EEG-based ADHD Detection Using Autoregressive Model Parameters and Gaussian Mixture Models: A Speaker Verification Analogy

Abstract

Attention Deficit Hyperactivity Disorder (ADHD) is a prevalent condition affecting a significant number of children, primarily boys. The diagnosis of ADHD relies mainly on subjective observations and interviews, necessitating the development of objective tests for accurate detection. In this study, we propose an ADHD detection method using EEG data collected from multiple channels. By employing autoregressive model parameters as features and drawing inspiration from the imposter problem in speaker verification, we employ Gaussian mixture models to define ADHD and universal background models. Subsequently, a likelihood ratio detector is designed. The effectiveness of this approach is evaluated using traditional performance measures such as the area-under-the-curve and equal-error-probability. Our results, obtained from a limited male database of approximately 6 years of age, demonstrate that the proposed approach achieves high detection probability and low equal error rate simultaneously. Notably, EEG data collected during an attention network task are utilized for analysis. Additionally, we explore the impact of contaminated data on the detection process. This research contributes to the advancement of objective ADHD detection methods and highlights the potential of EEG-based approaches in improving diagnostic accuracy.