We have a question concerning which option represents the best estimate of baseline, during calculating % signal change:
Since we use polort = 2 in 3dDeconvolve (which better fits our data)
would the model be accurately represented by :
z(t)= b0 + b1*t + b2*(t^2) + y(t) + error(t)
We have read in a previous message, baseline represented as:
bave = b0 + b1*(NT/2), where NT is the number of time points
but we assume this baseline represents using polort = 1
so, if our model assumes polort = 2 then what would the baseline become?
We have guessed it to be:
b0 + b1*(NT/2) + b2*(NT+1)(2*NT+1)/6
approximately = b0 + b1*(NT/2) + b2*(NT^2)/3
In an earlier message, B.Ward stated "the linear trend is small relative to the baseline (otherwise, the percent signal change relative to baseline wouldnt be very meaningful)". Our question is, does this remain the same given the addition of the polynomial trend in our equation above?
Because, we looked at data after running 3dDeconvolve, and we found for a highly active voxel, the baseline itself has a -1.5% ~ 4 % change relative to the average of the baseline across the complete time series.(E.g., the average baseline of 2233- the maximum baseline value is 2322 the minimum is 2200).
Or is this a better idea for calculating baseline for each time point t:
y(t)
---------------------------------
b0 + b1 * t + b2 * (t^2)
We think this may better reflect the real % change for each time point.
Any ideas?
