Generation, discrimination and Metrification of cerebral functional connectivity patterns

• Grant number: 20/04119-6
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Effective date (Start): July 01, 2020
• Effective date (End): June 30, 2021
• Field of knowledge: Physical Sciences and Mathematics - Computer Science
• Principal Investigator: João Ricardo Sato
• Grantee: Pedro Machado Nery dos Santos
• Host Institution: Centro de Matemática, Computação e Cognição (CMCC). Universidade Federal do ABC (UFABC). Ministério da Educação (Brasil). Santo André , SP, Brazil

Abstract

A considerable part of mental disorders can be contextualized as an inadequate establishment of connections in the human brain, especially in a context of atypical neurodevelopment. There is a strong interest in the characterization of such a typicality, as well as in biomarkers and metrics that allow the progression or risk of such diseases. The analysis of functional magnetic resonance imaging (fMRI) data is particularly interesting in this scenario due to the possibility of in vivo evaluation, granting the following-up of conditions, the outlining of intervention proposals and the delimiting of risk factors to patients. This project aims to utilize connectivity matrices resulting from and ROI (Region of Interest)-to-ROI analysis in fMRI exams to model the connectivity pattern of a typical brain. It is proposed the use of the model GAN (Generative Adversarial Network), capable of reproducing (generative part) and recognizing (discriminative part) the typicality pattern present in the data through a process of adversarial machine learning. The model analysis will support the proposal of a typicality metric posteriorly confronted with patient's cognitive indicators, such as DAWBA (Development and Well-Being Assessment) and CBCL (Child Behavior Checklist), in a process of validation and statistical analysis. The LIME and SHAP algorithms will be used in order to provide information of which features are more relevant to model predictions. Tests using data from typical and atypical individuals will be utilized for performance analysis.