摘要
目的探究基于皮层形态特征的机器学习模型在诊断早期帕金森病(PD)中的应用价值。资料与方法回顾性选取帕金森病进展标志物倡议数据库2014年1月—2017年12月100例早期PD患者和70例健康对照的MRI图像和临床资料。首先使用计算机解剖分析工具箱对图像进行预处理,提取大脑皮层分形维数(FD)和回指数(GI),并分析两种指标的组间差异。然后将所有受试者按照7∶3随机分为训练集和测试集,通过t检验和递归特征消除法筛选出最优特征。使用随机森林构建分类模型并绘制受试者工作特征曲线评估模型效能,应用决策曲线分析评估模型临床价值。结果与健康对照相比,早期PD患者双侧中央前回、双侧额中回前部、双侧额中回后部、双侧额下回三角部、双侧额下回岛盖部、双侧额下回眶部以及右侧额上回、右侧外侧眶额、右侧岛叶皮层GI值较低(P均<0.05),但FD值差异无统计学意义(P均>0.05)。模型评估结果显示,FD、GI以及联合模型在训练集的曲线下面积分别为0.860、0.895、0.939,在测试集分别为0.762、0.821、0.868。Hosmer-Lemeshow检验结果表明所有模型在训练集和测试集的拟合优度差异均无统计学意义(P均>0.05)。决策曲线分析曲线显示,当阈值概率为0.10~0.88时,联合模型的临床净效益最优。结论在疾病早期PD患者的大脑皮层形态已出现变化,基于皮层形态特征的机器学习模型具有良好诊断性能,在辅助临床诊断早期PD方面可能具有重要价值。
Purpose Toexplore the application value of machine learning models based on cortical morphological featuresin the diagnosis of early Parkinson's disease(PD).Materials and Methods MRI and clinical data of 170 subjects from January 2014 to December 2017,including 100 early PD patients and 70 healthy controls,were selected from the Parkinson's Progression Markers Initiative database.Firstly,computational anatomy toolbox was used to preprocess the images to extract the fractal dimension(FD)and gyrification index(GI)of the cerebral cortex,and the differences of two indexes between early PD and healthy controls were compared.Then,all subjects were randomly divided into the train set and the test set with a 7∶3 ratio,and the optimal features were selected by t-test and recursive feature elimination.The classification model was constructed by random forest and evaluated by the receiver operating characteristic curve,and the decision curve analysis was used to evaluate the clinical value of the model.Results Compared to healthy controls,early PD patients had reduced GI in the bilaterally precentral gyrus,bilaterallyrostral middle frontal cortex,bilaterally caudal middle frontal cortex,bilaterallytriangular part of inferior frontal gyrus,bilaterally opercular part of inferior frontal gyrus,bilaterally orbital part of inferior frontal gyrus,the right superior frontal gyrus,the right lateral orbitofrontal cortex and the right insula(all P<0.05),but there was no significant difference in the FD(all P>0.05).The results of model evaluation showed that the area under curve values of the FD,the GI and the combined model in the train set were 0.860,0.895 and 0.939,respectively,and those in the test set were 0.762,0.821 and 0.868,respectively.The Hosmer-Lemeshow test showed that there was no statistically significant difference in the goodness of fit between the train and test set(all P>0.05).The decision curve analysis curve showed that clinical net benefit of the combined model was optimal when the probability threshold was in the range of 0.10 to 0.88.Conclusion In the early stages of the disease,cortical morphology of PD patients have changed.Machine learning model based on cortical morphology features has good diagnostic performance,and may be of important value in assisting clinical early diagnosis of PD.
作者
饶定才
石采灵
岳文军
RAO Dingcai;SHI Cailing;YUE Wenjun(Department of Radiology,the Affiliated Hospital of North Sichuan Medical College,Nanchong 637000,China)
出处
《中国医学影像学杂志》
CSCD
北大核心
2024年第10期994-999,共6页
Chinese Journal of Medical Imaging
关键词
帕金森病
机器学习
磁共振成像
大脑皮层
诊断
基于表面的形态学
Parkinson's disease
Machine learning
Magnetic resonance imaging
Cerebral cortex
Diagnosis
Surface-based morphometry
作者简介
通信作者:岳文军.ywj820@163.com。
