基于PET/CT的影像组学联合模型预测结直肠癌肝转移的价值  

The Value of A Radiomics-Based Combined Model Using PET/CT in Predicting Liver Metastasis in Colorectal Cancer

作　　者：周军胜 郑静 徐阿巧 肖文波[3] 赵振华[2] ZHOU Junsheng;ZHENG Jing;XU Aqiao(Shaoxing Central Hospital,Shaoxing,Zhejiang Province 312030,P.R.China)

机构地区：[1]浙江省绍兴市中心医院医共体总院,312030 [2]浙江大学绍兴医院(绍兴市人民医院) [3]浙江大学医学院附属第一医院,杭州

出　　处：《临床放射学杂志》2024年第12期2101-2107,共7页Journal of Clinical Radiology

摘　　要：目的 探讨PET/CT联合模型预测结直肠癌肝转移的价值。方法 回顾性分析原发性结直肠癌患者73例,另纳入前瞻性验证队列53例患者。根据术后病理结果或MRI增强扫描将患者分为肝转移组和非肝转移组。在PET/CT影像中手动勾画病变,从预处理后的感兴趣中提取影像组学特征。利用组内相关系数分析、方差分析和Pearson相关性分析初步筛选稳定可靠的特征。然后通过结合多种特征选择手段,即最小冗余度最大相关度算法(mRMR)和最小绝对值收缩和选择算子(LASSO),在训练集上逐步筛选CT和PET组学特征。基于筛选后的特征,利用3种机器学习算法分别建立多个单模态影像组学标签。最后,LR算法对2个单模态最优组学标签进行联合建模,以探索最优的PET/CT结直肠癌肝转移影像组学联合模型。结果 在单模态影像组学模型中,CT和PET相关模型均分别纳入10个影像组学特征进行建模。其中,基于LR算法建模得到最优CT影像组学模型(MLR_CT),测试集上AUC、敏感度和特异度分别为0.701、66.7%和58.3%,前瞻性验证队列上AUC、敏感度和特异度分别为0.757、62.5%和62.1%。基于SVM算法建模得到最优PET影像组学模型(MSVM_PET),测试集上AUC、敏感度和特异度分别为0.811、72.7%和91.7%,前瞻性验证队列上AUC、敏感度和特异度分别为0.827、70.8%和75.9%。对于PET/CT联合影像组学模型(MCT+PET),其测试集上AUC、敏感度和特异度提升至0.92、81.8%和91.7%,前瞻性验证队列上AUC、敏感度和特异度分别提升至0.871、75.0%和72.4%。结论 PET/CT影像组学联合模型对结直肠癌肝转移患者的效能相比于单一影像模态大大提升,有望成为提高结直肠癌肝转移预测准确率的辅助工具。Objective To investigate the efficacy of a combined PET/CT radiomics model in predicting liver metastasis in patients with colorectal cancer.Methods A retrospective analysis was conducted on 73 patients with primary colorectal cancer treated at Shaoxing Central Hospital.Based on postoperative pathology results or MR enhancement tests,patients were divided into a liver metastasis group(28 cases)and a non-liver metastasis group(45 cases).Lesions were manually delineated in PET/CT images,from which radiomics features were extracted after preprocessing the area of interest.Preliminary selection of stable and reliable features was conducted using ICC analysis,analysis of univariate and variance.Then,CT and PET radiomics features were progressively selected on the training set by combining different feature selection methods(mRMR and Lasso).Based on the filtered features,multiple single-modality radiomics signatures were established using three machine learning algorithms.Lastly,the LR method was used to construct a combined model of the two optimal single-modality radiomics labels,to explore the most effective combined PET/CT radiomics model for colorectal cancer liver metastasis.Results In the single-modality radiomics models,10 radiomics features were respectively included in the CT and PET-related models for modeling.The optimal CT radiomics model based on LR(MLR_CT)achieved an AUC,sensitivity,and specificity of 0.701,66.7%,and 58.3%,respectively,on the test set.The optimal PET radiomics model based on the SVM(MSVM_PET)reached an AUC,sensitivity,and specificity of 0.811,72.7%,and 91.7%,respectively,on the test set.The combined PET/CT radiomics model(MCT+PET)attained an accuracy of 100%for all metrics on the training set,and high AUC,sensitivity,and specificity of 0.92,81.8%,and 91.7%respectively,on the test set.Conclusion The combined PET/CT radiomics model significantly enhances the efficacy in predicting liver metastasis in colorectal cancer patients compared to single imaging modalities,holding great potential a

关 键 词：结直肠癌 肝转移 正电子发射计算机断层显像/CT 影像组学 

分 类 号：R735.34[医药卫生—肿瘤] R730.44[医药卫生—临床医学]