目的:本研究旨在探讨结合影像组学和基因组学数据评估结直肠癌(colorectal cancer, CRC)预后的价值。方法:从癌症影像档案馆(The Cancer Imaging Archive, TCIA)中获取结直肠肝转移、TCGA-COAD和TCGA-READ数据集,包括225例结直肠癌患者...目的:本研究旨在探讨结合影像组学和基因组学数据评估结直肠癌(colorectal cancer, CRC)预后的价值。方法:从癌症影像档案馆(The Cancer Imaging Archive, TCIA)中获取结直肠肝转移、TCGA-COAD和TCGA-READ数据集,包括225例结直肠癌患者的影像学数据和654例基因组数据。手动勾画结直肠癌肿瘤边缘以定义感兴趣区域(the region of interest, ROI)。采用LASSO回归和10倍交叉验证进行特征选择及核心DEGs选择,后进行Kaplan Meier (K-M)生存分析和多变量COX回归分析。TCGA-COAD和TCGA-READ数据集以7:3的比例随机分为训练集和验证集。对训练集中选定的基因和放射组学特征进行多变量COX回归分析。通过前向–后向逐步回归选取关键指标。通过时间依赖ROC曲线和校准图评估验证集中的模型性能。结果:共提取5个关键特征。K-M生存分析显示,与高风险组相比,低风险组的总生存期明显更长(P P = 0.027, PTH1R: P = 0.009, UBQLNL: P = 0.008)组成的基因组学–影像组学联合模型。COX回归结果显示,以上模型的风险评分与CRC预后存在显著相关性(P Objective: This research endeavors to investigate the practical utility of combining radiomics and genetic data for assessing the prognosis of colorectal cancer (CRC). Methods: We obtained the Colorectal Liver Metastases, TCGA-COAD, and TCGA-READ datasets from The Cancer Imaging Archive (TCIA), including imaging data from 225 colorectal cancer patients and transcriptome data from 654 cases. CRC tumor margins were manually outlined to define the region of interest (ROI). LASSO regression and 10-fold cross-validation were used for feature and DEGs selection, followed by Kaplan Meier (K-M) survival analysis. and multivariate COX regression analysis. The TCGA-COAD and TCGA-READ datasets were randomly split into training and validation sets with a 7:3 ratio. We conducted multivariate COX regression analysis on the selected genes and radiomic features in the training set. Key indicators were chosen through forward-backward stepwise regression and visually presented with a forest plot. Model performance in the validation set was assessed through time-dependent ROC curves and calibration plots. Results: We’ve effectively filtered out 5 pivotal features. K-M survival analysis revealed a significantly longer overall survival in the low-risk group compared to the high-risk group (P P = 0.027, PTH1R: P = 0.009, UBQLNL: P = 0.008) was constructed using COX regression analysis on the selected radiomics features and key genes in the training set. The COX regression results indicated a significant association between these three genes, radiomics risk score, and the prognosis of CRC (P < 0.05). The calibration curve indicated a strong match between the predicted survival rate and the actual survival rate across these time points. Conclusion: The gene-radiomics combined model composed of three genes (PCOLCE2, PTH1R, and UBQLNL) demonstrates good predictive ability and application value for CRC.展开更多
文摘目的:本研究旨在探讨结合影像组学和基因组学数据评估结直肠癌(colorectal cancer, CRC)预后的价值。方法:从癌症影像档案馆(The Cancer Imaging Archive, TCIA)中获取结直肠肝转移、TCGA-COAD和TCGA-READ数据集,包括225例结直肠癌患者的影像学数据和654例基因组数据。手动勾画结直肠癌肿瘤边缘以定义感兴趣区域(the region of interest, ROI)。采用LASSO回归和10倍交叉验证进行特征选择及核心DEGs选择,后进行Kaplan Meier (K-M)生存分析和多变量COX回归分析。TCGA-COAD和TCGA-READ数据集以7:3的比例随机分为训练集和验证集。对训练集中选定的基因和放射组学特征进行多变量COX回归分析。通过前向–后向逐步回归选取关键指标。通过时间依赖ROC曲线和校准图评估验证集中的模型性能。结果:共提取5个关键特征。K-M生存分析显示,与高风险组相比,低风险组的总生存期明显更长(P P = 0.027, PTH1R: P = 0.009, UBQLNL: P = 0.008)组成的基因组学–影像组学联合模型。COX回归结果显示,以上模型的风险评分与CRC预后存在显著相关性(P Objective: This research endeavors to investigate the practical utility of combining radiomics and genetic data for assessing the prognosis of colorectal cancer (CRC). Methods: We obtained the Colorectal Liver Metastases, TCGA-COAD, and TCGA-READ datasets from The Cancer Imaging Archive (TCIA), including imaging data from 225 colorectal cancer patients and transcriptome data from 654 cases. CRC tumor margins were manually outlined to define the region of interest (ROI). LASSO regression and 10-fold cross-validation were used for feature and DEGs selection, followed by Kaplan Meier (K-M) survival analysis. and multivariate COX regression analysis. The TCGA-COAD and TCGA-READ datasets were randomly split into training and validation sets with a 7:3 ratio. We conducted multivariate COX regression analysis on the selected genes and radiomic features in the training set. Key indicators were chosen through forward-backward stepwise regression and visually presented with a forest plot. Model performance in the validation set was assessed through time-dependent ROC curves and calibration plots. Results: We’ve effectively filtered out 5 pivotal features. K-M survival analysis revealed a significantly longer overall survival in the low-risk group compared to the high-risk group (P P = 0.027, PTH1R: P = 0.009, UBQLNL: P = 0.008) was constructed using COX regression analysis on the selected radiomics features and key genes in the training set. The COX regression results indicated a significant association between these three genes, radiomics risk score, and the prognosis of CRC (P < 0.05). The calibration curve indicated a strong match between the predicted survival rate and the actual survival rate across these time points. Conclusion: The gene-radiomics combined model composed of three genes (PCOLCE2, PTH1R, and UBQLNL) demonstrates good predictive ability and application value for CRC.
