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

Abstract

There is increasing expectation that advanced, computationally expensive machine learning techniques, when applied to large population-wide neuroimaging datasets, will help to uncover key differences in the human brain in health and disease. We take a comprehensive approach to explore how multiple aspects of brain structural connectivity can predict sex, age, general cognitive function and general psychopathology, testing different machine learning algorithms from deep learning model (BrainNetCNN) to classical machine learning methods. We modelled N = 8, 183 structural connectomes from UK Biobank using six different structural network weightings obtained from diffusion MRI. Streamline count generally provided highest prediction accuracies in all prediction tasks. Deep learning did not improve on prediction accuracies from simpler linear models. Further, high correlations between model coefficients from deep learning and linear models suggested similar decision mechanisms were used. This highlights that model complexity is unlikely to improve detection of associations between structural connectomes and complex phenotypes.

Classification accuracy of structural and functional connectomes across different depressive phenotypes

“…In the current study, we chose healthy controls with age, sex and ICV matched (or closest to) with the cases in order to minimize the confounding effect and this could be the reason for having results inconsistent with the previous literature. In addition, a similar phenomenon was seen in our previous paper where model classification performances on general cognitive function and general psychopathology were comparable for all the structural connectivity modalities when age and sex were added to the models (Yeung et al, 2022).…”

“…Moreover, work by Schulz et al (2020) indicate that simple linear models were just as competitive as non-linear models, sometimes even outperforming non-linear models, in predicting common phenotypes from brain scans (Schulz et al, 2020). We have found similar results in a previous study of structural connectomes that ridge regression tends to be more consistent than other models in terms of model coefficients (Yeung et al, 2022). Therefore, a logistic ridge regression model was chosen for classification modelling in the current study.…”

“…At the time of processing, N = 9, 858 participants with compatible T1-weighted and dMRI data were available from the UKB, and the structural connectomes for these participants were derived locally. A full description of structural connectome processing can be found in Buchanan et al (2020) and Yeung et al (2022). These processes are described briefly below.…”