ADHD detection in children using feature extraction and classification of EEG signals

Abstract

Attention deficit hyperactivity disorder (ADHD) is a neurological condition that is diagnosed based on the evaluation of a number of symptoms of excessive and impairing levels of inattention, hyperactivity and impulsivity. Electroencephalography (EEG) tests are used to diagnose ADHD, but they serve as a supplement to the main clinical and psychological evaluation. This work presents a method for the classification of ADHD and control cases, with the use of EEG signals. The initial data set is formed from EEG records of 47 children including 22 diagnosed with ADHD and 25 in a control group. The system consists the following stages: signal acquisition, pre-processing and filtering, feature extraction and selection, and final classification. The 2-tone oddball paradigm was used to elicit auditory event-related potentials (ERP). The filtering stage includes wavelet filtering and synchronized averaging. For feature extraction, different measures were selected, including amplitude and latency of cognitive evoked potentials, frequency bands power, and entropy and chaos quantification. For the classification process, two previously selected subsets were passed through the same process of classification, where first was applied two different dimensionality reduction methods, and finally, multiple machine learning algorithms were applied. The best performance was obtained when using the Support Vector Machines algorithm, with a maximum accuracy value of 86.84$\%$ using a subset obtained though ridge regression, and both dimensionality reduction algorithms.