基于高分辨率T_(2)WI的影像组学列线图预测直肠癌脉管侵袭的研究

High resolution T_(2)WI-based radiomics nomogram for prediction of lymphovascular invasion in rectal cancer

目的:探讨高分辨率MR影像组学列线图预测直肠癌脉管侵袭(LVI)的价值。方法:回顾性入组本院2016年12月—2019年12月的直肠癌患者140例,选取病灶最大截面的斜轴位勾画病灶,提取影像组学特征。采用最大相关最小冗余对影像组学特征进行初步筛选,然后进行LASSO回归分析降维,计算影像组学标签。单因素和多因素Logistic回归分析临床特征、MR影像表现、影像组学标签与LVI的关系并构建预测LVI的模型。结果:LVI发生率约20.7%(29/140),LVI组和无LVI组两组间年龄((61.7±11.7)vs(65.8±9.6),P=0.049)、组织学分级(9/83/19 vs 8/19/2,P=0.011)、磁共振肿瘤T分期(79/32 vs 19/19,P<0.001)、磁共振肿瘤N分期(72/39 vs 11/18,P=0.011)、环周切缘状态(27/84 vs 14/15,P=0.020)、壁外血管侵犯状态(21/90 vs16/13,P<0.001)有统计学差异,余指标无统计学差异。最终预测LVI的列线图包括年龄、组织学分级、磁共振肿瘤T分期、影像组学标签,训练集曲线下面积(AUC)=0.84(95%CI:0.73~0.94),验证集AUC=0.81(95%CI:0.62~1.00)。结论:基于高分辨率MR影像组学列线图能较好的预测直肠癌LVI。

Objective:To establish and verify a radiomic nomogram for prediction of lymphovascular invasion in rectal cancer.Methods:One hundred and forty rectal cancer patients in our hospital from December 2016 to December 2019 were retrospectively enrolled.The lesions were delineated on high-resolution oblique axial T2 WI and radiomic features were extracted.Firstly,the radiomic features were initially screened by maximum correlation and minimum redundancy,then LASSO regression analysis was performed to screen the features again and radiomic signature was calculated.Univariate analysis was conducted on the radiomic features,clinical risk factors and MRI findings.Multivariate logistic analysis was carried out to in-vestigate the final feature subset and thus the predicting model was established.Results:LVI was present in 20.7%(29/140)of rectal cancer patients.There were statistically significant differences in age(61.7±11.7 vs 65.8±9.6,P=0.049),histological grade(9/83/19 vs 8/19/2,P=0.011),MRI reported T stage(79/32 vs 19/19,P<0.001),MRI reported N stage(72/39 vs 11/18,P=0.011),circumferential resection margin status(27/84 vs 14/15,P=0.020)and extramural vascular invasion(21/90 vs 16/13,P<0.001)between the LVI positive group and LVI negative group.The predictive nomogram of LVI included the age,histological grade,MRI reported T stage and radiomic signature.The AUC was 0.84(95%CI,0.73~0.94)in the training cohort,which was 0.81(95%CI,0.62~1.00)in the validation cohort.Conclusion:MRI-based radiomic nomogram could be helpful for the prediction of LVI preoperatively in rectal cancer patients.