Denoising for Modeling: Questions About WM/CSF Regressors

[Szymon Tyras]

Hello,

I sincerely apologize for posting a question that is only partially related to modeling. However, I am using this approach for my TVB-related study, and this group has been incredibly helpful so fa.  

I am trying to calculate 5 aCompCor components for CSF and WM, but I’m not entirely sure if the steps I’m following are correct. Here’s what I’ve been doing:  

1. Start with CSF and WM probability maps in MNI space at a resolution of 1×1×1 mm, as well as the fMRI signal in the same space but at a lower resolution, such as 3×3×4 mm. 

2. Binarize the probability maps at their higher resolution using a chosen threshold. I erode the binary masks at the same resolution multiple times (I aim for heavily eroded masks, so I set the erosion iterations to 4× for WM and 2× for CSF unless the number of voxels in each mask falls below 5 after resampling to the fMRI resolution). 

3. Resample the eroded masks to the fMRI’s lower resolution. 

4. Using the WM or CSF mask, extract the fMRI signal. 

5. Detrend and z-score the signal without averaging (from my understanding, there are different approaches in the literature - some include high-pass filtering here, while others do not) 

5. Calculate the first 5 principal components from the extracted signal. 

6. Regress these components, along with other confounds, while applying filtering and detrending to the main signal (I am using Nilearn’s masker function with a Butterworth filter. How will orthogonalization interact with WM/CSF filtering and detrending from the earlier steps?) 

I would greatly appreciate it if you could point out any errors in my approach or discuss potential pitfalls. 

Thank you so much in advance for your time and insights!  

Best regards, 

Szymon

[Daniele Marinazzo]

That's ok. But I would mostly correct for systemic low frequency fluctuations and blood arrival time. If your target is the similarity between modelled and empirical FC, you can obtain good values even by using a fake signal shifted in time according to the blood arrival time.

--
You received this message because you are subscribed to the Google Groups "TVB Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tvb-users+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/tvb-users/279feff3-98b2-4302-95a7-ba798e31c405n%40googlegroups.com.

[Szymon Tyras]

Dear Daniele,  

Thank you so much for answering and reassuring me about my pipeline. Although I am targeting FCD distribution, I am aware of the systemic fluctuation problems from our previous discussions. I will definitely take it into consideration.