This document provides instructions for automatically registering two structural MR datasets using a mutual information algorithm in 3D Slicer. It describes loading the datasets, visualizing the initial misalignment, defining an initial manual transformation, running the automatic registration using mutual information to maximize similarity, optionally refining the registration interactively, and applying the final transformation to spatially align the datasets. The goal is to demonstrate how to register a pre-operative MR with an intra-operative MR acquired with different scanners.

1. Slicer Advanced Training 11: Registration Sonia Pujol, Ph.D. Surgical Planning Laboratory Radiology, Brigham and Women’s Hospital Harvard Medical School Randy Gollub, M.D., Ph.D. Athinoula A. Martinos Center Psychiatry, Massachusetts General Hospital Harvard Medical School

5. Goal of the tutorial Guiding you step-by-step through the process of automatically registering two structural MR datasets using a mutual information algorithm. In this tutorial, an example of registration of a pre-operative MR dataset with an intra-operative MR dataset is used.

7. Processing pipeline Automatic registration Final Transform Semi-automatic refinement of the registration no yes Manual registration Initial transform Result OK ? (Step 2) (Step 3) (Step 4) Data loading (Step 1)

10. Loading dataset 1 Click on Add Volume in the Main Panel

11. Loading dataset 1 Select Properties Nrrd Reader Browse to load the file reg.nhdr Click on Apply to load the volume

12. Loading dataset 1 Slicer loads the volume reg.nhdr

13. Loading dataset 2 Click on AddVolume to load the dataset 2

14. Loading dataset 2 Browse to load the image I.001 Click on Apply to load the volume Select the Properties Basic

15. Loading dataset 2 Slicer loads the volume I

16. Initial mis-alignment Left-click on Fg to display the volume reg-nhdr in foreground.

17. Initial mis-alignment Click on Fade and use the slider to visualize the initial mis-alignment between the two volumes

18. Initial mis-alignment Observe the misalignment on the occipital lobe (axial slice 0) using the Fade function. I2 I1

19. Initial mis-alignment Observe the misalignment on the boundaries between the cerebrum and the cerebellum (sagittal slice 0). I2 I1

20. Initial mis-alignment Observe the misalignment on the lateral edge of the brain (axial slice 30). I2 I1

23. Rigid Transformation Image Space 1 Image Space 2 I 1 T(I 1 )

24. Rigid Transform By applying the registration transform to the initial volume I 1, we’ll generate a new volume spatially aligned with the volume I 2. This allows the extraction of complementary information from the two volumes. Image Space 1 Image Space 2 I 2 I 1

25. Adding a transformation To perform an initial manual registration between the two volumes, select the volume reg-nhdr and click on Add Transform. You will manually define the parameters of the initial registration matrix by using the mouse to superimpose the two volumes.

26. Adding a transformation Slicer adds the transform transform0 defined by the Identity matrix manual0. Double-click on manual0 to display the translation and rotation elements.

27. Adding a transformation Slicer displays the three translation parameters and the three rotation angles of the matrix manual0 (identity). The six degrees of freedom are defined in the anatomical directions Left-Right (LR), Posterior-Anterior (PA) and Inferior-Superior (IS).

28. Processing pipeline Automatic registration Final Transform Semi-automatic refinement of the registration no yes Manual registration Initial transform Result OK ? (Step 2) (Step 3) (Step 4) Data loading (Step 1)

29. Defining an initial transformation Click on Local and set the Mouse Action to Translate

30. Defining an initial transformation Hold the left mouse button down while clicking in the in the Axial view, and translate the slice in the anterior direction by 10 mm.

31. Defining an initial transformation Slicer displays the value of the applied manual translation in the PA direction.

32. Defining an initial transformation Click on Rotate to define the rotation component of the initial transformation.

33. Defining an initial transformation Hold the left mouse button down while clicking in the coronal view. Use the mouse to rotate the slice until you see the value of 3 degrees (counterclockwise) in the coronal view.

34. Defining an initial transformation Slicer displays the value of the applied manual rotation.

36. Similarity Measure I 2 T(I 1 ) The registration algorithm computes the parameters of the transformation T that optimizes a measure of similarity between the target image I 2 and the initial image that has been manually transformed T(I 1 ). T

38. Processing pipeline Final Transform Semi-automatic refinement of the registration no yes Manual registration Initial transform Result OK ? (Step 2) (Step 3) (Step 4) Data loading (Step 1) Automatic registration

39. Automatic registration Select the panel Auto in the module Alignments. Set the Volume to Move to reg-nhdr (I 1 ) and the Reference Volume to I (I 2 ). Select the Registration Mode to Intensity

43. Automatic registration Choose the Run Objective ‘Good and Slow’ and click on the button Start.

44. Automatic registration A Rigid Registration window appears, and Slicer displays the progress of the registration process.

45. Registration result Slicer displays the result of the automatic registration of the two volumes.

46. Registration result Slice through the volume to visualize the result of the registration

47. Observe the results of the registration in the occipital bone (axial slice 0). I2 T(I1) Registration result

48. Registration result Observe a better alignment of the boundaries between the cerebrum and the cerebellum (sagittal slice 0). I2 T(I1)

49. Registration result Observe the results of the registration on the lateral edge of the brain (axial slice 30). I2 T(I1)

50. Registration result: summary Before registration After registration

51. Registration result Click on the Props tab to display the parameters of the resulting rigid transformation T between the two datasets.

53. Registration result Note a tilt and a misalignment in the Inferior-Superior direction: observe the difference in shape of the ventricles in T(I1) and I2. I2 T(I1)

54. Registration result Note a misalignment in the Inferior-Superior direction: observe the difference in white matter localization on the middle line in T(I1) and I2. I2 T(I1)

55. Processing pipeline Automatic registration Final Transform Semi-automatic refinement of the registration no yes Manual registration Initial transform Result OK ? (Step 2) (Step 3) (Step 4) Data loading (Step 1)

56. Refine the registration Click on the tab Auto and select the mode Coarse to refine the result of the registration. Click on Start to launch the algorithm.

57. Refine the registration Left-click in the sagittal view, and slightly move the slice with the mouse to correct the tilt.

58. Refine the registration Left-click in the saggital view and slightly move the slice down with the mouse to correct the vertical misalignment

59. Refine the registration Observe Slicer iterating the registration algorithm, and updating the position of the volume in the three anatomical views. Iterate the process until you are satisfied with the alignment of the two volumes.

60. Refine the registration Click on Stop to terminate the semi-automatic registration process Information on details and performances of the registration algorithm are available at http://www.itk.org/HTML/MutualInfo.htm

61. Example of registration result Before registration After automatic registration The results might differ very slightly: these pictures show an example of a good outcome. After semi-automatic refinement

62. Processing pipeline Automatic registration Final Transform Semi-automatic refinement of the registration no yes Manual registration Initial transform Result OK ? (Step 2) (Step 3) (Step 4) Data loading (Step 1)

64. Apply the registration transform By applying the registration transform to the initial volume I 1, we’ll generate a new volume spatially aligned with the volume I 2. This allows the extraction of complementary information from the two volumes. Image Space 1 Image Space 2 I 2 I 1

65. Apply the registration transform Click on Modules  Examples and select the module TransformVolume. In the following section, we’ll use the transform Volume module to resample the initial volume reg-nhdr through the transform transform0 calculated by the registration.

66. Apply the registration transform Select the Reference Volume reg-nhdr and the Resample Mode ReferenceVolume Choose the Interpolation Mode Cubic Click on Show Preview to visualize a preview of the transformed volume.

67. Apply the registration transform A pop-up window displays a preview of the resampled volume, after applying transform0 . Click on DoTransform to apply the final transform calculated through the registration to the volume reg-nhdr.

68. Apply the registration transform Slicer generates the final volume xformed-reg-nhdr, which has the same orientation and spacing as the volume reg-nhdr . (See ‘ SlicerTraining7: Saving Data ’ to save the volume on disk.)

70. Appendix: TransformVolume The TransformVolume module offers the possibility to resample several volumes using the same transform. All the volumes will then be aligned to the same voxel space.