[Insight-users] Re: Mutual Information and Image Registration

[Luis Ibanez]

Hi Carole,

You can do it by integrating two metric classes into a single one.

One easily option is to create a new ImageMetric that internally
instantiates one of the ITK MutualInformation metrics, and also
computes your own metric, then make a linear combination of both
in order to return that in the GetValue() method.

This will require that you look at the internal code of the metrics.

This will be much simpler if you only implement GetValue() and
ignore GetDerivatives() and GetValueAndDerivatives().  The resulting
metric will be usable by optimizers that do not require derivatives,
such as the OnePlusOneEvolutionary algorithm.

Your metric may take the gradient magnitude of both Fixed and Moving
images and compute a Mean Squared difference on them, for example.
Then combine this value with the one returned from Mutual Information.
Note that you will have to compensate for the relative dynamic range
of each metric in the combination.


   Please let us know if you have further questions,


     Thanks


        Luis


-------------
张 卡璐 wrote:
> Hi Luis
> 
>   Thank you very much for your advice.According to your advice, I red 
> some papers on mutual information. Mutual information ignored spatial 
> information of images.
>   But how can i incorporate spatial information?? using gradient????   I 
> am puzzled, please help me!
> 
>   Thanks,
> 
>   carole
>