多参数3D MRI影像组学模型预测结直肠癌肝转移及相关临床危险因素的分析

Multi-parametric 3D MRI Radiomics Model for Predicting Liver Metastasis of Colorectal Cancer and Related Clinical Risk Factors Analysis

目的探讨多参数3D MRI影像组学模型对于结直肠癌肝转移(CCLM)的预测价值及相关临床危险因素的分析。方法收集2019-2022年在本院术前行MRI检查且术后病理证实的198例结直肠癌患者的临床及影像资料。按照8:2的比例随机分为训练集及验证集(n=159及39)。对T_(2)WI、DWI、e-THRIVE+序列分别逐层勾画,得到3D VOI,提取组学特征并降维处理、构建影像组学模型。采用逻辑回归分析,筛选出CCLM的独立危险因素并构建临床模型。结果三序列联合的影像组学模型预测CCLM的效能高于单一序列模型,其训练集与验证集AUC值分别为0.919及0.896。逻辑回归分析得到CCLM的独立危险因素分别为CEA、CA199及NLR(P<0.05)。三者联合的临床模型效能亦高于单一临床模型,训练集与验证集的AUC值分别为0.800及0.791。结论多参数3D MRI影像组学模型对于CCLM具有较高预测效能。基于CEA、CA199、NLR的临床模型对于CCLM具有一定预测效能。

Objective To explore the predictive value of multi-parameter 3D MRI radiomics model for colorectal cancer liver metastases(CCLM)and the analysis of related clinical risk factors.Methods From January 2019 to June 2022,the clinical and imaging data of 198 patients with colorectal cancer were collected from our hosipital who underwent preoperative MRI examination and were confirmed by postoperative pathology.It was divided into training set(n=159)and verification set(n=39)according to the ratio of 8:2.The T_(2)WI,DWI and e-THRIVE+sequences were delineated layer by layer to obtain 3D VOI,and the omics features were extracted and dimensionality reduction was performed to construct the radiomics model.The independent risk factors of CCLM were screened out by logistic regression analysis and the clinical model was constructed.Results The combined T_(2)WI,DWI and e-THRIVE+imaging model was more effective in predicting CCLM than the single sequence model.The AUC values of the training set and validation set were 0.919 and 0.896,respectively.Multivariate logistic regression analysis showed that CEA,CA199 and NLR were the independent risk factors for liver metastasis of colorectal cancer(P<0.05),and the AUC values of the clinical models constructed by the three were 0.800 and 0.791 in the training set and validation set,respectively.Conclusion Multi-parameter 3D MRI imaging model is effective in predicting liver metastasis of colorectal cancer.The clinical models based on clinical indicators CEA,CA199 and NLR have certain predictive efficacy for liver metastasis of colorectal cancer.