A method for removing arm backscatter from EPID images

To develop a method for removing the support arm backscatter from images acquired using current Varian electronic portal imaging devices (EPIDs).

The effect of arm backscatter on EPID images was modeled using a kernel convolution method. The parameters of the model were optimized by comparing on-arm images to off-arm images. The model was used to develop a method to remove the effect of backscatter from measured EPID images. The performance of the backscatter removal method was tested by comparing backscatter corrected on-arm images to measured off-arm images for 17 rectangular fields of different sizes and locations on the imager. The method was also tested using on- and off-arm images from 42 intensity modulated radiotherapy (IMRT) fields.

Images generated by the backscatter removal method gave consistently better agreement with off-arm images than images without backscatter correction. For the 17 rectangular fields studied, the root mean square difference of in-plane profiles compared to off-arm profiles was reduced from 1.19% (standard deviation 0.59%) on average without backscatter removal to 0.38% (standard deviation 0.18%) when using the backscatter removal method. When comparing to the off-arm images from the 42 IMRT fields, the mean γ and percentage of pixels with γ < 1 were improved by the backscatter removal method in all but one of the images studied. The mean γ value (1%, 1 mm) for the IMRT fields studied was reduced from 0.80 to 0.57 by using the backscatter removal method, while the mean γ pass rate was increased from 72.2% to 84.6%.

A backscatter removal method has been developed to estimate the image acquired by the EPID without any arm backscatter from an image acquired in the presence of arm backscatter. The method has been shown to produce consistently reliable results for a wide range of field sizes and jaw configurations.