Fast reconstructed radiographs from octree-compressed volumetric data

Area of Science:

• Medical Imaging
• Computer Graphics
• Radiotherapy

Background:

• Digitally reconstructed radiographs (DRRs) are vital for medical image registration and digital tomosynthesis (DTS).
• Conventional ray-casting for DRR generation from CT data is computationally intensive and slow.
• Real-time DRR generation is needed for image-guided radiation therapy (IGRT) systems.

Purpose of the Study:

• To develop a faster algorithm for rendering DRRs from octree-compressed CT data.
• To improve the efficiency of DRR generation for applications in medical image registration and DTS.
• To address the computational bottleneck in precomputing large numbers of DRRs for IGRT.

Main Methods:

• Utilized graphics processing units (GPUs) with vertex and fragment shaders to render DRRs.
• Employed octree-compressed CT data, projecting internal octree spaces via textured particle sprites.
• Procedurally generated textures dependent on the CT pose for direct projection onto the view plane.

Main Results:

• Achieved a 4x speed improvement compared to CUDA-based ray-casting on equivalent GPUs with ~95% octree compression.
• Observed rendering artifacts, but image quality improved with higher compression.
• Peak signal-to-noise ratio analysis showed marginally better quality than Gaussian footprints.

Conclusions:

• The octree-encoded DRR rendering approach shows promise for accelerating automatic image registration.
• GPU-based rendering of octree-compressed CT data offers a significant performance enhancement.
• This method could facilitate real-time DRR generation for advanced radiotherapy applications.