Enhancing Computational Diagnostic Tools for Major Depressive Disorder in Children and Adolescents: Deep Learning Analysis of Resting-State and Functional Connectivity from High-Density EEG

Abstract — This chapter explores the application of deep learning techniques—specifically convolutional neural networks—to identify neurological biomarkers associated with major depressive disorder in children and adolescents. It reviews foundational studies and highlights recent advancements in the field, with a focus on using resting-state functional brain connectivity derived from electroencephalography data to distinguish individuals diagnosed with major depressive disorder from healthy controls matched by age and sex. A convolutional neural network model based on the VGG16 architecture is employed for data analysis, following a comprehensive preprocessing workflow and the estimation of brain connectivity using a multivariate autoregressive model combined with independent component analysis. The model is designed to identify meaningful neurological biomarkers across different frequency ranges and specific regions of interest within the brain. The chapter also explores various methods for interpreting the deep learning model’s decision-making process, and compares the insights obtained from the convolutional neural network with results from conventional statistical approaches based on power spectral density and functional connectivity measurements analyzed separately. Furthermore, the potential use of this model as a diagnostic support tool is discussed, including its ability to classify individual participants based on short segments of their resting-state electroencephalography recordings. To enhance the model’s accuracy and reliability, the chapter addresses advanced strategies such as refined evaluation methods, robust cross-validation procedures, and ensemble learning approaches. Ultimately, this work contributes to the growing body of research on the neurobiological underpinnings of major depressive disorder and supports the development of electroencephalography-based diagnostic tools tailored for pediatric and adolescent populations.Keywords: Electroencephalography, Children, Adolescents, Major Depressive Disorder, Resting-state Functional Connectivity, Power Spectra Density, MVARICA, CNN, Deep Learning.