An R Package for Generating Data-Informed GLM Models in Task-BasedfMRI Data Analysis
Analysis of task-related fMRI activity at the level of individualparticipants is commonly based on general linear modelling (GLM) thatallows us to estimate to what extent the BOLD signal can be explained bytask response predictors specified in the GLM model. The predictors areconstructed by convolving the hypothesised timecourse of neural activitywith an assumed hemodynamic response function (HRF). To get valid andprecise estimates of task response, it is important to construct a modelof neural activity that best matches actual neuronal activity. Theconstruction of models is most often driven by predefined assumptions onthe components of brain activity and their duration based on the taskdesign and specific aims of the study. However, our assumptions aboutthe onset and duration of component processes might be wrong and canalso differ across brain regions. This can result in inappropriate orsuboptimal models, bad fitting of the model to the actual data andinvalid estimations of brain activity. Here we present an approach inwhich theoretically driven models of task response are used to defineconstraints based on which the final model is derived computationallyusing the actual data. Specifically, we developed autohrf — a packagefor the R programming language that allows for data-driven estimation ofHRF models. The package uses genetic algorithms to efficiently searchfor models that fit the underlying data well. The package uses automatedparameter search to find the onset and duration of task predictors whichresult in the highest fitness of the resulting GLM based on the fMRIsignal under predefined restrictions. We evaluate the usefulness of theautohrf package on publicly available datasets of task-related fMRIactivity. Our results suggest that by using autohrf users can findbetter task related brain activity models in a quick and efficientmanner.
This work was supported by the Slovenian Research Agency (YoungResearcher program and the research grants J3-9264, P3-0338,P5-0110).