目的: 探讨MRI图像纹理分析在胰腺神经内分泌肿瘤分级中的应用价值。方法: 回顾性分析我院64例经手术或病理活检确诊为神经内分泌肿瘤患者的MRI图像,运用MaZda软件通过手动勾画感兴趣区（region of inte-rest,ROI）的方式提取病变的纹理特征,通过特征选择方法选出最具鉴别胰腺神经内分泌肿瘤病理分级的纹理特征,方法包括Fisher系数（Fisher coefficient,Fisher）、分类错误概率联合平均相关系数（classification error probability combined average correlation coefficients,POE+ACC）、互信息（mutual information,MI）及上述3种方法的联合法（FPM）。用特征分类统计方法,判别3种病理分级（G1、G2、G3）,结果以判错率形式表示。结果: 特征选择方法中,3种方法联合法（FPM）选择的纹理特征鉴别3种病理分级的判错率最低,为9.38%（6/64）;特征统计方法中,非线性判别分析（nonlinear discriminant analysis,NDA）区分G1、G2、G3 3种分级的判错率要低于原始数据分析（raw data analysis,RDA）、主成分分析（principal component analysis,PCA）、线性判别分析（linear discriminant analysis,LDA）。各序列中,表观弥散系数(apparent diffusion coefficient, ADC)图纹理分析判错率最低,但各序列之间的差异尚无统计学意义。结论: MRI图像纹理分析可作为胰腺神经内分泌肿瘤术前分级的辅助工具,采用Fisher系数、分类错误概率联合平均相关系数（POE+ACC）和互信息（MI）3种方法联合（FPM）的MRI图像纹理特征选择技术,结合非线性判别分析（NDA）方法,有较低的错判率。
Objective: To investigate the application of texture analysis derived from MR imaging in grading of pancreatic neuroendocrine neoplasms(PNENs). Methods: MR imagings of 64 patients with pathologically confirmed PNENs admitted to Ruijin Hospital were enrolled. Texture features were extracted from manually drawn ROIs by using MaZda software, and were selected according to the pathological grade by the feature selection methods. Statistical methods including Fisher coefficient(Fisher), classification error probability combined with average correlation coefficients(POE+ACC), mutual information(MI), and combination of above three methods(FPM) were used to classify the pathological grading of PNENs. The results were shown by misclassification rate. Results: For feature selection methods, FPM had the lowest misclassification rate. Among the statistical methods, the misclassification of NDA was lower than those of RDA, PCA, and LDA. Among the MRI sequences, the ADC map obtained the lowest misclassification rate of 9.38%(6/64), but there was no significant difference between sequences. Conclusions: Texture analysis of MR imaging can be used as an assistant tool for preoperative grading of pancreatic neuroendocrine neoplasms, and when it comes to statistical methods, FPM has the lowest misclassification rate.
[1] Fesinmeyer MD, Austin MA, Li CI, et al.Differences in survival by histologic type of pancreatic cancer[J]. Cancer Epidemiol Biomarkers Prev,2005,14(7):1766-1773.
[2] Yao JC, Hassan M, Phan A, et al.One hundred years after "carcinoid": epidemiology of and prognostic factors for neuroendocrine tumors in 35,825 cases in the United States[J]. J Clin Oncol,2008,26(18):3063-3072.
[3] 蔺武军, 毕玉田, 陈东风. 胰腺神经内分泌肿瘤的诊治进展[J]. 胃肠病学和肝病学杂志,2017,26(2):234-238.
[4] 中国临床肿瘤学会神经内分泌肿瘤专家委员会. 中国胃肠胰神经内分泌肿瘤专家共识(2016年版)[J]. 临床肿瘤学杂志,2016,21(10):927-946.
[5] Tan EH, Tan CH.Imaging of gastroenteropancreatic neuroendocrine tumors[J]. World J Clin Oncol,2011,2(1):28-43.
[6] Panzuto F, Nasoni S, Falconi M, et al.Prognostic factors and survival in endocrine tumor patients: comparison between gastrointestinal and pancreatic localization[J]. Endocr Relat Cancer,2005,12(4):1083-1092.
[8] 段小玲, 陈自谦, 许尚文, 等. 胰腺神经内分泌肿瘤的影像表现及病理分级[J]. 医学影像学杂志,2016,26(3):464-467.
[9] Klimstra DS, Modlin IR, Coppola D, et al.The pathologic classification of neuroendocrine tumors: a review of nomenclature, grading, and staging systems[J]. Pancreas,2010,39(6):707-712.
[10] Brown AM, Nagala S, McLean MA, et al. Multi-institutional validation of a novel textural analysis tool for preoperative stratification of suspected thyroid tumors on diffusion-weighted MRI[J]. Magn Reson Med,2016,75(4):1708-1716.
[11] Zacharaki EI, Wang S, Chawla S, et al.Classification of brain tumor type and grade using MRI texture and shape in a machine learning scheme[J]. Magn Reson Med,2009, 62(6):1609-1618.
[12] 陈鑫, 魏新华, 杨蕊梦, 等. 常规MRI纹理分析鉴别脑胶质母细胞瘤和单发转移瘤的价值[J]. 中华放射学杂志,2016, 50(3):186-190.
[13] Waugh S A, Lerski R A, Bidaut L, et al.The influence of field strength and different clinical breast MRI protocols on the outcome of texture analysis using foam phantoms[J]. Medical physics,2011,38(9):5058-5066.
