多参数MRI影像组学模型预测直肠癌Ki-67表达的价值

Value of multiparametric MRI radiomics model in predicting Ki-67 expression in rectal cancer

目的探讨基于多参数磁共振成像(MP-MRI)影像组学模型预测直肠癌Ki-67表达的价值。方法回顾分析109例直肠癌患者的Ki-67表达水平,并分为高Ki-67表达和低Ki-67表达,并且将病理组织Ki-67表达作为金标准。同时回顾性分析所有患者盆腔DCE-MRI扫描的DCE-T1WI、T2WI、DWI序列图像,提取影像组学特征。所有患者按7∶3比例分为训练组(n=73)和测试组(n=36),训练组用于特征筛选和建立影像组学模型,测试组用于验证所建立模型的可靠性。特征筛选由Spearman相关分析和最小绝对收缩与选择算子回归(LASSO)完成,应用logistic回归构建影像组学模型预测直肠癌Ki-67表达指数。绘制受试者工作特征(ROC)曲线评价模型的预测效能。通过Delong检验、重分类改善指标(NRI)和综合判别改善指数(IDI)比较不同模型间的性能差异。结果联合模型对直肠癌Ki-67指数表达具有较好的预测效能,在训练组中AUC为0.89,在测试组中AUC为0.86,而单序列模型中,DCE-T1WI模型的性能优于T2WI模型和DWI模型,在训练组和测试组中的AUC分别为0.81、0.77。结论基于动态对比增强MRI的影像组学模型可以在术前预测直肠癌Ki-67表达,其中基于DCE-T1WI+T2WI+DWI的联合模型预测效能最佳。

Objective To investigate the value of the multiparameter magnetic resonance imaging(MP-MRI)radiomics model in predicting the expression of Ki-67 in rectal cancer.Methods The expression levels of Ki-67 in 109 rectal cancer patients were retrospectively analyzed and divided into high Ki-67 expression and low Ki-67 expression.The expression of Ki-67 in pathological tissue was as the gold standard.At the same time,the DCE-T1WI,T2WI,and DWI sequence images of pelvic DCE-MRI scans of all patients were retrospectively analyzed,and radiomic features were extracted.All patients were divided into training group(n=73)and test group(n=36)according to the ratio of 7∶3.The training group was used for feature screening and establishment of the radiomics model,and the test group was used to verify the reliability of the established model.Feature screening was done by Spearman correlation analysis and least absolute shrinkage and selection operator regression(LASSO).Logistic regression was used to construct a radiomics model to predict the expression index of Ki-67 in rectal cancer.The performance differences among different models were compared by Delong test,reclassification improvement index(NRI)and integrated discriminant improvement index(IDI).Results The combined model had better predictive performance for Ki-67 index expression in rectal cancer,with AUC of 0.89 in the training group and 0.86 in the test group,while in the single sequence model,the DCE-T1WI model outperformed the T2WI model and the T2WI model.For the DWI model,the AUCs in the training and testing groups were 0.81 and 0.77 respectively.Con⁃clusion The radiomics model based on dynamic contrast-enhanced MRI can predict the expression of Ki-67 in rectal cancer before surgery,and the com-bined model based on DCE-T1WI+T2WI+DWI has the best predictive performance.