[MINC-users] running classify unsupervised

[D. Louis Collins]

Anders,

I concur with Andrew.  The -mm, -knn, -ann, -hcm, -fcm and -bayes all 
need estimates of the intensities to build the internal structure 
required to do the segmentation.  So, they all need training data. The 
training data can come from a training file (containing the intensity 
info like mean, sd) or a tag file (then the program extracts the 
mean/sd from the volumes.

We do not have an EM technique coded up in classify.  However, there is 
a bit of EM code written by Sylvain Prima.  I think that it is in the 
CVS repository.

-Louis


On May 4, 2006, at 2:35 AM, Andrew Janke wrote:

> Hi Anders,
>
> As far as I am aware classify doesn't do unsupervised classification.
> In all cases it will train from an input set of tags/classes. I
> suspect by unsupervised this is merely describing the method of
> clustering that is used from data generated from the input tag points.
>
> I may be wrong and you can do real unsupervised learning with
> classify, so I shall watch with interest to see if anyone corrects me!
> :)
>
>
> a
>
>
> On 5/3/06, Anders Bertil Rodell <rodell at pet.auh.dk> wrote:
>> Does anyone have some examples of how to use the unsupervised
>> -fcm,-hcm, -apriory and -fuzzy options in classify?
>> I have tried a multitude of combinations of the options and the only 
>> way
>> I can get fuzzy results seems to be through first creating a training
>> file using a tagfile.. this seems a bit supervised to me
>>
>> Thanks for any help.
>> Anders
>>
>> --
>> Anders Bertil Rodell <rodell at pet.auh.dk>
>> PET Centre Aarhus General Hospital
>>
>> _______________________________________________
>> MINC-users at bic.mni.mcgill.ca
>> http://www.bic.mni.mcgill.ca/mailman/listinfo/minc-users
>>
>
>
> --
> Andrew Janke   (a.janke at gmail.com || http://a.janke.googlepages.com/)
> Canada->Montreal                                   Cell: +1 (514) 924 
> 2012
>
> _______________________________________________
> MINC-users at bic.mni.mcgill.ca
> http://www.bic.mni.mcgill.ca/mailman/listinfo/minc-users
>