Graph Neural Networks applied to classification of autism symptoms in fMRI data

• Grant number: 23/06737-7
• Support Opportunities: Scholarships in Brazil - Master
• Effective date (Start): August 01, 2023
• Status: Discontinued
• Field of knowledge: Physical Sciences and Mathematics - Computer Science
• Principal Investigator: André Carlos Ponce de Leon Ferreira de Carvalho
• Grantee: Ana Letícia Garcez Vicente
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Host Company: Universidade de São Paulo (USP). Instituto de Ciências Matemáticas e de Computação (ICMC)
• Associated research grant: 20/09835-1 - IARA - Artificial Intelligence in the Remaking of Urban Environments, AP.PCPE
• Associated scholarship(s): 24/03245-9 - Graph Neural Networks applied to classification of autism and symptoms in fMRI data, BE.EP.MS

Abstract

Science has been seeking tools to assist in the early diagnosis of diseases, especially developmental disorders in children. The study and diagnosis of psychiatric disorders benefit from tools such as functional magnetic resonance imaging (fMRI), which enables the capture of brain activation through the BOLD contrast. However, these data are complex, and machine learning tools, particularly deep learning, can be useful in assisting these studies. Among the methods, we highlight Graph Neural Networks (GNNs), which allow the manipulation of complex data, such as brain images represented as graphs. The aim of this master's project is to study different GNN models for fMRI studies. First, we will apply the studied methods to classify fMRIs based on the activities performed by individuals at the time of data collection. Then, we will apply the same methods to classify neurotypical (NT) individuals and individuals with Autism Spectrum Disorder (ASD). Finally, we will propose an interpretable approach for classifying autism symptoms in order to identify which areas of the brain exhibit differentiated activation during the occurrence of each symptom. This focus on interpretability is a fundamental characteristic for future applications in the medical field, as it ensures greater confidence in the results and facilitates professionals' understanding of the network-generated outcome. This project is associated with the international project 1kD, in which 10 universities worldwide will monitor the first 1000 days of life of 1000 children in different countries.