Hi Steve & Jeff,
Adding the derivatives does not hurt the model (usually) but the question is rather what you'll be doing with the parameters of those extra-regressors.
At the first level, i.e. fMRI time series of each subject, the HRF derivatives will capture the variance associated with the HRF shape (earlier/later, thinner/fatter) but since these regressors are orthogonal to the one containing canonical HRF, the amplitude of the HRF response is still solely captured by the canonical HRF regressor! The inclusion of the derivatives will thus affect stats on the canonical parameter performed at the 1st level (FFX) : the parameter estimate (the numerator in the t-test) will remain the same but potentially more variance is captured and the residual variance is reduced, so reducing the denominator in the t-test. At the 2nd level (RFX), if you only pass the beta's of the canonical HRF from each subject, it will not matter for your stat results if you included or not the derivatives at the 1st level.
Now if the question you have is about the difference in amplitude *and/or shape* of the response between your 2 groups, you need to include the beta's for the canonical *and* derivates in your GLM. Then, when building contrasts be careful not to mix up the different types of beta's as it does NOT make sense to average (or differentiate) canonical with derivatives. To look for any difference (
in amplitude and/or shape) between the 2 groups, you would need to use an F-test with one row per image type, i.e. one for the difference in canonical, one for the 1st derivative, and one more for the 2nd derivative.
BTW more flexible HRF models (gammas or FIR) will operate like canonical+derivatives but with a less constrained interpretation of the meaning and interpretation of the associated parameters.
HTH,
Chris
Christophe Phillips, Ir., Ph.D.
FRS-FNRS Senior Research Associate
& Associate Professor
GIGA in silico medicine &
Cyclotron Research Centre
University of Liège, B30
4000 Liège, Belgium
T: +3243662316
F: +3243662946
From: "Jeff Browndyke" <
[log in to unmask]>
To: [log in to unmask]Sent: Wednesday, 29 April, 2020 23:00:09
Subject: Re: [SPM] Event-related design in aging study: Temporal and dispersion derivatives?
Hi, Steve.
I think it is a wise idea to include the 2nd order derivatives, particularly if your older adult cohort has a history of cardiac, statin use or hypertension-related issues. Inclusion of these derivatives probably should also be guided by your task structure and event timing, but that’s a bit beyond my wheelhouse.
Hope this helps,
Jeff
Duke University Medical Center
Div. of Geriatric Behavioral Health
Dear SPM experts,
Please, could you give to me some suggestions about my previous email?.
Thanks for your time.
Best regards,
Dear all,
I am conducting a fMRI study (event-related design) in old adults. As I understand, aging can induce changes in latency and amplitude of the haemodinamuc response (that´s right?).
Taking into account this, I was wondering if it is a good choice uses the Canonical HRF with time and dispersion derivatives in my first level. More generally speaking, in which cases is it advisable to include derivatives?
Thank you in advance.
Best regards,
Steve.