Region of interest identification in collimated x-ray images utilizing nonlinear preprocessing and the Radon transform

Abstract

The identification of the region of interest (ROI) in collimated x-ray images remains an important, open problem. The x-ray scattering present in all practical cases complicates the identification of a clearly defined ROI. Such scattering degrades image quality by producing a blurring of the boundaries delineating the ROI as well as corrupting the underlying gray-scale histogram. The construction of a proper identification scheme to locate the ROI in x-ray images will result in x-rays with improved visual clarity. A robust two-stage method of nonlinear preprocessing in the spatial domain to enhance the ROI boundary followed by extraction in the projection domain is proposed as an efficient means of ROI identification. The method proves effective for a wide variety of x-ray images taken with varying dosages. The results of simulations run in comparison with commonly used snake-based methods, which attempt to identify arbitrary ROIs using force on control points, show the proposed method yields better success rates in high- and low-SNR cases.