1. Validation of a Contour Based Deformable
Registration Method for Efficient Integration
of MR Imaging into HDR Brachytherapy
Matthew Strugari
December 2014

2. Outline
• Background
• Workflow
• Calculations
• Results
• Conclusions
• Future Work
Prostate
Deformable
Registration

3. Background

4. Registration
Ultrasound Image Magnetic Resonance Image
A method of mapping information between two or more imaging data sets
Axial View
Prostate

5. Background – Project
• To evaluate the feasibility for
online integration of magnetic
resonance imaging (MRI) into the
real-time ultrasound (US) based
high dose rate (HDR) prostate
Brachytherapy treatment workflow
using open-source software
Goal

6. Workflow

9. Initial DICOM Data
Ultrasound Data Magnetic Resonance Data
Axial View
Prostate
DIL

10. Labelmaps to Distance Maps
Prostate Labelmap Prostate Distance Map
Axial View
MR
US

11. B-Spline Deformable Registration
Deformation Vector Field
MR Prostate Distance Map US Prostate Distance Map
Axial View

12. Labelmap Transformation
MR Prostate Labelmap Deformed MR Prostate Labelmap
MR DIL Labelmap Deformed MR DIL Labelmap
Deformation Vector Field
BEFORE AFTER
Axial View

13. Labelmap Smoothing
Deformed Labelmap Gaussian Filter Smoothed Labelmap
Prostate
DIL
Axial View

14. Export to DICOM
US DICOM Image
DICOM RT Structures

15. Calculations

16. Dice Similarity Coefficient (DSC)
“Hausdorff Distance Sample”. Wikipedia. 6 May 2008. 1 May 2014. <http://en.wikipedia.org/wiki/File:Hausdorff_distance_sample.svg>.
Deformed Prostate
US Prostate US DIL
Deformed DIL

17. Hausdorff Distance
“Hausdorff Distance Sample”. Wikipedia. 6 May 2008. 1 May 2014. <http://en.wikipedia.org/wiki/File:Hausdorff_distance_sample.svg>.
Deformed Prostate
US Prostate US DIL
Deformed DIL

18. Results

19. Results
Structure Modality DSC Hausdorff
Distance (mm)
Prostate
MR - US 0.81 ± 0.05 7.48 ± 1.15
Deformed - US 0.95 ± 0.01 2.91 ± 0.86
Deformed - MR 0.82 ± 0.06 7.30 ± 1.31
DIL
MR - US 0.37 ± 0.26 11.27 ± 4.73
Deformed - US 0.41 ± 0.28 11.25 ± 5.09
Deformed - MR 0.57 ± 0.25 6.93 ± 1.85
Table 1. Comparisons Between Structures
Average Registration Time: 51.22 ± 11.10 s

20. Result Improvements

21. Reduce Intra-Observer Variability
Ultrasound Magnetic Resonance

22. DSC: Intra-Observer Variability

23. Hausdorff Distance: Intra-Observer Variability

24. Table 2. DSC and Hausdorff Distances for
Intra-Observer Variability
Structure Modality DSC Hausdorff Distance
(mm)
Prostate
US 0.86 ± 0.07 6.11 ± 3.06
MR 0.87 ± 0.03 5.82 ± 1.02
Deformed MR 0.86 ± 0.07 5.93 ± 2.22
DIL
US 0.35 ± 0.29 11.29 ± 5.02
MR 0.50 ± 0.21 10.15 ± 4.94
Deformed MR 0.43 ± 0.23 11.42 ± 5.42

25. Conclusion

26. Conclusion
 It is possible to incorporate MRI into the real-time US
based HDR prostate Brachytherapy treatment
workflow
 Deformable registration is useful for the accurate
identification of the DIL position in order to ensure a
precise setup and treatment for the patient
 MR information is maintained throughout the
registration process, mainly when considering the
position of the DIL

27. Conclusion
 The quantification of results supports the quality of
the registration output
 Visual inspection of the registration output shows
satisfactory results between the deformed and US
structure sets
 This method can be applied to a number of different
modalities such as CT, CBCT, MR, and PET

28. Future Work

29. Future Work
 Increase the registration speed using the NVIDIA
GeForce GTX 780 GPU in conjunction with the Cuda
capabilities of plastimatch
 Implement the deformable registration workflow at
the Allan Blair Cancer Centre (ABCC), and Odette
Cancer Centre, Sunnybrook

30. Special Thanks
Saskatchewan Cancer Agency
 Niranjan Venugopal
 Andrew Alexander
 Chris Newcomb
 Craig Beckett (ABCC)
3D Slicer Group
 Greg Sharp
 Csaba Pinter
 Steve Pieper
 Adam Rankin
 Bradley Lowekamp
 Andras Lasso
Odette Cancer Centre, Sunnybrook HSC
 Ananth Ravi
 Gerard Morton
 Hans Chung
 Andrew Loblaw