摘要
目的:探讨MRI纹理分析评估肝外胆管癌淋巴结转移的价值。方法:回顾性搜集我院2011年1月至2020年03月经病理证实的肝外胆管癌患者110例,根据病理结果分为淋巴结转移组(n=31)和无转移组(n=79)。所有患者术前均行腹部3.0T MRI扫描,于轴面T1WI、T2WI及扩散加权成像(DWI)图像上选取病灶最大层面,使用MaZda软件手工勾画感兴趣区并提取肿瘤的纹理特征(300种)。每种MRI序列利用Fisher系数、分类错误概率联合平均相关系数及交互信息的方法分别选择30种最佳纹理特征。采用多层感知器整合三种MRI序列上选择的90种最佳纹理特征。以80%作为训练集,20%作为测试集,重复操作100次,建立预测肝外胆管癌淋巴结转移的模型并评估其性能。结果:110例患者中31例被病理学证实有淋巴结转移,79例无淋巴结转移。淋巴结转移组与无转移组的病灶病理分化程度(P=0.036)及大小(P=0.002)差异均有统计学意义,而两组患者的年龄、性别及肿瘤位置差异均无统计学意义。两位放射科医师对提取的纹理特征进行相关性分析的组内相关系数为0.97 (0.774~1.000,P<0.001),组间相关系数为0.93 (0.751~1.000,P<0.001),表明本研究具有良好的组内及组间一致性。多层感知器模型能较好地预测肝外胆管癌的淋巴结转移,平均曲线下面积为0.895(0.748~0.996),在训练集及测试集中的预测准确率分别为83.4%(67.8%~100%)、85.0%(64.0%~100%)。结论:基于多层感知器的MRI纹理分析能较好地评估肝外胆管癌的淋巴结转移,有助于最佳治疗方案的选择并评估其预后。
Objective:To explore the value of texture analysis of magnetic resonance imaging in evaluating lymph node metastases(LNM) of extrahepatic cholangiocarcinoma(ECC).Methods:A total of 110 patients with ECC confirmed by pathology from January 2011 to March 2020 in our hospital were retrospectively collected and analyzed.All subjects were divided into LNM(n=31) and non-LNM group(n=79) based on pathological status of the lymph nodes.All patients underwent preoperative abdominal MRI scan on a 3.0T MRI scanner.Regions of interest(ROIs, placed on the maximum section of the tumor) were delineated manually on axial T1-precontrast weighted imaging(T1WI),T2-weighted imaging(T2WI) and diffusion weighted imaging(DWI) and 300 texture features were extracted from each ROI using MaZda software.Thirty optimal texture features for each MRI sequence were selected based on Fisher coefficients, minimization of both classification error probability and average correlation coefficients(POE+ACC),and mutual information(MI) coefficients.Multi-layer perceptron(MLP) was used to integrate 90 optimal texture features selected from three MRI sequences.With 80% of the samples as the training set and 20% as the test set, the above steps were repeated for 100 times to establish a model for predicting LNM of ECC and evaluate its performance.Results:Of the 110 patients, 31 were diagnosed with lymphatic metastasis by pathological examination and 79 were diagnosed without lymphatic metastasis.There were statistically significant differences in differentiation degree(P=0.036) and lesion size(P=0.002) between LNM and non-LNM group, while there were no statistically significant differences in the age, gender and tumor location between the two groups.The mean intraclass correlation coefficient(ICC) was 0.97(0.774~1.000,P<0.001) in intra-observer agreement and the mean ICC was 0.93(0.751~1.000,P<0.001) in inter-observer agreement between two radiologists, which showed an intra-observer and inter-observer satisfactory consistency.The model based on MLP had a better performance for predicting LNM of ECC.It showed an average area under the curve of 0.895(0.748~0.996),with the prediction accuracy of 83.4%(67.8%~100%) and 85.0%(64.0%~100%) in the training and test cohorts, respectively.Conclusion:Texture analysis based on MRI performed well in predicting LNM of ECC through MLP,which may guide optimal treatment planning and help to determine prognosis.
作者
杨春梅
舒健
陆笑非
苏松
周铁军
YANG Chun-mei;SHU Jian;LU Xiao-fei(Department of Radiology,the Affiliated Hospital of Southwest Medical University,Sichuan 646000,China)
出处
《放射学实践》
CSCD
北大核心
2023年第1期33-38,共6页
Radiologic Practice
基金
国家自然科学基金资助项目(82272077)
西南医科大学校级科研项目资助项目(2020ZRQNA041)。
关键词
磁共振成像
纹理分析
多层感知器
肝外胆管癌
淋巴结转移
Magnetic resonance imaging
Texture analysis
Multi-layer perceptron
Extrahepatic cholangiocarcinoma
Lymph node metastases
