Multiple comparison using Monte Carlo

List overview All Threads
Download

newer

older

surfaceholes in aseg.stats

negative SUVr using PETsurfer

Ting Li

1 Dec 2019 1 Dec '19

11:34 p.m.

External Email - Use Caution

Dear Freesurfer Expert:

I have used the simulation code as show below:

mri_glmfit-sim \ --glmdir lh.gender_age.glmdir \ --cache 4 neg \ --cwp 0.05\ --2spaces My question is how many iterations are done during this simulation? I didn’t specify the iteration number.

Why I didn’t have the csd file?

Thanks a lot for your response!

Best regards, Ting

Attachments:

attachment.html (text/html — 1.4 KB)

Show replies by date

Greve, Douglas N.,Ph.D.

2 Dec 2 Dec

11:18 a.m.

That will use the pre-computed ("cached") CSD files distributed with FreeSurfer. This used 10,000 iterations.

On 12/1/2019 11:34 PM, Ting Li wrote:

External Email - Use Caution

Dear Freesurfer Expert:

I have used the simulation code as show below:

mri_glmfit-sim \ --glmdir lh.gender_age.glmdir \ --cache 4 neg \ --cwp 0.05\ --2spaces

My question is how many iterations are done during this simulation? I didn’t specify the iteration number.

Why I didn’t have the csd file?

Thanks a lot for your response!

Best regards, Ting

_______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edumailto:Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

Ting Li

11:58 a.m.

External Email - Use Caution

Dear Dr. Douglas,

Thank you so much for your quick response! You saved my half life.

From the introduction of clusterwise correction for multiple comparisons, I have a few questions.

1, What is the null hypothesis here ? 2, What is a z map? Do you mean synthesize cortical datasets? 3. Smooth z map, what FWHM do you use, if it is not consistent with my FWHM, how will it affect the results? 4, Threshold z map, what does level mean? Does it mean the vertex p-value?

Clusterwise Correction for Multiple Comparisons To perform a cluster-wise correction for multiple comparisons, we will run a simulation. The simulation is a way to get a measure of the distribution of the maximum cluster size under the *null **hypothesis*. This is done by iterating over the following steps: 1. Synthesize a z map 2. Smooth z map (using residual FWHM) 3. Threshold z map (level and sign) 4. Find clusters in thresholded map 5. Record area of maximum cluster 6. Repeat over desired number of iterations (usually > 5000)

Thank you so much!

Best regards, Ting

On Mon, Dec 2, 2019 at 10:20 AM Greve, Douglas N.,Ph.D. < DGREVE@mgh.harvard.edu> wrote:

...

That will use the pre-computed ("cached") CSD files distributed with FreeSurfer. This used 10,000 iterations.

On 12/1/2019 11:34 PM, Ting Li wrote:
    External Email - Use Caution
Dear Freesurfer Expert:

I have used the simulation code as show below:

mri_glmfit-sim \ --glmdir lh.gender_age.glmdir \ --cache 4 neg \ --cwp 0.05\ --2spaces

My question is how many iterations are done during this simulation? I didn’t specify the iteration number.

Why I didn’t have the csd file?

Thanks a lot for your response!

Best regards, Ting

Freesurfer mailing listFreesurfer@nmr.mgh.harvard.eduhttps://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

Greve, Douglas N.,Ph.D.

12:56 p.m.

On 12/2/2019 11:58 AM, Ting Li wrote:

External Email - Use Caution

Dear Dr. Douglas,

Thank you so much for your quick response! You saved my half life.

From the introduction of clusterwise correction for multiple comparisons, I have a few questions.

1, What is the null hypothesis here ? In the tutorial, the null hypothesis is that there is no effect of age (age slope = 0) 2, What is a z map? Do you mean synthesize cortical datasets? A z-map is just a map where all the values are from a gaussian distribution with mean=0 and stddev=1. This just assigns a random z-value to each vertex on fsaverage 3. Smooth z map, what FWHM do you use, if it is not consistent with my FWHM, how will it affect the results? The FWHM comes from an estimate from the residuals of the analysis. The residuals are everything that does not fit the linear model and represent noise. We use the residuals to estimate the underlying smoothness (FWHM), 4, Threshold z map, what does level mean? Does it mean the vertex p-value? Basically, it is the vertex p-value, but we use the "sig" = -log10(p). You also have to specify the sign. If you have an a priori hypothesis about the direction, then you can specify pos (positive) or neg (negative). If you do not have an a priori hypothesis, then use abs (for absolute, for an unsigned test).

Clusterwise Correction for Multiple Comparisons To perform a cluster-wise correction for multiple comparisons, we will run a simulation. The simulation is a way to get a measure of the distribution of the maximum cluster size under the null hypothesis. This is done by iterating over the following steps: 1. Synthesize a z map 2. Smooth z map (using residual FWHM) 3. Threshold z map (level and sign) 4. Find clusters in thresholded map 5. Record area of maximum cluster 6. Repeat over desired number of iterations (usually > 5000)

Thank you so much!

Best regards, Ting

On Mon, Dec 2, 2019 at 10:20 AM Greve, Douglas N.,Ph.D. <DGREVE@mgh.harvard.edumailto:DGREVE@mgh.harvard.edu> wrote: That will use the pre-computed ("cached") CSD files distributed with FreeSurfer. This used 10,000 iterations.

On 12/1/2019 11:34 PM, Ting Li wrote:

External Email - Use Caution

Dear Freesurfer Expert:

I have used the simulation code as show below:

mri_glmfit-sim \ --glmdir lh.gender_age.glmdir \ --cache 4 neg \ --cwp 0.05\ --2spaces

My question is how many iterations are done during this simulation? I didn’t specify the iteration number.

Why I didn’t have the csd file?

Thanks a lot for your response!

Best regards, Ting

Ting Li

2:03 p.m.

External Email - Use Caution

Dear Dr. Douglas,

Thank you so much for your detailed response.

The simulation here is to get the probability of a maximum cluster that size or larger in the cached CSD files. It checks the probability of the cluster size but have nothing to do with the analysis metrics that I have used. Do I understand correctly?

Best regards, Ting Li

On Mon, Dec 2, 2019 at 11:57 AM Greve, Douglas N.,Ph.D. < DGREVE@mgh.harvard.edu> wrote:

...

On 12/2/2019 11:58 AM, Ting Li wrote:
    External Email - Use Caution
Dear Dr. Douglas,

Thank you so much for your quick response! You saved my half life.

From the introduction of clusterwise correction for multiple comparisons, I have a few questions.

1, What is the null hypothesis here ?

In the tutorial, the null hypothesis is that there is no effect of age (age slope = 0)

2, What is a z map? Do you mean synthesize cortical datasets?

A z-map is just a map where all the values are from a gaussian distribution with mean=0 and stddev=1. This just assigns a random z-value to each vertex on fsaverage

Smooth z map, what FWHM do you use, if it is not consistent with my

FWHM, how will it affect the results?

The FWHM comes from an estimate from the residuals of the analysis. The residuals are everything that does not fit the linear model and represent noise. We use the residuals to estimate the underlying smoothness (FWHM),

4, Threshold z map, what does level mean? Does it mean the vertex p-value?

Basically, it is the vertex p-value, but we use the "sig" = -log10(p). You also have to specify the sign. If you have an a priori hypothesis about the direction, then you can specify pos (positive) or neg (negative). If you do not have an a priori hypothesis, then use abs (for absolute, for an unsigned test).

Clusterwise Correction for Multiple Comparisons To perform a cluster-wise correction for multiple comparisons, we will run a simulation. The simulation is a way to get a measure of the distribution of the maximum cluster size under the *null **hypothesis*. This is done by iterating over the following steps:

Synthesize a z map

Smooth z map (using residual FWHM)

Threshold z map (level and sign)

Find clusters in thresholded map

Record area of maximum cluster

Repeat over desired number of iterations (usually > 5000)

Thank you so much!

Best regards, Ting

On Mon, Dec 2, 2019 at 10:20 AM Greve, Douglas N.,Ph.D. < DGREVE@mgh.harvard.edu> wrote:

...
That will use the pre-computed ("cached") CSD files distributed with FreeSurfer. This used 10,000 iterations.

On 12/1/2019 11:34 PM, Ting Li wrote:
    External Email - Use Caution
Dear Freesurfer Expert:

I have used the simulation code as show below:

mri_glmfit-sim \ --glmdir lh.gender_age.glmdir \ --cache 4 neg \ --cwp 0.05\ --2spaces

My question is how many iterations are done during this simulation? I didn’t specify the iteration number.

Why I didn’t have the csd file?

Thanks a lot for your response!

Best regards, Ting

Freesurfer mailing listFreesurfer@nmr.mgh.harvard.eduhttps://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
Freesurfer mailing listFreesurfer@nmr.mgh.harvard.eduhttps://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

Greve, Douglas N.,Ph.D.

2:59 p.m.

Correct. This is no different than, eg, a t-test. Look in the back of a stats book from the 1950s and you will find a table of t-ratios and corresponding p-values. Obviously, they did not have access to any of the data being analyzed today:), but the tables are still valid.

On 12/2/2019 2:03 PM, Ting Li wrote:

External Email - Use Caution

Dear Dr. Douglas,

Thank you so much for your detailed response.

Best regards, Ting Li

On Mon, Dec 2, 2019 at 11:57 AM Greve, Douglas N.,Ph.D. <DGREVE@mgh.harvard.edumailto:DGREVE@mgh.harvard.edu> wrote:

On 12/2/2019 11:58 AM, Ting Li wrote:

External Email - Use Caution

Dear Dr. Douglas,

Thank you so much for your quick response! You saved my half life.

From the introduction of clusterwise correction for multiple comparisons, I have a few questions.

Clusterwise Correction for Multiple Comparisons To perform a cluster-wise correction for multiple comparisons, we will run a simulation. The simulation is a way to get a measure of the distribution of the maximum cluster size under the null hypothesis. This is done by iterating over the following steps: 1. Synthesize a z map 2. Smooth z map (using residual FWHM) 3. Threshold z map (level and sign) 4. Find clusters in thresholded map 5. Record area of maximum cluster 6. Repeat over desired number of iterations (usually > 5000)

Thank you so much!

Best regards, Ting

On 12/1/2019 11:34 PM, Ting Li wrote:

External Email - Use Caution

Dear Freesurfer Expert:

I have used the simulation code as show below:

mri_glmfit-sim \ --glmdir lh.gender_age.glmdir \ --cache 4 neg \ --cwp 0.05\ --2spaces

My question is how many iterations are done during this simulation? I didn’t specify the iteration number.

Why I didn’t have the csd file?

Thanks a lot for your response!

Best regards, Ting

Ting Li

3:38 p.m.

External Email - Use Caution

Dear Dr. Douglas,

Thank you so much for your explanation. You really helped me a lot!

Best regards, Ting

On Mon, Dec 2, 2019 at 1:59 PM Greve, Douglas N.,Ph.D. < DGREVE@mgh.harvard.edu> wrote:

...

Correct. This is no different than, eg, a t-test. Look in the back of a stats book from the 1950s and you will find a table of t-ratios and corresponding p-values. Obviously, they did not have access to any of the data being analyzed today:), but the tables are still valid.

On 12/2/2019 2:03 PM, Ting Li wrote:
    External Email - Use Caution
Dear Dr. Douglas,

Thank you so much for your detailed response.

The simulation here is to get the probability of a maximum cluster that size or larger in the cached CSD files. It checks the probability of the cluster size but have nothing to do with the analysis metrics that I have used. Do I understand correctly?

Best regards, Ting Li

On Mon, Dec 2, 2019 at 11:57 AM Greve, Douglas N.,Ph.D. < DGREVE@mgh.harvard.edu> wrote:

...
On 12/2/2019 11:58 AM, Ting Li wrote:
    External Email - Use Caution
Dear Dr. Douglas,

Thank you so much for your quick response! You saved my half life.

From the introduction of clusterwise correction for multiple comparisons, I have a few questions.

1, What is the null hypothesis here ?

In the tutorial, the null hypothesis is that there is no effect of age (age slope = 0)

2, What is a z map? Do you mean synthesize cortical datasets?

A z-map is just a map where all the values are from a gaussian distribution with mean=0 and stddev=1. This just assigns a random z-value to each vertex on fsaverage

Smooth z map, what FWHM do you use, if it is not consistent with my

FWHM, how will it affect the results?

The FWHM comes from an estimate from the residuals of the analysis. The residuals are everything that does not fit the linear model and represent noise. We use the residuals to estimate the underlying smoothness (FWHM),

4, Threshold z map, what does level mean? Does it mean the vertex p-value?

Basically, it is the vertex p-value, but we use the "sig" = -log10(p). You also have to specify the sign. If you have an a priori hypothesis about the direction, then you can specify pos (positive) or neg (negative). If you do not have an a priori hypothesis, then use abs (for absolute, for an unsigned test).

Clusterwise Correction for Multiple Comparisons To perform a cluster-wise correction for multiple comparisons, we will run a simulation. The simulation is a way to get a measure of the distribution of the maximum cluster size under the *null **hypothesis*. This is done by iterating over the following steps:

Synthesize a z map

Smooth z map (using residual FWHM)

Threshold z map (level and sign)

Find clusters in thresholded map

Record area of maximum cluster

Repeat over desired number of iterations (usually > 5000)

Thank you so much!

Best regards, Ting

On Mon, Dec 2, 2019 at 10:20 AM Greve, Douglas N.,Ph.D. < DGREVE@mgh.harvard.edu> wrote:

...
That will use the pre-computed ("cached") CSD files distributed with FreeSurfer. This used 10,000 iterations.

On 12/1/2019 11:34 PM, Ting Li wrote:
    External Email - Use Caution
Dear Freesurfer Expert:

I have used the simulation code as show below:

mri_glmfit-sim \ --glmdir lh.gender_age.glmdir \ --cache 4 neg \ --cwp 0.05\ --2spaces

My question is how many iterations are done during this simulation? I didn’t specify the iteration number.

Why I didn’t have the csd file?

Thanks a lot for your response!

Best regards, Ting

Freesurfer mailing listFreesurfer@nmr.mgh.harvard.eduhttps://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
Freesurfer mailing listFreesurfer@nmr.mgh.harvard.eduhttps://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
Freesurfer mailing listFreesurfer@nmr.mgh.harvard.eduhttps://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

Ting Li

3 Dec 3 Dec

1:36 p.m.

External Email - Use Caution

Dear Dr. Douglas,

Thank you so much for all the answers. I have two more questions.

1, When the Monte Carlo simulation is doing, how does one simulation is done?

- synthesize z map (synthesize z map for how many subjects?) - smooth z map (You have mentioned the FWHM comes from an estimate from the residuals of the analysis, which analysis? or the way we process the real data with the command mris_preproc with FWHM value? ) - threshold z map (Does it mean a GLM analysis so we can have threshold z map?)

2, When we reference to the CSD files in freesurfer, will it correct the cluster size p-value based on DOF (degree of freedom), like the t-ratios table including a DOF? Thanks a lot!

Best regards, Ting

On Mon, Dec 2, 2019 at 1:59 PM Greve, Douglas N.,Ph.D. < DGREVE@mgh.harvard.edu> wrote:

...

Correct. This is no different than, eg, a t-test. Look in the back of a stats book from the 1950s and you will find a table of t-ratios and corresponding p-values. Obviously, they did not have access to any of the data being analyzed today:), but the tables are still valid.

On 12/2/2019 2:03 PM, Ting Li wrote:
    External Email - Use Caution
Dear Dr. Douglas,

Thank you so much for your detailed response.

The simulation here is to get the probability of a maximum cluster that size or larger in the cached CSD files. It checks the probability of the cluster size but have nothing to do with the analysis metrics that I have used. Do I understand correctly?

Best regards, Ting Li

On Mon, Dec 2, 2019 at 11:57 AM Greve, Douglas N.,Ph.D. < DGREVE@mgh.harvard.edu> wrote:

...
On 12/2/2019 11:58 AM, Ting Li wrote:
    External Email - Use Caution
Dear Dr. Douglas,

Thank you so much for your quick response! You saved my half life.

From the introduction of clusterwise correction for multiple comparisons, I have a few questions.

1, What is the null hypothesis here ?

In the tutorial, the null hypothesis is that there is no effect of age (age slope = 0)

2, What is a z map? Do you mean synthesize cortical datasets?

A z-map is just a map where all the values are from a gaussian distribution with mean=0 and stddev=1. This just assigns a random z-value to each vertex on fsaverage

Smooth z map, what FWHM do you use, if it is not consistent with my

FWHM, how will it affect the results?

The FWHM comes from an estimate from the residuals of the analysis. The residuals are everything that does not fit the linear model and represent noise. We use the residuals to estimate the underlying smoothness (FWHM),

4, Threshold z map, what does level mean? Does it mean the vertex p-value?

Basically, it is the vertex p-value, but we use the "sig" = -log10(p). You also have to specify the sign. If you have an a priori hypothesis about the direction, then you can specify pos (positive) or neg (negative). If you do not have an a priori hypothesis, then use abs (for absolute, for an unsigned test).

Clusterwise Correction for Multiple Comparisons To perform a cluster-wise correction for multiple comparisons, we will run a simulation. The simulation is a way to get a measure of the distribution of the maximum cluster size under the *null **hypothesis*. This is done by iterating over the following steps:

Synthesize a z map

Smooth z map (using residual FWHM)

Threshold z map (level and sign)

Find clusters in thresholded map

Record area of maximum cluster

Repeat over desired number of iterations (usually > 5000)

Thank you so much!

Best regards, Ting

On Mon, Dec 2, 2019 at 10:20 AM Greve, Douglas N.,Ph.D. < DGREVE@mgh.harvard.edu> wrote:

...
That will use the pre-computed ("cached") CSD files distributed with FreeSurfer. This used 10,000 iterations.

On 12/1/2019 11:34 PM, Ting Li wrote:
    External Email - Use Caution
Dear Freesurfer Expert:

I have used the simulation code as show below:

mri_glmfit-sim \ --glmdir lh.gender_age.glmdir \ --cache 4 neg \ --cwp 0.05\ --2spaces

My question is how many iterations are done during this simulation? I didn’t specify the iteration number.

Why I didn’t have the csd file?

Thanks a lot for your response!

Best regards, Ting

Freesurfer mailing listFreesurfer@nmr.mgh.harvard.eduhttps://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
Freesurfer mailing listFreesurfer@nmr.mgh.harvard.eduhttps://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
Freesurfer mailing listFreesurfer@nmr.mgh.harvard.eduhttps://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

Greve, Douglas N.,Ph.D.

3:13 p.m.

On 12/3/2019 1:36 PM, Ting Li wrote:

External Email - Use Caution

Dear Dr. Douglas,

Thank you so much for all the answers. I have two more questions.

1, When the Monte Carlo simulation is doing, how does one simulation is done?

* synthesize z map (synthesize z map for how many subjects?)

The Z is essentailly an infinite # of subjects (it is just a single map).

* smooth z map (You have mentioned the FWHM comes from an estimate from the residuals of the analysis, which analysis? or the way we process the real data with the command mris_preproc with FWHM value? )

The GLM analysis, from the residual.

* threshold z map (Does it mean a GLM analysis so we can have threshold z map?)

No, the z-map is synthesized directly

The simulation is done in mri_glmfit-sim and in mri_mcsim.

2, When we reference to the CSD files in freesurfer, will it correct the cluster size p-value based on DOF (degree of freedom), like the t-ratios table including a DOF? Thanks a lot! I don't know what you mean.

Best regards, Ting

On Mon, Dec 2, 2019 at 1:59 PM Greve, Douglas N.,Ph.D. <DGREVE@mgh.harvard.edumailto:DGREVE@mgh.harvard.edu> wrote: Correct. This is no different than, eg, a t-test. Look in the back of a stats book from the 1950s and you will find a table of t-ratios and corresponding p-values. Obviously, they did not have access to any of the data being analyzed today:), but the tables are still valid.

On 12/2/2019 2:03 PM, Ting Li wrote:

External Email - Use Caution

Dear Dr. Douglas,

Thank you so much for your detailed response.

Best regards, Ting Li

On Mon, Dec 2, 2019 at 11:57 AM Greve, Douglas N.,Ph.D. <DGREVE@mgh.harvard.edumailto:DGREVE@mgh.harvard.edu> wrote:

On 12/2/2019 11:58 AM, Ting Li wrote:

External Email - Use Caution

Dear Dr. Douglas,

Thank you so much for your quick response! You saved my half life.

From the introduction of clusterwise correction for multiple comparisons, I have a few questions.

Clusterwise Correction for Multiple Comparisons To perform a cluster-wise correction for multiple comparisons, we will run a simulation. The simulation is a way to get a measure of the distribution of the maximum cluster size under the null hypothesis. This is done by iterating over the following steps: 1. Synthesize a z map 2. Smooth z map (using residual FWHM) 3. Threshold z map (level and sign) 4. Find clusters in thresholded map 5. Record area of maximum cluster 6. Repeat over desired number of iterations (usually > 5000)

Thank you so much!

Best regards, Ting

On 12/1/2019 11:34 PM, Ting Li wrote:

External Email - Use Caution

Dear Freesurfer Expert:

I have used the simulation code as show below:

mri_glmfit-sim \ --glmdir lh.gender_age.glmdir \ --cache 4 neg \ --cwp 0.05\ --2spaces

My question is how many iterations are done during this simulation? I didn’t specify the iteration number.

Why I didn’t have the csd file?

Thanks a lot for your response!

Best regards, Ting

Ting Li

9:24 p.m.

External Email - Use Caution

Dear Dr. Douglas,

Thank you so much.

I have difficulty to understand z map is essentially an infinite # of subjects. Before smooth, there is no GLM analysis. Does the smooth include in the GLM process?

Best regards, Ting

On Tue, Dec 3, 2019 at 2:14 PM Greve, Douglas N.,Ph.D. < DGREVE@mgh.harvard.edu> wrote:

...

On 12/3/2019 1:36 PM, Ting Li wrote:
    External Email - Use Caution
Dear Dr. Douglas,

Thank you so much for all the answers. I have two more questions.

1, When the Monte Carlo simulation is doing, how does one simulation is done?

synthesize z map (synthesize z map for how many subjects?)

The Z is essentailly an infinite # of subjects (it is just a single map).

smooth z map (You have mentioned the FWHM comes from an estimate

from the residuals of the analysis, which analysis? or the way we process the real data with the command mris_preproc with FWHM value? )

The GLM analysis, from the residual.

threshold z map (Does it mean a GLM analysis so we can have

threshold z map?)

No, the z-map is synthesized directly

The simulation is done in mri_glmfit-sim and in mri_mcsim.

2, When we reference to the CSD files in freesurfer, will it correct the cluster size p-value based on DOF (degree of freedom), like the t-ratios table including a DOF? Thanks a lot!

I don't know what you mean.

Best regards, Ting

On Mon, Dec 2, 2019 at 1:59 PM Greve, Douglas N.,Ph.D. < DGREVE@mgh.harvard.edu> wrote:

...
Correct. This is no different than, eg, a t-test. Look in the back of a stats book from the 1950s and you will find a table of t-ratios and corresponding p-values. Obviously, they did not have access to any of the data being analyzed today:), but the tables are still valid.

On 12/2/2019 2:03 PM, Ting Li wrote:
    External Email - Use Caution
Dear Dr. Douglas,

Thank you so much for your detailed response.

The simulation here is to get the probability of a maximum cluster that size or larger in the cached CSD files. It checks the probability of the cluster size but have nothing to do with the analysis metrics that I have used. Do I understand correctly?

Best regards, Ting Li

On Mon, Dec 2, 2019 at 11:57 AM Greve, Douglas N.,Ph.D. < DGREVE@mgh.harvard.edu> wrote:

...
On 12/2/2019 11:58 AM, Ting Li wrote:
    External Email - Use Caution
Dear Dr. Douglas,

Thank you so much for your quick response! You saved my half life.

From the introduction of clusterwise correction for multiple comparisons, I have a few questions.

1, What is the null hypothesis here ?

In the tutorial, the null hypothesis is that there is no effect of age (age slope = 0)

2, What is a z map? Do you mean synthesize cortical datasets?

A z-map is just a map where all the values are from a gaussian distribution with mean=0 and stddev=1. This just assigns a random z-value to each vertex on fsaverage

Smooth z map, what FWHM do you use, if it is not consistent with my

FWHM, how will it affect the results?

The FWHM comes from an estimate from the residuals of the analysis. The residuals are everything that does not fit the linear model and represent noise. We use the residuals to estimate the underlying smoothness (FWHM),

4, Threshold z map, what does level mean? Does it mean the vertex p-value?

Basically, it is the vertex p-value, but we use the "sig" = -log10(p). You also have to specify the sign. If you have an a priori hypothesis about the direction, then you can specify pos (positive) or neg (negative). If you do not have an a priori hypothesis, then use abs (for absolute, for an unsigned test).

Clusterwise Correction for Multiple Comparisons To perform a cluster-wise correction for multiple comparisons, we will run a simulation. The simulation is a way to get a measure of the distribution of the maximum cluster size under the *null **hypothesis*. This is done by iterating over the following steps:

Synthesize a z map

Smooth z map (using residual FWHM)

Threshold z map (level and sign)

Find clusters in thresholded map

Record area of maximum cluster

Repeat over desired number of iterations (usually > 5000)

Thank you so much!

Best regards, Ting

On Mon, Dec 2, 2019 at 10:20 AM Greve, Douglas N.,Ph.D. < DGREVE@mgh.harvard.edu> wrote:

...
That will use the pre-computed ("cached") CSD files distributed with FreeSurfer. This used 10,000 iterations.

On 12/1/2019 11:34 PM, Ting Li wrote:
    External Email - Use Caution
Dear Freesurfer Expert:

I have used the simulation code as show below:

mri_glmfit-sim \ --glmdir lh.gender_age.glmdir \ --cache 4 neg \ --cwp 0.05\ --2spaces

My question is how many iterations are done during this simulation? I didn’t specify the iteration number.

Why I didn’t have the csd file?

Thanks a lot for your response!

Best regards, Ting

Freesurfer mailing listFreesurfer@nmr.mgh.harvard.eduhttps://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
Freesurfer mailing listFreesurfer@nmr.mgh.harvard.eduhttps://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
Freesurfer mailing listFreesurfer@nmr.mgh.harvard.eduhttps://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
Freesurfer mailing listFreesurfer@nmr.mgh.harvard.eduhttps://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

Greve, Douglas N.,Ph.D.

5 Dec 5 Dec

12:38 p.m.

I've attached a couple of papers

On 12/3/19 9:24 PM, Ting Li wrote:

...

External Email - Use Caution

Dear Dr. Douglas,

Thank you so much.

I have difficulty to understand z map is essentially an infinite # of subjects. Before smooth, there is no GLM analysis. Does the smooth include in the GLM process?

Best regards, Ting

On Tue, Dec 3, 2019 at 2:14 PM Greve, Douglas N.,Ph.D. <DGREVE@mgh.harvard.edu mailto:DGREVE@mgh.harvard.edu> wrote:

On 12/3/2019 1:36 PM, Ting Li wrote:

...

        External Email - Use Caution

Dear Dr. Douglas,

Thank  you so much for all the answers. I have two more questions.

1, When the Monte Carlo simulation is doing, how does one
simulation is done?

  * synthesize z map (synthesize z map for how many subjects?)

The Z is essentailly an infinite # of subjects (it is just a
single map).

...

  * smooth z map (You have mentioned the FWHM comes from an
    estimate from the residuals of the analysis, which analysis?
    or the way we process the real data with the command
    mris_preproc with FWHM value? )

The GLM analysis, from the residual.

...

  * threshold z map (Does it mean a GLM analysis so we can have
    threshold z map?)

No, the z-map is synthesized directly

The simulation is done in mri_glmfit-sim and in mri_mcsim.

...

2, When we reference to the CSD files in freesurfer, will it
correct the cluster size p-value based on DOF (degree of
freedom), like the t-ratios table including a DOF? Thanks a lot!

I don't know what you mean.

...

Best regards,
Ting

On Mon, Dec 2, 2019 at 1:59 PM Greve, Douglas N.,Ph.D.
<DGREVE@mgh.harvard.edu <mailto:DGREVE@mgh.harvard.edu>> wrote:

    Correct. This is no different than, eg, a t-test. Look in 
    the back of a stats book from the 1950s and you will find a
    table of t-ratios and corresponding p-values. Obviously, they
    did not have access to any of the data being analyzed
    today:), but the tables are still valid.

    On 12/2/2019 2:03 PM, Ting Li wrote:

...

            External Email - Use Caution

    Dear Dr. Douglas,

    Thank you so much for your detailed response.

    The simulation here is to get the probability of a maximum
    cluster that size or larger in the cached CSD files. It
    checks the probability of  the cluster size but have nothing
    to do with the analysis metrics that I have used. Do I
    understand correctly?

    Best regards,
    Ting Li

    On Mon, Dec 2, 2019 at 11:57 AM Greve, Douglas N.,Ph.D.
    <DGREVE@mgh.harvard.edu <mailto:DGREVE@mgh.harvard.edu>> wrote:



        On 12/2/2019 11:58 AM, Ting Li wrote:

...

                External Email - Use Caution

        Dear Dr. Douglas,

        Thank you so much for your quick response! You saved my
        half life.

        From the introduction of clusterwise correction for
        multiple comparisons, I have a few questions.

        1, What is the null hypothesis here ?

        In the tutorial, the null hypothesis is that there is no
        effect of age (age slope = 0)

...

        2, What is a z map? Do you mean synthesize cortical
        datasets?

        A z-map is just a map where all the values are from a
        gaussian distribution with mean=0 and stddev=1. This
        just assigns a random z-value to each vertex on fsaverage

...

        3. Smooth z map, what FWHM do you use, if it is not
        consistent with my FWHM, how will it affect the results?

        The FWHM comes from an estimate from the residuals of
        the analysis. The residuals are everything that does not
        fit the linear model and represent noise. We use the
        residuals to estimate the underlying smoothness (FWHM),

...

        4, Threshold z map, what does level mean? Does it mean
        the vertex p-value?

        Basically, it is the vertex p-value, but we use  the
        "sig" = -log10(p). You also have to specify the sign. If
        you have an a priori hypothesis about the direction,
        then you can specify pos (positive) or neg (negative).
        If you do not have an a priori hypothesis, then use abs
        (for absolute, for an unsigned test).

...

        ClusterwiseCorrection for Multiple Comparisons
        To perform a cluster-wise correction for multiple
        comparisons, we will run a simulation. The simulation
        isa way to get a measure of the distribution of the
        maximum cluster size under the _null __hypothesis_.
        This isdone by iterating over the following steps:
        1. Synthesize a z map
        2. Smooth z map (using residual FWHM)
        3. Threshold z map (level and sign)
        4. Find clusters in thresholded map
        5. Record area of maximum cluster
        6. Repeat over desired number of iterations (usually >
        5000)

        Thank you so much!

        Best regards,
        Ting

        On Mon, Dec 2, 2019 at 10:20 AM Greve, Douglas N.,Ph.D.
        <DGREVE@mgh.harvard.edu
        <mailto:DGREVE@mgh.harvard.edu>> wrote:

            That will use the pre-computed ("cached") CSD files
            distributed with FreeSurfer. This used 10,000
            iterations.

            On 12/1/2019 11:34 PM, Ting Li wrote:

...

                    External Email - Use Caution

            Dear Freesurfer Expert:

            I have used the simulation code as show below:

            mri_glmfit-sim \
               --glmdir lh.gender_age.glmdir \
               --cache 4 neg \
               --cwp  0.05\
               --2spaces
            My question is how many iterations are done during
            this simulation? I didn’t specify the iteration
            number.

            Why I didn’t have the csd file?

            Thanks a lot for your response!

            Best regards,
            Ting

            _______________________________________________
            Freesurfer mailing list
            Freesurfer@nmr.mgh.harvard.edu  <mailto:Freesurfer@nmr.mgh.harvard.edu>
            https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

            _______________________________________________
            Freesurfer mailing list
            Freesurfer@nmr.mgh.harvard.edu
            <mailto:Freesurfer@nmr.mgh.harvard.edu>
            https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer


        _______________________________________________
        Freesurfer mailing list
        Freesurfer@nmr.mgh.harvard.edu  <mailto:Freesurfer@nmr.mgh.harvard.edu>
        https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

        _______________________________________________
        Freesurfer mailing list
        Freesurfer@nmr.mgh.harvard.edu
        <mailto:Freesurfer@nmr.mgh.harvard.edu>
        https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer


    _______________________________________________
    Freesurfer mailing list
    Freesurfer@nmr.mgh.harvard.edu  <mailto:Freesurfer@nmr.mgh.harvard.edu>
    https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

    _______________________________________________
    Freesurfer mailing list
    Freesurfer@nmr.mgh.harvard.edu
    <mailto:Freesurfer@nmr.mgh.harvard.edu>
    https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer


_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu  <mailto:Freesurfer@nmr.mgh.harvard.edu>
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu <mailto:Freesurfer@nmr.mgh.harvard.edu>
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

Ting Li

1:17 p.m.

External Email - Use Caution

Dear Dr. Douglas,

Thank you so much!

Best regards, Ting

On Thu, Dec 5, 2019 at 11:45 AM Greve, Douglas N.,Ph.D. < DGREVE@mgh.harvard.edu> wrote:

...

I've attached a couple of papers

On 12/3/19 9:24 PM, Ting Li wrote:

...

    External Email - Use Caution

Dear Dr. Douglas,

Thank you so much.

I have difficulty to understand z map is essentially an infinite # of subjects. Before smooth, there is no GLM analysis. Does the smooth include in the GLM process?

Best regards, Ting

On Tue, Dec 3, 2019 at 2:14 PM Greve, Douglas N.,Ph.D. <DGREVE@mgh.harvard.edu mailto:DGREVE@mgh.harvard.edu> wrote:

On 12/3/2019 1:36 PM, Ting Li wrote:

...

        External Email - Use Caution

Dear Dr. Douglas,

Thank  you so much for all the answers. I have two more questions.

1, When the Monte Carlo simulation is doing, how does one
simulation is done?

  * synthesize z map (synthesize z map for how many subjects?)

The Z is essentailly an infinite # of subjects (it is just a
single map).

...

  * smooth z map (You have mentioned the FWHM comes from an
    estimate from the residuals of the analysis, which analysis?
    or the way we process the real data with the command
    mris_preproc with FWHM value? )

The GLM analysis, from the residual.

...

  * threshold z map (Does it mean a GLM analysis so we can have
    threshold z map?)

No, the z-map is synthesized directly

The simulation is done in mri_glmfit-sim and in mri_mcsim.

...

2, When we reference to the CSD files in freesurfer, will it
correct the cluster size p-value based on DOF (degree of
freedom), like the t-ratios table including a DOF? Thanks a lot!

I don't know what you mean.

...

Best regards,
Ting

On Mon, Dec 2, 2019 at 1:59 PM Greve, Douglas N.,Ph.D.
<DGREVE@mgh.harvard.edu <mailto:DGREVE@mgh.harvard.edu>> wrote:

    Correct. This is no different than, eg, a t-test. Look in
    the back of a stats book from the 1950s and you will find a
    table of t-ratios and corresponding p-values. Obviously, they
    did not have access to any of the data being analyzed
    today:), but the tables are still valid.

    On 12/2/2019 2:03 PM, Ting Li wrote:

...

            External Email - Use Caution

    Dear Dr. Douglas,

    Thank you so much for your detailed response.

    The simulation here is to get the probability of a maximum
    cluster that size or larger in the cached CSD files. It
    checks the probability of  the cluster size but have nothing
    to do with the analysis metrics that I have used. Do I
    understand correctly?

    Best regards,
    Ting Li

    On Mon, Dec 2, 2019 at 11:57 AM Greve, Douglas N.,Ph.D.
    <DGREVE@mgh.harvard.edu <mailto:DGREVE@mgh.harvard.edu>>

wrote:

...

        On 12/2/2019 11:58 AM, Ting Li wrote:

...

                External Email - Use Caution

        Dear Dr. Douglas,

        Thank you so much for your quick response! You saved my
        half life.

        From the introduction of clusterwise correction for
        multiple comparisons, I have a few questions.

        1, What is the null hypothesis here ?

        In the tutorial, the null hypothesis is that there is no
        effect of age (age slope = 0)

...

        2, What is a z map? Do you mean synthesize cortical
        datasets?

        A z-map is just a map where all the values are from a
        gaussian distribution with mean=0 and stddev=1. This
        just assigns a random z-value to each vertex on fsaverage

...

        3. Smooth z map, what FWHM do you use, if it is not
        consistent with my FWHM, how will it affect the results?

        The FWHM comes from an estimate from the residuals of
        the analysis. The residuals are everything that does not
        fit the linear model and represent noise. We use the
        residuals to estimate the underlying smoothness (FWHM),

...

        4, Threshold z map, what does level mean? Does it mean
        the vertex p-value?

        Basically, it is the vertex p-value, but we use  the
        "sig" = -log10(p). You also have to specify the sign. If
        you have an a priori hypothesis about the direction,
        then you can specify pos (positive) or neg (negative).
        If you do not have an a priori hypothesis, then use abs
        (for absolute, for an unsigned test).

...

        ClusterwiseCorrection for Multiple Comparisons
        To perform a cluster-wise correction for multiple
        comparisons, we will run a simulation. The simulation
        isa way to get a measure of the distribution of the
        maximum cluster size under the _null __hypothesis_.
        This isdone by iterating over the following steps:
        1. Synthesize a z map
        2. Smooth z map (using residual FWHM)
        3. Threshold z map (level and sign)
        4. Find clusters in thresholded map
        5. Record area of maximum cluster
        6. Repeat over desired number of iterations (usually >
        5000)

        Thank you so much!

        Best regards,
        Ting

        On Mon, Dec 2, 2019 at 10:20 AM Greve, Douglas N.,Ph.D.
        <DGREVE@mgh.harvard.edu
        <mailto:DGREVE@mgh.harvard.edu>> wrote:

            That will use the pre-computed ("cached") CSD files
            distributed with FreeSurfer. This used 10,000
            iterations.

            On 12/1/2019 11:34 PM, Ting Li wrote:

...

                    External Email - Use Caution

            Dear Freesurfer Expert:

            I have used the simulation code as show below:

            mri_glmfit-sim \
               --glmdir lh.gender_age.glmdir \
               --cache 4 neg \
               --cwp  0.05\
               --2spaces
            My question is how many iterations are done during
            this simulation? I didn’t specify the iteration
            number.

            Why I didn’t have the csd file?

            Thanks a lot for your response!

            Best regards,
            Ting

            _______________________________________________
            Freesurfer mailing list
            Freesurfer@nmr.mgh.harvard.edu  <mailto:

Freesurfer@nmr.mgh.harvard.edu>

...

...
...
...
...

https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

...

            _______________________________________________
            Freesurfer mailing list
            Freesurfer@nmr.mgh.harvard.edu
            <mailto:Freesurfer@nmr.mgh.harvard.edu>

https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

...

        _______________________________________________
        Freesurfer mailing list
        Freesurfer@nmr.mgh.harvard.edu  <mailto:

Freesurfer@nmr.mgh.harvard.edu>

...

...
...
...

https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

...

        _______________________________________________
        Freesurfer mailing list
        Freesurfer@nmr.mgh.harvard.edu
        <mailto:Freesurfer@nmr.mgh.harvard.edu>

https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

...

    _______________________________________________
    Freesurfer mailing list
    Freesurfer@nmr.mgh.harvard.edu  <mailto:

Freesurfer@nmr.mgh.harvard.edu>

...

    https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

    _______________________________________________
    Freesurfer mailing list
    Freesurfer@nmr.mgh.harvard.edu
    <mailto:Freesurfer@nmr.mgh.harvard.edu>
    https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer


_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu  <mailto:

Freesurfer@nmr.mgh.harvard.edu>

...

https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu <mailto:

Freesurfer@nmr.mgh.harvard.edu>

...

https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

2382

Age (days ago)

2385

Last active (days ago)

freesurfer@nmr.mgh.harvard.edu

11 comments

2 participants

tags (0)

participants (2)

Greve, Douglas N.,Ph.D.
Ting Li