The h int 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 h int representation of a mammogram. We investigate the major contributions to noise and the steps in the h int 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 h int 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 h int representation is verified using experimental results on phantom images and mammograms with microcalcifications.
[1]
Michael Brady,et al.
A Representation for Mammographic Image Processing
,
1995,
CVRMed.
[2]
S. Webb.
The Physics of Medical Imaging
,
1990
.
[3]
Robert M. Nishikawa,et al.
Characterization of the mammographic appearance of microcalcifications: applications in computer-aided diagnosis
,
1993
.
[4]
Jong-Sen Lee,et al.
Digital Image Enhancement and Noise Filtering by Use of Local Statistics
,
1980,
IEEE Transactions on Pattern Analysis and Machine Intelligence.
[5]
G T Barnes,et al.
Radiographic mottle: a comprehensive theory.
,
1982,
Medical physics.