CT纹理分析对局部晚期胰腺癌化疗疗效的预测价值

Efficacy prediction of neoadjuvant chemoradiotherapy for local advanced pancreatic cancer:the value of CT texture analysis

目的探讨CT纹理分析对局部晚期胰腺癌(LAPC)患者化疗疗效的预测价值。方法收集2015年8月至2019年5月在浙江省人民医院肝胆胰外科行新辅助化疗治疗的LAPC患者104例。所有患者均行3次CT检查(化疗开始前1周、化疗开始第3周、化疗结束后4周)。根据化疗结果,将化疗后完全缓解、部分缓解、好转和稳定的35例患者纳入有效组(EG组),将化疗后病变进展的69例患者纳入进展组(PG组)。同时将所有患者按照7:3比例分为训练组和测试组,通过Anolysis Kit软件对训练组中确定的肿瘤区域CT的纹理特征进行提取和分析,采用logistic回归分析选择化疗前纹理特征作为结果的潜在预测因子,最后采用ROC曲线分别评判训练组和测试组的诊断准确性。结果经过计算及降维后最终得到10个特征参数。logistic回归模型评价指标的准确性、AUC、特异度、灵敏度、阳性预测值、阴性预测值分别为0.69、0.86、0.78、0.85、0.68、0.72。结论基于胰腺癌CT图像所得的纹理特征可预测LAPC的化疗疗效,可为临床实践中肿瘤生物学反应提供新的生物学标志。

Objective To analyze the texture characteristics of local advanced pancreatic cancer(LAPC)patients before chemotherapy,to establish a chemotherapeutic image classification model,and to explore its value in predicting the therapeutic efficacy.Methods FromAugust 2015 to May 2019,104 patients with pancreatic cancer undergoing neoadjuvant chemotherapy(NACT)in Department of Hepatobiliary Surgery of our hospital were enrolled.All patients received three CT enhanced scans(before chemotherapy,3 weeks after the start of chemotherapy and 4 weeks after the end of chemotherapy).The efficacy was evaluated based on the determination by MDT team and related postoperative pathology.Anolysis Kit software was used to extract and analyze the texture features of the determined tumor area on CT images.The prediction factors of CT texture features before chemotherapy were analyzed by logistic regression method.ROC curve was used to evaluate the prediction ability of texture features in pre-chemotherapy images.Results After calculation and dimensionality reduction,10 characteristic parameters were obtained.The accuracy,area under curve,specificity,sensitivity,positive predictive value and negative predictive value of single variable logistic regression model were 0.69,0.86,0.78,0.85,0.68 and 0.72,respectively.Conclusion The texture obtained from CT images of pancreatic cancer can predict the chemotherapy efficacy of local advanced pancreatic cancer,and it may provide new biological indicators for tumor biological response in clinical practice.