Implementation of GPU-accelerated back projection for EPR imaging.

Area of Science:

• Medical Imaging
• Biophysics
• Computational Science

Background:

• Electron paramagnetic resonance (EPR) Imaging (EPRI) is a key technique for in vivo oxygen partial pressure (pO2) measurement.
• The filtered backprojection (FBP) algorithm, commonly used for 3D EPRI reconstruction, is computationally intensive on CPUs.
• Existing acceleration methods for FBP are often inflexible or still time-consuming.

Purpose of the Study:

• To accelerate the computationally intensive backprojection step in 3D EPRI reconstruction.
• To implement and evaluate a single-stage backprojection algorithm on a Graphics Processing Unit (GPU).
• To assess the performance gains of GPU acceleration for EPRI image reconstruction.

Main Methods:

• Implementation of a single-stage backprojection algorithm on a multi-core GPU (1152 cores).
• Utilized Graphics Processing Units (GPUs) for parallel processing to speed up computations.
• Compared GPU implementation performance against traditional CPU-based methods.

Main Results:

• Achieved an overall acceleration factor exceeding 200x for EPRI image reconstruction.
• Observed a computing time acceleration factor of over 3500x solely from the GPU implementation.
• Summarized key implementation experiences for GPU-accelerated backprojection.

Conclusions:

• GPU acceleration offers a significant speedup for EPRI image reconstruction.
• The accelerated method enables real-time monitoring and supports time-sensitive EPRI applications.
• This advancement enhances the utility of EPRI for in vivo pO2 measurements.