell carcinoma (ccRCC) based on ensemble machine learning (ML) and contrast-enhanced CT, and to investigate its predictive value for survival. Methods: The clinical, CT imaging and RNAseq information of 113 ccRCC patients were collected from TCGA database. Radiomic features were extracted from whole tumor. ImRad predictors were constructed on tumor immuneinfiltration, tumor mutational burden, immune exhaustion gene expression after feature selection based on ensemble ML. Further, the predictive value of ImRad for overall survival (OS) was assessed using multivariate Cox regression analysis. Results: Among the 30 ImRad contructed by ensemble ML and validated by 5 folds cross validation, Naive Bayes algorithm achieved the generally best performance (area under the receiver operating characteristic curve was from 0.724 to 0.956). The median OS of 113 patients was 31 (10 to 54) months. Clinicopathologic immune comprehensive model on predicting OS achieved best performance, surpassing single modality indicators based on clinical, pathological and ImRad (C index: 0.938, 0.756 and 0.924, respectively). Among ImRad features, RadMast_cells_activated was one of eight independent predictive factors for patients’ prognosis. Conclusions: CT radiomics based on ensemble ML can predict immune microenvironment and improve the prediction efficiency of postoperative survival of ccRCCs.

    参考文献
    相似文献
    引证文献
引用本文
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2023-03-01
  • 最后修改日期:2023-05-16
  • 录用日期:2023-08-29
  • 在线发布日期:
  • 出版日期: