食管鳞状细胞癌病理图像特征预后价值分析

Application of pathological image features of 73 esophageal squamous cell carcinoma patients in predicting the prognosis

目的探讨食管鳞状细胞癌(ESCC)患者病理图像特征与临床病理特征及生存期的关系,以及在ESCC预后诊断中的价值。方法选取2009-01-01-2010-12-31山西省肿瘤医院ESCC患者73例为研究对象,使用简单随机化方法分为训练组(51例)和验证组(22例)。免疫组织化学法检测患者肿瘤组织中氨基甲酰磷酸合成酶Ⅱ、天冬氨酸转氨甲酰酶和二氢乳清酸酶(CAD)蛋白的表达水平,并采集病理图像,Python编程提取图像特征,单因素方差分析和最小绝对收缩和选择算子(LASSO)回归筛选特征、构建预后模型并验证。Cox回归分析预后相关因素,构建组合模型列线图,评估其对ESCC预后的预测能力。结果训练组CAD质(χ^(2)=20.011)和CAD核(χ^(2)=21.400)高和低表达患者的生存率差异无统计学意义,均P>0.05;验证组CAD质(χ^(2)=20.949)和CAD核(χ^(2)=20.458)高和低表达患者生存率差异无统计学意义,均P>0.05。训练组中,高图像分数和低图像分数患者生存率分别为8.0%和42.3%,差异有统计学意义,χ^(2)=210.476,P=0.001;验证组中,高图像分数和低图像分数患者生存率分别为33.3%和100.0%,差异有统计学意义,χ^(2)=24.183,P=0.041。Cox回归分析显示,图像分数是ESCC独立预后风险因素,采用图像分数和传统TNM分期构建组合模型列线图。训练组中,TNM分期、图像分数和组合模型的一致性指数(C-index)分别为0.602(95%CI为0.507~0.696)、0.664(95%CI为0.556~0.772)和0.664(95%CI为0.556~0.772);验证组中,TNM分期、图像分数和组合模型的C-index分别为0.560(95%CI为0.421~0.699)、0.647(95%CI为0.466~0.828)和0.668(95%CI为0.488~0.849)。图像分数和组合模型的预测性能优于传统TNM分期。结论由CAD免疫组化的组织病理学图像提取的图像特征在ESCC中具有潜在的预后价值,基于图像分数和TNM分期构建的列线图可作为ESCC的预后模型。

Objective To explore the relationship of pathological image features and clinical pathological information with the prognosis of ESCC.Methods The 73 ESCC patients in Shanxi Provincial Cancer Hospital from January 12009 to December 312010 were selected for the study,and were divided into training group(n=51)and validation group(n=22)by using a simple randomization method.Immunohistochemistry was performed to detect the expression of Carbamoyl-Phosphate Synthetase 2,Aspartate Transcarbamylase,and Dihydroorotase(CAD),and pathological images were collected.Python was used to extract image features.The one-way analysis of variance and least absolute shrinkage and selection operator(LASSO)regression were utilized to select features,then to construct and validate the prognostic model.The Cox regression was used to analyze the prognostic factors,then construct the nomogram and evaluate its predictive performance in ESCC.Results In the training group,there were no statistically significant difference in survival rates of patients between CAD cytoplasm and CAD nucleus high or low expression(χ^(2)values were 20.011 and 21.400,both P>0.05).In the validation group,there were no statistically significant difference in survival rates of patients between CAD cytoplasm and CAD nucleus high or low expression(χ^(2)values were 20.949 and 20.458,both P>0.05).In the training group,the survival rates of patients with high image-score and low image-score were 8.0%and 42.3%,which the difference was statistically significant(χ^(2)=210.476,P=0.001).In the training group,the survival rates of patients with high image-score and low image-score were 33.3%and 100.0%,the difference was statistically significant(χ^(2)=24.183,P=0.041).However,Cox regression analysis showed that image-score was an independent prognostic risk factor for ESCC.Image-score and TNM stage were used to construct nomogram.The concordance index(C-index)for TNM stage,image-score and combined model in the training group were 0.602(95%CI:0.507-0.696),0.664(95%CI:0.556-0.772),0.664(95%CI:0.556-0.772).The concordance index(C-index)for TNM stage,image-score and combined model in the validation group were0.560(95%CI:0.421-0.699),0.647(95%CI:0.466-0.828),0.668(95%CI:0.488-0.849).The predictive performance of the image-score and combined model was better than TNM stage.Conclusions The pathological image features extracted from CAD immunohistochemical staining pictures are potential for the ESCC prognosis.The nomogram constructed on the image-score and TNM stage can be used as a predictive model for prognosis of ESCC.