Investigating Whether Behavioral Reaction Time Patterns can Predict Structural Brain Phenotypes Using Models

Abstract

The individual extent to the content of behavioral data, and neuroimaging is rapidly growing everyday. However, the notice of information about neuroimaging and its variations from behavioral data remains unclear. This study investigated whether it is possible, and if so, to what extent can we take it to, of finding an output of structural brain anatomy from behavioral performance using machine learning models. We analyzed data from 204 participants from the behavioral capacity and performance, as well as their relative 3D MRI (Magnetic Resonance Imaging) scan. There were a total of eleven structural targets examined, these targets are the total brain volume, gray-matter volume, white-matter volume, frontal, parietal, temporal, occipital, cerebellar, subcortical, and ventricular volumes, while also adding a left-right asymmetry index. We created 5 domains of the specific behavior variables/predictors, GRT (general reaction time), visual performance, ETS, CPTS, and some oddball task measures. The variables derived from MRI were removed so it wouldn't provide any leakage in the data. The GRT performances provided the best values. We ran 13 featured GRT performance to lead us to an R^2 of 0.0723 for subcortical volume, 0.0667 for frontal volume, and 0.0660 for temporal volume.The relationships between the value that we found were 0.2710, 0.2623, and 0.2571, respectively. After testing permutation, we got the value of p = 0.004975 which was the value found for all 3 relationships. However, the individual reaction times did not seem to prove their capabilities compared to the multivariate model. After these results, I was confident that the behavior reaction times gave us decent information of the changes in brain phenotypes. They don’t represent casualty, nor can the behavioral reaction times be used to create a perfect depiction of the user’s brain. Instead, they demonstrate that multivariate behavioral measurements can capture a small but measurable component of inter-individual variation in structural neuroanatomy.