help
help > RE: Compcor:Is linear trend removed prior to pca?
Sep 17, 2015 05:09 PM | Alfonso Nieto-Castanon - Boston University
RE: Compcor:Is linear trend removed prior to pca?
Hi Immanuel,
Yes, CONN will actually go a little bit further and, by default, it will remove, in addition to linear trends, any effect already modeled by your first-level covariates and condition task effects from the White/CSF BOLD signal before computing its principal component decomposition. None of this is strictly necessary but it helps in focusing the resulting components on 'unknown' sources of BOLD signal noise and it makes the aCompCor procedure more robust (linear trends and other 'known' sources of BOLD signal confounding effects are going to be explicitly removed in the denoising step by adding specific regressors that account for these effects together with the aCompCor regressors, so aCompCor components within the same 'known' subspace would have simply no additional effect)
Hope this helps
Alfonso
Originally posted by Immanuel Elbau:
Yes, CONN will actually go a little bit further and, by default, it will remove, in addition to linear trends, any effect already modeled by your first-level covariates and condition task effects from the White/CSF BOLD signal before computing its principal component decomposition. None of this is strictly necessary but it helps in focusing the resulting components on 'unknown' sources of BOLD signal noise and it makes the aCompCor procedure more robust (linear trends and other 'known' sources of BOLD signal confounding effects are going to be explicitly removed in the denoising step by adding specific regressors that account for these effects together with the aCompCor regressors, so aCompCor components within the same 'known' subspace would have simply no additional effect)
Hope this helps
Alfonso
Originally posted by Immanuel Elbau:
Hi Alfonso,
I have a question regarding the aCompCor implementation in CONN:
In the Bhezadi et al.(2007) paper the authors describe that they removed the linear trend from the time series prior to performing pca within the respective ROI, wm or csf: ("Voxel time series from the noise ROI (either anatomical or tSTD) were placed in a matrix M of size N×m, with time along the row dimension and voxels along the column dimension. The constant and linear trends of the columns in the matrix M were removed prior to column-wise variance normalization. The covariance matrix C=MMT was constructed and decomposed into its principal components using a singular value decomposition.")
Is this step implemented in CONN? Also, is it necessary?
Thank you in advance!
Best,
Immanuel
I have a question regarding the aCompCor implementation in CONN:
In the Bhezadi et al.(2007) paper the authors describe that they removed the linear trend from the time series prior to performing pca within the respective ROI, wm or csf: ("Voxel time series from the noise ROI (either anatomical or tSTD) were placed in a matrix M of size N×m, with time along the row dimension and voxels along the column dimension. The constant and linear trends of the columns in the matrix M were removed prior to column-wise variance normalization. The covariance matrix C=MMT was constructed and decomposed into its principal components using a singular value decomposition.")
Is this step implemented in CONN? Also, is it necessary?
Thank you in advance!
Best,
Immanuel
Threaded View
| Title | Author | Date |
|---|---|---|
| Immanuel Elbau | Sep 17, 2015 | |
| Alfonso Nieto-Castanon | Sep 17, 2015 | |
| Immanuel Elbau | Nov 15, 2015 | |
| Alfonso Nieto-Castanon | Nov 16, 2015 | |
| Immanuel Elbau | Oct 21, 2015 | |