Detection of ADHD from Electroencephalogram Using Independent Component Analysis

Attention deficit hyperactivity disorder (ADHD) is one of the most common mental disorders affecting children. Delayed diagnosis and treatment of ADHD might result in poor school performance. In this paper, an intelligent system is proposed for detecting ADHD from electroencephalograms. The system first preprocesses the electroencephalograms to remove the powerline and blinking interferences. Then, it applies independent component analysis to separate the signals of the sources that generate the preprocessed electroencephalograms. Later, 3 features of Shannon entropy, kurtosis, and skewness are extracted from the independent components. A total of 57 features are obtained for 19 electroencephalogram channels. Finally, two classifiers (i.e., the k-nearest neighbors with k=3, and an artificial neural network) were used to detect ADHD from the extracted features. We used a public database containing signals from 61 children with ADHD and 60 healthy children to evaluate the performance of the proposed system. The accuracy, sensitivity, and specificity of the artificial neural network were 99.92%, 99.89%, and 99.95%, respectively, while these metrics were all 100% for the k nearest neighbors. The results imply that the features extracted from independent components of electroencephalograms are more powerful than other linear/nonlinear features directly extracted from the signals.