Proposing Two Different Feature Extraction Methods from Multi-fractal Detrended Fluctuation Analysis of Electroencephalography Signals: A Case Study on Attention-Deficit Hyperactivity Disorder

This study presents two pipelines of extracting multi-fractal detrended fluctuation analysis (MFDFA) features, to diagnose ADHD in children of age 7–13 thorough application of feature vectors and an auto-regression on the features. The features are q-order Hurst exponent, classical scaling exponent, and singularity spectrum. The MFDFA features of Electroencephalography (EEG) signals in the rest state of 16 ADHD and 16 normal children were extracted. Beside the calculated feature matrix (A), a second matrix (B) consisting of auto-regression coefficients of A was produced. Prominent features of A and B were selected and then classified using a linear support vector machine (SVM) algorithm, resulting in 94% and 97% accuracy respectively. Channels 16 and 2 in the EEG played the most important roles in discriminating the two groups, as it was proved in the literature. This study, though, depicted the individual differences in fractal properties of these two regions for the first time, which could be used as a bio-marker or a diagnosis tool for ADHD.