Brian, OK. Does one specify the model with the -m flag in addition to the -intref option? In essence, what is correct the command line format when using -intref? What do you mean by "leave-one-out cross-validation " Thx...........J ----- Original Message ----- From: "Brian Patenaude" <[log in to unmask]> To: <[log in to unmask]> Sent: Tuesday, July 08, 2008 8:13 PM Subject: Re: [FSL] FIRST Hippocampus Segmentation > Hi, > > There are 2 different models for the hippocampus, caudate, and amygdala. > The shape model portions are identical, the differences are in the > intensity model, more precisely, in its normalization. The models that > are located in ${FSLDIR}/data/first/models_317_bin/ are all normalized by > subtracting the mode of the intensity distribution within that structure. > When fitting the model, this normalization constant is estimated from the > data. > > The problem with the 3 structures mentioned above is that in the case of > severe atrophy, the initial structure estimate may encompass a > significant amount of CSF, thereby confounding the normalization constant > estimate. To get around this, models have been created that are > normalized by the mode of the intensity distribution of the thalamus > (which is more robustly estimated). > > Based on leave-one-out cross-validation using the intensity reference > models perform, on average, slightly better. > The general recommendation for the hippocampus is to use the intensity > reference model (in ${FSLDIR}/data/first/models_317_bin/ intref_thal ) > with 30 modes of variation (this is what is done in run_first_all). > > For the example image you provided, it is seems to do better in capturing > the lateral portion of the body of the hippocampus (though a few voxels > may still be missing). Similar results can be obtain for the standard > model if you increase the number of modes to 60 or a bit above. > > Hope this helps and clarifies things. > > cheers, > > Brian > > > > > > > > On 7 Jul 2008, at 14:10, Jay Ives wrote: > >> Hi Brian, >> >> Thanks for your interest. The files are ROA.hdr and ROA.img. The upload >> session reference number is 994681. >> I converted the DICOM image set to ANALYZE format with MRIConvert so as >> to use MRIcro to clip and convert to axial format to use in FIRST. >> I find the lateral aspect of the left hippocampus body is not completely >> within the model. I don't think there is anything else there such as >> choroid plexus. >> I look forward to your comments. >> >> Cheers................J >> >> ----- Original Message ----- From: "Brian Patenaude" >> <[log in to unmask]> >> To: <[log in to unmask]> >> Sent: Monday, July 07, 2008 5:59 PM >> Subject: Re: [FSL] FIRST Hippocampus Segmentation >> >> >>> Hi, >>> >>> Could you upload an example image at http://www.fmrib.ox.ac.uk/ cgi- >>> bin/upload.cgi ? >>> >>> Cheers, >>> >>> Brian >>> >>> >>> >>> >>> On 6 Jul 2008, at 13:59, Jay Ives wrote: >>> >>>> The lateral aspect of the hippocampus body is occasionally not >>>> included >>>> within the surface. I've tried 30 and 40 modes and different p values >>>> in >>>> these cases, but without effect. Is there anything else that can be >>>> tried? >>>> Thx....J >>>> >> >