I've been working on an fMRI dataset with 80+ subjects. For second-level thresholding, I've started using AFNI's 3dClustSim to estimate the appropriate cluster extent threshold given a particular height threshold (e.g. p < .005). The best (i.e. least likely to inflate Type I errors) method requires specific smoothness estimates from the first-level analysis. The basic workflow is: Write residuals from model estimation Estimate smoothness from the residuals Average smoothness estimates across participants Use average estimates to condition 3dClustSim when determining thresholds Step 1. Write residuals The first step is the have SPM write out the residuals images during model estimation. There are two relatively straightforward was to accomplish this: If…
-
-
Visualizing signal dropout across subjects
One of my current projects involves working with some data that I did not collect. I have the smoothed and normalized NIfTI images and the first-level stats images, including SPM.mat and the con_*, beta_*, and spmT_* files. Generating the second-level images is straight-forward, but when I visualized simple effects, there was larger than expected signal dropout in vmPFC and temporal pole around the sinuses. It seemed plausible that across the 80+ subjects in the dataset, there might be a handful of subjects with abnormally high dropout. Losing one or two individual subjects seems like a reasonable price to pay to recover vmPFC signal, since the task involves social decision-making. Rather…