基于直肠癌系膜脂肪MRI影像组学构建的机器学习模型术前鉴别T_(2)、T_(3)期直肠癌

The machine learning model based on MRI radiomics of mesenteric fat in rectal cancer for preoperative identification of T_(2)and T_(3)stage rectal cancer

目的:通过比较不同的机器学习算法模型,建立和验证基于直肠系膜脂肪的MRI影像组学模型,用于术前鉴别T_(2)、T_(3)期直肠癌。方法:回顾性入组288例T_(2)、T_(3)期直肠癌患者的资料。分别从T2WI、表观扩散系数(ADC)、弥散加权成像(DWI)序列中病灶感兴趣区(ROI)提取放射组学特征。在使用组间一致性分析(ICC)及Pearson相关性分析降维后,采用最小绝对收缩和选择算子(LASSO)回归分析对每个序列来选择特征。然后使用逻辑回归(Logistic)、随机森林(Random Forest)、K最近邻(KNN)、决策树(Decision Tree)、朴素贝叶斯(Naive Bayes)、支持向量机(SVM)、极端梯度提升(XGBoost)7种不同的机器学习算法将LASSO回归筛选出的影像组学特征构建不同的预测模型。使用受试者工作特征曲线(ROC)下面积(AUC)来评估每个模型的性能,将最佳的机器学习模型与临床资料构建联合模型。做决策曲线分析(DCA)和校准曲线来评估联合模型的临床实用性和校准度。结果:基于Logistic算法的放射组学模型表现最稳定,训练集和测试集的AUC分别为0.876和0.807。基于MRI报告T分期和Logistic算法建立的联合模型显示出出色的辨别力,训练集和测试集的AUC分别为0.921和0.889。校准图和临床决策曲线显示出良好的临床校准度和临床实用性。结论:基于直肠系膜脂肪的MRI多序列的机器学习模型对术前鉴别T_(2)、T_(3)期直肠癌具有良好的预测性能。

Objective:To establish and validate an MRI radiomics model based on mesenteric fat by comparing different machine learning algorithm models for preoperative identification of T_(2)and T_(3)stage rectal cancer.Methods:The data of 288 patients with T_(2)and T_(3)stage rectal cancer were retrospectively enrolled.Radiomics features were extracted from the region of interest(ROI)of the lesion in T2WI,apparent diffusion coefficient(ADC),and diffusion-weighted imaging(DWI)sequences,respectively.After dimensionality reduction using between-group consistency analysis(ICC)and Pearson correlation analysis,features were selected for each sequence using minimum absolute contraction and selection operator(LASSO)regression analysis.Then,seven different machine learning algorithms including:Logistic,Random Forest,K-nearest neighbor(KNN),Decision Tree,Naive Bayes model,support vector machine(SVM),and extreme gradient boosting(XGBoost)were used to construct different prediction models of the radiomics features screened out by LASSO regression.The performance of each model was evaluated using the area under the receiver operating characteristic(ROC)curve(AUC),and the best machine learning model was combined with clinical data.Decision curve analysis(DCA)and calibration curve were performed to evaluate the clinical utility and calibration of the joint model.Results:The radiomics model based on Logistic algorithm performed the most stable,with AUC of 0.876 and 0.807 in the training and test sets,respectively.The joint model based on MRI report T staging and Logistic algorithm showed excellent discrimination,and the AUC of the training set and the test set was 0.921 and 0.889,respectively.The calibration plot and clinical decision curve showed good clinical calibration and clinical practicality.Conclusion:The MRI multi-sequence machine learning model based on mesenteric fat has good predictive performance in the preoperative identification of T_(2)and T_(3)stage rectal cancer.