Emma Elisabeth Engstrom BS, Abstract Co-Author: Nothing to Disclose

Robert Mark Nishikawa PhD, Presenter: Shareholder, Hologic, Inc, Bedford, MA Royalties, Hologic, Inc, Bedford, MA Research funded, Hologic, Inc, Bedford, MA Consultant, Fuji Photo Film Co, Ltd, Stamford, CT Scientific Advisory Board, Dexela Limited, United Kingdom

Ingrid Reiser PhD, Abstract Co-Author: Nothing to Disclose

To compare the breast structure power spectrum for mammograms, tomosynthesis projection images and tomosynthesis reconstruction images.

Burgess has shown that the breast structure power spectrum follows a power law of C/fB, where f is the radial spatial frequency, C determines the magnitude of the power spectrum and B has values between 2.0 and 4.0. His investigations indicate that B characterizes the amount of breast structure in a mammogram. Our evaluation was done on two data sets: 61 digitized screen-film mammographic exams and tomosynthesis projections and reconstructions (ML-EM) of 28 breasts. The different cases covered a range of breast thicknesses and different amounts of breast structure. For each projection and reconstructed slice, we selected a number of ROIs (12.8 mm x 12.8 mm) from the central portion of the image. The power spectrum of each ROI was determined by taking the square of the Fourier transform of the ROI. We were careful to avoid any obvious reconstruction artifacts, the skin line and the pectoralis muscle in the ROIs. To determine the value of B, we fit a first order polynomial to a log-log plot of power versus spatial frequency. The fits were performed using the frequency range [0.2, 1] cycles/mm for the mammogram ROIs, and [0.2, 0.7] cycles/mm for the projection and reconstruction ROIs. Mean B was computed for each breast.

The 61 mammogram B means had average 2.76 (standard dev. 0.19). For the 28 projections it was 2.79 (0.19), while it was 2.69 (0.26) for the corresponding reconstructions. The difference in B between the projection ROIs and the reconstructed ROIs averaged across the 28 cases was 0.093, which was statistically significant (p = 0.0002). The 95% CI for the difference between the mean value of B for the projections and reconstructions was [0.048, 0.138].

When compared to the projection images, the corresponding tomosynthesis reconstructions lowers the average B by 0.093. In the future we will examine if B and C can be used to optimize a DBT system for CAD.

Our results are consistent with the observation that DBTs improve detectability compared to mammography because of reduced breast structure.

Engstrom, E, Nishikawa, R, Reiser, I, Comparison of the Breast Tissue Power Spectrum for Mammograms, Tomosynthesis Projection Images, and Tomosynthesis Reconstruction Images.  Radiological Society of North America 2007 Scientific Assembly and Annual Meeting, November 25 - November 30, 2007 ,Chicago IL. http://archive.rsna.org/2007/5005011.html