Date(s) - 03/29/2012
2:00 pm - 4:00 pm

Stephanie Leon, BME-Medical Physics PhD Student

Chair: Libby Brateman

    Mammography is a successful screening mechanism that has saved many lives, but there is still room for improvement, particularly in the rate of false negative exams. The detection of cancers could be improved if the image quality were improved beyond that which is currently available. One of the major limitations of breast imaging is a poor contrast among breast tissues, which is further reduced by the presence of scattered radiation. In current clinical practice, the impact of scattered radiation is reduced by use of an antiscatter grid, but the grid increases the absorbed dose to the breast by a factor of 2-3 and does not remove all the scatter.
    This aim of this study was to reduce the presence of scatter in digital mammography by developing a point spread function (PSF), which is a way to describe scatter mathematically, and subsequently using it to deconvolve the PSF from the image. The PSF developed was a function of target/filter combination, tube potential, tissue thickness, and the presence or absence of a grid. The parameters used to describe the PSF, scatter fraction and mean radial extent, were obtained with physical measurements and verified with Monte Carlo simulations.
    A computer program was written to deconvolve the scatter from raw (unprocessed) mammography images. To account for variations in tissue thickness throughout the image, the apparent thickness of the tissue over each pixel was calculated, and the deconvolution was carried out on a pixel-by-pixel basis. The image quality before and after the deconvolution was assessed by measurement of the noise power spectrum (NPS), contrast-to-noise ratio (CNR), and resolution. The image noise decreased after scatter correction by about 72-87% with a grid and 78-93% without a grid, depending on frequency. The CNR improved by 42-71% with a grid and 28-126% without a grid. The resolution appeared unchanged. These results indicated not only the potential to improve clinical image quality using a grid, but also a possibility to remove the grid and retain image quality equal to or better than that currently available with a grid.
    IRB approval was acquired to collect the raw data from 100 clinical images obtained by routine mammography, process them with the scatter correction program, and use these images in a reviewer assessment study. To date, the program has been tested on several of these images, with promising results. The next step in this project is to apply the scatter correction program to a sample of the clinical images and to have the clinical image quality assessed by radiologists, to determine whether the measured changes in image quality translate to improved clinical images. A reader preferences study will be conducted to compare mammograms before and after scatter correction, in which three MQSA-qualified radiologists will rate their preference for the corrected or uncorrected image in the following categories: detection of the pectoral muscle, visibility of the nipple, visibility of the skin, sharpness, contrast, visibility of microcalcifications, visibility of masses, and overall image quality. The statistical significance of the results in each category will be assessed using the one-sample Wilcoxon signed rank test.