A model for predicting pathologist's velocity profiles when navigating virtual slides

Navigation through large microscopic images is a potential benefit for histology or pathology teaching, for improving the quality of diagnosis in pathology, or for communicating pathologists in some telemedicine applications. However, the size of this kind of images is prohibitive for navigation with conventional techniques. This article presents a soft computing model, which permits to anticipate the pathologist trajectories in diagnosis tasks when exploring virtual slides. The Bayesian strategy combines an offline model of a baseline pathologist knowledge (the prior) and a prediction online module (the likelihood) that captures a particular pathologist navigation pattern. While optimal parameters for the biologically inspired offline model are calculated using an Expectation-Maximization strategy, prediction is carried out by a particle filter. Parameters are estimated from several series of actual navigations performed by several pathologists in different virtual slides. The present approach is compared with other conventional prediction methods and decreases the calculated MSE in about a 50% for the entire group of pathologists.