Implementation of GPU accelerated SPECT reconstruction with Monte Carlo-based scatter correction.

Area of Science:

• Medical Imaging
• Computational Science

Background:

• Statistical SPECT reconstruction is computationally intensive, particularly with corrections for collimator/detector response, attenuation, and scatter.
• Existing methods often require substantial processing time, limiting clinical workflow efficiency.

Purpose of the Study:

• To develop and evaluate an accelerated SPECT reconstruction algorithm utilizing graphics processing unit (GPU) technology.
• To assess the feasibility of GPU-based reconstruction for routine clinical use.

Main Methods:

• Implemented the Ordered Subset Expectation Maximization (OSEM) algorithm with comprehensive compensation models (CT-based attenuation, depth-dependent collimator-detector response, Monte Carlo scatter) using OpenCL.
• Compared the GPU implementation against a multi-threaded central processing unit (CPU) OSEM version using phantoms and clinical SPECT/CT data.

Main Results:

• GPU and CPU implementations showed negligible differences in scatter-to-primary ratios, visual quality, and standardized uptake values (SUVs).
• The GPU implementation achieved up to a 24-fold speed increase over the CPU version for bone SPECT/CT datasets, including all compensation methods.

Conclusions:

• GPU-accelerated SPECT reconstruction is a viable and highly efficient method.
• This technology holds significant potential for integration into daily clinical practice, improving throughput and turnaround times.