基于静息态功能磁共振成像和机器学习的青少年抑郁症脑网络连接及分类模型研究  

Brain network connectivity and classification model of adolescent depression based on resting-state functional magnetic resonance imaging and machine learning

作　　者：申延蕊 李学坤 李众 曹城浩 郑卓 吴宝林 Shen Yanrui;Li Xuekun;Li Zhong;Cao Chenghao;Zheng Zhuo;Wu Baolin(Department of Radiology,West China Hospital,Sichuan University,Chengdu 610041,China;Department of Magnetic Resonance Imaging,First Affiliated Hospital of Xinxiang Medical University,Xinxiang 453100,China)

机构地区：[1]四川大学华西医院放射科,成都610041 [2]新乡医学院第一附属医院核磁共振科,新乡453100

出　　处：《中华神经医学杂志》2025年第3期260-266,共7页Chinese Journal of Neuromedicine

基　　金：四川大学华西医院专职博士后研发基金(2024HXBH145)。

摘　　要：目的探讨青少年抑郁症患者脑功能网络连接异常模式及其对青少年抑郁症的诊断价值。方法选择四川大学华西医院放射科精神影像门诊自2020年1月至2022年12月收治的94例青少年抑郁症患者作为青少年抑郁症组,同期通过当地社区广告招募年龄、性别与青少年抑郁症患者相匹配的78例健康青少年作为健康对照组。采集2组受试者的静息态功能磁共振图像,将图像预处理后,采用组水平空间独立成分分析识别脑网络连接,比较2组受试者间网络功能连接的差异。将功能连边作为分类特征并进行特征排序和筛选,使用线性核函数支持向量机(SVM)构建分类模型,采用受试者工作特征(ROC)曲线分析该分类模型对青少年抑郁症的诊断价值。结果2组受试者年龄、性别、受教育年限和体质量指数的差异均无统计学意义(P>0.05)。与健康对照组比较,青少年抑郁症组患者感觉运动网络(SMN)、视觉网络(VN)、听觉网络(AN)、默认网络(DMN)和认知控制网络(CCN)的网络内和网络间功能连接强度降低,CCN内的功能连接强度升高,差异均有统计学意义(P<0.05)。当使用前75个排序较高的功能连接特征时,线性核函数SVM分类器构建的模型具有最好的分类性能(准确率为70.35%,敏感度为70.21%,特异度为71.80%,P<0.001)。ROC曲线分析结果显示该分类模型诊断青少年抑郁症的曲线下面积为0.724(95%CI:0.648~0.800,P<0.001)。共识别出51个一致性功能连接特征,这些一致性连接主要位于SMN、VN、AN、DMN和CCN的网络内或网络间。结论青少年抑郁症患者的静息态脑功能网络连接异常可为其临床诊断提供影像学依据。ObjectiveTo explore the abnormal patterns of brain functional network connectivity in depression adolescents and their diagnostic value in adolescent depression.MethodsA total of 94 depression adolescents(adolescent depression group)admitted to Outpatient Department of Psychiatric Imaging,West China Hospital,Sichuan University from January 2020 to December 2022 were selected.In addition,78 age-and gender-matched healthy adolescents were recruited from local community advertisements at the same time-period as healthy control group.Resting-state functional magnetic resonance imaging was performed;after image preprocessing,group-level spatial independent component analysis was performed to identify the intrinsic network connectivity,and differences in network connectivity between the two groups were compared.Functional connectivity edges were employed as classification features,and feature ranking and screening were then performed.A support vector machine(SVM)with linear kernel function was used to construct a classification model,and receiver operating characteristic(ROC)curve was used to analyze the diagnostic value of this classification model in adolescent depression.ResultsNo significant difference was noted in age,gender,years of education,and body mass index between the two groups(P>0.05).Compared with the healthy control group,the adolescent depression group had significantly decreased functional connectivity intensity within and between the networks of sensorimotor network(SMN),visual network(VN),auditory network(AN),default mode network(DMN),and cognitive control network(CCN),and significantly increased functional connectivity intensity within CCN(P<0.05).When using the 75 top-ranked functional connectivity features,this classification model had the best performance(accuracy rate:70.35%,sensitivity:70.21%,specificity:71.80%,P<0.001).ROC curve showed that area under the curve of this classification model in diagnosing adolescent depression was 0.724(95%CI:0.648-0.800,P<0.001).A total of 51 consistent functi

关 键 词：抑郁症 青少年 独立成分分析 功能连接 静息态功能磁共振成像 机器学习 

分 类 号：R749.4[医药卫生—神经病学与精神病学]