Characterization of mammographic images from a signal processing point of view

Candidate level (15 HP)

Area: Automated image analysis in medical physics

When developing new x-ray imaging techniques the work in early phases rests heavily on computer simulations. While the imaging chain, from x-ray source to detector and image reconstruction, is relatively easy to implement on standard PC’s, the physiological characteristics of the object (patient) being imaged are more difficult to capture. Even for such a seemingly homogenous imaging task as the female breast, the focus of our mammography unit, there exists no general reference work as to general frequency domain behavior of anatomical noise in the images.

The anatomical structures of the breast, mainly glandular tissue embedded as lobes attached with Cooper’s ligaments in a surrounding volume of adipose tissue, result in a characteristic cloudy appearance of the mammography image. These anatomical structures are best analyzed in the Fourier domain where they appear as an A*f^(-B) distribution (f being the frequency).

For some imaging tasks, this slowly varying 1/f noise is secondary to quantum noise which appears similar at all frequencies. For some imaging task however, specifically the non-contrast enhanced dual energy method developed in the group which enhances the detectability of lesions in digital mammography by as much as 50% under certain circumstances (PhD-thesis 2009, Erik Fredenberg), this type of structured noise is dominant.

This candidate level thesis encompasses developing a (Matlab) program that analyzes a large number (>1000) randomly selected digital mammograms. The program should estimate the fraction glandular tissue automatically from the image and determine the coefficients A and B of its Fourier power spectrum. Any possible dependency on glandularity or interdependency between A and B need to be statistically evaluated. This work will then serve as reference work when image construction algorithms designed to suppress anatomical noise are developed.