De-noising hint surfaces: A physics-based approach

Abstract

The hint representation is a normalised, quantitative version of a mammogram which has substantial quantum noise components because of the way in which it is computed. This paper presents a physics-based approach to de-noising the hint representation of a mammogram. We investigate the major contributions to noise and the steps in the hint generation that amplify noise, such as removal of intensifying screen glare. Estimating the radiographic noise components using parameters derived from physics models, we filter the original mammographic images with an adaptive wiener filter, W. Generating the hint representation from the filtered images yields a de-noised version which has substantially improved signal-to-noise ratio, and which is far better to use for further-processing, such as microcalcification detection. The accuracy of the de-noised hint representation is verified using experimental results on phantom images and mammograms with microcalcifications.