Digital mammography is a key technology for improving breast cancer detection and diagnosis. Current soft-copy display technology is the weakest link in determining image quality in digital mammography, and is inhibiting the realization of the full potential of the modality. This is a collaborative venture between the University of Pennsylvania and Sarnoff Research Laboratory. It seeks to apply the powerful Visual Display Model (VDM) software, developed at Sarnoff over the last 20 years, to the problem of workstation optimization for digital mammography. The goal of the proposed work is to develop a reliable, cost-effective and sensitive methodology for predicting and optimizing the performance of mammographers using soft-copy displays. This investigation is focused on microcalcification related tasks: detection (screening mammography) and discrimination (diagnostic mammography). The applicants propose to perform state-of-the-art extensions of the VDM to allow for masking effects of normal background, the effects of zoom, and effects peculiar to Liquid Crystal Display (LCD) monitors. The applicants will develop novel evaluation methodology, which will enable application of the VDM to the soft-copy optimization problem without the need for expensive Receiver Operating Characteristic (ROC) studies. The VDM model will be tested with observer performance data acquired under a wide range of display conditions. The applicants will predict optimal window and level settings for individual images - thereby enabling subsequent work on its implementation. The predictions of the VDM model will be tested with several validation studies. Specifically, the model will predict if superior performance can result from an alternative technology LCD display vs. state-of-the-art CRT technology. The predictions of the model will be tested with a clinical ROC study comparing a CRT and a LCD monitor in microcalcification detection and discrimination tasks. Also to be generated are iso-contour plots of the effects of common workstation variables on image quality in both tasks. This will allow digital mammography workstations to be more intelligently used.
