rotational minctracc takes too long

[bcdarwin]

e.g., more than 12 hours on some 40um files ... I suspect something slightly more clever is possible.

We could also allow Pydpiper users to set the number of seeds somehow, e.g., through a config file (a bit less annoying than more Pydpiper flags and a bit more reproducible than $ENV_VARS).

[gdevenyi]

Alternatively:

$ antsAI 

COMMAND: 
     antsAI
          Program to calculate the optimallinear transform parameters for aligning two 
          images. 

OPTIONS: 
     -d, --dimensionality 2/3
          This option forces the image to be treated as a specified-dimensional image. If 
          not specified, we try to infer the dimensionality from the input image. 

     -m, --metric MI[fixedImage,movingImage,<numberOfBins=32>,<samplingStrategy={None,Regular,Random}>,<samplingPercentage=[0,1]>]
                  Mattes[fixedImage,movingImage,<numberOfBins=32>,<samplingStrategy={None,Regular,Random}>,<samplingPercentage=[0,1]>]
                  GC[fixedImage,movingImage,<radius=NA>,<samplingStrategy={None,Regular,Random}>,<samplingPercentage=[0,1]>]
          These image metrics are available: MI: joint histogram and Mattes: mutual 
          information and GC: global correlation. 

     -t, --transform Rigid[gradientStep]
                     Affine[gradientStep]
                     Similarity[gradientStep]
          Several transform options are available. The gradientStep or learningRate 
          characterizes the gradient descent optimization and is scaled appropriately for 
          each transform using the shift scales estimator. 

     -p, --align-principal-axes 
          Boolean indicating alignment by principal axes. Alternatively, one can align 
          using blobs (see -b option). 

     -b, --align-blobs numberOfBlobsToExtract
                       [numberOfBlobsToExtract,<numberOfBlobsToMatch=numberOfBlobsToExtract>]
          Boolean indicating alignment by a set of blobs. Alternatively, one can align 
          using blobs (see -p option). 

     -s, --search-factor searchFactor
                         [searchFactor,<arcFraction=1.0>]
          Incremental search factor (in degrees) which will sample the arc fraction around 
          the principal axis or default axis. 

     -c, --convergence numberOfIterations
                       [numberOfIterations,<convergenceThreshold=1e-6>,<convergenceWindowSize=10>]
          Number of iterations. 

     -x, --masks fixedImageMask
                 [fixedImageMask,movingImageMask]
          Image masks to limit voxels considered by the metric. 

     -o, --output outputFileName
          Specify the output transform (output format an ITK .mat file). 

     -v, --verbose (0)/1
          Verbose output. 

     -h 
          Print the help menu (short version). 

     --help 
          Print the help menu. 

[bcdarwin]

Do you have any parameter settings that make this work? We can get a principal axis transformation from it, but if you specify some nonzero --convergence it gives (or not; unsure why) many errors of form

Using the global correlation metric 
WARNING: In /axiom2/projects/software/arch/linux-precise/src/minc-toolkit-v2-1.9.11/minc-toolkit-v2/build/ITKv4/Modules/Registration/Metricsv4/include/itkCorrelationImageToImageMetricv4HelperThreader.hxx, line 85
CorrelationImageToImageMetricv4HelperThreader (0x1f317a0): collected only zero points

WARNING: In /axiom2/projects/software/arch/linux-precise/src/minc-toolkit-v2-1.9.11/minc-toolkit-v2/build/ITKv4/Modules/Numerics/Optimizersv4/include/itkObjectToObjectMetric.hxx, line 529
CorrelationImageToImageMetricv4 (0x1f05e70): No valid points were found during metric evaluation. For image metrics, verify that the images overlap appropriately. For instance, you can align the image centers by translation. For point-set metrics, verify that the fixed points, once transformed into the virtual domain space, actually lie within the virtual domain.

[gdevenyi]

Can you share your existing command line call? (And maybe the file?)

[gdevenyi]

Following up here, any progress on antsAI? I'm trying it for some work and having a tough time..

[mcvaneede]

I've had no luck with antsAI... Ben?

[cfhammill]

A couple thoughts about this - we could off early termination if rotational_minctracc finds a rotations that puts the cost function in a user-defined "acceptable" range. Alternatively we could try to train a neural net to predict approximate rotation. We could do this on down-sampled volumes for speed.

[gdevenyi]

Following up on antsAI, here's a working implementation from ANTs
https://github.com/ANTsX/ANTs/blob/master/Scripts/antsBrainExtraction.sh#L464-L471

[cfhammill]

Have you had good luck with antsAI now Gabe?

[gdevenyi]

I have not tried it recently, however it is standard in their pipeline, which lots of people use, I think