Analysis of electroencephalogram microstates and spectral power in children and adolescents with attention deficit hyperactivity disorder.

Attention-Deficit/Hyperactivity Disorder (ADHD) is a widely recognized neurodevelopmental disorder characterized by inattention, hyperactivity, and impulsivity. Behavioral assessment, the current gold standard for diagnosis, is subject to inaccuracies due to its subjective nature and is limited to post-symptom onset evaluations. Early diagnosis is pivotal for enhancing the quality of life in ADHD patients. This study introduces a novel machine learning (ML) approach for the objective diagnosis of ADHD in medicated and unmedicated individuals, leveraging electroencephalogram (EEG) data features. We also investigated the utility of resting-state EEG microstates in discerning ADHD patients from healthy controls and in predicting symptom severity. Our analysis revealed that resting-state EEG microstates effectively differentiated ADHD patients from controls. The 8-feature Linear Discriminant Analysis model emerged as the most predictive ML model for ADHD, achieving an AUC of 0.967 in the training set, with sensitivity = 0.914, specificity = 0.927, and accuracy = 0.921. In the validation set, the model's performance remained robust with an AUC of 0.747, sensitivity = 0.695, specificity = 0.716, and accuracy = 0.706. Additionally, these EEG features accurately forecasted the severity of ADHD symptoms. The development of an ML model utilizing EEG data features presents a promising diagnostic tool for early and objective identification of ADHD. This advancement may facilitate earlier clinical intervention and contribute to the improvement of ADHD patients' quality of life through targeted therapeutic strategies.