The use of conspicuous visible features in medical images for objective assessment of image quality

We developed and assessed the usefulness of an objective image quality measure that is correlated with perceived image quality, as a function of the most conspicuous features contained within an image. Our salience measure is useful for identifying image features that are particularly conspicuous across a range of spatial scales. This new measure was applied to medical images; salience features within mammograms were exclusively studied.;The notion of feature salience was thoroughly investigated. We developed a scale-based salience model to assess low-level visual features that are contained in digitized mammograms. We found these features to be correlated with observer performance during detection and discrimination tasks. Two studies were performed that correlated the salience measure with task-dependent observer performance. In the first study, we used eye-position data to develop a model that relates the salience measure to the time required to fixate a mass. The model is useful for predicting time-to-first-hit in less than two seconds, but can only be applied to trained readers. In the second study, we found that the salience measure was useful for discrimination of true positive, true negative and false positive cases across different levels of reader experience.;Using feature salience, we developed a quantitative measure of breast density as a percentage of the whole breast. Our measure is semi-automatic and shows good correlation with the adaptive thresholding technique. Adaptive thresholding is a popular percentage breast density measurement; however, more reader interaction is required than is needed for our technique.;We developed a new method for quantifying case difficulty for use in the training and testing of CAD algorithms using a small number of cases. Using a large data set, we clustered cases into groups of perceived difficulty as quantified by our salience measure. We showed that test subsets can be objectively constructed and related to reader perception of case difficulty.