help
help > RE: Order between denoising and smoothing
Aug 3, 2015 08:08 AM | Sami Abboud
RE: Order between denoising and smoothing
Dear Alfoonso,
It is perfectly clear now.
Thank you very much for taking the time to elaborate on this.
Have a good week!
Cheers,
Sami
Originally posted by Alfonso Nieto-Castanon:
It is perfectly clear now.
Thank you very much for taking the time to elaborate on this.
Have a good week!
Cheers,
Sami
Originally posted by Alfonso Nieto-Castanon:
Dear
Sami,
Yes, you are exactly right. To clarify, as long as H is the same (i.e. when the aCompCor components are extracted from the unsmoothed volumes), then the order of smoothing and regression does not affect the results. To be a bit more precise, among the following procedures:
1) extract the ROI signals (including White/CSF areas) from the unsmoothed volumes, smooth your functional volumes, and then regress the confounding/aCompCor effects from the smoothed functional volumes
2) extract the ROI signals (including White/CSF areas) from the unsmoothed volumes, regress the confounding/aCompCor effects from the unsmoothed volumes, and then smooth your residual functional volumes
3) smooth your functional volumes, extract the ROI signals (including White/CSF areas) from the smoothed volumes, and then regress the confounding/aCompCor effects
procedures (1) and (2) will lead to exactly the same results (in both cases H is the same), and those will be different from the results in procedure (3) (in this case H will be different since the White/CSF signals are now being extracted from the smoothed functional data). Also procedure (1) is the default pipeline in CONN. Following your notation the results of each procedure would be:
1) H(X_average1) * (X * C)
2) (H(X_average1) * X) * C
3) H(X_average2) * X * C
Hope this helps, and let me know your thoughts
Alfonso
Originally posted by Sami Abboud:
Yes, you are exactly right. To clarify, as long as H is the same (i.e. when the aCompCor components are extracted from the unsmoothed volumes), then the order of smoothing and regression does not affect the results. To be a bit more precise, among the following procedures:
1) extract the ROI signals (including White/CSF areas) from the unsmoothed volumes, smooth your functional volumes, and then regress the confounding/aCompCor effects from the smoothed functional volumes
2) extract the ROI signals (including White/CSF areas) from the unsmoothed volumes, regress the confounding/aCompCor effects from the unsmoothed volumes, and then smooth your residual functional volumes
3) smooth your functional volumes, extract the ROI signals (including White/CSF areas) from the smoothed volumes, and then regress the confounding/aCompCor effects
procedures (1) and (2) will lead to exactly the same results (in both cases H is the same), and those will be different from the results in procedure (3) (in this case H will be different since the White/CSF signals are now being extracted from the smoothed functional data). Also procedure (1) is the default pipeline in CONN. Following your notation the results of each procedure would be:
1) H(X_average1) * (X * C)
2) (H(X_average1) * X) * C
3) H(X_average2) * X * C
Hope this helps, and let me know your thoughts
Alfonso
Originally posted by Sami Abboud:
Dear Alphonso,
Thank you again for the very detailed response and another apology for my late one.
Following your notation, doesn't the projection H depend on your X_average? This means:
X_corrected_smoothed = H(X_average1) * X * C = H(X*w) * X * C
X_smoothed_corrected = X * C * H(X_average2) = H(X*C*w) * X * C
Where w is the WhiteMatter ROI's weight vector.
For your equation to hold, we need H(X*w)=H(X*C*w) - is it so?
(I hope that I'm not missing something that will deem my questions futile)
Cheers,
Sami
Thank you again for the very detailed response and another apology for my late one.
Following your notation, doesn't the projection H depend on your X_average? This means:
X_corrected_smoothed = H(X_average1) * X * C = H(X*w) * X * C
X_smoothed_corrected = X * C * H(X_average2) = H(X*C*w) * X * C
Where w is the WhiteMatter ROI's weight vector.
For your equation to hold, we need H(X*w)=H(X*C*w) - is it so?
(I hope that I'm not missing something that will deem my questions futile)
Cheers,
Sami
Threaded View
Title | Author | Date |
---|---|---|
Sami Abboud | Jun 24, 2015 | |
Alfonso Nieto-Castanon | Jun 29, 2015 | |
Sami Abboud | Jul 9, 2015 | |
Alfonso Nieto-Castanon | Jul 18, 2015 | |
Sami Abboud | Jul 18, 2015 | |
Alfonso Nieto-Castanon | Jul 18, 2015 | |
Sami Abboud | Jul 31, 2015 | |
Alfonso Nieto-Castanon | Aug 3, 2015 | |
Sami Abboud | Aug 3, 2015 | |