modeling and simulation schemes in radiography/ ctxraysim.sourceforge.net/nde09slides.pdf · 2012....
TRANSCRIPT
![Page 1: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/1.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Modeling and Simulation schemes inRadiography/ CT
Koushik ViswanathanDr. Krishnan Balasubramanian
Center for Non-Destructive EvaluationIndian Institute of Technology, Madras
January 12, 2010
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 2: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/2.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
OutlineSimulations on GPUs
Outline
Were going to look at two schemes pertaining to X-rayradiography and CT
I Data reconstruction involving Computed Tomography, mainlydealing with reconstruction of 2D cross-sections from X-rayprojections
I Forward simulation to simulate radiography/ tomographicimaging on CAD models
I All of this is done ENTIRELY on commodity graphicshardware!
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 3: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/3.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
OutlineSimulations on GPUs
Outline
Were going to look at two schemes pertaining to X-rayradiography and CT
I Data reconstruction involving Computed Tomography, mainlydealing with reconstruction of 2D cross-sections from X-rayprojections
I Forward simulation to simulate radiography/ tomographicimaging on CAD models
I All of this is done ENTIRELY on commodity graphicshardware!
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 4: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/4.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
OutlineSimulations on GPUs
Outline
Were going to look at two schemes pertaining to X-rayradiography and CT
I Data reconstruction involving Computed Tomography, mainlydealing with reconstruction of 2D cross-sections from X-rayprojections
I Forward simulation to simulate radiography/ tomographicimaging on CAD models
I All of this is done ENTIRELY on commodity graphicshardware!
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 5: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/5.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
OutlineSimulations on GPUs
Outline
Were going to look at two schemes pertaining to X-rayradiography and CT
I Data reconstruction involving Computed Tomography, mainlydealing with reconstruction of 2D cross-sections from X-rayprojections
I Forward simulation to simulate radiography/ tomographicimaging on CAD models
I All of this is done ENTIRELY on commodity graphicshardware!
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 6: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/6.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
OutlineSimulations on GPUs
Simulations on GPUs
I Graphics Processing Units are capable of large-scale parallelprocessing
I Low-end GPUs contain around 32 shader coresI Available high end cards feature upto 480 coresI Intel’s next-gen processor will have 16 cores
I Two approaches to parallelization, namely multi-coreprocessing (reconstruction) and traditional GPGPU techniques(forward simulation)
I Drawbacks with GPUs
I No double precision support (though that is changing)I Low clock speeds (unlikely to change soon)
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 7: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/7.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
OutlineSimulations on GPUs
Simulations on GPUs
I Graphics Processing Units are capable of large-scale parallelprocessing
I Low-end GPUs contain around 32 shader cores
I Available high end cards feature upto 480 coresI Intel’s next-gen processor will have 16 cores
I Two approaches to parallelization, namely multi-coreprocessing (reconstruction) and traditional GPGPU techniques(forward simulation)
I Drawbacks with GPUs
I No double precision support (though that is changing)I Low clock speeds (unlikely to change soon)
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 8: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/8.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
OutlineSimulations on GPUs
Simulations on GPUs
I Graphics Processing Units are capable of large-scale parallelprocessing
I Low-end GPUs contain around 32 shader coresI Available high end cards feature upto 480 cores
I Intel’s next-gen processor will have 16 cores
I Two approaches to parallelization, namely multi-coreprocessing (reconstruction) and traditional GPGPU techniques(forward simulation)
I Drawbacks with GPUs
I No double precision support (though that is changing)I Low clock speeds (unlikely to change soon)
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 9: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/9.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
OutlineSimulations on GPUs
Simulations on GPUs
I Graphics Processing Units are capable of large-scale parallelprocessing
I Low-end GPUs contain around 32 shader coresI Available high end cards feature upto 480 coresI Intel’s next-gen processor will have 16 cores
I Two approaches to parallelization, namely multi-coreprocessing (reconstruction) and traditional GPGPU techniques(forward simulation)
I Drawbacks with GPUs
I No double precision support (though that is changing)I Low clock speeds (unlikely to change soon)
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 10: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/10.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
OutlineSimulations on GPUs
Simulations on GPUs
I Graphics Processing Units are capable of large-scale parallelprocessing
I Low-end GPUs contain around 32 shader coresI Available high end cards feature upto 480 coresI Intel’s next-gen processor will have 16 cores
I Two approaches to parallelization, namely multi-coreprocessing (reconstruction) and traditional GPGPU techniques(forward simulation)
I Drawbacks with GPUs
I No double precision support (though that is changing)I Low clock speeds (unlikely to change soon)
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 11: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/11.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
OutlineSimulations on GPUs
Simulations on GPUs
I Graphics Processing Units are capable of large-scale parallelprocessing
I Low-end GPUs contain around 32 shader coresI Available high end cards feature upto 480 coresI Intel’s next-gen processor will have 16 cores
I Two approaches to parallelization, namely multi-coreprocessing (reconstruction) and traditional GPGPU techniques(forward simulation)
I Drawbacks with GPUs
I No double precision support (though that is changing)I Low clock speeds (unlikely to change soon)
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 12: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/12.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
OutlineSimulations on GPUs
Simulations on GPUs
I Graphics Processing Units are capable of large-scale parallelprocessing
I Low-end GPUs contain around 32 shader coresI Available high end cards feature upto 480 coresI Intel’s next-gen processor will have 16 cores
I Two approaches to parallelization, namely multi-coreprocessing (reconstruction) and traditional GPGPU techniques(forward simulation)
I Drawbacks with GPUsI No double precision support (though that is changing)
I Low clock speeds (unlikely to change soon)
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 13: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/13.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
OutlineSimulations on GPUs
Simulations on GPUs
I Graphics Processing Units are capable of large-scale parallelprocessing
I Low-end GPUs contain around 32 shader coresI Available high end cards feature upto 480 coresI Intel’s next-gen processor will have 16 cores
I Two approaches to parallelization, namely multi-coreprocessing (reconstruction) and traditional GPGPU techniques(forward simulation)
I Drawbacks with GPUsI No double precision support (though that is changing)I Low clock speeds (unlikely to change soon)
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 14: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/14.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Data reconstruction techniquesOur implementationResults
Image reconstruction schemes
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 15: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/15.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Data reconstruction techniquesOur implementationResults
Image-reconstruction
I Most commonly used techniques include FilteredBackprojection and Algebraic Reconstruction algorithms
I Typical implementations on a standard PC require approx. 20seconds for around 0.5 degree projections
I Reconstruction done layer-by-layer, full 3D Feldkampreconstruction takes much longer due to larger datasets
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 16: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/16.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Data reconstruction techniquesOur implementationResults
Image-reconstruction
I Most commonly used techniques include FilteredBackprojection and Algebraic Reconstruction algorithms
I Typical implementations on a standard PC require approx. 20seconds for around 0.5 degree projections
I Reconstruction done layer-by-layer, full 3D Feldkampreconstruction takes much longer due to larger datasets
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 17: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/17.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Data reconstruction techniquesOur implementationResults
Image-reconstruction
I Most commonly used techniques include FilteredBackprojection and Algebraic Reconstruction algorithms
I Typical implementations on a standard PC require approx. 20seconds for around 0.5 degree projections
I Reconstruction done layer-by-layer, full 3D Feldkampreconstruction takes much longer due to larger datasets
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 18: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/18.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Data reconstruction techniquesOur implementationResults
Our implementation
Our Reconstruction algorithms are implemented using a large-scalemulti-threading scheme with NVidia’s CUDA API
I 2D FFT is done initially on the CPU due to problems withdouble precision handling
I After pre-filtering on the CPU, the image is transferred ontothe GPU for backprojection
I For a large number of projections, Backprojection iscomputationally the most intensive step
I Done massively in parallel, spawn one thread per projection
I This technique scales very well with modern GPUs
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 19: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/19.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Data reconstruction techniquesOur implementationResults
Our implementation
Our Reconstruction algorithms are implemented using a large-scalemulti-threading scheme with NVidia’s CUDA API
I 2D FFT is done initially on the CPU due to problems withdouble precision handling
I After pre-filtering on the CPU, the image is transferred ontothe GPU for backprojection
I For a large number of projections, Backprojection iscomputationally the most intensive step
I Done massively in parallel, spawn one thread per projection
I This technique scales very well with modern GPUs
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 20: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/20.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Data reconstruction techniquesOur implementationResults
Our implementation
Our Reconstruction algorithms are implemented using a large-scalemulti-threading scheme with NVidia’s CUDA API
I 2D FFT is done initially on the CPU due to problems withdouble precision handling
I After pre-filtering on the CPU, the image is transferred ontothe GPU for backprojection
I For a large number of projections, Backprojection iscomputationally the most intensive step
I Done massively in parallel, spawn one thread per projection
I This technique scales very well with modern GPUs
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 21: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/21.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Data reconstruction techniquesOur implementationResults
Our implementation
Our Reconstruction algorithms are implemented using a large-scalemulti-threading scheme with NVidia’s CUDA API
I 2D FFT is done initially on the CPU due to problems withdouble precision handling
I After pre-filtering on the CPU, the image is transferred ontothe GPU for backprojection
I For a large number of projections, Backprojection iscomputationally the most intensive step
I Done massively in parallel, spawn one thread per projection
I This technique scales very well with modern GPUs
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 22: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/22.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Data reconstruction techniquesOur implementationResults
Our implementation
Our Reconstruction algorithms are implemented using a large-scalemulti-threading scheme with NVidia’s CUDA API
I 2D FFT is done initially on the CPU due to problems withdouble precision handling
I After pre-filtering on the CPU, the image is transferred ontothe GPU for backprojection
I For a large number of projections, Backprojection iscomputationally the most intensive step
I Done massively in parallel, spawn one thread per projection
I This technique scales very well with modern GPUs
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 23: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/23.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Data reconstruction techniquesOur implementationResults
Our implementation
Our Reconstruction algorithms are implemented using a large-scalemulti-threading scheme with NVidia’s CUDA API
I 2D FFT is done initially on the CPU due to problems withdouble precision handling
I After pre-filtering on the CPU, the image is transferred ontothe GPU for backprojection
I For a large number of projections, Backprojection iscomputationally the most intensive step
I Done massively in parallel, spawn one thread per projection
I This technique scales very well with modern GPUs
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 24: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/24.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Data reconstruction techniquesOur implementationResults
Quick comparisons
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 25: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/25.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Fundamental algorithmPossible optimizationsOur algorithm
Forward simulation schemes
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 26: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/26.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Fundamental algorithmPossible optimizationsOur algorithm
Fundamental algorithm
I Use Ray tracing, more commonly termed ray-casting, sinceonly primary rays are followed through the medium
I Spawn one ray per detector pixel, going from the source tothe deteector, through the medium
I Depending on the requirements, a triangulated mesh or avoxel mode is used
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 27: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/27.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Fundamental algorithmPossible optimizationsOur algorithm
Fundamental algorithm
I Use Ray tracing, more commonly termed ray-casting, sinceonly primary rays are followed through the medium
I Spawn one ray per detector pixel, going from the source tothe deteector, through the medium
I Depending on the requirements, a triangulated mesh or avoxel mode is used
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 28: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/28.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Fundamental algorithmPossible optimizationsOur algorithm
Fundamental algorithm
I Use Ray tracing, more commonly termed ray-casting, sinceonly primary rays are followed through the medium
I Spawn one ray per detector pixel, going from the source tothe deteector, through the medium
I Depending on the requirements, a triangulated mesh or avoxel mode is used
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 29: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/29.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Fundamental algorithmPossible optimizationsOur algorithm
Fundamental algorithm
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 30: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/30.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Fundamental algorithmPossible optimizationsOur algorithm
Possible optimizations
Ray-tracing optimization has been treated extensively, severaltechniques have been proposed
I Spatial sub-division techniques esp. Octrees and KD Trees
I Intersection tests can be reduced using Bounding VolumeHierarchies
I Parallel processing involving parallel ray-tracing algorithms
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 31: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/31.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Fundamental algorithmPossible optimizationsOur algorithm
Possible optimizations
Ray-tracing optimization has been treated extensively, severaltechniques have been proposed
I Spatial sub-division techniques esp. Octrees and KD Trees
I Intersection tests can be reduced using Bounding VolumeHierarchies
I Parallel processing involving parallel ray-tracing algorithms
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 32: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/32.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Fundamental algorithmPossible optimizationsOur algorithm
Possible optimizations
Ray-tracing optimization has been treated extensively, severaltechniques have been proposed
I Spatial sub-division techniques esp. Octrees and KD Trees
I Intersection tests can be reduced using Bounding VolumeHierarchies
I Parallel processing involving parallel ray-tracing algorithms
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 33: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/33.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Fundamental algorithmPossible optimizationsOur algorithm
Possible optimizations
Ray-tracing optimization has been treated extensively, severaltechniques have been proposed
I Spatial sub-division techniques esp. Octrees and KD Trees
I Intersection tests can be reduced using Bounding VolumeHierarchies
I Parallel processing involving parallel ray-tracing algorithms
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 34: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/34.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Fundamental algorithmPossible optimizationsOur algorithm
Parallel Processing tests
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 35: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/35.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Fundamental algorithmPossible optimizationsOur algorithm
Our algorithm
We went back to the drawing board and reworked the basic lawevaluation
I Developed a new algorithm to evaluate the Beer-Lambert law
I Instead of following ray-tracing based techniques, wedeveloped a rasterization based algorithm
I Intensity of every detector pixel computed in parallel, we makeGPUs do what they were meant to do!
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 36: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/36.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Fundamental algorithmPossible optimizationsOur algorithm
Our algorithm
We went back to the drawing board and reworked the basic lawevaluation
I Developed a new algorithm to evaluate the Beer-Lambert law
I Instead of following ray-tracing based techniques, wedeveloped a rasterization based algorithm
I Intensity of every detector pixel computed in parallel, we makeGPUs do what they were meant to do!
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 37: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/37.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Fundamental algorithmPossible optimizationsOur algorithm
Our algorithm
We went back to the drawing board and reworked the basic lawevaluation
I Developed a new algorithm to evaluate the Beer-Lambert law
I Instead of following ray-tracing based techniques, wedeveloped a rasterization based algorithm
I Intensity of every detector pixel computed in parallel, we makeGPUs do what they were meant to do!
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 38: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/38.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Fundamental algorithmPossible optimizationsOur algorithm
Our algorithm
We went back to the drawing board and reworked the basic lawevaluation
I Developed a new algorithm to evaluate the Beer-Lambert law
I Instead of following ray-tracing based techniques, wedeveloped a rasterization based algorithm
I Intensity of every detector pixel computed in parallel, we makeGPUs do what they were meant to do!
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 39: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/39.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Fundamental algorithmPossible optimizationsOur algorithm
Very promising results
I Other groups around the world have tackled this problem, butour algorithm works faster than all of the results reported sofar
I Works on any consumer-grade hardware, only OpenGL 2.0support needed!
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 40: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/40.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Fundamental algorithmPossible optimizationsOur algorithm
Very promising results
I Other groups around the world have tackled this problem, butour algorithm works faster than all of the results reported sofar
I Works on any consumer-grade hardware, only OpenGL 2.0support needed!
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 41: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/41.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
XRaySim
I Open source Radiography/ Computed Tomography simulationtool
I Developed from scratch at the Center for Non-DestructiveEvaluation, IIT Madras
I Seamless integration with commercial CAD packages throughstandard data exchange formats
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT
![Page 42: Modeling and Simulation schemes in Radiography/ CTxraysim.sourceforge.net/NDE09slides.pdf · 2012. 9. 20. · Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation](https://reader033.vdocuments.net/reader033/viewer/2022060917/60aa23863305033f2f32eccd/html5/thumbnails/42.jpg)
IntroductionImage-reconstructionForward Simulation
Conclusions
Thank you for your attention!Questions/ enquiries are welcome
Koushik Viswanathan, Dr. Krishnan Balasubramanian Modeling and Simulation schemes in Radiography/ CT