Supported diagnosis of ADHD from EEG signals based on Hidden Markov Models and Probability Product Kernels

As a neuropsychiatric disorder, Attention Deficit and Hyperactivity (ADHD) onsets on childhood inducing learning problems, low self-esteem, and substance use in adolescence.1 Usually, ADHD diagnosis relies on information collected by family members, teachers, and clinical observation. However, the analysis depends on the specialist expertise, leading to high misdiagnosis rates.2 To find a more specific and sensitive diagnosis, strategies integrating biomarkers and regular scientific assessments are required.3 In an attempt to describe reliable biomarkers for ADHD identification, current researches rely on the brain physiological responseassessed through electroencephalography (EEG) to better understand neuropathologies.4 For instance, the Theta-Beta power ratio (TBR) contrasts slow waves (4-7 Hz) and fast waves (13-30 Hz), under the assumption of more powerful slow waves in ADHD than in controls on resting state. Another proposed physiologically interpretable EEG feature corresponds to the P300 that expects a latency larger than the typical 300 ms after a stimulus is presented with lower amplitudes in ADHD subjects. Also, the error-related negativity (ERN) wave, evoked between 50-100 milliseconds at the frontal-central region when incorrect responses occur, has been considered to characterize ADHD children based on the impulsivity symptom.5 Despite their physiological ground, recent works challenge the robustness and significance of TBR, P300, and ERN as ADHD biomarkers under the non-stationary and non-linear temporal dependencies of the EEG..