Processing block: regress - viewing EPI data at various locations Background: Now that we have seen where the results are significant for Vrel, Arel and even the contrast V-A, let us view this in relation to the EPI time series data. Recall that the input to 3dDeconvolve is the scaled EPI data (pb04.*.scale), or catenated in the single dataset, all_runs. Since each voxel was scaled over each run so that the mean became 100, we can view the EPI data as a percentage of the mean. Recall that at this voxel we have: Vrel beta = 2.325 Arel beta = 3.403 V-A contr = -1.08 (2.325 - 3.403) So the magnitude of the BOLD response for Vrel is 2.3 (percent change above baseline), but the Arel response was 3.4 percent above baseline. =========================================================================== View the EPI time series in the afni GUI: Leave controller [A] as it is along the top of the screen, and view the EPI data with controller [B] along the bottom. Set the Underlay dataset to be all_runs, and open an image and graph window. We don't care too much about the image window, since the scaled EPI data has lost anatomical contrast. The controllers will lock once we jump to a new location. - in controller [B], set Underlay to all_runs - [B] open axial image and graph windows, close all others (from [B]) - [B] axial image: jump-to-xyz 52 3.6 12 - [B] graph image: hit 'A' for autoscale (or Opt->scale->AUTO) note that this is a toggle button, the message on the terminal window should read: ++ Graph Viewer: Autoscale forced ON (so if it is now OFF, toggle it back on) For now, focus on just one time series. - [B] graph image: hit 'm' a couple of times, widen the window The BOLD response pattern should be clear here. -------------------- Note the run breaks. There are 3 runs of 150 TRs each. Make this clear in the graph window. - [B] graph image: Opt->Grid->Choose, and set to 150 - [B] graph image: Opt->Colors->Grid: set to yellow-orange Supporting our choice of run breaks, we see that the BOLD responses seem to die off and start again at the run breaks. -------------------- Display the ideal curve (regressor) for the Vrel stimulus class. - [B] graph image: FIM->Pick Ideal: set to ideal_Vrel.1D Recall that afni_proc.py used 1dcat to extract this from column 12 of the X-matrix via the command: % 1dcat X.xmat.1D'[12]' > ideal_Vrel.1D At each stimulus time in stimuli/AV1_vis.txt, 3dDeconvolve put down our requested basis function, 'BLOCK(20,1)', forming column 12 of the X-matrix (after columns 0-11, the 12 polynomial baseline regressors). For run 1, the times were 60, 90, etc. We see the first bump in the ideal regressor start 60 seconds (TR index 30) into the first run. So in the EPI data, it is clear which responses were from the Vrel stimulus class, while the others were from the Arel class. -------------------- Display the model fit time series on top of the data. Start the Dataset #N plugin as a Double Plot, and plot the fitts data in blue over the scaled data (still in black). - [B] graph image: Opt->Tran 1D: set to Dataset #N - Dataset #N window: set Input #1 to be fitts.FT - Dataset #N window: set the Color for Input #1 to be dk-blue - Dataset #N window: Set (either Keep or Close) There are a few things to note here. 1. The blue fitts curve is smoother than the black EPI curve. Remember that the fitts curve is a linear sum of our regressors, which include a cubic polynomial (per run), 2 BOLD response regressors (Vrel and Arel) and 6 motion curves. In this model, only the motion will appear noisy, while the polort and BLOCK() regressors are smooth. So the noisier the fitts data, the bigger the effect of motion at the given voxel. 2. Note the heights of the BOLD responses in the fitts data. In the fitts data, the heights of all Vrel responses are the same (that is our assumption in the regression), as are the the responses of Arel. It is sometimes hard to see, including the polort and especially the motion, but the response heights do appear the same. 3. Note that the heights of the Arel responses look higher than those of the Vrel. The first 2 responses of the first run are Arel, and they are higher than the next 3 responses which are Vrel. Recall again that at this voxel (xyz 52 3.6 12), the Arel beta is 3.4 while the Vrel beta is only 2.3 (percent). So at this voxel the V-A contrast is -1.08, because the response to Arel was bigger than the response to Vrel. This makes sense since this part of the brain responds to audio, while the visual area is in the posterior (we can see positive contrast values at the bottom of the axial image, for example). 4. Again, note that the EPI and fitts values hover around 100, which is the mean of every run. 5. By following the bottoms of the responses, one can get a sense of the polort baseline. In some cases it is fairly steady, in other cases it is clearly higher degree than even quadratic (so it is good that we used a cubic polynomial to model it).