Abordagens de aprendizado de máquina e redes complexas na classificação de transtornos mentais

Abstract

Complex network science has emerged as a powerful tool for studying the structural and functional organization of the brain under normal conditions and in pathological contexts. In recent years, network-based approaches have been widely used to understand how mental and neurodevelopmental disorders influence brain organization and alter its functionality. The integration of graph theory, functional neuroimaging techniques, and computational methods enables the reconstruction of brain networks and contributes to understanding the impacts of various disorders at the individual level. Moreover, advances in automated diagnosis using machine learning have shown promising results in predicting and characterizing a range of disorders based on the analysis of functional connectivity obtained from time series of brain activity. This project aims to classify different mental disorders through the reconstruction of complex networks from time series extracted from neuroimaging data. The goal is to assess which connectivity metric is most suitable for reconstructing the brain network topology associated with each mental disorder, as well as the one that achieves the best classification performance. Machine learning models were used to analyze functional connectivity matrices in order to characterize the brain topology associated with each condition and to identify disorder-specific connectivity patterns. For major depressive disorder (MDD), Spearman correlation was the most suitable metric for reconstructing brain networks. The results suggest that alterations in connectivity are related to deficits in attention, memory, and decision-making. For autism spectrum disorder (ASD) and attention-deficit hyperactivity disorder (ADHD), transfer entropy was used to build the connectivity matrix, revealing altered patterns in brain regions associated with attention and impulse control in ADHD, and impairments in social and cognitive areas in ASD. Connectivity patterns were also analyzed to understand the progression of Parkinson’s disease (PD). In its early stages, the patterns resembled those of healthy individuals, while in later stages, a reduction in synchronization between brain regions and disease progression was observed. Therefore, the findings of this study contribute to more accurate diagnoses and to the development of personalized intervention strategies, promoting advances in the understanding and treatment of these conditions.