Predicting gender and cognitive scores from structural and functional connectome

• Main Author: Balehithlu, Ananya
• Other Authors: Jagath C Rajapakse
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science
• Online Access: https://hdl.handle.net/10356/175291
• Institution: Nanyang Technological University

Neuroimaging studies have extensively utilised graph convolution networks (GCNs) to model brain connectivity for predicting cognitive behaviours and gender. However, the integration of the structural connectivity (SC) and functional connectivity (FC) to enhance prediction accuracy remains relatively underexplored. We introduce the Structure-Function Interaction Embedded GCN (SFI-GCN), a framework that leverages the interplay between the SC and FC. Our model transcends traditional GCNs by embedding a novel interactive module, which utilises a Multi-Layer Perceptron (MLP) to discern the nuanced interactions between the SC and FC. This model thus enables the learning of interactive weights to foster a deeper understanding of the structural-functional connections within the brain’s connectome. Applied to an extensive dataset from the Human Connectome Project encompassing 839 participants, SFI-GCN showcases superior performance in predicting gender as well as cognitive scores across various domains, including episodic memory, language comprehension, and processing speed. Additionally, it also demonstrates the ability to capture and therefore visualise specific regions of structure-function coupling in the 116 Regions of Interest (ROIs). The model's adeptness at learning and visualising the interactive weights between the SC and FC reveals significant insights into the complex inter dependencies of brain connectivity. Our model surpasses linear regression and multi-view GCN benchmarks for cognitive score prediction, while performing comparatively or better than logistic regression and support vector classification for gender prediction. SFI-GCN enhances prediction performance and deepens our understanding of the interplay between SC and FC.