Hi Freesurfer team,
I am sorry for writing again. Would you please be able to answer my last question?
Best Regards,
Katarina Trojacanec, M.Sc. Teaching and research assistant
Faculty of Computer Science and Engineering Ss. Cyril and Methodius University - Skopje, Republic of Macedonia
________________________________ From: freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edu on behalf of Katarina Trojacanec katarina.trojacanec@finki.ukim.mk Sent: Friday, February 3, 2017 9:02 AM To: freesurfer@nmr.mgh.harvard.edu Subject: [Freesurfer] Fw: Analysis of rates or percent changes
This sender failed our fraud detection checks and may not be who they appear to be. Learn about spoofinghttp://aka.ms/LearnAboutSpoofing Feedbackhttp://aka.ms/SafetyTipsFeedback
Hi Martin,
Apologies for multiple e-mails, but I got a message that something might be wrong with the previous e-mail, so I forward it again.
Please find below my previous question.
Regards, Katarina
Katarina Trojacanec, M.Sc. Teaching and research assistant
Faculty of Computer Science and Engineering Ss. Cyril and Methodius University - Skopje, Republic of Macedonia
________________________________ From: Katarina Trojacanec Sent: Thursday, February 2, 2017 2:46 PM To: Freesurfer support list Subject: Re: [Freesurfer] Analysis of rates or percent changes
Hi Martin,
I dropped --generic-time flag as you suggested in the previous e-mail and run the long_stats_slopes command again like this:
long_stats_slopes --qdec ./qdec/long_AD_NL_TP1_2_3_4.qdec.table.dat --stats aseg.stats --meas volume --sd $SUBJECTS_DIR --do-rate --do-pc1fit --do-pc1 --do-spc --time years --stack-rate ./qdec/long_AD_NL_TP1_2_3_4_m.aseg-rate.stack.txt --stack-pc1fit ./qdec/long_AD_NL_TP1_2_3_4_m.aseg-pc1fit.stack.txt --stack-pc1 ./qdec/long_AD_NL_TP1_2_3_4_m.aseg-pc1.stack.txt --stack-spc ./qdec/long_AD_NL_TP1_2_3_4_m.aseg-spc.stack.txt
However, I got exactly the same results for stacked rate pc1/fit, spc (the files long_AD_NL_TP1_2_3_4_m.aseg-*.stack.txt) as in the case with the --generic-time flag. Is this reasonable?
As I read and understood, the --generic-time flag assumes that time difference between the time points is 1, which is not true in my case (the time difference between TP1 and TP2 is around six months, and between TP2 and TP3, as well as TP3 and TP4 is one year). This indicates that there should be some difference in the result.
Am I doing something wrong or misunderstood anything, or maybe there are some temporary files from the first running of the command that are used also in the second one (without --generic-time flag)?
Best, Katarina
Katarina Trojacanec, M.Sc. Teaching and research assistant
Faculty of Computer Science and Engineering Ss. Cyril and Methodius University - Skopje, Republic of Macedonia
________________________________ From: Katarina Trojacanec Sent: Monday, September 12, 2016 10:01:53 PM To: Freesurfer support list Subject: Re: [Freesurfer] Analysis of rates or percent changes
Hi Martin,
Thanks a lot.
Best, Katarina
Katarina Trojacanec, M.Sc. Teaching and research assistant
Faculty of Computer Science and Engineering Ss. Cyril and Methodius University - Skopje, Republic of Macedonia
________________________________ From: freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edu on behalf of Martin Reuter mreuter@nmr.mgh.harvard.edu Sent: Monday, September 12, 2016 7:58:14 PM To: freesurfer@nmr.mgh.harvard.edu Subject: Re: [Freesurfer] Analysis of rates or percent changes
Hi Katarina,
different time spacing is OK (not optimal, but OK). This 2-stage procedure first fits a line in each subject, independent on how many time points the subject has. Of course a linear fit from 4 time points will be more reliable than one from just 2 time points. This is not taken into consideration and therefore we usually recommend the Linear Mixed Effects modeling (where it is considered in the model). If most of your subjects have 4 time points and the spacing is similar (across subjects) you should be fine.
Other things I noticed:
- you should drop the --generic-time flag. I probably should change the help text to be more specific, but what it does it assigns a time of 1 2 3 4 to the time points (this is if you do repeated measures and there is no real time). So don't pass it.
- you can drop some of the --do... flags. Probably you are interested in the rate and one of the pct change flags. The average is just the average thickness (across time) for each subject. Not sure you want to analyze that. The --do-stack is a subject-specific stack of the time points, usually only used for debugging.
- the -stack-avg is not needed unless you want to analyze average thickness for each subject, rather you may want to create a stack for the rate or the pc1 or whatever, to pass it into the stats tool (like mri_glmfit, or R, SPSS or whatever you use).
Best, Martin
On 09/12/2016 12:40 PM, Katarina Trojacanec wrote:
Hi,
I have a question about the analysis of rates or percent changes.
I use data with available scans at baseline (TP1) and the 6-month (TP2), 12-month (TP3) and 24-month (TP4) follow-ups from ADNI dataset. An example of some of the data in the appropriate qdec table is given below:
fsid fsid-base years age weight diagnosis gender Glob_CDR NPI-Q_TotScr MMSE_TotScr FAQ_TotScr visit1_2_3_4 ADNI_sub1_sc ADNI_base1 0 81.3 y AD M x x x x sc ADNI_sub1_m06 ADNI_base1 0.528767123 81.9 y AD M x x x x m06 ADNI_sub1_m12 ADNI_base1 1.030136986 82.4 y AD M x x x x m12 ADNI_sub1_m24 ADNI_base1 2.030136986 83.4 y AD M x x x x m24 ADNI_sub2_sc ADNI_base2 0 74 y AD F x x x x sc ADNI_sub2_m06 ADNI_base2 0.501369863 74.5 y AD F x x x x m06 ADNI_sub2_m12 ADNI_base2 1.005479452 75 y AD F x x x x m12 ADNI_sub2_m24 ADNI_base2 2 76 y AD F x x x x m24
The base is constructed using all four time points. The time variable is given in years. I am using long_stats_slopes for aseg.stats as follows:
long_stats_slopes --qdec ./qdec/long_AD_NL_TP1_2_3_4.qdec.table.dat --stats aseg.stats --meas volume --sd $SUBJECTS_DIR --do-avg --do-rate --do-pc1fit --do-pc1 --do-spc --do-stack --generic-time --time years --stack-avg ./qdec/long_AD_NL_TP1_2_3_4
(similarly for ?h.aparc.stats)
Having in mind that TP1 and TP2, as well as TP2 and TP3 are separated approximately 6 months and TP3 and TP4 are separated approximately 12 months, are the annualized percent change or atrophy rates using this scenario reasonable (statistically?). Is it maybe more reasonable to apply the same scenario using three time points (for example TP1, TP3 and TP4 from the previous example and the template based only on these time points with the same approximate difference between all of them of 12 months, or TP1, TP2 and TP3 from the previous example and the template based only on these time points with the same approximate difference between all of them of 6 months)?
Best Regards,
Katarina Trojacanec, M.Sc. Teaching and research assistant
Faculty of Computer Science and Engineering Ss. Cyril and Methodius University - Skopje, Republic of Macedonia
_______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edumailto:Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
-- Martin Reuter, PhD Assistant Professor of Radiology, Harvard Medical School Assistant Professor of Neurology, Harvard Medical School A.A.Martinos Center for Biomedical Imaging Massachusetts General Hospital Research Affiliate, CSAIL, MIT Phone: +1-617-724-5652 Web : http://reuter.mit.edu