Tech Reports

Timecourse methods

Sometime around June 2007 the lab switched from using raw averages to compute timecourses to using the "-iresp" flag in afni's 3dDeconvolve command to create timecouses. To insure that the methods are compatible we directly compared the output from each method on a set of 53 healthy subjects who's age ranged form 20-80 on the MID task. (If you're looking for information on how to use iresp to make timecourses look here in the lab manual under "How to run IRESP").

Iresp method

The best information about computing timecourses in general and how the iresp method works can be found on the gablab site at MIT: http://mindhive.mit.edu/node/86

Here is a little info on the method taken from that site:

"The [Finite Impulse Response] FIR model is a modification of the standard GLM which is designed precisely to deconvolve different conditions' peristimulus timecourses from each other. The main modification from the standard GLM is that instead of having one column for each effect, you have as many columns as you want timepoints in your peristimulus timecourse. If you want a 30-second timecourse and have a 3-second TR, you'd have 10 columns for each condition. Instead of having a single model of activity over time in one column, such as a boxcar convolved with a canonical HRF, or a canonical HRF by itself, each column represents one timepoint in the peristimulus timecourse. So the first column for each condition codes for the onset of each trial; it has a single 1 at each TR that condition has a trial onset, and zeros elsewhere. The second column for each condition codes for the onset + 1 point for each trial; it has a single 1 at each TR that's right after a trial onset, and zeros elsewhere. The third column codes in the same way for the onset + 2 timepoint for each trial; it has a single 1 at each TR that's two after a trial onset, and zeros elsewhere. Each column is filled out appropriately in the same fashion.

With this very wide design matrix, one then runs a standard GLM in the multiple regression style. Given enough timepoints and a properly randomized design, the design matrix then assigns beta weights to each column in the standard way - but these beta weights each represent activity at a certain temporal point following a trial onset. So for each condition, the first column tells you the effect size at the onset of a trial, the second column tells you the effect size one TR after the onset, the third columns tells you the effect size two TRs after the onset, and so on. This clearly translates directly into a peristimulus timecourse - simply plot each column's beta weight against time for a given condition, and voila! A nice-looking timecourse."

Raw Average Method

The Raw average method of building a timecouse is simply to dump out the raw data from the complete experiment (after preprocessing and converting to % signal change) and take an average over all of a particular trial type. For example if you want to look at $5.00 gain trials and the subject had 15 of these trials over the experiment, you would pull out each of these 15 trials from the beginning TR to the end TR (or past the end) for each trial and average them together at each TR.

While the iresp method can easily incorporate motion correction in the model, the raw averaging method does not necessarily include motion correction.

Comparison

Over all it looks like the two methods produce nearly identical results even when the number of subjects is low (which is good news for our data consistency!). However, there may be slight differences on an individual subject level. For example in the gain $0 trials for subject #2.

Physiological Corrections

Stephanie Greer 3/19/09

Procedure

MID Task subjects: 4 healthy controls used in the smoke-mid study.
Affective Picture Task subjects: 4 healthy controls used in the neurofeedback study.

I processed one set of data using the regular preprocessing method and used 3 different techniques for correcting for physiological signals:

Regular (reg) (No Correction)
rvhrcor (rvhr): see Chang, Cunningham & Glover Neuroimage 2009
- Preprocessing: Performs a very slight (almost unnoticeable) correction.
- GLM: Creates convolved Heart rate (HR) and respiratory volume (RV) regressors.
retroicor (ret): see Glover, Li and Ress MR in Medicine 2000
- Preprocessing: Performs a fairly drastic correction.
- GLM: Does not use any extra regressors
global signal regressor (glob)
- Preprocessing: No correction
- GLM: Includes an extra regressor that consists of an average over the entire grey mater.

Summary

It looks like we do pretty well using no correction for physiological confounds in terms of identifying activation that is consistently related to gain anticipation in the MID task. However, incorporating the physiology regressors given by rvhrcor seems to reduce noise and reduces the number of significant voxels that are not reliably related to gain anticipation in the MID Task. The Global regressor does next to nothing.

Retroicor did not work at all in this data set. Using retroicor eliminated significant voxels in our regions of interest and increased noise (i.e. standard deviation) in the time courses. One thing to note about this is that the original paper uses single shot spiral scans rather than a spiral in and spiral out sequence. Since retroicor is essentially a timing correction it might not work right with the in/out sequence (surely Gary would know if this is true).

Gain Anticipation Maps

Data was thresholded at z = 2.282 (p < .025) due to low n

Regular

rvhrcor

Global

retroicor

Regular

rvhrcor

Global

retroicor

Afni Region Name (z score)	X	Y	Z
No Region Found ( 3.67)	60	-56	16
No Region Found ( -3.162)	-71	-41	-11
Right Middle Frontal Gyrus ( 2.948)	-30	-38	16
Left Caudate ( 4.001)	11	-15	-3
Right Caudate Head ( 3.212)	-4	-15	1
Right Lentiform Nucleus ( 3.639)	-19	-15	4
Right Insula ( 3.133)	-38	-11	19
Left Caudate ( 2.909)	8	-11	1
Right Lentiform Nucleus ( 3.331)	-11	-8	-3
Right Claustrum ( 4.001)	-30	-8	-3
Left Putamen ( 3.651)	22	-4	19
Right Thalamus ( 3.448)	-4	4	4
Left Thalamus ( 2.974)	11	8	12
Right Thalamus ( 3.551)	-19	11	16
Right Anterior Nucleus ( 2.653)	-8	11	19
Right Thalamus ( 3.332)	-11	19	12
Left Precentral Gyrus ( 3.205)	38	19	57
Left Caudate ( 3.087)	34	26	-7
Left Precuneus ( 3.252)	15	52	53
Left Posterior Cingulate ( 3.211)	26	68	12
Left Cuneus ( 2.833)	4	79	27
Left Cuneus ( 3.621)	19	82	27
Left Cuneus ( 3.425)	4	90	16
No Region Found ( -3.624)	30	105	38
No Region Found ( 2.874)	15	109	31

Afni Region Name (z score)	X	Y	Z
No Region Found ( -2.936)	-71	-41	-14
Right Caudate ( 3.376)	-4	-19	4
Left Caudate ( 3.312)	11	-15	1
Right Middle Frontal Gyrus ( 3.312)	-34	-15	23
Right Caudate ( 2.988)	-15	-15	1
Left Lentiform Nucleus ( 3.323)	19	4	-7
Right Thalamus ( 3.718)	-4	8	4
Right Superior Temporal Gyrus ( -3.123)	-60	34	8
Left Cuneus ( 3.072)	4	82	27
No Region Found ( -3.528)	30	105	38

Afni Region Name (z score)	X	Y	Z
No Region Found ( 3.844)	60	-56	16
Left Caudate ( 4.001)	11	-15	-3
Right Lentiform Nucleus ( 4.001)	-19	-15	4
Right Caudate ( 3.241)	-8	-11	4
Right Inferior Frontal Gyrus ( 3.155)	-38	-8	23
Right Lentiform Nucleus ( 3.231)	-11	-8	-3
Right Insula ( 4.001)	-34	-8	-3
Left Putamen ( 3.413)	22	-4	19
Right Anterior Cingulate ( 3.009)	-4	0	-3
Right Thalamus ( 3.556)	-4	4	4
Left Lentiform Nucleus ( 3.281)	19	4	-7
Left Thalamus ( 3.085)	11	8	8
Right Thalamus ( 4.001)	-19	11	16
Left Precentral Gyrus ( 2.938)	38	19	57
Left Caudate ( 3.254)	34	26	-7
Right Superior Temporal Gyrus ( -2.962)	-64	34	8
Left Parahippocampal Gyrus ( 3.194)	30	49	8
Left Precuneus ( 3.216)	15	52	53
Left Cuneus ( 3.725)	19	82	27
Left Cuneus ( 3.328)	4	86	16
No Region Found ( -3.59)	30	105	38

Afni Region Name (z score)	X	Y	Z
Left Superior Temporal Gyrus ( 3.24)	52	-11	-3
Right Insula ( 3.093)	-38	-11	16
Right Cingulate Gyrus ( 3.093)	-22	-11	27
Left Lentiform Nucleus ( 2.945)	15	0	-7
Right Lateral Globus Pallidus ( 3.093)	-19	8	12
Left Insula ( 2.792)	41	11	8
Left Cuneus ( 3.309)	22	79	23
Left Middle Occipital Gyrus ( 2.663)	19	98	16

It's difficult to tell the differences between the regular, rvhrcor and global signal methods of preprocessing. However, judging from the tables, it looks like the rvhrcor method results in a somewhat less noisy map. To test this I compared the number of voxels above threshold (z > 2.58 p < .01) that overlapped with a map taken form the MID task meta analysis for gain anticipation > loss anticipation (Knutson & Greer 2008). The idea is that any significant voxels inside the mask are "correct" and any outside the mask are "noise". Here is the plot:

From this it looks like the regular, rvhrcor and global signal methods are similar in terms of identifying "correct" voxels but the rvhrcor results in substantially fewer "noise" voxels. My intuition is that this effect might go away if there were many more subjects.

Timecourses

Since the rvhrcor and retroicor methods do some correction to the data before preprocessing, I compared the time courses for these methods. Here's the data:

Clearly the regular processing method and the rvhrcor method are pretty much identical. Retroicor, however, looks more noisy and does something very bizarre to subject 2. To see how noisy it was, I compared the standard deviations of the first 200 TRs of subject 1 (shown below). Indeed retroicor is much noisier and rvhrcor is slightly less noisy than the regular method.

Since it looked like rvhrcor was effective in terms of getting rid of noise, but it also slightly reduced the significance of the "correct" voxels, I wanted to see what the relationship was between the physiological signals and the task. To do this I made timecourses from the heart rate (HR) and respiratory volume (Resp) regressors. I used both the raw data (Plain) and the data after it was convolved by rvhrcor. There was indeed an effect by condition (which fits with what rvhrcor is doing) but I don't know what to make of it.

Gain	Loss

Gain Outcome Maps

Data was thresholded at z = 2.282 (p < .025) due to low n

Regular

rvhrcor

Global

retroicor

Affective Picture Task

In this task subjects passively viewed affective pictures form the IAPS picture set. They saw six "negative" pictures with high negative arousal scores and six "neutral" pictures with low negative arousal scores. this task appeared to be effective in activating the amygdala although we were originally hoping to localize the insula with this task.