Impact of brain regions on attention deficit hyperactivity disorder (ADHD) electroencephalogram (EEG) signals: Comparison of machine learning algorithms with empirical mode decomposition and time domain analysis.

Abstract

Objective: This study emphasizes the importance of using proper combinations of brain area, extraction of features, and machine learning (ML) techniques for electroencephalogram (EEG)-based attention deficit hyperactivity disorder (ADHD) identification. The effectiveness of EEG-based solutions is determined by the feature extraction method, selection of brain regions, and ML algorithms used.

Method: Empirical mode decomposition (EMD) was employed to identify and analyze ADHD-related abnormalities in EEG waveforms. An analysis of nonstationary and nonlinear time series data using EMD reduces an EEG waveform to a collection of intrinsic mode functions (IMFs). Random forest (RF), AdaBoost (AB), Naive Bayes (NB), support vector machine (SVM), k-nearest neighbor (KNN), decision tree (DT), linear discriminant analysis (LDA), and deep belief networks (DBNs) classification techniques were utilized.

Result: Our research showed that RF, EMD based NB, and DBN constantly outperformed on all combinations of brain regions. The EMD-based NB classifier obtained the best score at 87% accuracy on the Frontal Pole (FPO) of the brain as well as on frontal region; second, the RF achieved accuracy at 84% on the Fronto-Central Pole (FCP) of the brain and 83% on the frontal region. The precision, accuracy, and recall of the EMD-based DBN and NB algorithms outperformed those of the other models.

Significance: Our technique provides interpretable insights by focusing on particular regions of brain that makes it more applicable and relevant clinically.