A Supervised Machine Learning Approach to Classify Brain Morphology of Professional Visual Artists versus Non-Artists

Abstract

This study aimed to investigate whether there are structural differences in the brains of professional artists who received formal training in the visual arts and non-artists who did not have any formal training or professional experience in the visual arts, and whether these differences can be used to accurately classify individuals as being an artist or not. Previous research using functional MRI has suggested that general creativity involves a balance between the default mode network and the executive control network. However, it is not known whether there are structural differences between the brains of artists and non-artists. In this study, a machine learning method called Multi-Kernel Learning (MKL) was applied to gray matter images of 12 artists and 12 non-artists matched for age and gender. The results showed that the predictive model was able to correctly classify artists from non-artists with an accuracy of 79.17% (AUC 88%), and had the ability to predict new cases with an accuracy of 81.82%. The brain regions most important for this classification were the Heschl area, amygdala, cingulate, thalamus, and parts of the parietal and occipital lobes as well as the temporal pole. These regions may be related to the enhanced emotional and visuospatial abilities that professional artists possess compared to non-artists. Additionally, the reliability of this circuit was assessed using two different classifiers, which confirmed the findings. There was also a trend towards significance between the circuit and a measure of vividness of imagery, further supporting the idea that these brain regions may be related to the imagery abilities involved in the artistic process.