基于弥散张量磁共振脑影像的神经性厌食症机器学习分类研究  

Classification of Anorexia Nervosa Based on Magnetic Resonance Diffusion Tensor Brain Imaging Data:a Machine Learning Approach

作　　者：郑琳莉 李潇[1,2] 刘筱薇 马菁 王瑜 李涛[1] 张岚[1] Zheng Linli;Li Xiao;Liu Xiaowei;Ma Jing;Wang Yu;Li Tao;Zhang Lan(Center for Mental Health, West China Hospital of Sichuan University, Chengdu 610041, China;The Fourth People's Hospital of Chengdu, Chengdu 610036, China)

机构地区：[1]四川大学华西医院心理卫生中心,成都610041 [2]成都市第四人民医院,成都610036

出　　处：《成都医学院学报》2021年第4期490-494,共5页Journal of Chengdu Medical College

摘　　要：目的构建区分神经性厌食症(AN)患者和健康对照者的机器学习模型,探索其脑影像生物学标志物。方法选取2017年6月至2019年12月于四川大学华西医院就诊的AN患者24例为AN组;同时招募来自四川各地的健康对照者24例为对照组。采集两组一般人口学信息及大脑弥散磁共振图像,提取各向异性(FA)、平均扩散率(MD)、轴向扩散率(AD)、径向扩散率(RD)作为数据特征进行机器学习。结果采用5种模型机器学习区分AN患者和健康对照者的结果提示,FA模型AUC为0.663,差异脑区为左侧海马、中脑、丘脑;MD模型AUC为0.883,差异脑区为中脑、海马旁回、海马;AD模型AUC为0.888,差异脑区为中脑、左侧海马旁回、海马;RD模型AUC为0.920,差异脑区为中脑、左侧海马、脑胼胝体;FA+MD+AD+RD模型AUC为0.883,差异脑区为海马、中脑、脑干、丘脑、胼胝体。结论通过机器学习建立的数据模型能在一定程度区分AN患者和健康对照者,而海马、脑胼胝体、中脑、丘脑等大脑白质区域弥散张量参数有潜力作为AN的脑影像学标志物。Objective To establish a machine learning model to distinguish anorexia nervosa(AN)patients from healthy controls,and to explore brain imaging biomarkers of AN.Methods From June 2017 to December 2019,24 AN patients treated in West China Hospital were selected as AN group,and 24 healthy controls from all over Sichuan were recruited as control group.General demographic information and diffusion tensor magnetic resonance brain images were collected.Fractional anisotropy(FA),mean diffusivity(MD),axial diffusivity(AD),and radial diffusivity(RD)value of diffusion tensor imaging were extracted as features for machine learning.Results The results of distinguishing AN patients from healthy controls by machine learning of the five models were as follows.Of FA model,the area under ROC curve(AUC)was 0.663,and the different brain regions were left hippocampus,midbrain,and thalamus.Of MD model,the AUC was 0.883,and the different brain regions were midbrain,parahippocampal gyrus,and hippocampus.Of AD model,the AUC was 0.888,and the different brain regions were midbrain,left parahippocampal gyrus,and hippocampus.Of RD model,the AUC was 0.920,and the different brain regions were midbrain,left hippocampus,and corpus callosum.Of FA+MD+AD+RD model,the AUC was 0.883,and the different brain regions were hippocampus,midbrain,brainstem,thalamus,and corpus callosum.Conclusion The data models established by machine learning can distinguish between AN patients and healthy controls to a certain extent.The diffusion tensor parameters of the white matter areas such as hippocampus,corpus callosum,midbrain and thalamus are potential brain imaging biomarkers of AN.

关 键 词：神经性厌食症 弥散张量成像 机器学习 

分 类 号：R749[医药卫生—神经病学与精神病学]