Our abstract „Leveraging EEG for Age Modelling and Analysis of Developmental Age Disorders“ has been accepted to the symposium on conference XXXVIII. Kongress der deutschen Gesellschaft für Kinder- und Jugendpsychiatrie
Advancements in neuroimaging and electrophysiology have enhanced our understanding of developmental age disorders, particularly through EEG recordings that reveal evolving neural activity from infancy to adulthood. While ML methods have been applied to EEG data from adults to identify age-related brain patterns, no ML studies have focused on continuous age-related brain activity in children and adolescents. This study hypothesized that ML models could accurately estimate chronological age from EEG recordings of 1024 subjects aged 3 to 19 years. By using denoised EEG records and a Gradient Boost ML model, 30 out of 1220 features were identified as most relevant for age estimation. The model, evaluated through 10-fold cross-validation, achieved a Mean Absolute Error of 1.352 years and an R-squared of 0.741. An automated pipeline was created to transform raw clinical EEG data into age estimates, potentially identifying deviations from typical development. Key EEG features, such as spectral power in delta, beta, and gamma bands, and phase-lag-index between specific lobes, were crucial for accurate age regression. This ML pipeline, due to its non-invasive and cost-effective nature, could be widely used to support diagnostics and research, including studies on neurodegenerative diseases and the retrogenesis hypothesis.