Attention deficit hyperactivity disorder detection in children using multivariate empirical EEG decomposition approaches: A comprehensive analytical study. (1st March 2023)

• Record Type: Journal Article
• Authors: Sharma, Yogesh; Kumar Singh, Bikesh
• Is Part Of: Expert systems with applications. Volume 213:Part C(2023)
• Publication Date: 2023-03-01
• Subjects: Attention deficit hyperactivity disorder -- EEG signal processing -- Multivariate empirical mode decomposition -- Multivariate empirical wavelet transform -- Multivariate variational mode decomposition; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2022.119219
• Languages: English
• ISSNs: 0957-4174

Abstract

Highlights: First study to observe three multivariate decomposition tools for ADHD detection. Features extracted from EEG modes, exhibiting hidden distinctive patterns. Features were selected via genetic algorithm and neighborhood component analysis. An optimal framework for ADHD classification using raw EEG was suggested. Abstract: Early detection and timely therapeutic intervention are of prime importance to prevent the severity of attention deficit hyperactivity disorder (ADHD) in children. Conventional diagnostic methods are time-taking as they are based on subjective evaluations. The present work proposes utilities of three multivariate empirical-basis decomposition approaches (EDAs) - multivariate empirical mode decomposition (MEMD), multivariate empirical wavelet transform (MEWT), and multivariate variational mode decomposition (MVMD), for ADHD diagnosis using electroencephalography (EEG) signals. A set of 15-features were derived from each EDA-decomposed oscillatory EEG mode. Significant features were identified then by genetic algorithm (GA) and neighborhood component analysis (NCA). Finally, two models -support vector machine with Gaussian radial basis function (SVM-RBF) and artificial neural network (ANN), were employed to classify children into ADHD and control categories using the GA and NCA selected attributes. A publicly available ADHD dataset from the IEEE data portal was considered for this work. Our results have unveiled MEMD-GA-ANN as the optimal classification scheme yielding an accuracy of 96.16%, F1-score of 96.32%, and Matthews correlation coefficient (MCC) of 0.92. Moreover, this study is the first comprehensive experimental analysis to incorporate multivariate EDAs in the process of classifying ADHD children. We believe that the presented mechanisms can be effective for detecting several other neurodevelopmental disorders in children by using EEG signals. Also, it may act as an informative platform for future researchers in this field.