NITRC Resting State Hemodynamic Response Function Retrieval and Deconvolution (RS-HRF) Forum: help

RE: Assistance with rs-HRF Toolbox for HRF Estimation from ROI Time Series

daniele marinazzo — Thu, 20 Feb 2025 11:10:37 GMT

Indeed it's supposed that when the data has been already organized in ROIs, the nuisance regression has been made already. The temporal mask can be defined from the framewise displacement, as explained on page 8 of the matlab manual. It indicates if there are points in time that should be excluded (akin to the "scrubbing" approach by Power et al.)

RE: Assistance with rs-HRF Toolbox for HRF Estimation from ROI Time Series

abir — Thu, 20 Feb 2025 10:52:00 GMT

Dear Guorong Wu,
Thank you again for this information, I truly appreciate it.
After reviewing the documentation in detail, I have some ambiguity regarding the nuisance.txt file. Specifically, when working with ROI signals stored in .mat files, is it correct that there is no need to download this regression variable file?
Additionally, could you please clarify how I can define the temporal mask to exclude spurious events?
Thanks again,
Best, Abir

RE: Assistance with rs-HRF Toolbox for HRF Estimation from ROI Time Series

guorong wu — Tue, 18 Feb 2025 13:54:03 GMT

ensure the ROI signals matrix is structured as (time points)*(number of ROIs)

RE: Assistance with rs-HRF Toolbox for HRF Estimation from ROI Time Series

abir — Tue, 18 Feb 2025 13:34:24 GMT

Dear Guorong Wu,
Thank you for your clarifications, they are greatly appreciated.
I encountered another issue after preprocessing the fMRI data using the CONN toolbox. I obtained a single rp.txt file for regressors, and for each subject, I obtained an ROI.mat file containing time series for all segmented ROIs.
To model the HRF response for the DMN, I extracted the time series corresponding to the DMN from the ROI.mat file. However, when I ran the rsHRF toolbox, I encountered an error indicating a size mismatch between rp.txt and DMN_time_series.mat.
Could you please help me resolve this issue?
Best regards,
Abir.

RE: Assistance with rs-HRF Toolbox for HRF Estimation from ROI Time Series

guorong wu — Tue, 18 Feb 2025 5:34:58 GMT

you should select 'Signals' for ROI time series analysis.

Assistance with rs-HRF Toolbox for HRF Estimation from ROI Time Series

abir — Mon, 17 Feb 2025 13:35:05 GMT

Dear Experts,
I recently discovered the rs-HRF toolbox for estimating and modeling the resting-state HRF from resting-state fMRI data.
I have successfully installed rs-HRF on SPM in MATLAB and followed the tutorials and documentation. However, I couldn't find clear instructions on estimating the HRF from .mat files containing preprocessed time series for each ROI.
Could you please clarify which option I should select in the GUI (Voxels, ROI volume, Signals, Vertices, etc.) to estimate the HRF from my ROI time series?
Your guidance would be greatly appreciated.
Best regards,
Abir.

RE: Statistical analysis

daniele marinazzo — Wed, 26 Jun 2024 8:21:25 GMT

Hi
 
sorry for the late reply. You can extract some shape parameters (width, height, time to peak) and use them as (individual or multiple) voxel-wise dependent variables, or use the full shape of the HRF, for example using AFNI's 3dMVM or 3dMSS.

RE: Batch editing

daniele marinazzo — Wed, 26 Jun 2024 8:18:52 GMT

Following the idea of choosing the most agnostic function, would you suggest to use the smoothed FIR instead of the standard FIR?
Both options are possible, sFIR allows for more spatial smoothing.
Based on the function selected, could the default parameter (32) for HRF's length be ok as a standard value or does it need to be optimised?
This is measured in seconds, so it's automatically adjusted to the TR. If for some reason you expert your HRF to be longer than 32 seconds (or much shorter), you can change this.
 
I still have some doubt about the meaning of k. Could be the default value (k=2) a good choice?
It's the range over which we search for the optimal lag (in TRs) between a pseudo-event and the HRF onset. The default is ok, unless you have a very short TR (0.2s or so).
 
we preprocessed rs-fMRI data using CONN toolbox. Which nuisance covariates in your opinion should be insered in the batch?
 If CONN is used for denoising, simply skip the denoising step in rsHRF (again, the default opinion is to do nothing). The order of processing might have an influence https://royalsocietypublishing.org/doi/10.1098/rsta.2015.0190

RE: Batch editing

lxop — Tue, 25 Jun 2024 10:33:21 GMT

Dear Daniele, 
Thank you for your kind reply.
Based on your feedback we'll adopt the FIR function. Following the idea of choosing the most agnostic function, would you suggest to use the smoothed FIR instead of the standard FIR?
Based on the function selected, could the default parameter (32) for HRF's length be ok as a standard value or does it need to be optimised?
 
See eq. [3] of the tutorial paper, where K is defined as the range of x values in which we look for a peak.: 
<div style="padding-left: 40px;">y(i) is defined as peak: y(i)>y(i+x),  x=[-K: K];     </div>
<div style="padding-left: 40px;"> </div>
<div>Thank you for the explanation. However, I still have some doubt about the meaning of k. Could be the default value (k=2) a good choice?</div>
<div class="yj6qo ajU"> </div>
Then, before the estimation of the HRF, we preprocessed rs-fMRI data using CONN toolbox. Which nuisance covariates in your opinion should be insered in the batch?
Thank you so much for your help
Sincerely
 
Beatrice
 
 
Originally posted by daniele marinazzo:
<blockquote>
Dear Beatrice
thanks for your questions, please see some replies inline below
<blockquote>
 
First, is there a way to choose the best HRF basis function based on my data?
There's no universal rule. The gamma mixtures will best reproduce a wide variety of shapes. FIR is more agnostic and data driven, and possibly more suitable if your target property is the whole hrf shape, for example with afni 3dMSS. It's also more suitable for unusual shapes, like animal or pathology.
Could be a correct approach to estimate both canonical and gamma functions and then compare them? If so, how should I do it operatively?
Unless there's a "ground truth" to compare to, then it's difficult. Indicators such as spatial or temporal variance can be useful though.
In the case of gamma function, how can I estimate the number of basis functions?
It's a choice you make, rather than an estimation. To avoid overfitting, the number should not be too high.
In Threshold (SD) for event detection should I change the value (i.e 1)? (considering both the options: canonical and gamma function)
This is not necessary, but it's indeed a parameter that you can choose to tune.
Also, what is K (local peak)? How is it calculated?
See eq. [3] of the tutorial paper, where K is defined as the range of x values in which we look for a peak.: 
<div style="padding-left: 40px;">y(i) is defined as peak: y(i)>y(i+x),  x=[-K: K];     </div>
<div class="yj6qo ajU"> </div>
</blockquote>
I hope this helps
 
</blockquote>

RE: Batch editing

daniele marinazzo — Tue, 25 Jun 2024 7:32:59 GMT

Dear Beatrice
thanks for your questions, please see some replies inline below
<blockquote>
 
First, is there a way to choose the best HRF basis function based on my data?
There's no universal rule. The gamma mixtures will best reproduce a wide variety of shapes. FIR is more agnostic and data driven, and possibly more suitable if your target property is the whole hrf shape, for example with afni 3dMSS. It's also more suitable for unusual shapes, like animal or pathology.
Could be a correct approach to estimate both canonical and gamma functions and then compare them? If so, how should I do it operatively?
Unless there's a "ground truth" to compare to, then it's difficult. Indicators such as spatial or temporal variance can be useful though.
In the case of gamma function, how can I estimate the number of basis functions?
It's a choice you make, rather than an estimation. To avoid overfitting, the number should not be too high.
In Threshold (SD) for event detection should I change the value (i.e 1)? (considering both the options: canonical and gamma function)
This is not necessary, but it's indeed a parameter that you can choose to tune.
Also, what is K (local peak)? How is it calculated?
See eq. [3] of the tutorial paper, where K is defined as the range of x values in which we look for a peak.: 
<div style="padding-left: 40px;">y(i) is defined as peak: y(i)>y(i+x),  x=[-K: K];     </div>
<div class="yj6qo ajU"> </div>
</blockquote>
I hope this helps