 |
 |
Hi Rosalia, On 2 August 2016 at 08:48, Rosalia Dacosta Aguayo <[log in to unmask]> wrote: > Hi Anderson, > > You know me too much. I do not know if this is good ;) > As you surely have noticed, sometimes my questions are repeated because I > always have doubts and I need a professional feedback in order to feel more > secure I am doing things well in order to ensure that when I publish > something, the results are real and rigurous enough to be replicated. > > In this case, I have two options: > > 1. I have T1 MPRAGE images. The main of them are in saggital view oders > not...in axial...with different acquisition parameters and different > machines....so I finally thought not to conduct a longitudinal analysis but > a cross-sectional one, in order to minimize changes in time related to > other events that my variable of interest. I know I have to take into > account different machines in the GLM but not sure how to do this (I have > never been in that situation). By the other hand I believe that to ask my > question: which region of interest is mediating between X and Y...the best > option is mediation analysis...what not sure how to conduct the design > here, where I have to define X as independent variable, Y as dependent > variable, cov to enter all the covariates (age, gender, education and TIV) > and finally M as the mediator variable...here: the list of betted in MNI > space acquisitions. If I am interested to see not only this relations but > differences in this relations regarding three different groups...this is > worrying me a lot. > There isn't a stable tool in FSL doing mediation analysis at the moment so perhaps your best bet is Tor Wager's mediation toolbox. There is a script in PALM but it's still very experimental and not ready for prime time. I won't be able to help much here, perhaps others in the list. Same for the questions below, I leave for others to answer. All the best, Anderson > 2. The other question is that I would like to take into account some > regions of interest to see intrahemispheric as well as interhemispheric > connectivity with DTI data and compare connetivity patterns between the > same three groups regarding not only to connectivity but with some graph > metrics and see how these metrics are mediating again...as with structural > MRI and see the differences between three groups. In this case I know how > to arrive to bedpostx but not sure about probtrack to arrive to the > symmetric weighted matrices with Matlab. > > Regarding fMRI data...I have realized I do not have enough patients to > conduct this kind of a analysis. > > Finally, it would be enough to take into account only my three groups of > patients or should I better include MRI acquisitions from healthy controls > as reference group? > > Thanks in advance for your helping as well as your patience and your > confidence in my possibilities conducting some analysis with MRI. > > Rosalía > > El 2 ago. 2016 9:06, "Anderson M. Winkler" <[log in to unmask]> > escribió: > >> Hi Rosalia, >> >> So you have (let's simplify) three variables: X, Y, and imaging. What is >> your hypothesis related to all three? The association between X and Y is >> straightforward to do with the GLM, even in the repeated measures case, and >> I know that you know how to do it. I don't get what your hypothesis >> relating X, Y and the imaging data is. What and how to model depends on >> your answer. >> >> All the best, >> >> Anderson >> >> >> On 1 August 2016 at 08:48, Rosalia Dacosta Aguayo <[log in to unmask]> >> wrote: >> >>> Dear Anderson and FSL team, >>> >>> I have a doubt and I would be greatly appreciate your experience in >>> order to set the better design to study my hypothesis. >>> >>> I have one variable X that was measured only at Time0 (basal) and I have >>> another variable that was measured at Time0, Time1, Time2, Time3 and Time4 >>> (repeated measures during four years) over three different groups that has >>> been splitted taking into account variable X at Time0. I think I should use >>> a mixed effect model followed by Kaplan and Meyer graphics and Cox >>> Regression to calculate Hazard Ratio value as well as p value associated. >>> >>> From a neuroimaging point of view. I have only 45 MRI scans T1 (not very >>> well quality and I had problems with registration to template even after >>> removing artefacts and deleting several scans (from 404 I have only >>> 45...with same acquisition parameters and same machine). I had to performe >>> registration with another toolbox and the results were very good. The fact >>> is that I will like to conduct and analysis to see the effect of variable X >>> over another variables >>> >>> I have the hypothesis that status at variable X will have effects over >>> another variable Y that was measured at Time0, Time1, Time2, Time3 and >>> Time4 (repeated measures during four years). I will like to test this >>> hypothesis either with MRI and fMRI. Here, my sample is small enough to >>> prevent me to split it in three groups, so I consider all my sample as just >>> one group. My first thought was to conduct this kind of analysis >>> taking into account variable X at Time0 and variable Y at time4. But I >>> think it is far more interesting to conduct a repeated measures analysis in >>> order to see the changes in structural MRI and fMRI at rest regarding to >>> see the effects of variable X over variable Y in structural MRI and >>> functional fMRI along 4 years. But, I have never done this kind of analysis >>> and I am pretty lost regarding the design I would have to conduct using GLM >>> (in the case of structural MRI) and FEAT in the case of fMRI at rest). I >>> know that with fMRI I have to correct by motion parameters (6) and by CSF >>> as well as WM signal...but I am not able to imagine how to conduct this >>> kind of analysis taking into account that, for example, motion parameters, >>> CSF and WM signals will be different for every individual scan at every >>> year.... >>> >>> Any help with this issue would be highly appreciate. >>> >>> Now I am going to pre-process fMRI data with MELODIC...just for T0 >>> scans...but do not know if pre-process the other scans related to T1, T2, >>> T3 and T4 separately..... >>> >>> Thank you a lot, >>> >>> Rosalia >>> >> >>