Each subject performed 8 runs. Since the scanning time was too long, the 8 runs tasks were divided into two scan time. The subjects performed 4 runs+1 high resolution anatomical images at first. After some rest “outside the scan room”, subjects performed other 4 runs+1 high resolution anatomical images. The question is how to combine these two scanned data at firstlevel in SPM. Do I model all the 8 runs into one design matrix? Or do I model two 4 runs into two design matrix respectively, and then combine these two scanned data using second level? Please give me any suggestion.
The main problem is that two scanning sessions have two different brain calibration. After the rest outside the scan room, subjects need another brain calibration before the second scanning session. Since the head movements (realignment parameters) are based on the first brain calibration, I wonder if the two scanning sessions (include realignment parameters) can be modeled into a design matrix.
A quick answer is you should simply combine the two scanning sessions into one design matrix. As long as you model each run separately (with run-specific regressors controlling for baseline differences between runs), things will be fine.
A longer answer: in theory it should make little difference between the two approaches. But including both scanning sessions will give you a bit edge on statistical power, since you're having more data points in one grand model. Also the smoothness estimation on your data should be better with more data point, thereby reducing the threshold for multiple comparison correction using random field theory.
You should include all your runs in one design matrix. If you will combine two sessions from each subject in the second level design you will double the number of subjects and the results would be not valid because it will reduce the variance of your sample and therefore it will effect on the results of statistic tests.
The main problem is that two scanning sessions have two different brain calibration. After the rest outside the scan room, subjects need another brain calibration before the second scanning session. Since the head movements (realignment parameters) are based on the first brain calibration, I wonder if the two scanning sessions (include realignment parameters) can be modeled into a design matrix.
Just for further discussion: if you have two separate models (aka design matrices) and therefore two beta image for the same condition, in the 2nd level you could use flexible factorial design and specify them as within-subject conditions and the problem raise by Roza can be solved.
It is definitely possible to have 2 different scan sessions in 1 design matrix and there are many ways to do it. Here are some ideas: in a GLM regression of individual sessions use the obtained constants to normalize sessions, or use a regression on both sessions simultenously, containing a factor that represents each session (0,1) and remove the obtained fit.
Cheers, Wouter
Dear Ping, regarding your question:
"The main problem is that two scanning sessions have two different brain calibration. Since the head movements (realignment parameters) are based on the first brain calibration, I wonder if the two scanning sessions (include realignment parameters) can be modeled into a design matrix."
- You need to preprocess your data separately for each experimental session (4runs+1 anatomical / then 4runs+1 anatomical). This is important not only for realigment but also for co-registration. It will cover your brain callibrations (two independent preproc) and then you can insert all 8 runs into a single design matrix in first level analysis. Hope it helps!
how the 8 runs into a single design matrix in the first level analysis? In my study, ı have 2 runs. I dont know how to combine them into a single design matrix. When I enter 2 sessions (even if the condition names in the different runs are the same) in the first level analysis section, although there are 4 conditions, it seems like there are 8 conditions. Could you help me?
Met the same problem but didn't find an answer about how to combine two scanned data (with two different timing) at first level in SPM
I encountered the same problem. I had two sessions with 3 runs in each session, and the subjects came out for a rest during the sessions. However, because I was doing the fMRI data of the dogs, I preprocessed and did slice-timing, realignment, coregister, and smooth, did not do segmentation and normlization. In the first-level analysis, when the 6 runs of two sessions were placed in a design matrix, an error was reported: "The images do not all have the same orientation and/ or voxel sizes..." What should I do, thank you
- Badges
- Science topic