Dear Stuart,
Glad to hear the cerebellum segmentation is working better.
It is difficult to accurately quantify grey and white matter in the cerebellum
with the kind of resolution that MRI is capable of in-vivo at present. The
best you could do would be to run FAST tissue segmentation and take
the partial volume estimates for grey and white matter and sum these up
within the cerebellum mask. This should be a reasonable estimate of
the tissue type, but it is unlikely to be super accurate.
All the best,
Mark
On 19 Jul 2011, at 22:25, Stuart Currie wrote:
> Dear Mark,
>
> Thanks for your help on the cerebellar volume problem. I tried it on the 36 subjects over the w/e and it was a vast improvement.
>
> I wonder if you could help me with another query? Is it possible to obtain grey and white matter volumes for the cerebellum on FSL? i.e. if we know the cerebellum shows atrophy is it grey or white matter that is predominantly affected?
>
> Thanks Mark. Your help is greatly appreciated.
>
> Best wishes
>
> Stuart
>
> On 14 July 2011 12:49, Stuart Currie <[log in to unmask]> wrote:
> Thanks Mark,
>
> I'll try that and let you know.
>
> Best wishes
>
> Stuart
>
> On 13 July 2011 23:05, Mark Jenkinson <[log in to unmask]> wrote:
> Dear Stuart,
>
> We (Brian Patenaude and I) have had a look at this and you get much better
> results in this case if you use *many* more degrees of freedom. So when I
> tried the following:
>
> first_flirt $filename ${filename}_to_std_sub -cort
> run_first -i $filename -t ${filename}_to_std_sub_cort.mat -n 320 -o ${filename}_L_Cereb -m ${FSLDIR}/data/first/models_336_bin/intref_puta/L_Cereb.bmv -intref ${FSLDIR}/data/first/models_336_bin/05mm/L_Puta_05mm.bmv
>
> I got a reasonable, though not perfect, results. Much better than with the
> smaller -n though.
>
> I also suspect that the cerebellum is the structure which is the most sensitive to
> resolution, since our training data was at best 1mm and often a bit worse on
> average, then the detail that you can see with your 0.8mm will be considerably
> better and this might have a slightly detrimental effect on the results. This is
> only a hypothesis that I haven't carefully tested, but it may be the case that
> the cerebellum needs more work.
>
> Anyway, try the higher number of modes (and make sure you use the exact
> options as above, since the commands you sent in the previous emails were
> different, possibly just because of typos) and see if the results are acceptable.
>
> All the best,
> Mark
>
>
>
>
>
> On 13 Jul 2011, at 10:37, Stuart Currie wrote:
>
> > Dear Mark,
> >
> > Thanks for taking a look.
> >
> > The zip file 'RC' has been given the ref no. 436669.
> >
> > There are 36 subjects (some with more than 1 MR brain hence co16, co16b for example).
> >
> > Originally I ran the programme as a job lot using the following:
> >
> > Registration:
> >
> >
> >
> > for filename in co01 co02 co03 co04 co05 co07 co08 co09 co09b co10 co11 co11b co12 co13b co14 co14b co15 co16 co16b co17 co17b co18 co19 co20 co21 co21b co22 co23 co24 co25 co25b co25c co26 co27 co28 co29b co30 co31 co32 co32b co33 co34 co35 co36 ; do first_flirt $filename ${filename}_to_std_sub –cort ; done
> >
> >
> > Left Cerebellar Segmentation (I found -n 33 gave the best fit):
> >
> >
> >
> > for filename in co01 co02 co03 co04 co05 co07 co08 co09 co09b co10 co11 co11b co12 co13b co14 co14b co15 co16 co16b co17 co17b co18 co19 co20 co21 co21b co22 co23 co24 co25 co25b co25c co26 co27 co28 co29b co30 co31 co32 co32b co33 co34 co35 co36 ; do run_first –i $filename –t ${filename}_to_std_sub.mat –n 33 –o ${filename}_L_Cereb –m ${FSLDIR}/data/first/models_336_bin/intref_puta/L_Cereb.bmv –intref ${FSLDIR}/data/first/models_336_bin/05mm/L_Cereb_05mm.bmv ; done
> >
> >
> > Right C Segmentation (found -n 31 gave the best fit):
> >
> >
> >
> > for filename in co01 co02 co03 co04 co05 co07 co08 co09 co09b co10 co11 co11b co12 co13b co14 co14b co15 co16 co16b co17 co17b co18 co19 co20 co21 co21b co22 co23 co24 co25 co25b co25c co26 co27 co28 co29b co30 co31 co32 co32b co33 co34 co35 co36 ; do run_first –i $filename –t ${filename}_to_std_sub –n 31 –o {filename}_R_Cereb –m ${FSLDIR}/data/first/models_336_bin/intref_puta/R_Cereb.bmv –intref ${FSLDIR}/data/first/models_336_bin/05mm/R_Cereb_05mm.bmv ;
> >
> > done
> >
> >
> > Export to excel:
> >
> >
> >
> > for filename in co01 co02 co03 co04 co05 co07 co08 co09 co09b co10 co11 co11b co12 co13b co14 co14b co15 co16 co16b co17 co17b co18 co19 co20 co21 co21b co22 co23 co24 co25 co25b co25c co26 co27 co28 co29b co30 co31 co32 co32b co33 co34 co35 co36 ; do LCereb=`fslstats ${filename}_L_Cereb -v | awk '{print $2}'`; RCereb=`fslstats ${filename}_R_Cereb -v | awk '{print $2}'`; echo "${filename},${LCereb},${RCereb}" >> Cerebellum1.csv ; done
> >
> >
> > The last command gave huge values for the cerebellum, some twice as big as expected. I then ran the command adding intensity values instead (in bold below) which gave more realistic volumes but still too small (which I think is due to the original segmentation missing some peripheral cerebellar tissue):
> >
> >
> >
> > for filename in co01 co02 co03 co04 co05 co07 co08 co09 co09b co10 co11 co11b co12 co13b co14 co14b co15 co16 co16b co17 co17b co18 co19 co20 co21 co21b co22 co23 co24 co25 co25b co25c co26 co27 co28 co29b co30 co31 co32 co32b co33 co34 co35 co36 ; do LCereb=`fslstats ${filename}_L_Cereb -l 6.5 -u 7.5 -v | awk '{print $2}'`; RCereb=`fslstats ${filename}_R_Cereb -l 46.5 -u 47.5 -v | awk '{print $2}'`; echo "${filename},${LCereb},${RCereb}" >> Cerebellum1.csv ; done
> >
> >
> >
> > Thanks again for taking a look.
> >
> >
> >
> > Best Wishes
> >
> >
> >
> > Stuart
> >
> >
> >
> > On 12 July 2011 17:38, Mark Jenkinson <[log in to unmask]> wrote:
> > Dear Stuart,
> >
> > This looks correct.
> > Possibly it is something unusual with your data or some
> > difference with the manual outlining protocol used by the
> > CMA to train with versus what you'd prefer to see.
> >
> > I can't tell without seeing the data, so can you please upload
> > it to:
> > http://www.fmrib.ox.ac.uk/cgi-bin/upload.cgi
> > and send me the reference number.
> >
> > All the best,
> > Mark
> >
> >
> > On 12 Jul 2011, at 17:33, Stuart Currie wrote:
> >
> > > Hi Mark,
> > >
> > > Thanks for replying.
> > >
> > > Using first_flirt and after initial registration:
> > >
> > > First_flirt input image subject1_to_std_sub_cort
> > >
> > >
> > > I'm trying to calculate cerebellar volume by calculating right and left hemispheres:
> > >
> > > e.g. for the left hemisphere:
> > >
> > > run_first -i file name -t subject1_to_std_sub.mat -n 33 -o cerebellum -m ${FSLDIR}/data/first/models_336_bin/intref_puta/L_Cereb.bmv -intref ${FSLDIR}/data/first/models_336_bin/05mm/L_Puta_05mm.bmv
> > >
> > >
> > >
> > > when I view this over the original nii.gz images on fslview it is clear that the cerebellar model is missing some of the peripheral tissue.
> > >
> > >
> > >
> > > I have tried changing the -n (number of modes of variation) up and down from 33 but 33 seems to be the best fit. Similarly 31 seems to work best for the right hemisphere. Unfortunately when I finally receive the volume data the total cerebellar volumes are ~ 50 cm3, no where near the ~ 110 cm3 expected.
> > >
> > >
> > >
> > > Is there a way of modifying the cerebellar model further to gain a better fit?
> > >
> > >
> > >
> > > Cheers
> > >
> > >
> > >
> > > Stuart
> > >
> > >
> > > On 12 July 2011 16:57, Mark Jenkinson <[log in to unmask]> wrote:
> > > Can you be a bit more specific?
> > > All the best,
> > > Mark
> > >
> > > On 12 Jul 2011, at 16:39, Stuart Currie wrote:
> > >
> > > > Trying to run the cerebellar volume command but seems to be underestimating volumes despite changing the -n value to attempt best fit. Any suggestions?
> > > >
> > >
> > >
> > >
> > > --
> > >
> > > Dr Stuart Currie
> > > BSc, MB ChB, MRCS, FRCR
> > > Clinical Research Fellow/Honorary Neuroradiology Fellow
> > > Academic Unit of Radiology
> > > University of Sheffield
> > > C Floor
> > > Royal Hallamshire Hospital
> > > Sheffield
> > > S10 2JF
> > >
> >
> >
> >
> > --
> >
> > Dr Stuart Currie
> > BSc, MB ChB, MRCS, FRCR
> > Clinical Research Fellow/Honorary Neuroradiology Fellow
> > Academic Unit of Radiology
> > University of Sheffield
> > C Floor
> > Royal Hallamshire Hospital
> > Sheffield
> > S10 2JF
> >
>
>
>
> --
>
> Dr Stuart Currie
> BSc, MB ChB, MRCS, FRCR
> Clinical Research Fellow/Honorary Neuroradiology Fellow
> Academic Unit of Radiology
> University of Sheffield
> C Floor
> Royal Hallamshire Hospital
> Sheffield
> S10 2JF
>
>
>
>
> --
>
> Dr Stuart Currie
> BSc, MB ChB, MRCS, FRCR
> Clinical Research Fellow/Honorary Neuroradiology Fellow
> Academic Unit of Radiology
> University of Sheffield
> C Floor
> Royal Hallamshire Hospital
> Sheffield
> S10 2JF
>
|
