脑功能连接特征判别青少年注意缺陷多动障碍的探索

Preliminary study on diagnosis of adolescents with attention-deficit/hyperactivity disorder (ADHD) by brain functional connection characteristics

目的:基于静息态脑功能连接特征建立注意缺陷多动障碍(ADHD)的智能诊断模型。方法:收集符合DSM-IV标准的60例8~16岁的ADHD患者和92例正常对照的静息态脑功能核磁共振成像数据,比较两组脑区间的功能连接,运用机器学习,探讨与机器学习理论的双重信息特征筛选方法得到的30个最优特征集合对ADHD诊断的预测性能。结果:ADHD在默认网络和感觉运动网络之间功能连接增强,默认网络内部,以及默认网络和小脑网络间的功能连接减弱。ADHD距离较远脑区间功能连接减弱,而距离较近的脑区间功能增强。模型的灵敏度和特异度分别可达80.3%和90.0%。结论:结合新的机器学习模型,运用ADHD脑功能连接特征的诊断方法可以获得较高的准确率。

Objective:To identify neurobiological markers and constructing classification models in adolescents with attention-deficit/hyperactivity disorder(ADHD)to provide a more powerful and objective tool for diagnosis.Methods:The sample was composed of 60 ADHD patients and 92 age-matched,typically developing healthy individuals(8-16 years)with IQ of>80 from a homogeneous Han background.Based on statistics and machine learning algorithms,30 optimal feature collection was selected and applied to Extreme Gradient Boosting(XGBoost).Results:Among patients with ADHD,although part of functional connectivity between default network and sensorimotor network was significantly increased,part of functional connectivity within default network and connectivity between default network and cerebellum network were decreased.Besides,functional connectivity with higher strength in patients with ADHD than healthy controls was mainly among nearby brain regions,while functional connectivity among distant brain regions was weaker in patients with ADHD.The sensitivity and specificity of the model based on static resting-state brain network connectivity obtained 80.3%and 90%,respectively.The AUC for the model was 89.4%,indicating the ideal classification ability.Conclusion:It suggests that together with machine learning,static functional brain connectivity features could classify the male teenage ADHD effectively.