 |
 |
Hi FSL experts, Using dtifit: what method is better to calculate the tensor? The default ( that uses least squares on the log-transformed signal) or the weighted least squares option (—was) . 2016-09-22 21:30 GMT+02:00 Saad Jbabdi <[log in to unmask]>: > Hi - DTIFIT uses least squares on the log-transforned signal by default > (indeed using pinv). > The weighted least squares option (—wls) uses the approach suggested in > http://onlinelibrary.wiley.com/wol1/doi/10.1002/hbm.20076/full > No need for the user to specify “weights”. > > However the user can also optionally specify a confounds file. This has > been used, e.g., to diminish the effect of a once infamous artefact > http://www.ncbi.nlm.nih.gov/pubmed/19603408 > > > Cheers, > Saad > > > > > > > On 22 Sep 2016, at 16:21, Michael Knight <[log in to unmask] > <[log in to unmask]>> wrote: > > That unless you give dtifit a set of weights, I think it goes ahead with > ordinary least squares. > > type dtifit -h to get the help, or see the wiki page > > Here's a tutorial page (one of many) on an easy approach to diffusion > tensor fitting, easy to follow and implement in Matlab/Python etc. At the > final step, I'm pretty certain dtifit uses the Moore-Penrose psuedo-inverse > of the design matrix, not simply the inverse (inverting a design matrix is > not 100% sensible). > > http://www.diffusion-imaging.com/2014/04/from-diffusion- > weighted-images-to.html > > On 22 September 2016 at 15:52, Ana Maria Escorza < > [log in to unmask]> wrote: > >> what you mean with "But the weights are user-supplied if WLLS, default >> OLLS"? >> >> thanks >> >> cheers, >> >> Ana. >> >> 2016-09-22 16:44 GMT+02:00 Michael Knight <[log in to unmask]>: >> >>> OLLS or WLLS (by moore-penore psuedo-inverse I think). But the weights >>> are user-supplied if WLLS, default OLLS >>> >>> On 22 September 2016 at 15:41, Ana E. <[log in to unmask]> >>> wrote: >>> >>>> What diffusion tensor estimation does "dtifit" use? >>>> OLLS,WLLS,NLLS??? >>>> >>> >>> >>> >>> -- >>> Michael John Knight, PhD >>> Elizabeth Blackwell Institute Early Career Fellow >>> School of Experimental Psychology and School of Clinical Sciences >>> University of Bristol >>> Tel 07920 113060 >>> >>> >> > > > -- > Michael John Knight, PhD > Elizabeth Blackwell Institute Early Career Fellow > School of Experimental Psychology and School of Clinical Sciences > University of Bristol > Tel 07920 113060 > > >