摘要
目的通过分析胰腺神经内分泌肿瘤(pancreatic neuroendocrine neoplasm,pNEN)增强CT纹理特征,预测其分化程度。方法将102例pNEN患者的CT资料导入ITK-SNAP软件绘制感兴趣区并提取动静脉期纹理特征。运用R软件行最优化处理,采用Pearson及Mann-Whitney U检验去除冗余特征,再利用向后逐步回归法建立最佳模型。绘制特征曲线(receiver operating characteristic curve,ROC),计算曲线下面积(the area under curve,AUC)、准确度、灵敏度及特异度。结果筛选出动脉期有效特征2个,静脉期有效特征1个,联合动静脉期资料共同分析,筛选出有效特征9个。单独动脉期有效特征建立模型1,曲线下面积为0.776(准确度0.737,灵敏度0.814,特异度0.625);单独静脉期有效特征建立模型2,曲线下面积为0.753(准确度0.710,灵敏度0.683,特异度0.750);联合动静脉期纹理特征共同分析建立模型3,曲线下面积为0.825(准确性0.768,灵敏度0.847,特异度0.650)。结论增强CT多个纹理参数在不同病理级别的pNEN间有显著差异,可用于预测其病理分级,联合动静脉期参数共同分析,其曲线下面积、准确性及灵敏性明显高于单一期参数。
Objective To explore the predictive efficacy of enhanced CT texture features in the pathological grades of pancreatic neuroendocrine neoplasms(PNENs).Methods A total of 102 cases of pancreatic neuroendocrine neoplasms,diagnosed by postoperative pathology,were retrospectively enrolled in our study.Arterial and portal venous image data were derived from PACS system in DICOM format,the region of interest(ROI)was delineated with ITK Snap software,and then the texture parameters were extracted in A.K.software.Feature dimension reduction was performed by using R software(version 4.0.3).The redundant features were screened out with Pearson correlation analysis and Mann-whitney U test.Then the optimal model was constructed according to Akaike Information Criterion(AIC)with backward stepwise regression method.The receiver operating characteristic(ROC)curves of arterial phase,portal venous phase and arterial combined with portal venous phase were drawn,and the area under curve(AUC),accuracy,sensitivity,specificity were calculated.Results Two texture features of arterial phase,one texture feature of portal venous phase and nine texture parameters of arterial combined with venous phase were selected respectively.Arteria model(model 1),portal venous model(model 2)and arterial combined with portal venous model((model3))were constructed.The area under the curve,accuracy,sensitivity,specificity in model 1 were 0.770.737,0.814,0.625,respectively.The area under the curve,accuracy,sensitivity,specificity in model 2were0.753,0.710,0.683,0.750,respectively.The area under the curve,accuracy,sensitivity,specificity in model 3were 0.825,0.768,0847,0.650,respectively.Conclusion Enhanced CT texture features in the arterial and portal venous stages showed certain efficacy in predicting the pathological grading of pancreatic neuroendocrine tumors.Combined with the texture data of arterial and portal venous stages,the predictive efficacy can be further improved.
作者
刘群
单海荣
罗一烽
史红媛
徐青
张艳
LIU Qun;SHAN Hai-Rong;LUO Yi-Feng;SHI Hong-Yuan;XU Qing;ZHANG Yan(Department of Radiology,the Affiliated Yixing Hospital of Jiangsu University,Yixing 214200,Jiangsu Province,China;Department of Medical Imaging,the First Affiliated Hospital Of Nanjing Medical University,Nanjing 210029,Jiangsu Province,China;Department of Gastroenterology,the Affiliated Yixing Hospital of Jiangsu University,Yixing 214200,Jiangsu Province,China)
出处
《中国CT和MRI杂志》
2023年第12期122-124,共3页
Chinese Journal of CT and MRI
基金
江苏大学临床医学科技发展基金资助项目(JLY2021042)。
关键词
增强CT
纹理特征
胰腺
神经内分泌肿瘤
病理分级
Enhanced CT
Texture Features
Pancreas
Neuroendocrine Neoplasm
Pathological Grade
