基于贝叶斯先验NMF的ADHD儿童脑网络重叠社区检测

OVERLAPPING COMMUNITY DETECTION OF ADHD CHILDREN S BRAIN NETWORK BASED ON BAYESIAN PRIOR NON-NEGATIVE MATRIX FACTORIZATION

为探索视觉刺激下ADHD(Attention Deficit Hyperactivity Disorder)儿童与正常儿童脑功能活动的差异性,对两组儿童脑功能网络的功能重叠社区开展研究。获取ADHD儿童与正常儿童的任务态fMRI(functional Magnetic Resonance Imaging)数据,进行数据预处理;采用自适应稀疏表示法分别构建脑功能网络;采用基于贝叶斯先验的非负矩阵分解(Nonnegative Matrix Factorization,NMF)方法。通过预设不同的重叠社区数目,对两组儿童的脑功能网络进行重叠社区检测。实验结果显示,ADHD儿童的脑功能重叠比指标为10.7%,略低于正常儿童,表明ADHD儿童在任务中脑功能协同效率较低,且ADHD儿童的额叶-杏仁核-枕叶网络具有连接异常性。将两组儿童的各重叠社区值作为特性进行分类,其分类精度高于传统方法,达到96.6%。

In order to explore the differences in brain function activities between ADHD children and normal children under visual stimulation,a study is carried out on the functional overlapped communities of the two groups of children s brain function networks.Task-state fMRI data of ADHD children and normal children was obtained,and data preprocessing was performed.Adaptive sparse representation was used to construct brain function networks.Bayesian prior-based non-negative matrix factorization(NMF)method was used.By presetting different numbers of overlapping communities,the brain function networks of the two groups of children were tested for overlapping communities.The experimental results show that the brain function overlap ratio index of ADHD children is 10.7%,which is slightly lower than that of normal children,indicating that ADHD children have a lower brain function coordination efficiency in the task,and the frontal lobe-amygdala-occipital lobe network of ADHD children is connected abnormally.The overlapping community values of the two groups of children are classified as characteristics,and the classification accuracy is higher than that of the traditional method,reaching 96.6%.