Generation of planar radiographs from 3D anatomical models using the GPU
Generation of planar radiographs from 3D
anatomical models using the GPU
André dos Santos Cardoso
Supervisor: Jorge M. G. Barbosa
University of Porto
Faculty of Engineering of University of Porto
July 14, 2010
André dos Santos Cardoso DRR Generation 1 / 19
2 Why is it Important?
3 Our Speciﬁc Case
4 What Has Been Done?
5 Current Solution
6 What’s expected?
7 Involved Technologies
8 Work Plan
André dos Santos Cardoso DRR Generation 2 / 19
Digitally Reconstructed Radiographs
Taking a radiography from 3D digital
anatomical models of vertebrae
Form of depth peeling, using
Key component in 2D/3D registration
André dos Santos Cardoso DRR Generation 3 / 19
DRRs are built from vertebrae models represented by 3D
DRR generation as mean to validate and/or reﬁne 3D
reconstructions of the spine from multi-planar radiography
Vertebrae Shape Recovery Using 2D/3D Non-Rigid
Important techniques for Scoliosis therapy and follow-ups
André dos Santos Cardoso DRR Generation 4 / 19
Build Fast DRR Algorithms
DRR calculation is a bottleneck
3D reconstruction usage in a
routine clinical environment
requires high performances
Take advantage of processing
power of new GPUs and APIs
Common workstations could do
André dos Santos Cardoso DRR Generation 5 / 19
Why is it Important?
DRR generation key component
in many 2D/3D registration
Allows to compare/use data
from different sources and times
The CAS model versus real-time
imagery from the patient
Many applications in medical area
CAS, Radiotherapy, Volume
Known to be a common bottleneck
André dos Santos Cardoso DRR Generation 6 / 19
Why is it Important?
Daily work on the ﬁeld demands
on-the-ﬂy results, and high
Advantages on using GPUs
versus Hardware solutions
General Purpose Computing on
GPUs gaining increasing interest
André dos Santos Cardoso DRR Generation 7 / 19
Our Speciﬁc Case
Where will the Optimized
Shape Recovery of human
spine – attaining a 3D model of
Viable alternative to MRIs and
CTs – why?
Expensive, Amount of
Procedures, Require lying
Not the scope of this project!!
Moura, D. et al 
André dos Santos Cardoso DRR Generation 8 / 19
What has been done in this area?
Attenuation Law – monochromatic x-ray radiation
Nout (E ) = Nin (E ) × e − µ(E ,ρ(x ),Z (x ))dx
Focus on GPU Implementations!
Monte Carlo Volume Shear-Warping
Rendering Viewing transformations
Volumetric integral for Splatting
each pixel Throw voxels into the
Fourier Volume Rendering viewing pane
Inverse 2D Fourier Ray Casting
transform of a slice Shoot rays to each pixel
André dos Santos Cardoso DRR Generation 9 / 19
Attenuation law for bone material
Few Applications of DRR to 3D Meshes (most work on CT
data – voxels)
Using OpenGL Shading Language (GLSL)
Multi Pass Algorithm is available
Single Pass Algorithm is considered the state of the art, but
no applied implementation exists
Compute Uniﬁed Device Architecture (CUDA) peeling
applications exist (but not for computing DRRs)
André dos Santos Cardoso DRR Generation 10 / 19
CAD model of vertebrae
Camera position, object positions, object orientation, …
André dos Santos Cardoso DRR Generation 11 / 19
Ray Casting and Depth
Why use CAD models?
Problem context requires
deformations to the 3D
Faster to deform CAD
amount of computation
André dos Santos Cardoso DRR Generation 12 / 19
Enhance the current solution
1 Modify code to implement the reported single-pass approach
2 Port solution to CUDA
3 Test and compare solutions
Signiﬁcant speed-ups are expected
André dos Santos Cardoso DRR Generation 13 / 19
Technologies – GLSL
Allows the modiﬁcation
of ﬁxed functionality of
the GPU pipeline
Similar syntax to C/C++
Modules called Shaders
André dos Santos Cardoso DRR Generation 14 / 19
Technologies – CUDA
Compute Uniﬁed Device
Parallel Computing Architecture
Allows direct access to parallel
processors and memory
Kernel function executed on
conﬁguration of threads upon
Massive data parallelism
Allows versatile and more
André dos Santos Cardoso DRR Generation 15 / 19
André dos Santos Cardoso DRR Generation 16 / 19
Thank You for Listening!
André dos Santos Cardoso DRR Generation 17 / 19
The opengl shading language.
Lisa Gottesfeld Brown.
A survey of image registration techniques.
ACM Comput. Surv., 24(4):325–376, 1992.
David B. Kirk and Wen mei W. Hwu.
Programming Massively Parallel Processors - A Hands-on Approach.
Morgan Kaufmann, 2010.
A. Mitulescu, W. Skalli, D. Mitton, and J. A. De Guise.
Three-dimensional surface rendering reconstruction of scoliotic vertebrae using a non stereo-corresponding
European Spine Journal, 2002.
Shinichiro Mori, Masanao Kobayashi, Motoki Kumagai, and Shinichi Minohara.
Development of a gpu-based multithreaded software application to calculate digitally reconstructed radiographs
Radiological Physics and Technology, 2009.
Daniel C. Moura, Jonathan Boisvert, Jorge G. Barbosa, and João Manuel Tavares.
Fast 3d reconstruction of the spine using user-deﬁned splines and a statistical articulated model.
In ISVC ’09: Proceedings of the 5th International Symposium on Advances in Visual Computing, pages
586–595, Berlin, Heidelberg, 2009. Springer-Verlag.
Scott D. Roth.
Ray casting for modeling solids.
j-CGIP, 18(2), 1982.
F. P. Vidal, M.Garnier, N. Freud, J. M. Létang, and N.W. John.
Simulation of x-ray attenuation on the gpu.
In Proceeding of TCPG’09 - Theory and Practice of Computer Graphics, pages 25–32. Eurographics, 2009.
André dos Santos Cardoso DRR Generation 18 / 19
Full Bibliography Listed in:
André dos Santos Cardoso DRR Generation 19 / 19