Overlooking: The nature of gaze behavior and anomaly detection in expert dentists Nora Castner Perception Engineering, University of Tübingen Tübingen, Germany castnern@informatik.uni-tuebingen. de Solveig Klepper Computer Science Institute, University of Tübingen Tübingen, Germany solveig.klepper@student. uni-tuebingen.de Lena Kopnarski Computer Science Institute, University of Tübingen Tübingen, Germany lena.kopnarski@student. uni-tuebingen.de Fabian Hüttig ∗ University Hospital Tübingen Tübingen, Germany fabian.huettig@med.uni-tuebingen. de Constanze Keutel 2 University Hospital Tübingen Tübingen, Germany constanze.keutel@med. uni-tuebingen.de Katharina Scheiter Leibniz-Institut für Wissensmedien Tübingen, Germany k.scheiter@iwm-tuebingen.de Juliane Richter Leibniz-Institut für Wissensmedien Tübingen, Germany j.richter@iwm-tuebingen.de Thérése Eder Leibniz-Institut für Wissensmedien Tübingen, Germany tf.eder@iwm-tuebingen.de Enkelejda Kasneci Perception Engineering, University of Tübingen Tübingen, Germany enkelejda.kasneci@uni-tuebingen.de ABSTRACT The cognitive processes that underly expert decision making in medical image interpretation are crucial to the understanding of what constitutes optimal performance. Often, if an anomaly goes undetected, the exact nature of the false negative is not fully under- stood. This work looks at 24 experts’ performance (true positives and false negatives) during an anomaly detection task for 13 images and the corresponding gaze behavior. By using a drawing and an eye-tracking experimental paradigm, we compared expert target anomaly detection in orthopantomographs (OPTs) against their own gaze behavior. We found there was a relationship between the number of anomalies detected and the anomalies looked at. How- ever, roughly 70% of anomalies that were not explicitly marked in the drawing paradigm were looked at. Therefore, we looked how often an anomaly was glanced at. We found that when not explicitly marked, target anomalies were more often glanced at once or twice. In contrast, when targets were marked, the number of glances was higher. Furthermore, since this behavior was not similar over all images, we attribute these diferences to image complexity. ∗ Department of Prosthodontics 2 Department of Radiology, Center of Dentistry, Oral Medicine and Maxillofacial Surgery Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for proft or commercial advantage and that copies bear this notice and the full citation on the frst page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specifc permission and/or a fee. Request permissions from permissions@acm.org. MCPMD’18 , October 16, 2018, Boulder, CO, USA © 2018 Association for Computing Machinery. ACM ISBN 978-1-4503-6072-2/18/10. . . $15.00 https://doi.org/10.1145/3279810.3279845 CCS CONCEPTS · Applied computing → Psychology; Education; · Human- centered computing → Interactive systems and tools; Visualiza- tion design and evaluation methods; KEYWORDS Remote Eye Tracking, Medical image interpretation, Cognitive Mod- elling, Expertise ACM Reference Format: Nora Castner, Solveig Klepper, Lena Kopnarski, Fabian Hüttig, Constanze Keutel, Katharina Scheiter, Juliane Richter, Thérése Eder, and Enkelejda Kasneci. 2018. Overlooking: The nature of gaze behavior and anomaly detection in expert dentists. In Workshop on Modeling Cognitive Processes from Multimodal Data (MCPMD’18 ), October 16, 2018, Boulder, CO, USA. ACM, New York, NY, USA, 6 pages. https://doi.org/10.1145/3279810.3279845 1 INTRODUCTION Expertise in any domain is what many strive for. It is known that these skills are established through practice. Yet, there are still mechanisms that are not fully understood. Mainly, how experts process their visual input such that their domain knowledge is efectively applied. In general, experts are not easily available due to time and work constraints. Therefore, the majority of the literature measures ex- pertise with small samples of experts. Such small caches can lead to an insufcient understanding of expertise. In the literature review from Gegenfurtner et al., [4], across all expertise domains evaluated, mean expert sample sizes ranged from six to 17 experts; with the medical profession having approximately eight experts. More re- cently, van der Gijp et al. [10] provided a similar review that focused solely on radiology. Of the 26 studies evaluated in the meta-analysis,