Gaze and visual scanpath features for data-driven expertise recognition in medical image inspection

Abstract

Expert medical professionals must visually examine medical images (MRI and CT scans, radiographs, ultrasounds etc.) with the utmost concern for a patient’s health. Developing the perceptual abilities to distinguish an atypical shadow from an anatomical structure involves considerable training and time. Although students view a multitude of these images in their studies, often, they must receive further supervision upon entering their residencies or even early on in their careers. This current approach can exhaust expert resources allocated for supervision and leaves room for error. This thesis sets out to investigate the gaze behavior as an effective tool for expert and novice anomaly recognition, specifically in the context of dental image inspection (Technical term: orthopantomograms, or OPTs). Our ability to go deeper into the predictive aspect of scanpath analysis makes our research truly innovative. Much of the current literature regarding experts and novices has found that domain specific tasks evoke different eye movements. However, research has yet to predict these behaviors and guide students towards expert behavior strategies. More important, advanced pattern recognition and analysis algorithms have not yet been employed to identify and quantify differences in the visual search strategy between advanced learners, residents, and expert practitioners. The potential to integrate expertise model development from scanpath features into intelligent tutoring systems is the ultimate inspiration for our research. This novel approach to training dentistry students with gaze-based learning environments can offer insight into the training of students in other medical domains. Currently, the training of OPT interpretation in dental students exhibits a deficit of systematic learning approaches and can vary between universities. Moreover, there are no known user-aware intervention techniques that address the improvement of image reading performance in students or advanced learners. By employing machine learning-based scanpath classification, we found features in the gaze indicative of expertise and expert cognitive processes. We were also able to distinguish gaze behavior related to a student’s level of understanding. The culmination of these findings provide support for a robust classification algorithm we developed to extract semantic features of the gaze and cluster experts and novices based on feature similarities in the scanpath with high accuracy.