Observers screening mammograms for breast cancer, are estimated to have a sensitivity of about 75 percent and a specificity of about 95 percent. A 25 percent miss rate is a clinically significant level of error that deserves study to find the causes and to develop methods for reducing error. While studying observers searching chest images for lung nodules, we observed that unreported nodules received prolonged eye-fixation dwell time. This phenomenon of perceptual attraction to occult lesions has been used to provide visual cues for revising decision making. Eye-position recording during the initial viewing is used to identify areas that receive prolonged visual dwell time. These areas are immediately circled on the image and the observer is instructed to re-examine the cued areas and revise his or her initial decision. A controlled study using receiver operating characteristic (ROC) methods have shown an absolute improvement in detection performance. We propose to verify the psychophysical relationship between visual dwell time and the detection of clinically-occult breast tumors as measured by decision outcome, and to extend the basic observations made using pulmonary nodules to breast lesions. The value of perceptual cues for improving the detection of cancer in screening mammograns will be evaluated. The main purpose of this research is to determine how to effectively use programmable soft copy displays in conjunction with perceptual feedback to decrease observer error rates during the screening of mammograms for cancer.
Showing the most recent 10 out of 12 publications