Gaze-entropy as a task load index for safety-critical operators: military pilots and surgeons.

Abstract

Background: Task (over)load imposed on operators who work at safety-critical environments is a main contributing factor to the occurrence of catastrophic errors. Laboratory research has shown that gaze metrics represent a valid tool to asses operator task load. Thus, gaze metrics have the potential to improve operator and system safety by providing accurate and objective measurements of task load variations and subsequent operational recommendations. However, in many of the diverse safety-critical environments where this information could be of use, such as aviation or surgical domains, the direct relationship between gaze metrics and task load has not been investigated yet. Here, we studied the effects of task complexity on the gaze metrics of combat helicopter pilots and surgical trainees. Methods: We recorded the eye movements of the entire attack helicopter battalion of the Spanish Army (n=13), and of surgical trainees from the Andalusian healthcare system (n=25), using a mobile eye-tracker system (Tobii Glasses 2.0), while performing several high-fidelity virtual simulation exercises of different complexity levels. We also measured performance and collected subjective ratings of complexity. Results: Among helicopter pilots, gaze entropy decreased with increased task complexity: visual exploration pattern became more stereotyped (i.e. less random) during the more complex tasks (i.e. flight emergencies). Among surgeon trainees, we found the opposite trend. In both cases, performance and perceived task complexity differed as expected. Conclusions: Our results show, for the first time, that gaze metrics are a valid index of task load in safety-critical environments. Task features and visual display solutions might explain different trends in pilots and surgeons. These findings have the potential to improve flight and patient safety by providing accurate measurements of operator task (over)load. These findings might also lead to future indices for the assessment of operators' learning curves without the need for time-consuming expert evaluation.

Meeting abstract presented at VSS 2016