基于增强CT门静脉期影像组学模型对局部进展期结直肠癌新辅助化疗疗效的预测研究

Predictive of a radiomic model based on enhanced CT portal vein images for predicting neoadjuvant chemotherapy in patients with locally advanced colorectal cancer

目的探讨基于CT图像的影像组学模型预测局部进展期结直肠癌(LACRC)患者新辅助化疗(NAC)疗效的潜在价值。方法回顾性分析。纳入2014年1月—;2019年9月山西省肿瘤医院接受手术治疗、术前均行NAC的局部进展期结直肠腺癌患者181例,其中男96例、女85例,年龄23~85岁。181例患者按7∶;3的比例随机分为训练集(127例)、验证集(54例),依据肿瘤退缩分级(TRG)标准分为疗效反应良好组(TRG 0~1级,81例)、反应不良组(TRG 2~3级,100例)。所有患者在治疗前均行增强CT检查。提取门脉期CT图像的1037个影像组学特征,通过以最小绝对收缩与选择算子算法(LASSO)为主的四步法进行特征降维,然后采用多因素logistic回归对筛选出的特征构建影像组学模型;在训练集中,通过单因素及多因素logistic回归筛选预测LACRC患者NAC疗效的临床病理独立危险因素并构建临床模型;联合临床病理独立危险因素及影像组学特征构建融合模型并绘制列线图。绘制受试者工作特征曲线(ROC曲线)、校正曲线和决策曲线分析(DCA),评估各模型对LACRC患者NAC疗效的预测性能、校准性能及其临床效益。结果验证集患者的年龄大于训练集,差异有统计学意义(Z=-3.47,P<;0.05);两组患者性别分布,肿瘤临床T分期、N分期、病理分化程度,以及TRG级别等基线资料比较,差异均无统计学意义(P值均>;0.05)。训练集中,疗效反应良好组(57例)和反应不良组(70例)患者的性别以及肿瘤临床T分期、N分期、病理分化程度的差异均有统计学意义(P值均<;0.05);验证集中的反应良好组(24例)和反应不良组(30例)患者的肿瘤临床T分期、N分期的差异均有统计学意义(P值均<;0.05)。基于门脉期CT图像降维选择后共得到4个关键影像组学特征(P值均<;0.05),用于构建影像组学模型。临床模型包括临床T分期和病理分化程度2个独立危险因素(P<;0.05)。影像组学模型、临床模型和融合模型在训练集的ROC曲线下面积(AUC)分别为0.822、0.702、0.850,验证集对应的AUC分别为0.757、0.706、0.824。校正曲线分析显示,影像组学模型和融合模型均有良好的校准性能。DCA曲线分析显示,3种预测模型均有一定的临床效益,其中融合模型净收益值最大。结论基于增强CT图像的影像组学特征结合相关临床因素构建的融合模型在预测LACRC患者NAC疗效方面有一定的价值。

Objective This study aimed to investigate the potential value of a radiomic model based on computed tomography(CT)for predicting neoadjuvant chemotherapy(NAC)in patients with locally advanced colorectal cancer(LACRC).Methods The clinical and CT imaging data of 181 patients(96 males and 85 females,23-85 years old)with colorectal adenocarcinoma who underwent preoperative NAC followed by surgery in Shanxi Province Tumor Hospital from January 2014 to September 2019 were retrospectively analyzed.Using a random method,127 patients were classified into training cohort,and 54 patients were classified into validation cohort at a ratio of 7:3.These patients were divided into the good response cohort(0-1 grade,81 patients)and non-good response cohort(2-3 grade,100 patients)in accordance with the tumor regression grade(TRG)standard.All patients underwent enhanced CT examination before treatment.A total of 1037 imaging features were extracted from portal venous-phase CT images,and four steps,particularly the least absolute shrinkage and selection operator,were applied for feature extraction.Subsequently,the selected features were used to construct a radiomic model by using multivariate logistic regression.Then,clinicopathological independent risk factors were selected by using univariate and multivariate logistic regression and were used to construct a clinical model.Finally,the combined model integrating the radiomic signature and clinicopathological independent risk factors and the corresponding nomogram were constructed.Respectively,the predictive and calibration performances of the three models were evaluated by analyzing the receiver operating characteristic(ROC)curve and calibration curve analysis(CCA).Finally,decision curve analysis(DCA)was used to determine the clinical importance of the three models.Results No statistically significant differences were found in gender,clinical T stage,degree of N-stage pathological differentiation,and TRG were found between the training cohort and validation cohort(all P values>0.05).However,age showed statistically significant differences(Z=-3.47,P<0.05).In the training cohort,gender,clinical T stage,N stage,and degree of pathological differentiation between patients in the good response cohort(57 patients)and non-good response cohort(70 patients)were statistically significant(all P values<0.05).In the validation cohort,clinical T stage and N stage between patients in the good response cohort(24 patients)and non-good response cohort(30 patients)were statistically significant(all P values<0.05).Four key radiomic features derived from portal venous-phase CT images were selected for constructing the radiomic model.The clinical model included two independent risk factors,clinical T stage and pathological differentiation.The area under the ROC curve of the radiomic model,clinical model,and combined model was 0.822,0.702,and 0.850 in the training cohort and 0.757,0.706,and 0.824 in the validation cohort,respectively.The CCA showed that the radiomic model and the clinical model had good calibration.The DCA showed that the three predictive models had certain clinical importance,among which the combined model had the largest net profit.Conclusion The combined model integrating the radiomic signature based on contrast-enhanced CT and clinicopathological independent risk factors exhibit a potential value for predicting NAC outcomes in LACRC.