Personalized modeling of human gaze: Exploratory investigation on mammogram readings

Eye tracking studies in medical imaging typically focus on studying radiologists' visual search process and how it relates to the clinical interpretation task at hand. In this pilot study, we have investigated gaze patterns to gain insight into their association with radiologists' expertise level as well as the presence of individual differences to facilitate personalized modeling and recognition of radiologists. First, we collected gaze data from six radiologists viewing 40 mammographic images each. Then, the collected gaze data were analyzed with two different approaches: 1) using a multilayer perceptron and 2) using a hidden Markov model. Both approaches confirmed that the experience level of a radiologist can be inferred with high accuracy by simply studying their gaze pattern. Personalized modeling and identification of radiologists was successful with both approaches with accuracy significantly higher than random guessing. The results of this pilot study confirm that a radiologist's perceptual behavior is not only a function of clinical training and level of experience, but there are individual aspects that could serve as a personal biomarker when developing models of human perception and cognition in medical image interpretation.