|Research Laboratory for Computer Graphics & Virtual/Augmented Reality (LIRV)||[ Français ]|
GPU Acceleration of Dose Deposition Calculation for Radiotherapy and Brachytherapy
( #simulations physiques, #calculGPU )
This research in computer engineering, undertaken since 2008, focuses on dose calculation when developing a treatment plan in radiotherapy. The quality of the calculations depends on various quantities which, among other things, the actual density of the patient to be treated (previously acquired by axial tomography medical imaging) and the modeling of the irradiation apparatus used to evaluate the behavior of the photon beam at the output. The integration of all these information leads to calculations that can take considerable time. So we developed a new software platform using the power of graphics cards (GPU) to perform very quickly highly parallel calculations.
Radiation therapy is one of the possible treatments for attempt the elimination of cancerous tissue. The methodology consists of direct ionizing radiation, most often emitted by an accelerator linear to cancerous tissue while avoiding as much as possible healthy tissues and organs.
Calculating the received dose is at the heart of treatment planning and aims to find the best configuration of ionizing beams for respond to the prescription, in terms of tissue dose, developed by the radiation oncologist. These calculations involve the physical principles of energy transport in the material as well as information on the composition of the medium previously obtained by medical imaging. At the dose calculation step, several algorithms can be used, of the fastest, making several approximations as to the nature of the medium and particle-matter interactions, to the most accurate ones that make the compromise of a long calculation time for fidelity to the principles physical and augmented environment. In the family of fast techniques, we find the commonly used methods of pencil beam and convolution superposition. For the most accurate methods, it includes stochastic solutions (Monte Carlo simulation) and deterministic of coupled Boltzmann equations. Differences of more 10% can be observed when comparing algorithms accurate and fast.
The current version of the resulting software running on GPU tends to show some accuracy equivalent to existing versions while being 250x to 1200x faster.
Highly Qualified Personnel Training (HQP) :