Questions about SurfClust and slow_surf_clustsim.py

Hi all. I recently conducted a typical GLM on a set of fMRI data. The data is on the surface space (std.141 surface generated by SUMA). I am particularly interested in whether the regression coefficient of a task-related regressor is statistically significant different from zero at the group-level. I first used 3dttest++ to obtained p value at the node level, and clustered contiguous nodes that pass this first-level alpha value into cluster. I then used slow_surf_clustsim.py to estimate the cluster-level size threshold. I had used several online tutorials as my reference and believe that the steps I took are reasonable.

I am aware of the issue about the inflation of false-positive rate related to cluster simulations (e.g., Eklund, Nichols, and Knutsson, 2016 PNAS; Cox et al., 2017 Brain connectivity), and I am aware of the newer methods provided in function 3dClustSim (e.g., the -acf option) which leads to better control of false-positive rate. My question is how this issue apply to the cluster simulations conducted on the surface space, especially those implemented using SurfClust and slow_surf_clustsim.py.

  1. I had ensured that the targeted smoothness level are set at the same value (fwhm) for my data, and for cluster simulation. Is this good enough? That is, I should not be too worried about the mismatch between the noise spatial extent in the real data, and that of the simulated data.
  2. Are there any references/papers that provide more detailed descriptions about what's implemented in SurfClust and slow_surf_clustsim.py. Are they the same as the simulations conducted in 3dClustSim, but only that it's on the surface? Or there are critical differences that one should be aware of?

Thanks.

Hi Shin,

The main difference is that slow_surf_clustsim.py uses the simple Gaussian ACF function, not the more recommended mixed-model form that combines both linear and squared exponential decay. But we do not have a mixed-model ACF method for surface analysis right now.

  • rick