NA-MIC National Alliance for Medical Image Computing Registering Image Volumes in Slicer Steve Pieper
National Alliance for Medical Image Computing Requirements The tools described here are part of Slicer 2.5
National Alliance for Medical Image Computing Data to Use Get the two datasets: These are the same subject, taken on two different scanners –1.5 Tesla diagnostic MR scanner (low resolution for better interactivity) –0.5 Tesla intraoperative MR scanner
National Alliance for Medical Image Computing Load First Data Volume 1: Click Add Volume 2: Pick Nrrd 3: Browse to reg.nhdr 4: Click Apply
National Alliance for Medical Image Computing Load the Second Volume 1: Click Add Volume 4: Click Apply 3: Browse to I.001 in mrt images dir 2: Use Basic Reader
National Alliance for Medical Image Computing View Initial (Mis)Registration Put One Volume in Background with Bg and One in Foreground with Fg Use Fade Slider and Toggle to Compare
National Alliance for Medical Image Computing Add a Transformation 1: Select Volume 2: Click Add Transform Transform and Matrix are Created Around Selected Volume 3: Double-Click Matrix To Enter Alignments Module
National Alliance for Medical Image Computing Transformation Notes Slicer Transforms are 4x4 matrices –Support Linear Affine Transformations Transformations can be Nested in a Hierarchy –But Automated Registration Requires Only One Level (like current example) Transformations are RAS RAS –RAS = Right, Anterior, Superior –Meaning they are in Patient Space (mm) –Scan Order (Axial, Coronal, etc) Already Taken into Account Before Transformations are Applied Transformations are Stored in MRML Scene File
National Alliance for Medical Image Computing Manual Transform Local: Moves Relative to Volume Global: Moves Relative to World Use the Sliders, Buttons and Mouse Actions and Watch How the Volume Moves Use Left Mouse Button in Slice Views to Edit Transform Reset to Identity
National Alliance for Medical Image Computing See/Edit the Actual Matrix Matrix is Shown in the Props Tab You Can Edit the Elements (If you know what you are doing!) The Last Column is the Translation
National Alliance for Medical Image Computing ITK-Based Automated Registration Go to the Auto Tab Volume to Move is the One Inside the Transformation Reference is the One Outside the Transformation Use Intensity Registration, with Mutual Information Method, Normal Interface. Optimized for MR-MR, but Works Well for Other Tasks Too.
National Alliance for Medical Image Computing Registration Modes Scroll to See the Start Button if needed Pre-Programmed Modes: -Coarse: large steps, animated -Fine: small steps, animated -Good and Slow: run to completion, usually gets good result -Very Good and Very Slow: run to completion; if this doesn’t work, use manual to get closer Runs the Registration
National Alliance for Medical Image Computing Registration Result After Running Very Good and Very Slow (actually takes less than a minute on a laptop for this data)
National Alliance for Medical Image Computing Resampling Volumes Use the TransformVolume Module –New in Slicer 2.5 Makes a New Slicer Volume with Dimensions, Spacing, and Orientation to Match Registration Target
National Alliance for Medical Image Computing TransformVolume Select Transform Node from Data Tree (NB: May Contain Multiple Volumes to Transform) Deformation Not Yet Supported Select ReferenceVolume Resample Mode and the Volume the Result Should be Like (In Terms of Orientation, Dimensions, Spacing) Generally Cubic for Image Volumes, Nearest Neighbor for Labels Creates New Volumes with Given Prefix
National Alliance for Medical Image Computing Nonrigid Registration Uses the AG Module –Developed by Alexandre Guimond, PhD while working at the BWH Center for Neurological Imaging (Charles Guttmann, M.D., Director); Slicer Integration by Lifeng Liu, PhD. –Based on Original Work Done at Inria, France –A. Guimond, A. Roche, N. Ayache, and J. Meunier. Multimodal brain warping using the demons algorithm and adaptative intensity corrections. Technical Report 3796, Institut National de Recherche en Informatique et en Automatique, Sophia Antipolis, France, November Multimodal brain warping using the demons algorithm and adaptative intensity corrections –ftp://ftp.inria.fr/INRIA/publication/publi-pdf/RR/RR pdf Implemented Using Custom VTK Classes (not ITK) Nonlinear Registration Not Supported in MRML Tree, so Resampled Volumes are Created
National Alliance for Medical Image Computing AG Basics Select Reference Volume and Volume to Move Create New Resampled Output Optional Label Map (e.g. ICC) Optional Second Channel Run Registration
National Alliance for Medical Image Computing AG Methods Methods: -Initial (Optional) Uses Slicer Matrix as Starting Point; Select Matrix in Transform Tab -Linear Run Linear Alignment as Selected in Expert Tab -Non-linear Run Warp Transform Registration as Selected in Expert Tab NB: Registration is Multi-Step: First Linear (Translation, Rigid, Affine) Then Non-Linear Warp
National Alliance for Medical Image Computing AG to Transform Volumes Co-Registration: -Resamples Volumes using Currently Calculated Transformation (Linear and Non-linear -Run Volumes One-at-a-Time -Useful for Atlas Registration Scroll Down
National Alliance for Medical Image Computing AG Transform Tab Select Matrix and Toggle On/Off Button to Use Initial Transform Load and Use Existing Non-Linear Warp File (One Displacement Vector per Voxel) Export Current Linear Transform to Slicer Matrix Save Non-Linear Warp to a File (.vtk format)
National Alliance for Medical Image Computing Some AG Expert Settings Where to Stop the Linear Step Type of Non-Linear Intensity Compensation Function Registration Metric (GCR is Generalized Correlation Ratio; Robust for Multi-Modality)