基于卷积神经网络的ADHD的判别分析

Classification of ADHD based on CNN

提出了利用卷积神经网络(CNN)对儿童注意缺陷与多动症(ADHD)的功能性核磁共振图像(f MRI)进行特征分析,从而对ADHD患者进行判别分析。针对ADHD-200全球竞赛的三个数据库的f MRI数据,首先利用快速傅里叶变换将数据从时域转换到频域,并利用Fisher线性判别方法选择有效的频率信息,然后利用改进的卷积神经网络对选择频域数据进行特征学习,自动提取出有效的特征并进行分类。实验结果表明,提出的方法有助于ADHD患者的判别,为基于f MRI数据的ADHD研究提供了新的手段。

The paper proposes to use convolutional neural network （CNN） to analyze the functional Magnetic Resonance Imaging （fMRI） data for Attention Deficit Hyperactive Disorder（ADHD） classification. The fMRI data from three dataset of ADHD -200 global competition is used in the paper. The fast Fourier transforming is applied for the whole brain fMRI data to transform data from time domain to the frequency do- main. Then we select useful features by the Fisher＇ s linear discriminant method and use the modified CNN to learn feature and analyze feature for ADHD classification. The experiment result illustrate that the proposed method is efficient for ADHD classification which can provide help for ADHD diagnosis in practical application.