Neural mass modelling for the masses: Democratising access to whole-brain biophysical modelling with FastDMF

Abstract

Different whole-brain models constrained by neuroimaging data have been developed during the last years to investigate causal hypotheses related to brain mechanisms. Among these, the Dynamic Mean Field (DMF) model is a particularly attractive model, combining a biophysically realistic single-neuron model that is scaled up via a mean-field approach and multimodal imaging data. Despite these favourable features, an important barrier for a widespread usage of the DMF model is that current implementations are computationally expensive - to the extent that the model often becomes unfeasible when no high-performance computing infrastructure is available. Furthermore, even when such computing structure is available, current implementations can only support simulations on brain parcellations that consider less than 100 brain regions. To remove these barriers, here we introduce a user-friendly and computationally-efficient implementation of the DMF model, which we call FastDMF, with the goal of making biophysical whole-brain modelling accessible to neuroscientists worldwide. By leveraging a suit of analytical and numerical advances - including a novel estimation of the feedback inhibition control parameter, and a Bayesian optimisation algorithm - the FastDMF circumvents various computational bottlenecks of previous implementations. An evaluation of the performance of the FastDMF showed that it can attain a significantly faster performance than previous implementations while reducing the memory consumption by several orders of magnitude. Thanks to these computational advances, FastDMF makes it possible to increase the number of simulated regions by one order of magnitude: we found good agreement between empirical and simulated functional MRI data parcellated at two different spatial scales (N=90 and N=1000 brain regions). These advances open the way to the widespread use of biophysically grounded whole-brain models for understanding the interplay among anatomy, function and brain dynamics in health and disease, and to provide mechanistic explanations of recent results obtained from empirical fine-grained neuroimaging data sets, such as turbulence or connectome harmonics. HighlightsO_LIWe present the FastDMF, a user-friendly and computationally efficient implementation of the Dynamic Mean Field model for simulations of whole-brain dynamics. C_LIO_LIUsing analytical and numerical tools, we develop a novel estimation of the feedback inhibition control based on the structural connectivity, bypassing an important computational bottleneck. C_LIO_LIThe FastDMF is coupled with a Bayesian Optimization algorithm significantly reducing the number of simulations required to fit the FastDMF to empirical neuroimaging data. C_LIO_LIOur advances open the possibility of simulating thousands of brain regions in a biophysically grounded whole-brain model. C_LI