MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel
  1. Home
  2. Meeting Program
  3. All
  4. GPU-Based Acceleration of FastEPID Image Simulation

GPU-Based Acceleration of FastEPID Image Simulation

M Shi1,2*, M Myronakis2 , M Jacobson2 , D Ferguson2 , C Williams2 , I Valencia Lozano2 , T Harris2 , M Lehmann3 , P Huber3 , R Fueglistaller3 , P Baturin4 , D Morf3 , R Berbeco2 , (1) University of Massachusetts Lowell, Lowell, MA, (2) Brigham and Women's Hospital, Dana-Farber Cancer Institute, and Harvard Medical School, Boston, MA, (3) Varian Medical Systems, Baden-Dattwil, Switzerland, (4) Varian Medical Systems, Palo Alto, CA

Presentations

(Wednesday, 7/17/2019) 10:30 AM - 11:00 AM

Room: Exhibit Hall | Forum 6

Purpose: Long computation time has hindered the use of Monte Carlo (MC) for image simulation. We have previously validated FastEPID, a novel method to significantly speed up image simulation. In this work, we have developed a graphic processing unit (GPU) based FastEPID technique to further shorten simulation time and enable local simulations.

Methods: The FastEPID method has been previously executed on central processing unit (CPU), simulating photons one after another. Detection of the photon incident on EPID surface is recorded if a newly generated random number is less than the photon energy deposition efficiency (η). The corresponding optical photon spread function (OSF) of the detected photon is then added to EPID image with center aligned to the incident position. OSFs and η of various photon energies are pre-calculated with a validated EPID model, and scaled based on photon energy during simulation. This method was implemented on GPU utilizing its parallel computation ability to improve simulation efficiency. The GPU-based simulation of a Las Vegas (LV) phantom was validated against the CPU-based simulation, while the mega-voltage cone beam CT (MVCBCT) images of an electron density phantom reconstructed from projections simulated by GPU- and CPU-based FastEPID were compared. Simulation time utilizing GPU was compared to the CPU simulation.

Results: The GPU-based simulation provided similar image quality with CPU-based. Contrast-to-noise ratio of LV phantom image and the overall performance of MVCBCT reconstruction showed excellent agreement with each other and with measurement. The GPU-based FastEPID has shortened simulation time by a factor up to 40.

Conclusion: Simulation of an EPID image can be performed in matter of hours or less with a single GPU card. Further acceleration for advanced applications will be performed by the addition of more GPU cards. Novel imager development and image simulation for clinical applications will be aided by this innovation.

Funding Support, Disclosures, and Conflict of Interest: NIH/NCI R01CA188446

Keywords

Simulation, Flat-panel Imagers, Cone-beam CT

Taxonomy

IM/TH- RT X-ray Imaging: CBCT imaging/therapy implementation

Contact Email

Additional Links