基于MRI的空间影像组学模型鉴别幕上毛细胞型星形细胞瘤与节细胞胶质瘤的价值  

Spatial radiomics model for identifying supratentorial pilocytic astrocytoma and ganglioglioma based on MRI

作　　者：詹天亮 李建瑞 许强 卓芝政 李俊杰[3] 陈浩辉 刘亚欧 张志强[1] Zhan Tianliang;Li Jianrui;Xu Qiang;Zhuo Zhizheng;Li Junjie;Chen Haohui;Liu Ya'ou;Zhang Zhiqiang(Department of Radiology,Jinling Clinical Medical College,Nanjing University of Chinese Medicine,Nanjing 210002,China;Department of Radiology,General Hospital of Eastern Theater Command,Nanjing 210002,China;Department of Radiology,Beijing Tiantan Hospital,Capital Medical University,Beijing 100070,China)

机构地区：[1]南京中医药大学金陵临床医学院放射诊断科,南京210002 [2]东部战区总医院放射诊断科,南京210002 [3]首都医科大学附属北京天坛医院放射科,北京100070

出　　处：《中华放射学杂志》2024年第12期1381-1387,共7页Chinese Journal of Radiology

基　　金：国家自然科学基金(82371951);国家重点研发计划(2018YFA0701703);国家科技创新2030—“脑科学与类脑研究”(2022ZD0211800)。

摘　　要：目的基于幕上毛细胞型星形细胞瘤(PA)与节细胞胶质瘤(GG)病变空间分布特征构建空间影像组学模型并评价其鉴别诊断效能。方法该研究为横断面研究。回顾性收集2016年6月至2022年6月在首都医科大学附属北京天坛医院(中心1)就诊244例幕上PA与GG患者和2019年3月至2022年10月在东部战区总医院(中心2)就诊的116例幕上PA与GG患者。将中心1的患者按照随机数表法以7∶3的比例分为训练集(171例)和验证集(73例),中心2的患者整体作为测试集。所有患者均接受MRI检查。基于增强T 1WI和T 2WI图像分割肿瘤病灶,配准到标准空间生成肿瘤位置统计参数图,并进行组间比较;采用约翰·霍普金斯大学模板提取189个肿瘤位置特征,构建肿瘤位置空间特征模型;采用PyRadiomic 3.0.1软件提取肿瘤影像组学特征构建影像组学模型;将两者融合构建空间影像组学模型。采用受试者操作特征曲线和曲线下面积(AUC)分析空间影像组学模型、空间模型、影像组学模型鉴别PA与GG的效能;以测试集与验证集间准确度的差值(ΔACC)评估模型的泛化能力;利用临床决策曲线和校准曲线对比模型的临床效用。结果病灶位置统计参数图显示,幕上PA以中线结构区为主分布在鞍上、丘脑、基底节以及额叶、颞叶、顶叶,而GG则分布在以双侧颞叶为主以及额叶、枕叶、顶叶。在测试集中空间影像组学模型、影像组学模型、空间模型鉴别PA与GG的AUC分别为0.876、0.785、0.819,准确度分别为77.59%、72.41%、77.14%,测试集与验证集的ΔACC分别为11.6%、15.43%、6.94%。临床决策曲线显示空间影像组学模型的临床获益整体高于传统影像组学模型、空间模型。结论含有病变位置空间信息的空间影像组学模型,可以提高幕上PA与GG的诊断效能,并增强预测模型的泛化性。Objective To construct a spatial radiomics model based on the spatial distribution characteristics of supratentorial pilocytic astrocytoma(PA)and ganglioglioma(GG)and to evaluate its differential diagnosis efficiency.Methods The study was a cross-sectional study.A retrospective collection of 244 patients with episodic PA and GG who attended Beijing Tiantan Hospital of Capital Medical University(Center 1)from June 2016 to June 2022 and 116 patients with episodic PA and GG who attended General Hospital of Eastern Theater Command(Center 2)from March 2019 to October 2022 was performed.The patients in Center 1 were divided into a training set(171 patients)and a validation set(73 patients)in a 7∶3 ratio according to the random number table method,and the patients in Center 2 as a whole were regarded as test sets.All patients underwent MRI.Segmentation of tumor based on enhanced T 1WI and T 2WI images,alignment to standard space to generate a statistical parametric mapping of tumor locations and intergroup comparison was conducted.The Johns Hopkins University template was used to extract 189 tumor location features to construct a spatial model of tumor location;PyRadiomic 3.0.1 software was used to extract tumor radiomics features to construct a radiomics model;and the two models were fused to construct a spatial radiomics model.The efficacy of spatial radiomics model,spatial model,and radiomics model to discriminate PA from GG was analyzed using receiver operating characteristic curves and area under the curve(AUC).The generalization ability of the model was assessed by the difference in accuracy between the test sets and the validation sets(ΔACC).The clinical utility of the model was compared using clinical decision curves and calibration curves.Results The statistical parametric mapping of lesions showed that supratentorial PA was vulnerable to medial structure areas such as suprasellar region,thalamus,basal ganglia and frontal lobe,temporal lobe,parietal lobe.GG was mainly distributed in bilateral temporal lobes,

关 键 词：星形细胞瘤 神经节神经胶质瘤 磁共振成像 位置 影像组学 

分 类 号：R445.2[医药卫生—影像医学与核医学] R739.4[医药卫生—诊断学]