Real-time Dose Computation for Radiation Therapy Using Graphics Processing Unit Acceleration of the Convolution/Superposition Dose Computation Method

C10271: Real-time Dose Computation for Radiation Therapy

Value Proposition:

This invention describes a method to obtain real time dose calculation through the use of graphics processing unit (GPU) to accelerate the convolution superposition dose calculation algorithm. This algorithm is commonly used in treatment planning systems due to its accuracy in the presence of tissue inhomogeneities, but its traditional implementation is not suited for the pipeline nature of the GPU. The enhanced performance of this method also facilitates quicker inverse method radiotherapy (IMRT) planning, enabling beam direction optimization as well as intensity modulation optimization. The main component of this invention is the restructuring of the convolution superposition algorithm for GPU based acceleration. The S/C algorithm was implemented using an inverse cumulative-cumulative kernel, as forward kernels suffer from parallel read-write conflicts. Ray directions were chosen using a uniform azimuth and non-uniform zenith sampling. In experimenting with multiple sampling methods (Figure 1), "Naïve" sampling visits every voxel along a ray. Alternatively a "Varable Step" length may be taken as individual contrbution from distant voxels is small. A novel "Multi-resolution" method is introduced, which uses an array of sub-sampled total energy released per unit mass (TERMA) volumes (i.e. a volumetric mip-map) to approximate each ray as a true solid angle instead of a line. This is achieved by stepping to a coarser resolution as distance increases, which effectively increases the ray width.

Technical Details:

Radiation therapy treatment planning utilizes a computed tomography (CT) scan of the patient and the geometric and dosimetric characteristics of a medical linear accelerator to determine the appropriate beam directions and apertures to deliver high doses of radiation to targeted volumes and minimize dose to critical or avoidance regions. Accurate dose computation for these systems often takes up to a minute per beam forcing the user to wait after small changes to review the resulting dose from the change. Thus, an unmet need exists to all for real-time dose computation during treatment planning of radiation therapy treatments while maintaining accuracy in the presence of tissue inomogeneities. Such a performance enhancement would improve the quality of treatment plans by enabling greater versatility in the treatment planning process. This would provide the added capabilities of inverse planning of beam directions while optimizing intensity modulation. Treatment planning efficiency would thereby be improved allowing patients to begin therapy at an earlier stage.