Hi Eelke,
Thanks for your helpful suggestions about the input of prior probability maps in FAST segmentation. But I still have a confusion: the order of prior maps effects the segmentation result much.
For example:
"fast -a prior2input.mat -A prior_prob_0 prior_prob_1 prior_prob_2 [-P] <other options>" v.s. "fast -a prior2input.mat -A prior_prob_2 prior_prob_1 prior_prob_0 [-P] <other options>"
Both of the corresponding segmentation files use the same label with 1-CSF, 2-GM and 3-WM. But the segmentation results are different.
What order is better? When I process a group of data, the segmentation results are not consistent. The input order is" CSF GM WM".
Thanks,
Zheng
-----Original Message-----
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Zheng Liu
Sent: Wednesday, September 30, 2015 8:57 AM
To: [log in to unmask]
Subject: Re: [FSL] fast segmentation
Hi Eelke,
Your understanding is exact right. Yes, this is what I am trying to do. I hope FAST do not do any registration and only use these input maps as they are.
Thanks very much for your explains for the -P option.
Zheng
-----Original Message-----
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Eelke Visser
Sent: Wednesday, September 30, 2015 8:49 AM
To: [log in to unmask]
Subject: Re: [FSL] fast segmentation
Hi Zheng,
I am not sure that I understand your problem completely. If you have already used nonlinear registration to warp the probability maps to your input image, specifiying an identity matrix essentially tells FAST to not apply any other registration and keep the maps as they are. That sounds fine to me - is this what you are trying to do?
I think the text in the manual about the -P option is not completely accurate. The option doesn’t take any arguments as it is just a flag that controls whether the priors specified with -a/-A are used in the final segmentation as well.
Hope that helps,
Eelke
> On 30 Sep 2015, at 16:30, Zheng Liu <[log in to unmask]> wrote:
>
> Hi Eekle,
>
> Thanks for your helpful reply.
> We generated the probability maps from atlas by non-linear registration and it is hard to form the transform matrix as specified by the default ident.mat.
> In the manual, the sentence "-P <file> or --Prior use prior probability maps at all segmentation stages" seems that the prior probability maps is not only used for initialisation.
> My idea is that can we set a weighting term to weight the prior maps in the whole segmentation procedure?
>
> Thanks,
> Zheng
>
> -----Original Message-----
> From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Eelke Visser
> Sent: Wednesday, September 30, 2015 3:33 AM
> To: [log in to unmask]
> Subject: Re: [FSL] fast segmentation
>
> Hi Zheng,
>
> To use prior images you need to specify both the -a and -A options:
>
> fast -a prior2input.mat -A prior_prob_0 prior_prob_1 prior_prob_2 [-P] <other options>
>
> The -a option specifies the affine transformation (from FLIRT) that registers the priors to the input image. If you have already registered them, you can specify $FSLDIR/etc/flirtsch/ident.mat for the matrix. The optional -P option tells fast to use the priors in the final segmentation as well; by default it only uses them for initialisation (see the manual for some more details).
>
> You can specify the prior images in any order - the order of the outputs will correspond to the input order.
>
> Cheers,
> Eelke
>
>
>> On 29 Sep 2015, at 22:50, Zheng Liu <[log in to unmask]> wrote:
>>
>> Hello:
>>
>>
>>
>> How to input the prior probability maps in fast command?
>>
>> For example:
>>
>> fast -t 1 -n 3 -H 0 -I 4 -l 20.0 -g -p -o output-withprior input.nii.gz -P prior_prob_0.nii.gz prior_prob_1.nii.gz prior_prob_2.nii.gz
>>
>> that’s correct? And the order of probability maps is CSF, GM and WM?
>>
>>
>>
>> Thanks,
>>
>> Zheng
>>
>
|