[Freesurfer] Question regarding mri_glmfit DODS vs SPSS interaction terms in GLM analyses
How are the DODS analyses in Freesurfer comparable to interaction terms in GLM analyses in SPSS? I notice that others have had similar questions, but it is still unclear to me. To get comparable interaction terms in SPSS GLM analyses and individual slopes in mri-glmfit Freesurfer DODS analyses, I have used the following independent variables in the SPSS GLM: group, gender, age*group*gender (age was centralized before calculating the interaction term). Another alternative suggested by a colleague was to include all main effects and possible interaction effects, thus group, gender, age, age*group, age*gender, group*gender, age*group*gender. I also saw an earlier reply on the support page suggesting the following independent variables: group*age, gender*age, group*gender*age, but I am uncertain whether the main effects should also be included. Thus, I wonder which independent variables you would suggest should be included in the GLM analyses in SPSS to be as comparable to mri_glmfit DODS analyses in Freesurfer as possible? Below is more information about what analyses I have done: Overview analyses: We investigate difference between two groups. We have to sets of data, one with skewed age and gender distribution between groups (n = 89) (older participants and more girls in the risk group), and one with 33 extra control participants (n = 122) making the risk and control group having similar age and gender distribution. The ROI area and thickness from APARC exported files are analyzed using GLM analyses in SPSS. We have done the analyses both without any covariate, and controlling for gender or/and age. We have also run analyses controlling for gender and the interaction term group*gender*age (with age being centralized). Here is an example of the SPSS syntax for the full model trying to be as similar as possible to the vertex based mri_glmfit DODS analyses in Freesurfer as possible: GLM lh_bankssts_area BY group gender WITH zage /PRINT=DESCRIPTIVE PARAMETER /DESIGN = gender group gender*group*zage. The vertex based analyses are done with mri_glmfit in Freesurfer version 5.3. I have analyzed the data both with DOSS and DODS. Both without covariates, controlling for age (demeaned continuous variable) or controlling for gender, or controlling for both age and gender. We have used smoothing fwhm 10. The results were controlled for multiple comparisons with the standard mri_glmfit-sim method in Freesurfer with cache 1.3 abs (we did not use 2spaces because we will also compare the results from mri_glmfit with results from Qdec). We have used tksurfer (DOSS and DODS) and qdec (DODS) to visualize the results, but mainly used results from cache.th13.abs.sig.cluster.summary files to compare with the analyses from SPSS. The contrast in the full DODS model in the mri_glmfit is 1 1 -1 -1 0 0 0 0, with the fsgd file being: GroupDescriptorFile 1 Title n89-33 Class group1risk_gender1girl Class group1risk_gender2boy Class group2control_gender1girl Class group2control_gender2boy Variables age Input UN001 group1control_gender2boy -1.0172131148 Etc Preproc was done with: mris_preproc --target fsaverage --hemi lh --meas area --fsgd $glm/demographicn122.fsgd --out $preproc Smoothing was done with: mri_surf2surf --hemi lh --s fsaverage --sval $preproc --fwhm 10 --cortex --tval $smoothing10 GLM was done with: mri_glmfit --glmdir $contrast --y $smoothing10 --fsgd $glm/demographicn122.fsgd --C $glm$contrast_kjal.mat --surf fsaverage lh --cortex Correction for multiple comparisons was done with: mri_glmfit-sim --glmdir $contrast --cache 1.3 abs Kind regards, Egil Nygaard ___ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer The information in this e-mail is intended only for the person to whom it is addressed. If you believe this e-mail was sent to you in error and the e-mail contains patient information, please contact the Partners Compliance HelpLine at http://www.partners.org/complianceline . If the e-mail was sent to you in error but does not contain patient information, please contact the sender and properly dispose of the e-mail.
[Freesurfer] Question regarding Qdec vs mri_glmfit outputs
Dear Freesurfer support, I find the qdec to be very pedagogic in its outlay, and would love to use it. However, I find several discrepancies as compared to results from the mri_glmfit analyses, and wonder why (see description below). 1) One discrepancy is the placement of the clusters; X, Y, Z's. Should I use: * the figures from the result summary from mri_glmfit, * the result summary from Qdec (which is almost the same as from the results summary from mri_glmfit), * or from the “Find clusters and goto max” function in the display window of Qdec (which shows very different placements than the two others)? 2) Which size should I use? The sizes of the clusters from qdec are 1-2% smaller than what are reported from the mri_glmfit analyses. I expect that the reason for the discrepancies is that I probably do something different in the mri_glmfit analyses than the standard Qdec analyses, but I can not figure out what it is. Below is information about which analyses I have done and some more information about the Qdec output. Overview analyses: We investigate difference between two groups. We have two sets of data, one with skewed age and gender distribution between groups (n = 89) (older participants and more girls in the risk group), and one with 33 extra control participants (n = 122) making the risk and control group having similar age and gender distribution. The vertex based analyses are done with mri_glmfit in Freesurfer version 5.3. I have analyzed the data both with DOSS and DODS. Both without covariates, controlling for age (demeaned continuous variable) or controlling for gender, or controlling for both age and gender. We have used smoothing fwhm 10. The results were controlled for multiple comparisons with the standard mri_glmfit-sim method in Freesurfer with cache 1.3 abs and 2spaces (did not use 2spaces when comparing mri_glmfit results with Qdec results). We have used tksurfer (DOSS and DODS) and qdec (DODS) to visualize the results, but mainly used results from cache.th13.abs.sig.cluster.summary files to compare with the analyses from SPSS. The contrast in the full DODS model in the mri_glmfit is 1 1 -1 -1 0 0 0 0, with the fsgd file being: GroupDescriptorFile 1 Title n89-33 Class group1risk_gender1girl Class group1risk_gender2boy Class group2control_gender1girl Class group2control_gender2boy Variables age Input UN001 group1control_gender2boy -1.0172131148 Etc Preproc was done with: mris_preproc --target fsaverage --hemi lh --meas area --fsgd $glm/demographicn122.fsgd --out $preproc Smoothing was done with: mri_surf2surf --hemi lh --s fsaverage --sval $preproc --fwhm 10 --cortex --tval $smoothing10 GLM was done with: mri_glmfit --glmdir $contrast --y $smoothing10 --fsgd $glm/demographicn122.fsgd --C $glm$contrast_kjal.mat --surf fsaverage lh --cortex Correction for multiple comparisons was done with: mri_glmfit-sim --glmdir $contrast --cache 1.3 abs Qdec: There are discrepancies between results from qdec and mri_glmfit. I am unsure why. I notice for example that there are small discrepancies in the fwhm.dat files from Qdec and mri_glmfit. The most important for my use is that when I compare the clusters in the file: cache.th13.abs.sig.cluster.summary from the mri_glmfit, and the clusters from the file: mc-z.abs.th13.sig.cluster.summary from qdec, I find the same four clusters of group differences in cortex areas with the same Max number and Vtxmax. However, the sizes are marginally different, with the sizes from qdec being 1-2% smaller. Most importantly for me are the differences in the placement of the clusters. The MNIX, Y and Z from mri_glmfit are not similar to the TalX, Y and Z from Qdec. The differences are not big, but still significant for us. I also note that the TalX, Y and Z's from the result file from Qdec is very different from the TalX, Y and Z's from the output one get from Qdec when one use the button “Find clusters and got max” in the visual Display window in Qdec. It seems the output from Qdec's “Find clusters and got max” shows the mean cluster placement (all Max figures are 4), whereas the output from mri_glmfit and from the resultsfile from Qdec shows the placement of masked clusters (the Max figures vary)? Kind regards, Egil Nygaard ___ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer The information in this e-mail is intended only for the person to whom it is addressed. If you believe this e-mail was sent to you in error and the e-mail contains patient information, please contact the Partners Compliance HelpLine at http://www.partners.org/complianceline . If the e-mail was sent to you in error but does not contain patient information, please contact the sender and properly dispose of the e-mail.
Re: [Freesurfer] Question regarding Qdec vs mri_glmfit outputs
Ty Doug, Kind regards, Egil From: freesurfer-boun...@nmr.mgh.harvard.edu on behalf of Douglas N Greve Sent: 23 March 2015 17:45 To: freesurfer@nmr.mgh.harvard.edu Subject: Re: [Freesurfer] Question regarding Qdec vs mri_glmfit outputs Use the result from mri_glmfit. doug On 03/20/2015 10:20 AM, Egil Nygaard wrote: > > Dear Freesurfer support, > > I find the qdec to be very pedagogic in its outlay, and would love to > use it. However, I find several discrepancies as compared to results > from the mri_glmfit analyses, and wonder why (see description below). > > 1) One discrepancy is the placement of the clusters; X, Y, Z's. Should > I use: > > * > the figures from the result summary from mri_glmfit, > * > the result summary from Qdec (which is almost the same as from the > results summary from mri_glmfit), > * > or from the “Find clusters and goto max” function in the display > window of Qdec (which shows very different placements than the two > others)? > > 2) Which size should I use? The sizes of the clusters from qdec are > 1-2% smaller than what are reported from the mri_glmfit analyses. > > I expect that the reason for the discrepancies is that I probably do > something different in the mri_glmfit analyses than the standard Qdec > analyses, but I can not figure out what it is. > > > > Below is information about which analyses I have done and some more > information about the Qdec output. > ** > > *Overview analyses:* > > We investigate difference between two groups. We have two sets of > data, one with skewed age and gender distribution between groups (n = > 89) (older participants and more girls in the risk group), and one > with 33 extra control participants (n = 122) making the risk and > control group having similar age and gender distribution. > > The vertex based analyses are done with mri_glmfit in Freesurfer > version 5.3. I have analyzed the data both with DOSS and DODS. Both > without covariates, controlling for age (demeaned continuous variable) > or controlling for gender, or controlling for both age and gender. We > have used smoothing fwhm 10. The results were controlled for multiple > comparisons with the standard mri_glmfit-sim method in Freesurfer with > cache 1.3 abs and 2spaces (did not use 2spaces when comparing > mri_glmfit results with Qdec results). > > We have used tksurfer (DOSS and DODS) and qdec (DODS) to visualize the > results, but mainly used results from > cache.th13.abs.sig.cluster.summary files to compare with the analyses > from SPSS. > > The contrast in the full DODS model in the mri_glmfit is 1 1 -1 -1 0 0 > 0 0, with the fsgd file being: > > GroupDescriptorFile 1 > Title n89-33 > Class group1risk_gender1girl > Class group1risk_gender2boy > Class group2control_gender1girl > Class group2control_gender2boy > Variables age > Input UN001 group1control_gender2boy -1.0172131148 > Etc > > Preproc was done with: > > mris_preproc --target fsaverage --hemi lh --meas area --fsgd > $glm/demographicn122.fsgd --out $preproc > > Smoothing was done with: > mri_surf2surf --hemi lh --s fsaverage --sval $preproc --fwhm 10 > --cortex --tval $smoothing10 > > GLM was done with: > > mri_glmfit --glmdir $contrast --y $smoothing10 --fsgd > $glm/demographicn122.fsgd --C $glm$contrast_kjal.mat --surf fsaverage > lh --cortex > > Correction for multiple comparisons was done with: > > mri_glmfit-sim --glmdir $contrast --cache 1.3 abs > > *Qdec:* > There are discrepancies between results from qdec and mri_glmfit. I am > unsure why. I notice for example that there are small discrepancies in > the fwhm.dat files from Qdec and mri_glmfit. The most important for my > use is that when I compare the clusters in the file: > cache.th13.abs.sig.cluster.summary from the mri_glmfit, and the > clusters from the file: mc-z.abs.th13.sig.cluster.summary from qdec, I > find the same four clusters of group differences in cortex areas with > the same Max number and Vtxmax. However, the sizes are marginally > different, with the sizes from qdec being 1-2% smaller. Most > importantly for me are the differences in the placement of the > clusters. The MNIX, Y and Z from mri_glmfit are not similar to the > TalX, Y and Z from Qdec. The differences are not big, but still > significant for us. I also note that the TalX, Y and Z's from the > result file from Qdec is very different from the TalX, Y and Z's from > the output one get from Qdec when one use the button “Find clusters > and got max” in the visual Display window in Qdec. It seems the output > from Qdec's “Find clusters and got max” shows the mean clus
Re: [Freesurfer] Question regarding mri_glmfit DODS vs SPSS interaction terms in GLM analyses
Ty Doug, That gives 8 regressors, so it seems reasonable. Kind regards, Egil From: freesurfer-boun...@nmr.mgh.harvard.edu on behalf of Douglas N Greve Sent: 23 March 2015 17:44 To: freesurfer@nmr.mgh.harvard.edu Subject: Re: [Freesurfer] Question regarding mri_glmfit DODS vs SPSS interaction terms in GLM analyses I don't use SPSS so I can't really comment with any authority. But I think it is group*gender and group*gender*age doug On 03/20/2015 10:08 AM, Egil Nygaard wrote: > > How are the DODS analyses in Freesurfer comparable to interaction > terms in GLM analyses in SPSS? I notice that others have had similar > questions, but it is still unclear to me. > > To get comparable interaction terms in SPSS GLM analyses and > individual slopes in mri-glmfit Freesurfer DODS analyses, I have used > the following independent variables in the SPSS GLM: group, gender, > age*group*gender (age was centralized before calculating the > interaction term). > > Another alternative suggested by a colleague was to include all main > effects and possible interaction effects, thus group, gender, age, > age*group, age*gender, group*gender, age*group*gender. > > I also saw an earlier reply on the support page suggesting the > following independent variables: group*age, gender*age, > group*gender*age, but I am uncertain whether the main effects should > also be included. > > Thus, I wonder which independent variables you would suggest should be > included in the GLM analyses in SPSS to be as comparable to mri_glmfit > DODS analyses in Freesurfer as possible? > > Below is more information about what analyses I have done: > > *Overview analyses:* > > We investigate difference between two groups. We have to sets of data, > one with skewed age and gender distribution between groups (n = 89) > (older participants and more girls in the risk group), and one with 33 > extra control participants (n = 122) making the risk and control group > having similar age and gender distribution. > > The ROI area and thickness from APARC exported files are analyzed > using GLM analyses in SPSS. We have done the analyses both without any > covariate, and controlling for gender or/and age. We have also run > analyses controlling for gender and the interaction term > group*gender*age (with age being centralized). Here is an example of > the SPSS syntax for the full model trying to be as similar as possible > to the vertex based mri_glmfit DODS analyses in Freesurfer as possible: > > GLM lh_bankssts_area BY group gender WITH zage > /PRINT=DESCRIPTIVE PARAMETER > /DESIGN = gender group gender*group*zage. > > The vertex based analyses are done with mri_glmfit in Freesurfer > version 5.3. I have analyzed the data both with DOSS and DODS. Both > without covariates, controlling for age (demeaned continuous variable) > or controlling for gender, or controlling for both age and gender. We > have used smoothing fwhm 10. The results were controlled for multiple > comparisons with the standard mri_glmfit-sim method in Freesurfer with > cache 1.3 abs (we did not use 2spaces because we will also compare the > results from mri_glmfit with results from Qdec). > > We have used tksurfer (DOSS and DODS) and qdec (DODS) to visualize the > results, but mainly used results from > cache.th13.abs.sig.cluster.summary files to compare with the analyses > from SPSS. > > The contrast in the full DODS model in the mri_glmfit is 1 1 -1 -1 0 0 > 0 0, with the fsgd file being: > > GroupDescriptorFile 1 > Title n89-33 > Class group1risk_gender1girl > Class group1risk_gender2boy > Class group2control_gender1girl > Class group2control_gender2boy > Variables age > Input UN001 group1control_gender2boy -1.0172131148 > Etc > > Preproc was done with: > > mris_preproc --target fsaverage --hemi lh --meas area --fsgd > $glm/demographicn122.fsgd --out $preproc > > Smoothing was done with: > mri_surf2surf --hemi lh --s fsaverage --sval $preproc --fwhm 10 > --cortex --tval $smoothing10 > > GLM was done with: > > mri_glmfit --glmdir $contrast --y $smoothing10 --fsgd > $glm/demographicn122.fsgd --C $glm$contrast_kjal.mat --surf fsaverage > lh --cortex > > Correction for multiple comparisons was done with: > > mri_glmfit-sim --glmdir $contrast --cache 1.3 abs > > Kind regards, > > Egil Nygaard > > > > ___ > Freesurfer mailing list > Freesurfer@nmr.mgh.harvard.edu > https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer -- Douglas N. Greve, Ph.D. MGH-NMR Center gr...@nmr.mgh.harvard.edu Phone Number: 617-724-2358 Fax: 617-726-7422 Bugs: surfer.nmr.mgh.harv
