Dear all
I'm fairly new to Freesufer and have started a cortical thickness group analysis. I had a few doubts/questions which I found most the answers for on previous emails on the FS archives - thank you! But I have a few more questions, I hope one of the FS experts can help me out. I will briefly describe my analysis - I have 2 groups - controls and patients. I am interested in the simple questions - do controls have higher GM thickness than patients and where? Do patients have higher GM thickness than controls and where? I expect controls to have higher GM thickness than the patients in cortical areas, but I expect the patients to have higher volumes of certain subcortical structures.
I am not interested in the effects of age or gender on GM thickness although I would like to control for these as covariates of no interest. In the example below I have not considered gender for simplicity's sake. Because I assume a similar slope among both groups, I decided to go with DOSS and not DODS.
I set out my FSGD file as below:
GroupDescriptorFile 1 Title lh.group_age.glmdir MeasurementName thickness Class Control Class Patient Variables age Input Subject_02 Control 29 Input Subject_04 Control 22 Input Subject_05 Control 23 Input Subject_06 Control 29 Input Subject_07 Control 18 Input Subject_08 Control 20 Input Subject_09 Control 36 Input Subject_10 Control 24 Input Subject_14 Control 29 Input Subject_15 Control 26 Input Subject_16 Control 39 Input Subject_17 Control 40 Input Subject_18 Control 26 Input Subject_19 Control 44 Input Subject_20 Control 34 Input Subject_22 Control 28 Input Subject_23 Control 26 Input Subject_31 Control 46 Input Subject_32 Control 40 Input Subject_41 Control 38 Input Subject_01 Patient 35 Input Subject_03 Patient 28 Input Subject_11 Patient 46 Input Subject_12 Patient 22 Input Subject_21 Patient 32 Input Subject_25 Patient 33 Input Subject_26 Patient 40 Input Subject_27 Patient 30 Input Subject_28 Patient 31 Input Subject_29 Patient 42 Input Subject_30 Patient 39 Input Subject_33 Patient 43 Input Subject_34 Patient 44 Input Subject_35 Patient 36 Input Subject_36 Patient 32 Input Subject_37 Patient 37 Input Subject_38 Patient 27
And my contrast files as: lh-diff-ControlvsPatient.mtx: +1 -1 0 0 lh-diff-PatientvsControl.mtx: -1 1 0 0
For GLM analysis I used the following command: "mri_glmfit --y lh.group_age.10.mgh --fsgd lh.group.fsgd doss --C lh-diff-ControlvsPatient.mtx --surf fsaverage lh --cortex --glmdir lh.group_age.glmdir"
My questions are:
Surface-based group analysis:
1. Can I ignore the error message: 'WARNING: unrecognized mri_glmfit cmd option doss'
2. Do I need 2 contrasts as above or should I be running only the first contrast with the 'abs' thresh-sign in order to obtain a double-sided t-test?
3. If my alternative hypothesis is that Controls>Patients and I specify a 'negative' thresh-sign - does that mean I will only see clusters where Controls<Patients (the reverse of my contrast)? The command I run is "mri_glmfit-sim --glmdir lh.group_age_precachedsim.glmdir--cache 1.3 neg"
4. I'm a bit confused about the vertex-wise threshold - if I specify it to be 3 (p=0.001), is that more or less stringent than specifying it to be 1.3 (p=0.05). I thought that the later (p=0.05) would allow for the consideration of more vertices and therefore would be less stringent, however when I applied both to try and test my theory, no clusters survived multiple comparison correction with p=0.05 but 1 cluster survived when I specified p=0.001. Could someone please shed some light on this?
Subcortical volume group analysis:
1. Can I just perform statistics in a program such as SPSS on the subcortical structure volumes I obtain from the 'aseg.stats' file within each subject's stats sub-directory? Or are there other steps that need to be undertaken in FS beforehand?
I'm sorry about the long email, I really appreciate your advice so thank you in advance
Kind regards Reem
Reem Jan BPharm (Hons), RegPharmNZ
PhD Student / Pharmacist School of Pharmacy, Faculty of Medical & Health Sciences, The University of Auckland, Private Bag 92019, Auckland, New Zealand. Ph: +64 9 373 7599 ext 81138. DDI: +64 9 923 1138 F: +64 9 367 7192
Hi Reem,
#1. Is this from mri_glmfit-sim? If so, you can ignore it #2. You do not need both contrasts. Use abs to get a double sided test #3. The first contrast will be positive for Control>Patient. If you specify neg, it will only show clusters where Control<Patient #4. Depends on what you mean by "stringent". There is nothing more or less conservative about any particular voxel-wise threshold since you are not drawing conclusions based on the voxel-wise threshold but on the cluster-wise p-value. In general, the clusterwise correction is probably more reliable at higher voxel-wise thresholds.
Subcortical: yes, just create a table and load it into SPSS. The only other thing is that you may want to normalize the volumes for ICV or brain volume, depending upon your hypothesis.
doug
On 05/09/2012 08:44 PM, Reem Jan wrote:
Dear all
I’m fairly new to Freesufer and have started a cortical thickness group analysis. I had a few doubts/questions which I found most the answers for on previous emails on the FS archives – thank you! But I have a few more questions, I hope one of the FS experts can help me out.
I will briefly describe my analysis – I have 2 groups – controls and patients. I am interested in the simple questions – do controls have higher GM thickness than patients and where? Do patients have higher GM thickness than controls and where? I expect controls to have higher GM thickness than the patients in cortical areas, but I expect the patients to have higher volumes of certain subcortical structures.
I am not interested in the effects of age or gender on GM thickness although I would like to control for these as covariates of no interest. In the example below I have not considered gender for simplicity’s sake. Because I assume a similar slope among both groups, I decided to go with DOSS and not DODS.
I set out my FSGD file as below:
GroupDescriptorFile 1
Title lh.group_age.glmdir
MeasurementName thickness
Class Control
Class Patient
Variables age
Input Subject_02 Control 29
Input Subject_04 Control 22
Input Subject_05 Control 23
Input Subject_06 Control 29
Input Subject_07 Control 18
Input Subject_08 Control 20
Input Subject_09 Control 36
Input Subject_10 Control 24
Input Subject_14 Control 29
Input Subject_15 Control 26
Input Subject_16 Control 39
Input Subject_17 Control 40
Input Subject_18 Control 26
Input Subject_19 Control 44
Input Subject_20 Control 34
Input Subject_22 Control 28
Input Subject_23 Control 26
Input Subject_31 Control 46
Input Subject_32 Control 40
Input Subject_41 Control 38
Input Subject_01 Patient 35
Input Subject_03 Patient 28
Input Subject_11 Patient 46
Input Subject_12 Patient 22
Input Subject_21 Patient 32
Input Subject_25 Patient 33
Input Subject_26 Patient 40
Input Subject_27 Patient 30
Input Subject_28 Patient 31
Input Subject_29 Patient 42
Input Subject_30 Patient 39
Input Subject_33 Patient 43
Input Subject_34 Patient 44
Input Subject_35 Patient 36
Input Subject_36 Patient 32
Input Subject_37 Patient 37
Input Subject_38 Patient 27
And my contrast files as:
lh-diff-ControlvsPatient.mtx: +1 -1 0 0
lh-diff-PatientvsControl.mtx: -1 1 0 0
For GLM analysis I used the following command:
“mri_glmfit --y lh.group_age.10.mgh --fsgd lh.group.fsgd doss --C lh-diff-ControlvsPatient.mtx --surf fsaverage lh --cortex --glmdir lh.group_age.glmdir”
My questions are:
_Surface-based group analysis:_
1.Can I ignore the error message: *‘WARNING: unrecognized mri_glmfit cmd option doss’*
2.Do I need 2 contrasts as above or should I be running only the first contrast with the ‘abs’ thresh-sign in order to obtain a double-sided t-test?
3.If my alternative hypothesis is that Controls>Patients and I specify a ‘negative’ thresh-sign – does that mean I will only see clusters where Controls<Patients (the reverse of my contrast)? The command I run is *“mri_glmfit-sim --glmdir lh.group_age_precachedsim.glmdir--cache 1.3 neg”*
4.I’m a bit confused about the vertex-wise threshold – if I specify it to be 3 (p=0.001), is that more or less stringent than specifying it to be 1.3 (p=0.05). I thought that the later (p=0.05) would allow for the consideration of more vertices and therefore would be less stringent, however when I applied both to try and test my theory, no clusters survived multiple comparison correction with p=0.05 but 1 cluster survived when I specified p=0.001. Could someone please shed some light on this?
_Subcortical volume group analysis:_
1.Can I just perform statistics in a program such as SPSS on the subcortical structure volumes I obtain from the ‘aseg.stats’ file within each subject’s stats sub-directory? Or are there other steps that need to be undertaken in FS beforehand?
I’m sorry about the long email, I really appreciate your advice so thank you in advance
Kind regards
Reem
*Reem Jan***
BPharm (Hons), RegPharmNZ
PhD Student / Pharmacist
School of Pharmacy, Faculty of Medical & Health Sciences, The University of Auckland, Private Bag 92019, Auckland, New Zealand.
Ph: +64 9 373 7599 ext 81138. DDI: +64 9 923 1138
F: +64 9 367 7192
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
freesurfer@nmr.mgh.harvard.edu
-
Douglas N Greve -
Reem Jan