ADHD-Net: A Robust Deep Learning-based Framework for Detecting ADHD Patients using EEG

Attention Deficit Hyperactivity Disorder (ADHD) is a prevalent neurodevelopmental condition that affects a significant proportion of the global population. In modern civilization, many individuals, including adults and children, are diagnosed with ADHD. ADHD can substantially affect the lives of individuals in diverse domains. The condition can make it challenging for a person to maintain focus, organizational skills, and impulse control, leading to difficulties in academic, social, and professional environments. It is much better to identify ADHD during the earlier stages in order to arrange tailored treatments. The goal of this study is to design a deep learning-based framework known as ADHD-Net for the precise diagnosis of ADHD from electroencephalogram (EEG) signals of the human brain. This study utilizes a publicly available dataset consisting of EEG data from 60 healthy controls and 61 ADHD patients aged 7-12 years. First, filtering of the EEG signal is performed to remove noise and artifacts. Subsequently, Independent Component Analysis (ICA) is performed to remove the noise generated by muscle movements and eye blinking. Principal component analysis (PCA) is utilized to reduce the dimensionality of the preprocessed EEG data. A deep learning model called ‘ADHD-Net’ is designed for automated detection of ADHD patients by analyzing EEG signals. The PCA-enhanced EEG data obtained is used to train the ‘ADHD- Net’ model. The trained model can classify an EEG data sample into two classes: ‘ADHD Patient’ and ‘Healthy Person’, with a high classification accuracy of 97.96%. The novelty of this research lies in: i) designing a deep learning model named ADHD-Net for automated ADHD detection, and ii) incorporating the attention mechanism into the designed ADHD-Net model to enhance its efficacy.