This document describes the Bayesian Fast Accurate Spatial Tricks in 3D (BFAST3D) code that can be used to run fMRI analysis with spatial priors using empricial Bayes and the Matérn prior (EBMatern, Sidén et al. 2020) and Spatial variational Bayes (SVB) and Markov chain Monte Carlo (MCMC) methods for fMRI analysis in Sidén et al. (2017). The code is an add-on to the SPM12 software (Ashburner et al., 2013) and to its Bayesian single subject method (Penny et al., 2003, 2005b,a, 2007; Penny and Flandin, 2005). Posterior standard deviations and PPMs are computed using the simple RBMC method described in Sidén et al. (2018). To use the code, follow these steps:
(1) Download/duplicate your spm12-directory (downloadable at http://www.fil.ion.ucl.ac.uk/spm/software/spm12/).
(2) Delete the spm_spm_vb.m-file in the new spm12-directory.
(3) Copy the svb-folder from this package into the new spm12-directory.
(4) Run mexAll.m to compile the C++ code needed for EBMatern.
runExample.m gives an example on how to run the code and calls the functions runEBMatern.m, runSVB.m and runMCMC.m which are adapted for the OpenNEURO dataset ds107, which is automatically downloaded in the example. Just specify the correct paths in settings_ds107_10.m and you are ready to go.
In the current version, various settings are defined in different files and some of these are described below:
- For the EB Matern-method, settings are changed in runEBMatern.m, e.g. the number of optimization iterations, warmup iterations and PCG tolerance d can be changed by changing the variables maxiter, nWarmupIterations, and PCGtol.
- For the EB Matern-method various spatial priors are available: Matern1Iso and Matern2Iso are isotropic Matérn priors with alpha = 1 or 2, Matern2Aniso is the anisotropic Matérn prior and Eye is an independent shrinkage prior. These can be changed by changing hrfPrior in runEBMatern.m. Hyperparameters can be chosen to be fixed or optimixed.
- For the MCMC-method, settings are changed in runMCMC.m, e.g. the number of MCMC iterations, warmup iterations, thinning factor and PCG tolerance d can be changed by changing the variables niter, warmup, thinningFactor and PCGTol.
- To not run SVB in parallel, in spm12/svb/2D/spm_spm_vb.m or spm12/svb/3D/spm_spm_vb.m, change the variable SPM.ParallelGMRFSampling to 0.
- Change the number of SVB iterations in spm12/svb/2D/spm_spm_vb.m or spm12/svb/3D/spm_spm_vb.m, variable SPM.PPM.maxits (default 50).
- In the beginning of spm12b/svb/spm_svb_init.m one can change some SVB settings, e.g. the number of expectation approximation samples Ns (default 100) and the PCG tolerance d (default 10^-8).
The code requires MATLAB_R2016a or later. Post-processing computePPMs.m depends on the Tools for NIfTI and ANALYZE imagepackage (http://www.mathworks.com/matlabcentral/fileexchange/8797-tools-for-nifti-and-analyze-image). To compute joint PPMs, R is required and so is the excursions-package (Bolin and Lindgren, 2015), and also the R-package R.matlab. Be aware that for large data sets the time and memory requirements can be quite demanding, especially for MCMC, which can be quick-fixed by limiting the number of iterations.
- Ashburner, J., Barnes, G., Chen, C.-c., Daunizeau, J., Moran, R., Henson, R., Glauche, V., and Phillips, C. (2013). SPM12 Manual The FIL Methods Group ( and honorary members ). Functional Imaging Laboratory, pages 475–1.
- Bolin, D. and Lindgren, F. (2015). Excursion and contour uncertainty regions for latent Gaussian models. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 77(1):85–106.
- Penny, W. and Flandin, G. (2005). Bayesian analysis of fMRI data with spatial priors. In Proceedings of the Joint Statistical Meeting (JSM). American Statistical Association.
- Penny, W., Flandin, G., and Trujillo-Barreto, N. (2007). Bayesian comparison of spatially regularised general linear models. Human Brain Mapping, 28(4):275–293.
- Penny, W., Kiebel, S., and Friston, K. (2003). Variational Bayesian inference for fMRI time series. NeuroImage, 19(3):727–741.
- Penny,W. D., Trujillo-Bareto, N., and Flandin, G. (2005a). Bayesian analysis of single-subject fMRI data: SPM implementation. Technical report, Wellcome Department of Imaging Neuroscience.
- Penny, W. D., Trujillo-Barreto, N. J., and Friston, K. J. (2005b). Bayesian fMRI time series analysis with spatial priors. NeuroImage, 24(2):350–362.
- Sidén, P., Eklund, A., Bolin, D., and Villani, M. (2017). Fast Bayesian whole-brain fMRI analysis with spatial 3D priors. NeuroImage, 146:211–225.
- Sidén, P., Lindgren, F., Bolin, D., and Villani, M. (2018). Efficient covariance approximations for large sparse precision matrices. Journal of Computational and Graphical Statistics, 27(4), 898-909.
- Sidén, P., Lindgren, F., Bolin, D., and Villani, M. (2020). Spatial 3D Matérn priors for fast whole-brain fMRI analysis. arXiv preprint.