DRR acceleration using inexpensive GPUs for model-image registration based joint kinematic measurements

Area of Science:

• Medical imaging
• Biomechanical engineering
• Computer science

Background:

• Model-image registration is crucial for measuring 3D joint kinematics from X-ray images in research.
• Current methods are often too slow or costly for clinical application.
• Digitally reconstructed radiographs (DRRs) are computationally expensive components of these methods.

Purpose of the Study:

• To develop a computationally efficient method for generating DRRs.
• To assess the feasibility of using consumer-grade graphics hardware for clinical joint kinematics analysis.

Main Methods:

• Implemented a numerically efficient Siddon-Jacobs algorithm for DRR computation.
• Utilized a programming language for parallel architectures on a consumer-grade graphics card.
• Compared performance against traditional CPU-only voxel projection algorithms.

Main Results:

• Achieved rendering speedups of 650-1546 times compared to CPU-only implementations.
• Maintained equivalent performance for joint kinematics measurements.
• Demonstrated the effectiveness of parallel computation on graphics hardware.

Conclusions:

• Optimized DRR computation using graphics cards drastically reduces processing time.
• This approach enhances the clinical practicality of model-image registration for joint kinematics.
• Consumer-grade hardware offers a viable solution for accelerating clinical biomechanical analysis.