A Hidden Markov Model to Identify Regions of Interest from Eye Movements, with an Application to Nodule Detection in Chest X-Rays

Nodules that may represent lung cancer are often missed in chest X-rays by radiologists. Recording of eye movements during search for nodules provides insights into the search process. We develop a Hierarchical Bayes Hidden Markov Model (HMM) and analyze eye tracking data of sixteen laymen looking at fourteen chest X-rays, of which seven contained a potentially cancerous nodule. We use the luminance of pixels in the X-ray image as prior information on the location of a nodule. Using a reversible jump Markov Chain Monte Carlo algorithm to estimate the HMM enables us to identify the number of regions of interest (ROIs) on each image, as well as their centers, sizes and orientations. In the application in most cases one of the ROIs covers the nodule precisely. Our study thus demonstrates that a HMM analysis of eye movements recorded on laymen may accurately reveal the location of nodules, which has significant implications. The HMM model may be useful in other applications to identify ROIs from eye movement data