Classification of brain networks

Abstract

The classification of brain networks has emerged as a critical area of research in neuroscience, with implications for understanding various neurological disorders such as ADHD and autism. This thesis focuses on developing algorithms to classify individuals into control and disease groups using network data, with the aim of identifying unique connectivity patterns that distinguish between these groups. The study leverages techniques from network science, machine learning, and deep learning to create robust and interpretable classification algorithms. The methodology involves the use of autoencoder technology to discern unique features between the control and disease groups. Autoencoders, a type of artificial neural network used to learn efficient representations of data, facilitate in-depth analysis and reconstruction of alterations in brain connectivity. By training the autoencoders on brain network data, the study aims to uncover latent structures that can be used to classify individuals with neurological disorders more accurately. Despite employing sophisticated techniques, the results indicate that differentiating between control and disease groups is not straightforward. The subtle and complex nature of brain connectivity patterns presents significant challenges for classification. While the autoencoder models demonstrated some ability to differentiate between the groups, the accuracy was not sufficient for reliable classification. This finding underscores the complexity of brain network data and highlights the need for further research and development of more advanced models and methodologies. The study contributes to the ongoing effort to develop automated diagnostic tools for neurological disorders, providing a foundation for future exploration in brain connectivity dynamics.