If you are getting negative values, it probably means that your model (instantiated by the X matrix) is not proper in some ranges. The formula below extrapolates the effects of the covariates in order to subtract them off of a subject's value. If the model is not quite right, then the extrapolation will not be quite right, and you can get uninterpretable numbers. I don't have a solution for it.
On 9/18/19 11:56 AM, Maximo, Jose Omar wrote: > > External Email - Use Caution > > Hi, > > I tried that and it seemed to work. My averages between the 2 groups > look different now instead of almost identical. Thanks! > > I have another similar question. I have a cluster that shows a > significant difference between the groups when looking at volume. When > I look at the y.ocn.dat file, the values are in the 0.74188 - 2.16321 > range, which are pretty low for volume. I found a post where it says > that you can multiply those values with the cluster size(mm^2) from > the summary file to convert them to volume. Then, I tried using the > formula you provided below (beta, res, yhat), but then I get some > negative values. Negative volumetric values do not make sense, so I > was thinking if using the absolute value would correct for that. > > Any thoughts? > > Thanks for all the help so far. > > >If you say y = ocn.dat, then > > >beta = (X'*X)*X'*y; > > >res = y - X*beta; > > >yhat = X(:,1:2)*beta(1:2) + res; > > >yhat will be a score for each subject with the nuisance variables removed. > > > > > >On 9/11/19 2:40 PM, Maximo, Jose Omar wrote: > > > > > > External Email - Use Caution > > > > > >Hi, > > > >I plotted the data from the *.y.ocn.dat file and the graph still shows > no apparent significant difference between the groups (see attached > pic). Then I >found a post with the following: > > > /the ocn.dat files have data that is uncorrected in that sense and > might need to nuisance factors removed before plotting./ > /There is a design matrix in there (Xg.dat). You can load that into > matlab along with the ocn.dat, compute beta = inv(X'*X)*(X'*ocn)/ > /to get the betas. You can then compute yhat = X2*beta2 where X2 has > nuisance columns removed and beta2 has the same nuisance coefficients > removed, then treat yhat as your data to be plotted./ > > > >I tried that since I have 2 nuisance factors (age and TICV) and want > to plot my significant cluster w/o any nuisance effects. This is what > I get > > > 1.8973 > 1.8973 > 1.8973 > … > 3.4728 > 3.4728 > 3.4728 > … > >Are these the mean averages for each group (cortical thickness). If so, is > there a way to get a score for each subject? > > > >Best, > >Omar > > > >Look in the *.y.ocn.dat file. > > > > From mri_glmfit-sim --help > > > >csdbase.y.ocn.dat - this is a summary of the input (y) over each > >cluster. It has a column for each cluster. Each row is a subject. The > >value is the average of the input (y) in each cluster. This is a > >simple text file. > > > > > > > >On 9/3/2019 5:16 PM, Maximo, Jose Omar wrote: > >/ External Email - Use Caution/ > >// > >/Hi,/ > >// > >/Which specific file should I load? I see cluster.mgh, > cluster.summary, sign.ocn.annot, sig.ocn.mgh, sig.vertex.mgh, and > pdf.dat./ > >// > >/How can I extract the values from fsaverage space?/ > >// > >/Basically, what is the correct way to extract values from these > significant clusters?/ > >// > >/Many thanks,/ > >/Omar/ > >// > >/ Date: Tue, 3 Sep 2019 15:15:02 +0000/ > >/ From: "Greve, Douglas N.,Ph.D." <DGREVE at mgh.harvard.edu > <https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer>>/ > >/ Subject: Re: [Freesurfer] qdec contrast and data extraction/ > >/ To: "freesurfer at nmr.mgh.harvard.edu > <https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer>" > <freesurfer at nmr.mgh.harvard.edu > <https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer>>/ > >/ Message-ID: <2286f9bf-37a4-1be9-2252-0bccc833a218 at > mgh.harvard.edu > <https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer>>/ > >/ Content-Type: text/plain; charset="utf-8"/ > >// > >/ You might have done something wrong along the way. When you run > the montecarlo correction, it will create a file with the thickness > values in it for each cluster. the first thing to do is to load that > and see if you see the expected differences. The other thing is to not > go back into native space to extract the numbers. There are several > operations that happen as it moves into fsaverage space and in > preparation for group analysis (interpolation, and smoothing); > sometimes these make a big difference. if the ROI is small, it may not > map accurately back into the native space (and you should not need to > draw it in the first place)/ > >// > >/ On 8/30/2019 11:36 AM, Maximo, Jose Omar wrote:/ > >// > >/ External Email - Use Caution/ > >/ Hi,/ > >// > >/ I have a question:/ > >// > >/ My design is 2 groups (HC and Patients in that respective > order) and 2 nuisance factors (age and eTIV). When I look at the > average volume difference between the 2 groups, I get blue and red > clusters. I presume the color coding is where each group is greater > than the other (Blue = Patients > HC and Red = HC > Patients)./ > >// > >/ Then, I processed to extract individual values from each > cluster in order to plot them. When I extract data from the blue > clusters and plot them, the two groups show no difference in thickness > at all, whereas when I look at volume, HC show more than patients in > blue clusters (see attached figure). I would assume that both figures > would show patients > HC based on the negative statistic./ > >// > >/ Am I interpreting the colors wrong? Or am I doing something wrong?/ > >// > >/ These are my steps 1) After applying montecarlo correction, I > drew my ROIs to extract the data from; 2) map it onto every single > subject; and then 3) used mris_anatomical_stats to extract the data > from each subject./ > >// > >/ Any suggestions are welcome./ > >// > >/ Best,/ > >/ Omar/ > >// > >/ Jose O. Maximo, Ph.D. | Postdoctoral Fellow/ > >/ Department of Psychiatry & Behavioral Neurobiology/ > >/ UAB School of Medicine/ > >/ University of Alabama at Birmingham/ > >/ Cell Phone: (619) 252-3492/ > >/ Email: jomaximo at uabmc.edu > <https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer><mailto:jomaximo > > at uabmc.edu > <https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer>>/ > > _______________________________________________ > Freesurfer mailing list > [email protected] > https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer _______________________________________________ Freesurfer mailing list [email protected] https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
