This work derives and assesses the usefulness of an objective image quality measure. The measure is correlated with perceived image quality as a function of the most salient features contained within a given image. They are determined by combining aspects of both visual discrimination theory and signal detection theory to define a new measure that quantifies the importance of contrast-based features as a function of spatial frequency. We discuss the development of a perceptually-correlated metric that is useful for quantifying the conspicuity of local, low-level or bottom-up visual cues, and the identification of those spatial frequencies that are most distinct and perhaps most relied upon by radiologists for decision-making. A parsimonious analysis of variance model is developed that accounts for the variance in the salience metric. The model is generalizable to a population of readers and to a population of cases. This work has application to the development of techniques to quantitatively assess breast density, to classify radiographic parenchymal patterns in mammograms, and to optimize compression techniques for task-based performance. The salience measure can be used also to assess data set difficulty for use in development and testing of computer-assisted-diagnosis algorithms, and to determine conspicuous regions-of-interest, which can be used to identify regions for higher compression.

Prof. Murray H. Loew
After receiving the Ph.D. in Electrical Engineering from Purdue University in 1972, Murray Loew spent six years in industry. He then joined the Department of Electrical and Computer Engineering at George Washington University, Washington, D.C., where he is now professor and director of the Biomedical Engineering program. His research interests include medical image processing, analysis, and compression, and pattern recognition. Recent work has included the derivation of measures of noise induced by image compression, and development of medical image registration methods and techniques for evaluating their performance in the absence of reference data. Prof. Loew is a Fellow of the IEEE and of AIMBE.