help
help > RE: Extremely high values of contrast estimates in SPM after using CONN preprocessed and denoised data
Sep 27, 2021 04:09 AM | Sneha Sheth - UBC
RE: Extremely high values of contrast estimates in SPM after using CONN preprocessed and denoised data
Hi everyone,
I am also facing the same issue of really high beta values as described by Avantika, and it isn't getting resolved even after switching off "grand mean scaling", "global normalization" and "Serial Correlations" in SPM.
I preprocessed the data in SPM first, then denoised in CONN and used the niftiDATA files (confound-corrected files generated by CONN) as input in SPM for first level analysis. The SPM second level output looks significantly cleaner after these steps as compared to the traditional approach of using only SPM preprocessed files for the GLM analysis (and skipping denoising in CONN).
However, when I try to conduct ROI analysis in MarsBar, I get really high beta values/parameter estimates from the denoised version of the SPM analysis. On the other hand, these values look normal on files both preprocessed & analyzed only in SPM (and without denoising in CONN).
Any ideas on what I can do?
Thanks
Sneha
Originally posted by avantika mathur:
I am also facing the same issue of really high beta values as described by Avantika, and it isn't getting resolved even after switching off "grand mean scaling", "global normalization" and "Serial Correlations" in SPM.
I preprocessed the data in SPM first, then denoised in CONN and used the niftiDATA files (confound-corrected files generated by CONN) as input in SPM for first level analysis. The SPM second level output looks significantly cleaner after these steps as compared to the traditional approach of using only SPM preprocessed files for the GLM analysis (and skipping denoising in CONN).
However, when I try to conduct ROI analysis in MarsBar, I get really high beta values/parameter estimates from the denoised version of the SPM analysis. On the other hand, these values look normal on files both preprocessed & analyzed only in SPM (and without denoising in CONN).
Any ideas on what I can do?
Thanks
Sneha
Originally posted by avantika mathur:
Hi
Alfonso,
The Global Normalization is already set to "None" in spm first level.
Attached is the image of the same.
Do you mean something else is supposed to be changed?
Avantika
Originally posted by Alfonso Nieto-Castanon:
The Global Normalization is already set to "None" in spm first level.
Attached is the image of the same.
Do you mean something else is supposed to be changed?
Avantika
Originally posted by Alfonso Nieto-Castanon:
Hi
Avantika,
I would suggest trying to set the "grand mean scaling" option in SPM first-level estimation off, since that looks like a possible culprit for this behavior (after band-pass filtering, the mean functional data is zero at every voxel, so global signal scaling -and similarly any other default mechanisms that rely on the average BOLD signal containing anatomical information/features- are likely to fail in rather unexpected ways). Let me know if that works
Best
Alfonso
Originally posted by avantika mathur:
I would suggest trying to set the "grand mean scaling" option in SPM first-level estimation off, since that looks like a possible culprit for this behavior (after band-pass filtering, the mean functional data is zero at every voxel, so global signal scaling -and similarly any other default mechanisms that rely on the average BOLD signal containing anatomical information/features- are likely to fail in rather unexpected ways). Let me know if that works
Best
Alfonso
Originally posted by avantika mathur:
Hi Conn users,
After following the following posts,
https://www.nitrc.org/forum/message.php?...
I used the alternative method to import conn preprocessed data in SPM which is the following :
Entering the preprocessed/denoised timeseries into SPM to perform the first-level analyses.
The data I am analyzing is children data thus, ART was used at liberal threshold in preprocessing [Global signal z value threshold 10, subject motion 5 mm]. I did not have the "effect of Condition X" entered as confounding effects during Denoising.
I used the file generated after conn preprocessing and denoising...the niftiDATA_Subject001_Condition000 and further defined first-level design matrices within SPM, specified masking threshold to -Inf in first level analysis [https://www.nitrc.org/forum/message.php?msg_id=14852].
After doing first level analysis and group level analysis [10 subjects], I get weird beta estimate values - which are extremely high . Attached are the bar plots for the same [1st bar-chart - single subject, 2nd bar chart - group of 10 subjects]. Beta values should not be this high.
Can someone direct me where I am going wrong?
Avantika
After following the following posts,
https://www.nitrc.org/forum/message.php?...
I used the alternative method to import conn preprocessed data in SPM which is the following :
Entering the preprocessed/denoised timeseries into SPM to perform the first-level analyses.
The data I am analyzing is children data thus, ART was used at liberal threshold in preprocessing [Global signal z value threshold 10, subject motion 5 mm]. I did not have the "effect of Condition X" entered as confounding effects during Denoising.
I used the file generated after conn preprocessing and denoising...the niftiDATA_Subject001_Condition000 and further defined first-level design matrices within SPM, specified masking threshold to -Inf in first level analysis [https://www.nitrc.org/forum/message.php?msg_id=14852].
After doing first level analysis and group level analysis [10 subjects], I get weird beta estimate values - which are extremely high . Attached are the bar plots for the same [1st bar-chart - single subject, 2nd bar chart - group of 10 subjects]. Beta values should not be this high.
Can someone direct me where I am going wrong?
Avantika
Threaded View
| Title | Author | Date |
|---|---|---|
| avantika mathur | Mar 2, 2019 | |
| Alfonso Nieto-Castanon | Mar 5, 2019 | |
| avantika mathur | Mar 7, 2019 | |
| Sneha Sheth | Sep 27, 2021 | |