Assessment of child neurodevelopment through neuroimages and deep neural network models

• Grant number: 22/07782-3
• Support Opportunities: Scholarships in Brazil - Doctorate (Direct)
• Effective date (Start): July 01, 2022
• Effective date (End): October 31, 2025
• Field of knowledge: Interdisciplinary Subjects
• Principal Investigator: João Ricardo Sato
• Grantee: Sergio Leonardo Mendes
• Host Institution: Centro de Matemática, Computação e Cognição (CMCC). Universidade Federal do ABC (UFABC). Ministério da Educação (Brasil). Santo André , SP, Brazil
• Associated research grant: 21/05332-8 - Brazilian high-risk cohort for psychiatric disorders: 10 years follow-up, AP.TEM

Abstract

The transition between childhood and adolescence is of extreme importance for neurodevelopment, as this phase presents intense changes that result in the consolidation of brain connectivity networks. However, these structures, which are accountable for behavior patterns, can mature atypically, resulting in psychiatric symptoms (e.g., depression, aggression, somatization and others) or psychopathologies. In this context, structural Magnetic Resonance Imaging (sMRI) is an important tool to provide relatively accurate characterizations of brain structures. The sMRI biomarkers can provide important information about pathological mechanisms to help understanding the nature of these diseases. However, most psychiatric disorders are still diagnosed exclusively by clinical interviews, and little is known about the etiology of these conditions. Therefore, this project aims to investigate sMRI biomarkers of psychiatric symptoms and/or psychopathologies in children and adolescents. For this purpose, normative models based on autoregressive transformers will be used. Currently, these deep neural networks are the state of the art in computer vision and anomaly detection from medical images. The data set which will be used for training the models comprises sMRI from 652 subjects, aged between 7 and 15 years, students from 57 Brazilian public schools, with different levels of psychiatric symptoms. The models will be trained from cross-sectional data, but if the data do not provide enough predictive power for the models' learning, then longitudinal data will be included in the study. The adopted approach is expected to allow for the prediction and characterization of mental health conditions from the sMRI of the studied subjects. (AU)