Identification of tRNS-induced EEG changes in children with ADHD using signal processing and machine learning

Abstract

Attention-Deficit/Hyperactivity Disorder (ADHD) is a prevalent neurodevelopmental condi- tion associated with alterations in neural dynamics and cortical excitability. Recent work has shown promising results using transcranial random noise stimulation (tRNS) combined with cognitive training (CT) to modulate ADHD symptoms and EEG activity, primarily through analyses of linear spectral features. This study aims to investigate whether non-linear EEG fea- tures provide complementary insight into tRNS-induced neurophysiological changes in chil- dren with ADHD. Using an existing data-set from a randomised sham-controlled trial, resting-state EEG record- ings from 23 unmedicated children (6–12 years) were analysed across three time points (base- line, post-treatment, and three-week follow-up). A comprehensive feature set was extracted, in- cluding linear features (spectral power, power ratios and aperiodic components) as in the orig- inal study, alongside an extended set of non-linear measures (entropy- and complexity-based metrics). Statistical analyses of intra-group and inter-group differences were conducted and complemented with machine-learning models to classify treatment conditions based on EEG- derived biomarkers. Results showed that non-linear features, particularly in frontal and central regions, revealed significant group differences that were not fully captured by linear metrics alone. Moreover, machine-learning models combining linear and non-linear features achieved up to 85% accu- racy and 100% sensitivity, demonstrating clear improvements over models relying solely on linear features. These findings indicate that non-linear EEG dynamics enhance the detection of tRNS-related neural reorganisation, especially at later follow-up stages. In conclusion, this project demonstrates that non-linear EEG features provide valuable and com- plementary insights for characterising neuromodulation effects in paediatric ADHD. The inte- gration of these features with machine-learning approaches offers a promising pathway towards developing objective biomarkers of treatment response, supporting future efforts towards per- sonalised neuromodulation therapies.

Outgoing