Registration:Categories

Back to registration portal page

Slicer Registration Case Library: Examples & Tutorials

looking for a set of features: 3DSlicer Registration Feature/Decision Matrix

• the matrix below shows which modules support which feature, intended to help you if looking for a method with a particular combination of features.
• if you would like to see additional criteria or a new module is not yet listed, please let us know: mailto:slicer-users at bwh.harvard.edu

looking for Speed

• Manual/interactive alignment can be done via the Transforms module, e.g. for initial alignment. See here for a tutorial and example dataset on Manual Registration
• Affine: The Fast Affine Registration module performs automated affine registration. This is being replaced by the Register Images Module that performs the same function.
• Bspline: The Fast Nonrigid BSpline Registration module performs non-rigid automated image registration. With default settings results within <1 minute are common.

looking for Robustness and/or Precision

• The Expert Automated Registration module performs automated image registration, rigid to affine, based on image intensity similarities. It allows to focus the registration on a region of interest
• The Robust Multiresolution Affine Registration module performs robust automated affine image registration employing a multi-resolution scheme.
• The Plastimatch module performs automated registration of images from rigid to affine to non-rigid. As a unique feature it provides non-rigid deformation from fiducials, which can be used to "edit/repair" a registration.

choose by DOF

• rigid 6 DOF:
  • Expert Automated Registration. Set DOF to PipelineRigid.
• similarity 9 DOF:
  • The BAINSfit module includes similarity transform; type ScaleVersor3D in the transform type field; provides masking support.
• affine 12 DOF:
  • Fast Affine Registration
  • The Robust Multiresolution Affine Registration module performs robust affine (12 DOF) image registration employing a multi-resolution scheme.
  • Expert Automated Registration. Set DOF to PipelineAffine.

• non rigid 27- 100s DOF:
  • The Fast Nonrigid BSpline Registration module performs non-rigid automated image registration.
  • The BAINSfit module includes a registration based on a Bspline transform. Initially designed for but not limited to brain images. Also includes many options such as masking support.
• non rigid (fluid) >100 DOF
  • The BRAINSDemonWarp module performs automated registration of brain MRI based on an optic flow mechanism. Deformations here are significantly more "fluid" (i.e. have more DOF and are less constrained) than for the nonrigid BSpline method.
  • The Plastimatch module performs automated registration of images from rigid to affine to non-rigid. As a unique feature it provides non-rigid deformation from fiducials, which can be used to "edit/repair" a registration.
  • The HAMMER Module performs elastic (non-rigid) alignment of brain images of different individuals based on tissue class segmentation and intensity (experimental stage).

choose byDatatype

• images, same modality: The Expert Automated Registration module performs automated image registration, rigid to affine, based on image intensity similarities. It allows to focus the registration on a region of interest
• images, different modality: The Expert Automated Registration module performs automated image registration, rigid to affine, based on image intensity similarities. Select Mutual Information as cost function.
• surfaces: The Surface Registration module performs automated registration of surfaces (not images). This is useful if image data directly is unreliable, but surfaces can be produced from segmentations that provide good information about desired alignment.
• fiducials: The Fiducial Alignment module can align images based on pairs of manually selected fiducial points (rigid and affine). Two sets of fiducials (fiducial lists) are required, forming matching pairs to be aligned. See Fiducials module below.

looking for Brain

• The AC-PC Transform module is used to orient brain images along predefined anatomical landmarks: (manually defined) fiducials for the inter-hemispheral midline, anterior- and posterior commissure are used to align an image such that these landmarks become vertical and horizontal, respectively.
  • The BAINSfit module includes a registration based on a Bspline transform. Initially designed for but not limited to brain images. Also includes many options such as masking support.
• non rigid (fluid) >100 DOF
  • The BRAINSDemonWarp module performs automated registration of brain MRI based on an optic flow mechanism. Deformations here are significantly more "fluid" (i.e. have more DOF and are less constrained) than for the nonrigid BSpline method.
  • The HAMMER Module performs elastic (non-rigid) alignment of brain images of different individuals based on tissue class segmentation and intensity (experimental stage).

looking for Tools for Preparing Data for Registration

• Intensity Normalization & Filtering

Intensity corrections are often the first processing step of choice. MRI bias field inhomogeneities can adversely affect registration accuracy and stability, as can large differences in the intensity ranges between the two images, or large amounts of noise

• Bias Field Correction to remove intensity drift across the image (e.g from variable coil sensitivity)
• Histogram Equalization for matching intensity ranges
• Median Filtering to remove speckle noise without smoothing edges
• Gaussian Blur filter to remove noise and smooth edges
• other filters see here

• Masking

Masks are an important component of many registration tasks. They allow to focus the algorithm on the region of interest (ROI) that is to be registered and prevent it from being distracted by image content outside this ROI.

• Skull Stripping extension module automatically builds a mask of the brain from an input MRI image (T1w is best). This is an extension module and needs to be installed via the Extension manager.
• Thresholding within the Interactive Editor or via the Threshold module.
• Otsu Threshold Segmentation for automated segmentation of main image object from background
• Fast Marching segmentation for interactive fiducial-based segmentation of small local regions.
• Mask image module to apply a mask and remove image content outside (if mask is not directly supported by the chosen registration method)
• Crop/SubvolumeExtraction module to extract a box region from the image.
• ROI module to quickly define a box region of interest that is supported as mask by Robust Multires Registration, for example.