Predicting cognitive scores with graph neural networks through sample selection learning

Martin Hanik,Mohammed Amine Gharsallaoui,Islem Rekik,Mehmet Arif Demirtaş

doi:10.1007/s11682-021-00585-7

Abstract

Analyzing the relation between intelligence and neural activity is of the utmost importance in understanding the working principles of the human brain in health and disease. In existing literature, functional brain connectomes have been used successfully to predict cognitive measures such as intelligence quotient (IQ) scores in both healthy and disordered cohorts using machine learning models. However, existing methods resort to flattening the brain connectome (i.e., graph) through vectorization which overlooks its topological properties. To address this limitation and inspired from the emerging graph neural networks (GNNs), we design a novel regression GNN model (namely RegGNN) for predicting IQ scores from brain connectivity. On top of that, we introduce a novel, fully modular sample selection method to select the best samples to learn from for our target prediction task. However, since such deep learning architectures are computationally expensive to train, we further propose a learning-based sample selection method that learns how to choose the training samples with the highest expected predictive power on unseen samples. For this, we capitalize on the fact that connectomes (i.e., their adjacency matrices) lie in the symmetric positive definite (SPD) matrix cone. Our results on full-scale and verbal IQ prediction outperforms comparison methods in autism spectrum disorder cohorts and achieves a competitive performance for neurotypical subjects using 3-fold cross-validation. Furthermore, we show that our sample selection approach generalizes to other learning-based methods, which shows its usefulness beyond our GNN architecture.

Highlights

Understanding how the structure of the brain influences cognitive scores such as intelligence quotient (IQ) plays a vital role in understanding the working principles of the human brain
We summarize the main contributions of our work as follows: 1. We introduce a novel, learning-based sample selection method for graph neural networks that helps to increase accuracy when predicting cognitive scores from connectomes
We observe that while the stateof-the-art machine learning model CPM surpasses naive applications of graph neural networks (GNNs) in the form of PNA, our RegGNN, paired with sample selection, outperforms CPM in all tasks according to both mean absolute error (MAE) and RMSE with the exception of the NT (FIQ) task

Summary

Introduction

Understanding how the structure of the brain influences cognitive scores such as IQ plays a vital role in understanding the working principles of the human brain. Building upon it, (Dryburgh et al, 2020) improved the results by evaluating negative and positive correlations of brain regions separately They performed their analysis on both neurotypical subjects and subjects with Autism Spectrum Disorder (ASD) in order to investigate how neural correlates of intelligence scores are altered by atypical neurodevelopmental disorders. Such works achieved significant success, they mainly relied on classical machine learning approaches, which do not incorporate the graph structure of the connectomes; the local and global topological properties of the connectomes are not leveraged. We introduce a novel, learning-based sample selection method for graph neural networks that helps to increase accuracy when predicting cognitive scores from connectomes. We design a pipeline, consisting of RegGNN with sample selection, which outperforms state-of-the-art models in predicting full scale intelligence and verbal intelligence quotients from functional brain connectomes in an autism spectrum disorder cohort and achieves a competitive performance in a neurotypical cohort

Methods

Qsi I Qhl

Results and Discussion

Method

Conclusion