SU‐E‐J‐16: Automatic Image Contrast Enhancement Based On Automatic Parameter Optimization for Radiation Therapy Setup Verification

Abstract

Purpose:In RT patient setup 2D images, tissues often cannot be seen well due to the lack of image contrast. Contrast enhancement features provided by image reviewing software, e.g. Mosaiq and ARIA, require manual selection of the image processing filters and parameters thus inefficient and cannot be automated. In this work, we developed a novel method to automatically enhance the 2D RT image contrast to allow automatic verification of patient daily setups as a prerequisite step of automatic patient safety assurance.Methods:The new method is based on contrast limited adaptive histogram equalization (CLAHE) and high‐pass filtering algorithms. The most important innovation is to automatically select the optimal parameters by optimizing the image contrast. The image processing procedure includes the following steps: 1) background and noise removal, 2) hi‐pass filtering by subtracting the Gaussian smoothed Result, and 3) histogram equalization using CLAHE algorithm. Three parameters were determined through an iterative optimization which was based on the interior‐point constrained optimization algorithm: the Gaussian smoothing weighting factor, the CLAHE algorithm block size and clip limiting parameters. The goal of the optimization is to maximize the entropy of the processed Result.Results:A total 42 RT images were processed. The results were visually evaluated by RT physicians and physicists. About 48% of the images processed by the new method were ranked as excellent. In comparison, only 29% and 18% of the images processed by the basic CLAHE algorithm and by the basic window level adjustment process, were ranked as excellent.Conclusion:This new image contrast enhancement method is robust and automatic, and is able to significantly outperform the basic CLAHE algorithm and the manual window‐level adjustment process that are currently used in clinical 2D image review software tools.