GPU-accelerated Monte Carlo based scatter correction in brain PET/MR

Michaela Gaens, Julien Bert, U. Pietrzyk, Awen Autret, N. Jon Shah, Dimitris VisvikisEJNMMI 2014 · DOI

TL;DR

This paper presents a novel GPU-accelerated Monte Carlo simulation (MCS) method for scatter correction in brain PET/MR imaging. By leveraging NVIDIA's CUDA technology, they achieve a significant speedup over traditional MCS, making it practical for clinical use, and demonstrate improved image quality, particularly in contrast recovery, compared to existing vendor-provided scatter correction methods.

What Problem Does This Paper Address?

Positron Emission Tomography (PET) imaging is susceptible to scatter, where photons deviate from their original path before detection. This scattering blurs images and can lead to inaccurate quantification of radiotracer uptake, especially in complex structures like the brain. Traditional scatter correction methods, such as Single Scatter Simulation (SSS), can be computationally intensive and sometimes struggle with accurately modeling the scatter distribution, particularly in the tails of the distribution. The integration of PET with Magnetic Resonance Imaging (MRI) in PET/MR scanners introduces additional challenges, such as the presence of MR coils that can affect photon detection and scattering. This paper aims to develop a faster and more accurate scatter correction method for PET/MR by utilizing the parallel processing power of Graphics Processing Units (GPUs) for Monte Carlo simulations.

Methods

The core of this work is an extension of an existing GPU-based Monte Carlo simulation (MCS) framework, originally developed by Bert et al. [1]. The authors adapted this framework to specifically model the detection process within the Siemens MR-BrainPET scanner. They utilized NVIDIA's CUDA architecture to parallelize photon tracking and detection, from their generation based on the activity distribution to their final registration as coincidences. The simulated coincidences are then stored in list-mode.

For scatter correction, the reconstructed emission image, a patient-specific MR-derived attenuation map, and a CT-derived attenuation template of the MR coils are used as inputs for the simulation. The simulated projection data are then scaled to match the measured projection data. This scaling approach circumvents the need for fitting the distribution tails, which is a known instability in some SSS methods. The resulting scaled scatter distribution is then incorporated into the image reconstruction process.

Datasets

The paper evaluates their method using:

  • Sphere phantom measurements: Used to evaluate the proposed scatter correction methodology and compare it against the standard SSS approach.
  • Patient data: Used to investigate the effect of the new scatter correction on real clinical images.

The paper does not introduce new benchmark datasets but rather uses established phantom and patient imaging protocols for evaluation.

Key Insights and Ideas

  • GPU Acceleration for Practical MCS: The most significant insight is that by leveraging GPUs, Monte Carlo simulations for scatter correction can become fast enough for routine clinical use, overcoming a major bottleneck of traditional MCS.
  • Circumventing SSS Instabilities: The proposed method's approach of scaling simulated scatter data to measured data elegantly bypasses the problematic tail-fitting step often required in SSS, leading to a more robust correction.
  • Improved Contrast Recovery: The integration of MR coil attenuation information into the simulation is crucial for accurate scatter correction in PET/MR, leading to demonstrably better contrast recovery, especially for smaller structures.
  • Synergy of MCS and PET/MR: The work highlights the power of combining advanced simulation techniques with the multi-modal capabilities of PET/MR scanners for enhanced image quality.

Key Results

  • Speedup: The new GPU implementation (GeForce GTX690) achieved a speedup factor of 135 compared to a single CPU implementation (Gate v6.2 on an Intel Core i7, 3.4GHz). The authors note that using multiple GPUs can further increase this acceleration.
  • Contrast Recovery (Phantom Data): For measurements including MR coils, the MC-based scatter correction improved contrast recovery by 10-14% compared to SSS. Table 1 shows improved recovery coefficients for hot spheres across various sizes, with differences ranging from 0.7% to 14.8%.
  • Qualitative Improvement (Patient Data): First patient images showed improved grey/white matter contrast after applying the new scatter correction method.

Limitations and Future Work

  • Limitations: The paper focuses on Compton scatter correction. While this is a major component of scatter, other scatter phenomena might not be fully addressed. The specific GPU hardware used (GeForce GTX690) might influence the reported speedup, and performance could vary with different GPU architectures.
  • Future Work: The authors suggest that the use of multiple GPUs can further enhance the acceleration. While not explicitly stated as future work, the improved contrast and accuracy suggest potential for broader clinical adoption and application to other PET/MR imaging scenarios.

Summary updated

Abstract

To develop and evaluate a fast and accurate Monte Carlo simulation (MCS) based scatter correction for PET/MR imaging, based on Geant4 using graphical processing units (GPUs). The GPU implementation by Bert et al. [1] including the photon tracking within a voxelized volume was extended with the modeling of the detection process in the Siemens MR-BrainPET. Using NVIDIA’s CUDA photon pairs undergo several kernel functions in parallel, from their generation according to the activity distribution up to their detection in the scanner. Subsequently, coincidences are sorted and stored in list-mode. The new GPU implementation (GeForce GTX690) was validated by comparison with Gate v6.2 on a single CPU (Intel Core i7, 3.4GHz), showing a speedup factor of 135. The use of multiple GPUs can further increase this acceleration. For Compton scatter correction the reconstructed emission image, the patient MR-derived attenuation map and a CT-derived attenuation template of the MR coils are used as input for the simulation. The resulting simulated projection data are scaled to the normalized measured projection data. Thus, the often unstable fit to the distribution tails, necessary for single scatter simulation (SSS)[2], is circumvented. The scaled simulated scatter distribution is then incorporated in the reconstruction. The proposed scatter correction methodology was evaluated and compared to the SSS approach using sphere phantom measurements. Finally, the effect on patient data was investigated. Phantom results suggest an improved contrast with our new MC-based scatter correction, especially for measurements including the MR coils, where the contrast recovery was improved by 10-14%. First patient images also show improved grey/white matter contrast. Table 1 Comparison of recovery coefficients after scatter correction based on MCS and SSS. MC-based scatter correction shows improved recovery coefficients for the hot spheres. Figure 1 Assessment of image quality according to NEMA NU 2-2007. Images show the phantom with a hot spheres/background ratio of 8:1 measured with MR head coils inside the BrainPET scanner. Figure 2 Comparison of reconstructed FDG images using the scatter estimate provided by the vendor’s SSS (left) and our MCS method (right). Images were reconstructed using the vendor provided 3D OP-OSEM algorithm with 2 subsets and 32 iterations. The new GPU-accelerated MCS allows for fast and accurate estimation of scattered coincidences in PET studies using the Siemens MR-BrainPET with a major speed-up compared to standard MCS and superior image quality compared with the standard correction based on SSS.
BibTeX
@inproceedings{gaens2014gpu,
  author = {Michaela Gaens and Julien Bert and U. Pietrzyk and Awen Autret and N. Jon Shah and Dimitris Visvikis},
  title = {{GPU-accelerated Monte Carlo based scatter correction in brain PET/MR}},
  booktitle = {EJNMMI},
  year = {2014},
  doi = {10.1186/2197-7364-1-s1-a32},
}
Figure