Abstract
Purpose: In vivo dosimetry using an electronic portal imaging device (EPID) may provide a uniquely effective means for preventing errors and may allow for real‐time beam interlocks. We present graphical processor (GPU)‐based fast EPID dose computation and simulate the sensitivity for detecting errors caused by patient mispositioning. Methods: We implemented an enhanced superposition convolution algorithm to compute the EPID dose. The patient was simulated using a digital cylinder phantom (15cm length and 15cm diameter) with air and bone structure built in. Setup parameters are: SAD 100cm, 12×12‐cm2 open field, 1.5×1.5×1.5‐mm3 dose grid, 6‐MV photon and isocenter at the center of cylinder. 2D dose images were obtained at 1.6‐cm depth of a water slab positioned 60‐cm below the isocenter to simulate the EPID readout. To examine errors caused by mispositioning, we shifted the phantom left‐and‐right and up‐and‐down and compared dose profiles to those with no shifts. Gamma value of each pixel was calculated as an index to identify dose mismatch. Dose failure was determined as gamma > 1 using 3‐mm distance‐to‐agreement and 3% dose difference criteria. Results: Dose failures occur mainly in air‐tissue and bone‐tissue interfaces and in regions with rapid thickness change. Gamma is sensitive to the left‐and‐right shift, with a 3‐mm shift causing dose failure in ∼10% pixels; but is much less sensitive to the up‐and‐down shift, with dose failure detected only after a shift of several centimeters. Analysis with patient CT data sets is underway and will be presented. Conclusions: In vivo EPID dosimetry enabled by GPU‐based dose computation offers the potential for real‐time beam interlocks. Future directions include simulation of other error scenarios, optimization of dose computation, improved EPID modeling and comparison with measured EPID profiles.
