Comprehensive Monte Carlo calculation of the point spread function for a commercial a-Si EPID

Images produced by commercial amorphous silicon electronic portal imaging devices (a-Si EPIDs) are subject to multiple blurring processes. Implementation of these devices for fluence measurement requires that the blur be removed from the images. A standard deconvolution operation can be performed to accomplish this assuming the blur kernel is spatially invariant and accurately known. This study determines a comprehensive blur kernel for the Varian aS500 EPID. Monte Carlo techniques are used to derive a dose kernel and an optical kernel, which are then combined to yield an overall blur kernel for both 6 and 15 MV photon beams. Experimental measurement of the line spread function (LSF) is used to verify kernel shape. Kernel performance is gauged by comparing EPID image profiles with in-air dose profiles measured using a diamond detector (approximating fluence) both before and after the EPID images have been deconvolved. Quantitative comparisons are performed using the chi metric, an extension of the well-known y metric, using acceptance criteria of 0.0784 cm (1 pixel width) distance-to-agreement (deltad) and 2% of the relative central axis fluence (deltaD). Without incorporating any free parameters, acceptance was increased from 49.0% of pixels in a cross-plane profile for a 6 MV 10 x 10 cm2 open field to 92.0%. For a 10 x 10 cm2 physically wedged field, acceptance increased from 40.3% to 73.9%. The effect of the optical kernel was found to be negligible for these chi acceptance parameters, however for (deltaD= 1%, deltad = 0.0784 cm) we observed an improvement from 66.1% (without) to 78.6% (with) of chi scores <1 (from 20.6% before deconvolution). It is demonstrated that an empirical kernel having a triple exponential form or a semiempirical kernel based on a simplified model of the detector stack can match the performance of the comprehensive kernel.