[ITK-users] [ITK] ITK 3D Registration parameters

[Matt McCormick]

Hi Denis,

Choosing the optimizer, metric, and interpolator is generally an open
problem that depends on the input data.  Preprocessing can also be
important.  For example, if a normalized correlation metric performs
slightly better than mean squares metric, histogram equalization or
bias correction may improve the mean squares result.

Plotting the progression of the optimizer and the shape of the metric
space can be helpful for choosing the best optimizer and the
hyperparameters of the optimizer.

A LinearInterpolateImageFunction is usually sufficient for most
desired registration precisions.

A transform such as the versor transform is usually performed over the
AffineTransform because is behaves better during optimization.

ITK-based tools such as ANTS [1] and elastix [2] can also be helpful
to quickly evaluate different combinations of optimizers, metrics, and
interpolators.

Hope this helps,
Matt

[1] http://stnava.github.io/ANTs/

[2] http://elastix.isi.uu.nl/

On Fri, Apr 25, 2014 at 8:18 AM, Dennis Zobnin <dennzo1993 at gmail.com> wrote:
> Hi everyone,
>
> I’m trying registration in ITK of 3D ANALYZE brain images and seem to get
> lost in various registration components. As I’ve read in some papers, each
> component should be chosen taking into account the features of my particular
> task. But still it is very hard for me to choose the right ones.
>
> Now I’m using RegularStepGradientDescentOptimizer with
> MeanSquaresImageToImageMetric and LinearInterpolateImageFunction
> (NearestNeighbor interpolator seems to produce worse result, BSpline
> interpolator makes the program too slow). As for the metric, I’ve also tried
> NormalizedCorrelation metric and it produces almost the same result, but
> sometimes is slower. Current transform is AffineTransform. I’m sure my
> images are correct (downloaded from IBSR). But I’m not satisfied with the
> results.
>
> Could you please explain how to choose an optimizer, metric and
> interpolator?
>
> Thanks,
> Denis
>
> _____________________________________
> Powered by www.kitware.com
>
> Visit other Kitware open-source projects at
> http://www.kitware.com/opensource/opensource.html
>
> Kitware offers ITK Training Courses, for more information visit:
> http://www.kitware.com/products/protraining.php
>
> Please keep messages on-topic and check the ITK FAQ at:
> http://www.itk.org/Wiki/ITK_FAQ
>
> Follow this link to subscribe/unsubscribe:
> http://www.itk.org/mailman/listinfo/insight-users
>
> _______________________________________________
> Community mailing list
> Community at itk.org
> http://public.kitware.com/cgi-bin/mailman/listinfo/community
>