A region-based Retinex with data filling for the enhancement of electronic portal images

Abstract

PurposePortal images are acquired by electronic portal imaging devices (EPID) with megavoltage (MV) x-ray, but they are inherently poor in terms of contrast, due to Compton Effect. In comparison with diagnostic x-ray images, portal images usually lack sufficient detail information for normal human vision. Therefore, an effective method of enhancing these images would be very useful. This paper proposes a new approach that combines global and local enhancement techniques. Materials and methodsA portal image usually has a high dynamic range (HDR) of up to 16bits, so it could records details that are imperceptible to the naked eye. However, this property provides the potential for enhancement of the portal image. In order to overcome the low contrast appearance caused by innate physical properties, two phases and four sequential steps were proposed. At phase 1, global enhancement, HE is used to stretch narrow range histogram of original raw image to reasonable wide range so that we can easily partition the image into regions for local enhancement. At phase 2, local enhancement, EPIs were first segmented into regions based on histogram distribution. Then a new concept of local enhancement, pseudo-data filling, in which enhancement is controlled by manipulating the pseudo-data, is proposed in order to maximize the regional enhancement. Finally each region of EPI is enhanced by Retinex with optimized parameter and synthesized as output image. ResultsAt phase 1, HE can successfully improve EPIs contrast at varies body sites by redistribution histogram. This step provides possibility of histogram analyzing at phase 2. Therefore, histogram-based segmentation is feasible for nearly every patient as we expected. Simulation of pseudo-data filling and region-based Retinex enhancement demonstrate that the proposed method provides a more detailed portal image, which is proved by objective evaluation of two groups of radiation oncology staffs. ConclusionsAn effective enhancement method was developed, which combines the use of both global and local algorithms for the enhancement of electronic portal images. The two contrast enhancement algorithms; HE and Retinex, were integrated to produce enhanced images with a strong contrast.