摘要
目的:探讨基于多模态MRI影像组学模型在辅助新生儿脑白质损伤(WML)诊断的价值。方法:回顾性分析2018年4月至2021年12月大连市妇女儿童医疗中心(集团)诊断为新生儿WML的91例患儿临床及MRI影像学资料,其中预后良好组58例,预后不良组33例。对各序列[T_(1)WI、T_(2)WI、弥散加权成像(DWI)、磁敏感加权成像(SWI)]图像病灶进行感兴趣区(ROI)勾画。分别构建T_(1)WI + T_(2)WI(模型1)、T_(1)WI+T_(2)WI+DWI(模型2)、T_(1)WI + T_(2)WI + SWI(模型3)三种模型。利用PyRadiomics工具对ROI区域进行重采样、图像预处理和特征提取,在深睿科研平台上对已提取的特征进行标准化处理。运用逻辑回归分类器建立影像组学模型和10次5折交叉验证,并筛选关键特征,绘制受试者工作特征(ROC)曲线评价模型的预测能力,并使用DeLong比较不同模型的效能。 结果:在三种模型中,模型2、模型3在验证集曲线下面积(AUC)值为0.95、0.93,高于模型1的AUC值(0.90);模型1、模型2、模型3的准确度分别为86%、89%、87%,灵敏度分别为79%、90%、88%,特异度分别为90%、88%、86%。DeLong检验模型1与模型2比较差异有统计学意义( P<0.05),而模型3与模型1、模型3与模型2比较差异无统计学意义( P>0.05)。 结论:基于多模态MRI影像组学模型在辅助新生儿WML诊断中有较高的应用价值。
Objective To investigate the effectiveness of multimodal magnetic resonance imaging(MRI)-based radiomics models in the auxiliary diagnosis of neonatal cerebral white matter lesion(WML).Methods Clinical and MRI imaging data were collected from 91 children diagnosed with WML in Dalian Women and Children's Medical Center from April 2018 to December 2021,including 58 cases in the good prognosis group and 33 cases in the poor prognosis group.The region of interest(ROI)was drawn for the lesion of each sequence image(T_(1)WI,T_(2)WI,DWI and SWI).Three models,T_(1)WI+T_(2)WI(model 1),T_(1)WI+T_(2)WI+DWI(model 2)and T_(1)WI+T_(2)WI+SWI(model 3),were created.Resample of the ROI,pre-processing of images and extraction of features were performed by using PyRadiomics,and the extracted features were standardized on the Dr.Wise research platform.A logistic regression classifier was used to create the radiomics model and 5-fold cross-validation was carried out 10 times.Key features were screened,and the receiver operating characteristic(ROC)curve was drawn to evaluate the predictive ability of the models.And DeLong was used to compare the effectiveness of different models.Results The AUC values of model 2(0.95)and model 3(0.93)in the validation set were higher than that of model 1(0.90).The accuracy,sensitivity and specificity of model 1,model 2,and model 3 were(86%,89% and 87%),(79%,88% and 88%)and(90%,90% and 86%),respectively.The differences between model 1 and model 2 by DeLong's test was statistically significant(P<0.05),whereas there was no statistically significant differences between model 3 and model 1 or between model 3 and model 2(P>0.05).ConclusionsMultimodal MRI-based radiomics model was proved to be an effective tool in the auxiliary diagnosis of neonatal WML.
作者
于泓
王静石
陈忠博
Yu Hong;Wang Jingshi;Chen Zhongbo(Department of Radiology,Dalian Municipal Women and Children's Medical Center(Group),Dalian 116033,China)
出处
《中国医师进修杂志》
2023年第9期810-815,共6页
Chinese Journal of Postgraduates of Medicine
基金
大连市医学科学研究计划项目(2011030)
广州市合生元营养与护理研究院“母婴临床研究专项基金”项目(2021BINCLC007)。
关键词
磁共振成像
新生儿
脑白质损伤
影像组学
Magnetic resonance imaging
Neonate
White matter lesions
Radiomics
