upsampling and runwiseFIR

[jongmin-529]

Dear developers,

I have a question about the upsampling.

function: GLMestimatesingletrial(design_matrix, original_data, STIMULUS_DURATION, INPUT_TR, save_here, opt)

My original TR is 2 s, stimulus duration is 0.5 s, and the onset of stimulus is NOT TR-locked.
When I ran GLMsingle based on the original information (STIMULUS_DURATION = 0.5; INPUT_TR = 2), I got the figure 1.

It seems bad.
So, I upsampled data ( upsample_data = tseriesinterp(original_data, 2, 1, 4) ; original TR = 2, new TR = 1) and adjusted design matrix, and I got the figure 2 (STIMULUS_DURATION = 0.5; INPUT_TR = 1).

it still looks bad. so I changed INPUT_TR to 2 and used upsampled data and adjusted design matrix. The figure 3 looks okay.

So, should I use the original TR as INPUT_TR after upsampling the original data?

Thank you very much in advance.

[kendrickkay]

I agree with your general assessments of quality/sanity in the various results. However, the last change you did is not conceptually correct (even though the figure might look better). Also, the last figure has FIR timecourses that are positive peaking, but the peak comes unexpectedly too late.

I suspect that there may be design matrix coding errors going on somewhere.

If you upsample to X, then you should set the input TR to X when you call GLMsingle.

You mention that your design isn't exactly TR locked. If that's the case, you should carefully create the design matrix for whatever upsampled TR you are going to ultimately use. And, you'll have to eventually round to the nearest TR. I would suggest double-checking that your design matrix is correct.

[nihaoxiaoli]

Hello, did you finally find out the cause? My runwiseFIR is also acting strangely at the moment, and I can't figure out the reason.