1. -1- Massachusetts General Hospital National Alliance for Medical Image Computing Using Plastimatch for Deformable Registration Gregory C. Sharp Department of Radiation Oncology Massachusetts General Hospital 3D Slicer Training Compendium Tutorial Version 1.0, Apr 26, 2010

2. -2- Massachusetts General Hospital National Alliance for Medical Image Computing Learning Objective This tutorial is a step-by-step guide, and includes: 1) Downloading the Plastimatch extension to 3D Slicer 2) Loading the sample images 3) Running deformable registration on the CPU 4) Running deformable registration on the GPU 5) Inspecting registration quality in 3D Slicer The plastimatch web site is: http://plastimatch.org

3. -3- Massachusetts General Hospital National Alliance for Medical Image Computing Prerequisites This tutorial assumes that you have already downloaded the sample data. You can get the data from here: http://forge.abcd.harvard.edu/gf/download/frsrelease/85/1004/rider-lung-images.tar.gz

4. -4- Massachusetts General Hospital National Alliance for Medical Image Computing Part 1: Downloading the Plastimatch Extension

5. -5- Massachusetts General Hospital National Alliance for Medical Image Computing (This part of the tutorial might not work correctly, pending the Slicer 3.6 release)

6. -6- Massachusetts General Hospital National Alliance for Medical Image Computing Start up 3D Slicer Choose “Extension Manager” from the “View” menu

7. -7- Massachusetts General Hospital National Alliance for Medical Image Computing Click “Next”

8. -8- Massachusetts General Hospital National Alliance for Medical Image Computing Find the plastimatch plugin, and click “Select” Then, click “Download and Install”

9. -9- Massachusetts General Hospital National Alliance for Medical Image Computing The “Status” should become green Click “Next”

10. -10- Massachusetts General Hospital National Alliance for Medical Image Computing Restart 3D Slicer

11. -11- Massachusetts General Hospital National Alliance for Medical Image Computing Part 2: Loading the example data

12. -12- Massachusetts General Hospital National Alliance for Medical Image Computing Start up 3D Slicer

13. -13- Massachusetts General Hospital National Alliance for Medical Image Computing Choose “Add data” from the menu

14. -14- Massachusetts General Hospital National Alliance for Medical Image Computing Choose “Add files” in dialog box

15. -15- Massachusetts General Hospital National Alliance for Medical Image Computing Select (highlight) both example files: fix.nrrd and mov.nrrd Then click “Open”

16. -16- Massachusetts General Hospital National Alliance for Medical Image Computing Click “Apply”

17. -17- Massachusetts General Hospital National Alliance for Medical Image Computing The images are now loaded

18. -18- Massachusetts General Hospital National Alliance for Medical Image Computing Part 3: Visualizing the example data

19. -19- Massachusetts General Hospital National Alliance for Medical Image Computing We want to look at how well the images are aligned before we start 3D Slicer can view a “foreground” (F) and “background” (B) image at the same time. After loading, (F) is set to “None” in all views.

20. -20- Massachusetts General Hospital National Alliance for Medical Image Computing Click, and select “fix” as the foreground image. Repeat for all three views.

21. -21- Massachusetts General Hospital National Alliance for Medical Image Computing Use the “Manipulate Slice Views” slider to blend between foreground and background

22. -22- Massachusetts General Hospital National Alliance for Medical Image Computing We can now see the alignment of the images. To see it better, we need to increase the viewport size. Click on the layout chooser button

23. -23- Massachusetts General Hospital National Alliance for Medical Image Computing Choose “Red slice only”

24. -24- Massachusetts General Hospital National Alliance for Medical Image Computing Much better! Next we're going to try color blending. Choose the “Volumes module.

25. -25- Massachusetts General Hospital National Alliance for Medical Image Computing We're going to modify the color of the moving volume. Choose “mov” as the active volume.

26. -26- Massachusetts General Hospital National Alliance for Medical Image Computing Set it to “Warm Tint 1”

27. -27- Massachusetts General Hospital National Alliance for Medical Image Computing

28. -28- Massachusetts General Hospital National Alliance for Medical Image Computing Part 4: Running Plastimatch

29. -29- Massachusetts General Hospital National Alliance for Medical Image Computing Go back to the module selector

30. -30- Massachusetts General Hospital National Alliance for Medical Image Computing Choose “B-spline deformable registration” from the “Plastimatch” section

31. -31- Massachusetts General Hospital National Alliance for Medical Image Computing Set “Fixed Volume” to “fix” Set “Moving Volume” to “mov” Set “Output Volume” to “Create New Volume”

32. -32- Massachusetts General Hospital National Alliance for Medical Image Computing Click “Apply” (You might need to scroll down)

33. -33- Massachusetts General Hospital National Alliance for Medical Image Computing Check the status in the status bar With a Tesla C1060 GPU, the registration takes 6 seconds A laptop might take 1 or 2 minutes

34. -34- Massachusetts General Hospital National Alliance for Medical Image Computing When the registration is complete, the warped image is automatically displayed

35. -35- Massachusetts General Hospital National Alliance for Medical Image Computing You have to set the foreground view again to see the registration quality

36. -36- Massachusetts General Hospital National Alliance for Medical Image Computing Your results should look like this.

37. -37- Massachusetts General Hospital National Alliance for Medical Image Computing Part 5: Optimizing Your Registration

38. -38- Massachusetts General Hospital National Alliance for Medical Image Computing We're going to try to improve the registration result.

39. -39- Massachusetts General Hospital National Alliance for Medical Image Computing Click on “Enable Stage 2” Then click “Apply” This takes 12 seconds on the Tesla C1060. Might be 3-4 minutes on a laptop.

40. -40- Massachusetts General Hospital National Alliance for Medical Image Computing Like before, the output is automatically loaded.

41. -41- Massachusetts General Hospital National Alliance for Medical Image Computing Your results should look like this. Note improvement in the alignment of the mediastinum

42. -42- Massachusetts General Hospital National Alliance for Medical Image Computing Part 6: Advanced Plastimatch Options

43. -43- Massachusetts General Hospital National Alliance for Medical Image Computing By default, plastimatch optimizes Mean-squared error (MSE). But you can choose Mutual Information (MI) instead

44. -44- Massachusetts General Hospital National Alliance for Medical Image Computing By default, plastimatch uses the GPU. But you can choose to use the CPU instead. Plastimatch CPU uses OpenMP to take advantage of modern multi-core systems However, in Plastimatch 1.4, mutual information does not take advantage of the GPU, nor is it multi-threaded.

45. -45- Massachusetts General Hospital National Alliance for Medical Image Computing In our tutorial, the images were sufficiently well aligned that we could use B-spline registration. But if they are not well aligned, you can do a “rough alignment” using translation, rigid, or affine registration. Click “Enable Stage 0” to enable the rough alignment.

46. -46- Massachusetts General Hospital National Alliance for Medical Image Computing For each stage, you can modify the subsampling rate, grid size, and maximum iterations Decreasing the subsampling rate improves accuracy Increasing the subsampling rate improves reliability

47. -47- Massachusetts General Hospital National Alliance for Medical Image Computing Decreasing max iterations improves registration speed Increasing max iterations improves registration accuracy

48. -48- Massachusetts General Hospital National Alliance for Medical Image Computing Decreasing the grid spacing improves accuracy Increasing the grid spacing improves reliability

49. -49- Massachusetts General Hospital National Alliance for Medical Image Computing Conclusion Congratulations! You have completed the tutorial. Please send corrections or suggestions to: Greg Sharp gcsharp@partners.org gcsharp@partners.org Or visit the web page at: http://plastimatch.org

50. -50- Massachusetts General Hospital National Alliance for Medical Image Computing National Alliance for Medical Image Computing NIH U54EB005149 Acknowledgements National Institutes of Health NIH / NCI 6-PO1 CA 21239 Federal share of program income earned by MGH on C06CA059267 Progetto Rocca Foundation A collaboration between MIT and Politecnico di Milano