 |
 |
Dear Helen, I'd just like to reinforce what Michael Harms was saying about reading the FBIRN papers. There are issues that will not be fully corrected by just including scanner regressors, especially if there are things that affect accurate registration (e.g. imperfectly corrected susceptibility distortions, gradient non-linearity distortions, concommitent gradient distortions) and these are likely to be different between scanner manufacturers _and_ between field strengths. These are not dealt with by voxel-wise regressors, as such regressors still assume that the registration is perfect. So be aware that there are still significant problems in combining this data together within a single analysis and that many reviewers are likely to be highly skeptical, especially if your subject groups are not completely balanced (in which case the list of potential problems is much worse). I would seriously consider just reporting the results from the individual scanner analyses side-by-side, as consistency in this does not have any of these problems and is likely to reinforce your point in a much stronger way. All the best, Mark On 18 Jan 2013, at 14:37, Helen Sawaya <[log in to unmask]> wrote: > Thank you Jesper, > > I have both patients and controls tested on both scanners. The number of participants across scanners differs, however. I guess I will try entering it as a covariate and see if I get anything meaningful. > > ________________________________________ > From: FSL - FMRIB's Software Library [[log in to unmask]] On Behalf Of Jesper Andersson [[log in to unmask]] > Sent: Friday, January 18, 2013 12:51 PM > To: [log in to unmask] > Subject: Re: [FSL] combine data from two different scanners and with different tesla > > Dear Helen, > >> I'm cutting in because I have the same question as Lorena. Given the problem with pooling datasets from different scanners, would it instead be possible to model this variable as a covariate? I'm not sure how this is done but as Lorena said could we add an EV in the glm design (for example coding scanner one as 0 and scanner two as 1). If we don't get a significant difference in results between the two datasets does that mean that controlling for the difference in scanner eliminated the differences in acquisition parameters and other potential differences? > > this will depend completely on how well you have managed to balance the two groups across the two scanners. If you scan all subject in group A on scanner 1 and all in group B on scanner 2 then your group-variable will be identical to your scanner-nuisance-variable and you will get only zeros. If on the other hand you manage to completely balance your groups across your scanners you will have no loss of power at all. > > Jesper > >> >> Thank you! >> Helen >> ________________________________________ >> From: FSL - FMRIB's Software Library [[log in to unmask]] On Behalf Of Michael Harms [[log in to unmask]] >> Sent: Friday, January 18, 2013 6:11 AM >> To: [log in to unmask] >> Subject: Re: [FSL] combine data from two different scanners and with different tesla >> >> Just to chime in, you might find it useful to read some of the papers from >> the FBIRN project, including those from G. Glover's group on multi-site >> differences in fMRI and analysis approaches to help ameliorate those >> differences. >> >> Note also that it is not trivial to truly match "acquisition parameters" >> across different scanner vendors. For example, differences in the timing >> order of the slices with interleaved acquisitions, the default phase >> encoding polarity, and the handling of bandwidth/echo spacing. And >> different vendors use different k-space filtering algorithms, which can >> make it difficult (or even impossible) to match the inherent spatial >> smoothness of the reconstructed images, even if you acquire at the same >> TR, TE, and voxel resolution. As for differences in field strength, the >> BOLD contrast is inherently field strength dependent, so "combining" BOLD >> data across field strengths is particularly problematic. >> >> cheers, >> -MH >> >> -- >> Michael Harms, Ph.D. >> >> ----------------------------------------------------------- >> Conte Center for the Neuroscience of Mental Disorders >> Washington University School of Medicine >> Department of Psychiatry, Box 8134 >> 660 South Euclid Ave. Tel: 314-747-6173 >> St. Louis, MO 63110 Email: [log in to unmask] >> >> >> >> >> On 1/17/13 9:53 PM, "Lorena Jimenez-Castro" <[log in to unmask]> wrote: >> >>> Hello Dr. Jenkinson, >>> >>> Thanks Chris Watson and Pablo Velasco for your response. I got a follow >>> up question for you all. >>> >>> Dr, Jenkinson Was your response taking into account that our groups are >>> going to be balance across scanners? Does your explanation applies for >>> scanners with different magnetic field strength or even for different >>> scanners from different manufactures but same tesla? >>> >>> Regarding the fMRI data, I thought, that if I the groups are balanced >>> across the two scanners and we use the same acquisition parameters in >>> both sites, We could model the difference of the scanners in the design >>> matrix (with one extra EV) and thus We would be able to analyze the >>> combined data from the two sites. On the other hand I thought that to >>> combine the DTI data could cause more problems in the analysis. >>> >>> So, since in our two data sets the groups (patients and controls) are >>> going to be balance across the two scanners. That is, the number of >>> patients and controls from one site are going to be the same number from >>> the other site. Beside, exactly same acquisition parameters are going >>> to be use in each site. All the above for the fMRI and for the DTI data. >>> >>> I just want to make sure that even so, it is not possible to combined the >>> data (neither DTI nor fMRI), because we are going to use different >>> scanners even if the scanners had the same magnetic field strength. Am I >>> understanding this rightt?? >>> >>> Thank you very very much. I greatly appreciate your help!! >>> >>> Lorena >>> >>> -- Lorena Jimenez-Castro, MD >>> Postdoctoral Fellow >>> Research Imaging Institute >>> University of Texas Health Science Center >>> 8403 Floyd Curl Drive >>> San Antonio, TX 78229 >>> (210) 567-8215- office phone >>> (210) 567-8103- fax >>> >>> >>> >>> >>> >>> >>> ________________________________________________ >>> Re: [FSL] combine data from two different scanners and with different >>> tesla >>> Thursday, January 17, 2013 6:38 PM >>> From: "Mark Jenkinson" <[log in to unmask]> >>> To:[log in to unmask] >>> >>> Dear Lorena, >>> >>> I agree with Chris Watson, who answered your original email. >>> It is very, very difficult to mix different scanners and different field >>> strengths and in general we would advise against it as there are likely >>> to be many non-biological factors that could result in changes that you >>> would detect. >>> >>> If you really want to combine things then I would do independent analyses >>> for the two different datasets and then combine the results, either >>> qualitatively or quantitatively, as having similar results on different >>> scanners is quite compelling. It also avoids the minefield of comparing >>> changes in the data that could be attributable to the scanners only, as >>> all your comparisons, to generate the individual statistical results, >>> would be done on within-scanner data. >>> >>> Sorry I don't have better news. >>> >>> All the best, >>> Mark