Assessing Viewing Pattern Consistency in Mammogram Readers

Abstract

Breast cancer screening programs typically require very large volumes of x-ray images (mammograms) to be viewed by highly experienced human readers. The readers can recognise a wide range of different visible features indicative of clinically abnormal situations, which they use as a basis to generate a report on their findings. Errors in reporting can occur if the readers fail to identify a particular feature of interest for further visual inspection during the viewing process. This risk is typically reduced by training readers to follow a particular viewing path through an image, which they should be able to apply consistently. Knowledge of the extent of consistency in this viewing behaviour within and between viewers would inform the development of an automated checking approach, based on monitoring of viewer visual attention. This paper presents an analysis of some reader viewing pattern profiles obtained using eye tracking with an infra red computer vision system, as a basis for developing a suitable consistency assessment model. .