SimpleITK/Tutorials/MICCAI2015

SimpleITK Registration: An Interactive, Python-Based Introduction to Registration with the Insight Toolkit (ITK)

Who Should Attend

• Do you want your students to gain practical experience with registration while minimizing their programming load?
• Do you want to easily experiment with various ITK registration configurations, or optimize the settings of a specific registration configuration?

If you answered yes to either of these questions, then this tutorial is for you.

Goals

The goal of this half-day tutorial is to introduce students and researchers to SimpleITK’s interface for the ITK version 4 registration framework. SimpleITK is, as the name suggests, a simpler interface to ITK. It provides a procedural interface and bindings to several interpreted languages, facilitating fast experimentation with ITK algorithms. In this tutorial we will use the Python programming language and the IPython Notebook interactive environment to explore the various features of the ITKv4 registration framework. Key features presented include: uniform treatment of linear, deformable and composite transformations, embedded multi-resolution registration and self calibrating optimizers. Using a hands on approach, participants will experiment with various registration tasks, learning how to use SimpleITK in order to gain insight into the effects of registration component selection and parameter settings on accuracy and running time of ITK based registration algorithms.

Tentative Program

• 8:30am: Setup and introduction.
• 9:00am: SimpleITK basics: loading data, image access, image transformations, image resampling, basic filters.
• 10:00am: Coffee break.
• 10:30am: Registration 1: composite transform, transformation initialization, embedded multi-resolution, scale parameter estimation, optimization termination criteria.
• 11:30am: Registration 2: nonrigid registration - Nonrigid registration, Bspline and displacement field transformations.
• 12:30pm: Lunch.

SimpleITK Registration, It's Really Simple

The following script illustrates the use of SimpleITK to perform rigid registration between a CT and MR (registration specific code is highlighted). By the end of the tutorial you will be familiar with the various components that are available as part of the SimpleITK registration framework, easily modifying this example to suite your specific needs.

<source lang="python" highlight="19-22, 28-42"> import SimpleITK as sitk

def save_combined_central_slice(fixed, moving, transform, file_name_prefix):

   
   Transform the moving image using the given transform and create an alpha
   blended slice from the fixed and moving volumes' central z slices. The
   result is saved to a jpg file with the given prefix.
   
   central_slice = fixed.GetSize()[2]/2
   alpha = 0.7
   moving_transformed = sitk.Resample(moving, fixed, transform, 
                                      sitk.sitkLinear, 0.0, 
                                      moving_image.GetPixelIDValue())
   combined_image = (1.0 - alpha)*fixed[:,:,central_slice] + \
                    alpha*moving_transformed[:,:,central_slice]
   sitk.WriteImage(sitk.Cast(sitk.RescaleIntensity(combined_image), sitk.sitkUInt8), 
                   file_name_prefix + '.jpg') 

1. read the images

fixed_image = sitk.ReadImage('training_001_ct.mha', sitk.sitkFloat32) moving_image = sitk.ReadImage('training_001_mr_T1.mha', sitk.sitkFloat32)

1. initial alignment of the two volumes

initial_transform = sitk.CenteredTransformInitializer(fixed_image,

                                                     moving_image, 
                                                     sitk.Euler3DTransform(), 
                                                     sitk.CenteredTransformInitializerFilter.GEOMETRY)

1. save alpha-blended central slice

save_combined_central_slice(fixed_image, moving_image,

                           initial_transform, 'initialAlignment')

1. multi-resolution rigid registration using Mutual Information

registration_method = sitk.ImageRegistrationMethod() registration_method.SetMetricAsMattesMutualInformation(numberOfHistogramBins=50) registration_method.SetMetricSamplingStrategy(registration_method.RANDOM) registration_method.SetMetricSamplingPercentage(0.01) registration_method.SetInterpolator(sitk.sitkLinear) registration_method.SetOptimizerAsGradientDescent(learningRate=1.0,

                                                 numberOfIterations=100, 
                                                 convergenceMinimumValue=1e-6, 
                                                 convergenceWindowSize=10)

registration_method.SetOptimizerScalesFromPhysicalShift() registration_method.SetShrinkFactorsPerLevel(shrinkFactors = [4,2,1]) registration_method.SetSmoothingSigmasPerLevel(smoothingSigmas=[2,1,0]) registration_method.SmoothingSigmasAreSpecifiedInPhysicalUnitsOn() registration_method.SetInitialTransform(initial_transform, inPlace=False) final_transform = registration_method.Execute(fixed_image, moving_image)

sitk.WriteTransform(final_transform, 'ct2mrT1.tfm')

1. save alpha-blended central slice

save_combined_central_slice(fixed_image, moving_image,

                           final_transform, 'finalAlignment')

</source>

To run this example you will need to download the CT and MR data. These are part of the training data provided by the Retrospective Image Registration Evaluation Project.