fNIRS Course – DAY 3 – October 21, 2020

DAY 3 – October 21
10:00 am – 12:00 pm ET

– Short separation regression and GLM
0
Please leave a feedback on thisx
LECTURE AND EXERCISE

DAY 3 Video and Zoom chat:

Logged Zoom chat

 

29 Comments
Oldest
Newest Most Voted
Inline feedback
View all comments

Yuanyuan Gao
Yuanyuan Gao (@yuanyuan216)
1 day ago

What is the advantage of using additional short-separation detectors over using additional short-separation sources? In my experience, sources are cheaper than detectors.

David Boas
David Boas (@dboas)
Reply to  Yuanyuan Gao
1 day ago

for us it was harder to have a short separation source. It made our experiments more complex. Indeed, adding short separation detectors is more expensive. Something to think about for hardware designers.

Sue Peters
Sue Peters (@speters9)
1 day ago

With the consecutive Gaussian basis function, how is the trange taken into account? Since you’re not modelling a specific shape, I would assume that the trange doesn’t impact the results very much. E.g if we have 1 min of resting before and after the trial, can we use a longer trange (-20s to 20 post task termination)? Is there a downside to using this long trange?

David Boas
David Boas (@dboas)
Reply to  Sue Peters
1 day ago

trange is very important! It tells you when to start the basis functions and when to end then. So, I would typically go from -2 to 15 sec. And if my gaussian width is 1sec and step is 1sec, this will give me 18 consecutive gaussians in my basis functions for the GLM
There is no major downside to using a longer tRange except that the baseline estimate (i.e. before t=0) can be corrupted by physiological fluctuations for the many seconds before t=0 that you include in tRange.

Sue Peters
Sue Peters (@speters9)
Reply to  David Boas
1 day ago

Thanks. So what does the GLM do with the tRange of -2 to t=0? Is there any concern that the participant may be ‘anticipating’ the onset of the stimulus so we might want to give a longer baseline timeperiod to reduce the effect of anticipation on the overall HRF?

David Boas
David Boas (@dboas)
Reply to  Sue Peters
1 day ago

the best way to reduce anticipation effects is to jitter the time interval between stimuli. But putting a longer baseline period is also a good idea to help you visualize if there is an anticipatory effect

Yuanyuan Gao
Yuanyuan Gao (@yuanyuan216)
1 day ago

In Homer2 we have segment and downsample functions. Are we going to have them in Homer3?

David Boas
David Boas (@dboas)
Reply to  Yuanyuan Gao
1 day ago

I don’t fully recall these functions in Homer2… but we can certainly add them again.
 Stephen Tucker , want to add this to our to do list

Yuanyuan Gao
Yuanyuan Gao (@yuanyuan216)
Reply to  David Boas
1 day ago

Thanks. I actually used the segment function a lot in Homer2 when the data collector forgot to hit the stop button and recorded fNIRS data for another 1 hour with the cap off the subject head. In that case segment function helped me to extract the time period I need, instead of process the 1 hour file every time. And the probe prune function would prune all channels just because the last 1 hour data was all noise.

David Boas
David Boas (@dboas)
Reply to  Yuanyuan Gao
1 day ago

We will likely add these functions into SNIRF related tools as they operate on the SNIRF files. This is instead of incorporating it into Homer.

Hila Gvirts
Hila Gvirts (@hila-gvirtsgmail-com)
1 day ago

Is it possible to extract the beta value ? for statistical analysis?

David Boas
David Boas (@dboas)
Reply to  Hila Gvirts
1 day ago

It is possible! These results are all in the groupResults.mat file. And we do have instructions on how to extract results from groupResults.mat on the wiki here at https://github.com/BUNPC/Homer3/wiki/Output-Files

Sue Peters
Sue Peters (@speters9)
1 day ago

Meryem – in the painful stim dataset, it looks like subject 2’s HRF for HbR goes up vs down where the other subjects it looks like HbR goes down like we would expect. Any thoughts on why this might occur? is it physiological or is it artifact?

David Boas
David Boas (@dboas)
Reply to  Sue Peters
1 day ago

there is always variability between subjects. We don’t always see the same responses across all subjects

Karla Holmboe
Karla Holmboe (@karlaholmboe)
1 day ago

I (eventually) managed to run SS Exercise 1. Without knowing much about the task design and ROI, I found it hard to judge which hand was being stimulated. It seems to me that there’s a larger response on the right side of the probe in the painful condition, so would this be the left hand? Thanks!

David Boas
David Boas (@dboas)
Reply to  Karla Holmboe
18 hours ago

you got it!

YILEI ZHENG
YILEI ZHENG (@zhengyilei)
1 day ago

Here is the wiki showing how to access data and extract HRF results from groupResults.mat file:
https://github.com/BUNPC/Homer3/wiki/Output-Files
if you are interested in.

Itai Gutman
Itai Gutman (@itai-gutman)
1 day ago

Can you saw the right solution for the problem that was at end of the lecture?

YILEI ZHENG
YILEI ZHENG (@zhengyilei)
Reply to  Itai Gutman
1 day ago

Can you reclarify the issue? Sorry I was in breakout room then and missed the end of lecture.

David Boas
David Boas (@dboas)
Reply to  YILEI ZHENG
18 hours ago

They are asking if we can post the right solution. Do you have it?

Sarah Levy
Sarah Levy (@sarahmarielevy)
1 day ago

Is it possible for the video for today to be uploaded before tomorrow? I would like to revisit some bits that I missed while I was in a breakout room. Thanks

David Boas
David Boas (@dboas)
Reply to  Sarah Levy
18 hours ago

it is uploaded now

Heather Kwan
Heather Kwan (@hkwan)
1 day ago

When you are performing the preprocessing steps, how do you decide which channels to select and monitor (beyond selecting the channel closest to the region of interest)? Also, since the preprocessing steps are being done uniformly for each participant, does the channel selected really matter?

David Boas
David Boas (@dboas)
Reply to  Heather Kwan
18 hours ago

the channel selected has no impact on analysis. It is only for visualizing your results. This is helpful for looking at different channels to check the quality of the data and whether or not you need to use pruneChannels to automatically remove channels with poor SNR

Karla Holmboe
Karla Holmboe (@karlaholmboe)
21 hours ago

What is the advantage of the GLM method over block average if you do not have short separations? Does the GLM account better for artifacts (e.g., systemic, motor) via the regressors? Thanks!

David Boas
David Boas (@dboas)
Reply to  Karla Holmboe
18 hours ago

GLM allows you to incorporate drift correction into the model which is generally better than doing a bandpass filter before hand. You can also analyze results when stimuli are presented more rapidly such that the HRFs may be overlapping in time. And then, of course, if you have an auxilliary measurements like motion and short separation, then you can include them as regressors as well.

Karla Holmboe
Karla Holmboe (@karlaholmboe)
Reply to  David Boas
17 hours ago

Thanks!

David Boas
David Boas (@dboas)
15 hours ago

Here is the run processing stream we used for the Day 3 exercise with SS regression
I tried attaching a screen shoot too..

% Homer3 (v1.26, R2019b)

% group
@ hmrG_SubjAvg [dcAvg,nTrials] (dcAvgSubjs,nTrialsSubjs
@ hmrG_SubjAvgStd dcAvgStd (dcAvgSubjs

% subj
@ hmrS_RunAvg [dcAvg,nTrials] (dcAvgRuns,mlActRuns,nTrialsRuns
@ hmrS_RunAvgStd2 dcAvgStd (dcAvgRuns,dcSum2Runs,mlActRuns,nTrialsRuns

% run
@ hmrR_Intensity2OD dod (data
@ hmrR_MotionArtifact tIncAuto (dod,probe,mlActMan,tIncMan tMotion %0.1f 0.5 tMask %0.1f 1 STDEVthresh %0.1f 50 AMPthresh %0.1f 5
@ hmrR_BandpassFilt dod (dod hpf %0.3f 0 lpf %0.3f 3
@ hmrR_OD2Conc_new dc (dod,probe ppf %0.1f_%0.1f 1_1

Screen Shot 2020-10-22 at 10.18.44 AM.png
Itai Gutman
Itai Gutman (@itai-gutman)
Reply to  David Boas
13 hours ago

Thanks