Microestados cerebrais no diagnóstico de distúrbios mentais: uma abordagem multivariada de clusterização e classificação de algoritmos

Abstract

Mental disorders are being diagnosed at alarming rates worldwide, taking on epidemic proportions that highlight the urgent need to develop effective methods of prevention and intervention. In this context, the electroencephalogram (EEG) stands out as one of the most widely used tools for analyzing brain electrical activity, enabling the study of brain microstates. These microstates, considered promising biomarkers, have proven useful in the diagnosis and monitoring of mental disorder progression, significantly contributing to the advancement of medical literature on these diseases. In this study, a comparative analysis of various clustering and classification algorithms was conducted, integrated into a performance validation flow, with the aim of providing a multivariate analysis of metrics in relation to different numbers of microstates. The results of the classification stage indicated that the Atomize and Agglomerative Hierarchical Clustering (AAHC) and the Modified K-Means algorithms stood out, reaching accuracy metrics around 80%. Based on these results, the numbers of microstates that presented the best performance were selected for a detailed evaluation of the clustering, using the Silhouette, Calinski-Harabasz, Davies-Bouldin, and Dunn Index metrics. The analyses converged on microstates 2 and 5 as those that provided the best outcomes, suggesting that these numbers are the most suitable for future investigations and clinical applications.