Modeling Brain Dynamics in Brain Tumor Patients Using the Virtual Brain

Abstract

Presurgical planning for brain tumor resection aims at delineating eloquent tissue in the vicinity of the lesion to spare during surgery. To this end, noninvasive neuroimaging techniques such as functional MRI and diffusion-weighted imaging fiber tracking are currently employed. However, taking into account this information is often still insufficient, as the complex nonlinear dynamics of the brain impede straightforward prediction of functional outcome after surgical intervention. Large-scale brain network modeling carries the potential to bridge this gap by integrating neuroimaging data with biophysically based models to predict collective brain dynamics. As a first step in this direction, an appropriate computational model has to be selected, after which suitable model parameter values have to be determined. To this end, we simulated large-scale brain dynamics in 25 human brain tumor patients and 11 human control participants using The Virtual Brain, an open-source neuroinformatics platform. Local and global model parameters of the Reduced Wong–Wang model were individually optimized and compared between brain tumor patients and control subjects. In addition, the relationship between model parameters and structural network topology and cognitive performance was assessed. Results showed (1) significantly improved prediction accuracy of individual functional connectivity when using individually optimized model parameters; (2) local model parameters that can differentiate between regions directly affected by a tumor, regions distant from a tumor, and regions in a healthy brain; and (3) interesting associations between individually optimized model parameters and structural network topology and cognitive performance.

Modeling brain dynamics after tumor resection using The Virtual Brain

“…MRI sequence details and preprocessing procedures for the post-operative data are mostly identical to those that we used before to collect and preprocess the pre-operative data. All details are described in Aerts et al (2018). In the following sections, we provide a summary of these procedures, as well as an overview of the minor modifications that were applied.…”

“…Preprocessing steps included correction for various artifacts (noise (Veraart et al, 2016), Gibbs ringing (Kellner, Dhital, Kiselev, & Reisert, 2016), motion and eddy currents (Andersson & Sotiropoulos, 2016), susceptibility induced distortions (Andersson, Skare, & Ashburner, 2003) and bias field inhomogeneities (Zhang, Brady, & Smith, 2001)), registration of subjects' high-resolution anatomical images to diffusion space (Jenkinson, Bannister, Brady, & Smith, 2002;Jenkinson & Smith, 2001), and segmentation of the anatomical images into gray matter, white matter and cerebrospinal fluid (Zhang et al, 2001). Further, quantitative whole-brain probabilistic tractography was performed using MRtrix3 (Tournier et al, 2019), resulting in 7.5 million streamlines per subject (more details are available in Aerts et al, 2018). Structural connectivity (SC) matrices were then constructed by transforming each individual's FreeSurfer parcellation scheme to the diffusion MRI data and calculating the number of estimated streamlines between each pair of brain regions.…”

“…Procedures for simulating large-scale brain dynamics and optimizing model parameters were also identical to those applied to the pre-operative data, as described in detail in Aerts et al (2018). Briefly, local dynamics for each of the 68 cortical brain regions were simulated using Reduced Wong-Wang neural mass models (Deco et al, 2014), which faithfully approximate the mean dynamics of interacting populations of excitatory and inhibitory spiking neurons.…”

“…In addition, inhibitory synaptic weights (Ji) -which control the strength of connections from inhibitory to excitatory mass models within each large-scale region i -were automatically tuned in each iteration of the parameter space exploration, to clamp the average firing rate at 3 Hz for each excitatory mass model (Deco et al, 2014;Schirner et al, 2018). After simulations, the obtained local inhibitory connection strengths were corrected for their respective region size for further analyses, since the need for local inhibition to balance global excitation depends on the total connection strength a brain region has, which tightly correlates with region size (Aerts et al, 2018). Median Ji values (both corrected for region size as well as uncorrected) across the entire brain and across tumor and non-tumor regions in brain tumor patients were then computed per subject.…”

“…In prior work (Aerts et al, 2018) we investigated brain dynamics before tumor resection in 25 brain tumor patients and 11 healthy control subjects using The Virtual Brain (TVB) (Sanz Leon et al, 2013); an open-source neuroinformatics platform that enables the construction, simulation and analysis of large-scale brain network models. In particular, we optimized model parameters of the Reduced Wong-Wang model (Deco et al, 2014) on an individual basis, after which we compared the fitted parameters between brain tumor patients and healthy control subjects.…”