Wei Chen MEng, Dipl Eng, Presenter: Nothing to Disclose

Daniel Kolditz, Abstract Co-Author: Employee, Artemis Imaging GmbH

Martin Hupfer DIPLPHYS, Abstract Co-Author: Employee, Artemis Imaging GmbH

Yulia Smal MSc, Abstract Co-Author: Nothing to Disclose

Willi A. Kalender PhD, Abstract Co-Author: Consultant, Siemens AG Consultant, Bayer AG Founder, CT Imaging GmbH Scientific Advisor, CT Imaging GmbH Shareholder, CT Imaging GmbH Founder, Artemis Imaging GmbH CEO, Artemis Imaging GmbH Shareholder, Artemis Imaging GmbH

Monte Carlo simulation is an established technique for dose calculation in diagnostic radiology. The main drawback is its high computational cost. The aim of this study was to develop a real-time Monte Carlo software tool for patient-specific dose calculations using graphics processing units (GPUs).

GPUs are powerful systems which are especially suited to problems that can be expressed as data-parallel computations. In Monte Carlo simulations each photon track is independent of the others; each launched photon can be mapped to one thread on the GPU, thousands of threads are executed in parallel in order to achieve the high performance. For a further comparison we considered multi-CPU and multi-GPU variants. The total computation was divided into different parts that can be calculated in parallel on multiple devices. In addtion, to validate the concept we tested it using the software ImpactMC (CT Imaging GmbH, Germany). Measurements were performed on a 64-slice CT system. The dose was measured using calibrated ionization chambers and thermoluminesence dosimeters (TLDs) in standard cylindrical CTDI phantoms. To estimate the performance gain we benchmarked dose calculation time on a 2.6 GHz Intel Xeon 5430 Quad core PC equipped with two NVIDIA GTX 285 cards. A 32 cm diameter CTDI phantom of (1.0 mm3) voxelsize was simulated. 1.8x109 photon histories were simulated resulting in a statistical accuracy of better than 1%.  

The dose values and distributions obtained by MC simulation were in good agreement with the CTDI measurements and the reference TLD profiles with differences less than 5%. Calculation times of about 5 hours and 1.6 hours were necessary with single-CPU and multi-CPU solution, respectively. Using GPUs the same MC calculations were done in 5 minutes (single card) and 2.7 minutes (two cards) without compromising quality. Simulations can thus be speeded up on GPUs up to 60 times and 35 times compared to single- and multiple-core CPUs, respectively.

The Monte Carlo software package provides modeling of 3D dose distributions in real time which makes it a practical tool for routine applications.

The acceleration provides the potential for real-time dose calculations. MC calculations offer a fast and flexible solution to assess patient-specific dose distributions in clinical practice.

Chen, W, Kolditz, D, Hupfer, M, Smal, Y, Kalender, W, Implementation and Validation of a Real-time Monte Carlo Software Tool for Patient-specific Dose Calculations.  Radiological Society of North America 2011 Scientific Assembly and Annual Meeting, November 26 - December 2, 2011 ,Chicago IL. http://archive.rsna.org/2011/11004865.html