Pipeline comparisons of convolutional neural networks for structural connectomes: predicting sex across 3,152 participants

Abstract

With several initiatives well underway towards amassing large and high-quality population-based neuroimaging datasets, deep learning is set to push the boundaries of what is possible in classification and prediction in neuroimaging studies. This includes those that derive increasingly popular structural connectomes, which map out the connections (and their relative strengths) between brain regions. Here, we test different Convolutional Neural Network (CNN) models in a benchmark sex prediction task in a large sample of N=3,152 structural connectomes acquired from the UK Biobank, and compare results across different connectome processing choices. The best results (76.5% test accuracy) were achieved using Fractional Anisotropy (FA) weighted connectomes, without sparsification, and with a simple weight normalisation through division by the maximum FA value. We also confirm that for structural connectomes, a Graph CNN approach, the recently proposed BrainNetCNN, outperforms an image-based CNN.

Predicting sex, age, general cognition and mental health with machine learning on brain structural connectomes

“…We then performed maximum value normalisation on the symmetric SC matrices. The maximum value normalisation was performed on all modalities as we previously found this to be the most effective normalisation (Yeung et al, 2020). Maximum value normalisation is formulated as follows: Since it is known that the head size has great influence on the streamline counts, we computed the correlation between the intracranial volume (ICV) and the total sum of entries in the SC matrices before and after normalisation.…”

“…Additionally, different graph-based neural network (GNN) models have been developed for predictive modelling based on connectivity matrices from diffusion MRI (dMRI) and resting state functional MRI (rs-fMRI) (Kawahara et al, 2017;Li and Duncan, 2020). In a previous study (Yeung et al, 2020), we have shown that the BrainNetCNN neural network architecture proposed by Kawahara et al (Kawahara et al, 2017) was more appropriate for sex classification based on brain connectome adjacency matrices compared with a naive image-based CNN architecture. Although there are several theoretical merits in the application of DL methods for understanding the neurobiological correlates of important between-person differences, the use of promising cutting-edge DL methods and its quantitative benefits beyond more conventional statistical methods remain moot.…”

Predicting sex, age, general cognition and mental health with machine learning on brain structural connectomes

“…We then performed maximum value normalisation on the symmetric SC matrices. The maximum value normalisation was performed on all modalities as we previously found this to be the most effective normalisation (Yeung et al, 2020). Maximum value normalisation is formulated as follows:…”

“…Additionally, different graph-based neural network (GNN) models have been developed for predictive modelling based on connectivity matrices from diffusion MRI (dMRI) and resting state functional MRI (rs-fMRI; Kawahara et al, 2017;Li et al, 2021). In a previous study (Yeung et al, 2020), we have shown that the BrainNetCNN neural network architecture proposed by Kawahara et al (2017) was more appropriate for sex classification based on brain connectome adjacency matrices compared with a naive image-based CNN architecture.…”