help
help > RE: Using CONN's ART output in other software
Aug 26, 2015 03:08 AM | Alfonso Nieto-Castanon - Boston University
RE: Using CONN's ART output in other software
Hi Darren,
Yes, ART was developed well before CONN and you should be able to use it in the context of other software processing pipelines without an issue. Of course, depending on the specific algorithm used for preprocessing, the estimated motion parameters and global BOLD signal used by ART to identify outliers will be slightly different but I would expect those differences to be relatively minor.
Regarding the specific placement of ART within your preprocessing pipeline, generally ART can be run at any point in the processing pipeline (after realignment, of course, since it uses the subject motion parameters generated at this step). The only difference in its output, depending on where exactly within the pipeline it is run, arises from potential differences in the estimated global BOLD signal at those different points in the pipeline. The effect of coregistration and normalization will typically be very small since those have very small impact on the global signal, followed probably by smoothing, which as a small but noticeable effect, and slice-timing correction, which can have a larger effect due to the slice-specific temporal interpolation procedure. There are arguments to place it as close to the end as possible (e.g. so that the identified outliers most closely represent outliers in the actual data that will be entered into the first-level models) but also to place it as close to the realignment step as possible (so that the expected distribution of motion/global measures more closely resembles that used to determine the standard/recommended thresholds). The default placement, after normalization but before smoothing, is a compromise between these two points.
Last, regarding the specific order of the preprocessing steps, note that the default CONN pipeline uses realignment before slice-timing correction (while you listed the opposite order). A general recommendation is to use the one that is expected to have the largest impact first (so for example I would use your order only in cases where the TR is relatively high and I am confident that there is low in-scanner motion).
Hope this helps
Alfonso
Originally posted by Darren Yeo:
Yes, ART was developed well before CONN and you should be able to use it in the context of other software processing pipelines without an issue. Of course, depending on the specific algorithm used for preprocessing, the estimated motion parameters and global BOLD signal used by ART to identify outliers will be slightly different but I would expect those differences to be relatively minor.
Regarding the specific placement of ART within your preprocessing pipeline, generally ART can be run at any point in the processing pipeline (after realignment, of course, since it uses the subject motion parameters generated at this step). The only difference in its output, depending on where exactly within the pipeline it is run, arises from potential differences in the estimated global BOLD signal at those different points in the pipeline. The effect of coregistration and normalization will typically be very small since those have very small impact on the global signal, followed probably by smoothing, which as a small but noticeable effect, and slice-timing correction, which can have a larger effect due to the slice-specific temporal interpolation procedure. There are arguments to place it as close to the end as possible (e.g. so that the identified outliers most closely represent outliers in the actual data that will be entered into the first-level models) but also to place it as close to the realignment step as possible (so that the expected distribution of motion/global measures more closely resembles that used to determine the standard/recommended thresholds). The default placement, after normalization but before smoothing, is a compromise between these two points.
Last, regarding the specific order of the preprocessing steps, note that the default CONN pipeline uses realignment before slice-timing correction (while you listed the opposite order). A general recommendation is to use the one that is expected to have the largest impact first (so for example I would use your order only in cases where the TR is relatively high and I am confident that there is low in-scanner motion).
Hope this helps
Alfonso
Originally posted by Darren Yeo:
Hi Alfonso,
I hope you're doing well. I would like to use CONN's ART output, in particular, the number of outlying volumes for each subject, as an exclusion criterion. While it is straightforward to make use of that information within CONN, I would also like to use it with other neuroimaging software such as BrainVoyager to cross-validate my results.
To ensure that the cross-platform analyses are valid, I tried to match the preprocessing pipelines implemented in CONN and BrainVoyager:
1) functional slice-timing correction
2) functional realignment & unwarp
3) functional coregistration to structural
4) structural segmentation & normalization
5) functional normalization
6) functional outlier detection (ART-based)
7) functional smoothing
My concerns are: Is it fine to use the ART output from this preprocessing pipeline in other software given the same order of preprocessing steps (albeit different algorithms for interpolation and others)? Generally, to what extent of preprocessing of functional data (i.e., after slice-timing correction, after realignment or after both) be used as input for the ART-based functional outlier detection?
I would greatly appreciate your advice.
Thank you!
Best,
Darren
I hope you're doing well. I would like to use CONN's ART output, in particular, the number of outlying volumes for each subject, as an exclusion criterion. While it is straightforward to make use of that information within CONN, I would also like to use it with other neuroimaging software such as BrainVoyager to cross-validate my results.
To ensure that the cross-platform analyses are valid, I tried to match the preprocessing pipelines implemented in CONN and BrainVoyager:
1) functional slice-timing correction
2) functional realignment & unwarp
3) functional coregistration to structural
4) structural segmentation & normalization
5) functional normalization
6) functional outlier detection (ART-based)
7) functional smoothing
My concerns are: Is it fine to use the ART output from this preprocessing pipeline in other software given the same order of preprocessing steps (albeit different algorithms for interpolation and others)? Generally, to what extent of preprocessing of functional data (i.e., after slice-timing correction, after realignment or after both) be used as input for the ART-based functional outlier detection?
I would greatly appreciate your advice.
Thank you!
Best,
Darren
Threaded View
| Title | Author | Date |
|---|---|---|
| Darren Yeo | Aug 25, 2015 | |
| Alfonso Nieto-Castanon | Aug 26, 2015 | |
| Darren Yeo | Aug 27, 2015 | |
| Alfonso Nieto-Castanon | Aug 28, 2015 | |