Computer-Assisted Visual Search/Decision Aids as a Training Tool for Mammography

摘要

This project focuses on the perceptual training of diagnostic interpretation skills in mammography which are acquired mainly as a result of experience reading mammograms. The primary aim of this project is to develop a computer-assisted mammography training tool that will act as a surrogate mentor in aiding radiologists in making plausible diagnostic decisions. We propose to provide a computer aid that will interact with the radiologist immediately after image interpretation by providing systematic feedback about how the mammogram was searched for abnormalities and what features received prolonged visual attention indicating potential lesions during scanning. The eye-position parameter, visual dwell, is used to predict the locations of suspicious lesions on the mammogram (Krupinski, Nodine, Kundel, 1998; Nodine, Kundel, Mello-Thoms et al., 1999). The radiologist is then asked to re-examine the entire Image including the highlighted areas, determine if any abnormal features are present, and re-evaluate the original diagnostic decision. This re-evaluation of suspicious regions with visual feedback provides a perceptually-guided basis for a plausible problem-solving diagnostic solution. We showed in 1990 (Kundel, Nodine, Krupinski, 1990) that computer-assisted visual search (CAVS) is effective in improving the detection of lung nodules, and Krupinski (1996) showed that visual dwell predicts the location of true and false, positive and negative decision outcomes. Our goal is to determine if CAVS improves the detection and interpretation of breast cancers.