EYE TRACKING AND ELECTROENCEPHALOGRAM (EEG) MEASURES FOR WORKLOAD AND PERFORMANCE IN ROBOTIC SURGERY TRAINING
Robotic-assisted surgery (RAS) is one of the most significant advancements in surgical techniques in the past three decades. It provides benefits of reduced infection risks and shortened recovery time over open surgery as well as improved dexterity, stereoscopic vision, and ergonomic console over laparoscopic surgery. The prevalence of RAS systems has increased over years and is expected to grow even larger. However, the major concerns of RAS are the technical difficulty and the system complexity, which can result in long learning time and impose extra cognitive workload and stress on the operating room. Human Factor and Ergonomics (HFE) perspective is critical to patient safety and relevant researches have long provided methods to improve surgical outcomes. Yet, limited studies especially using objective measurements, have been done in the RAS environment.
With advances in wearable sensing technology and data analytics, the applications of physiological measures in HFE have been ever increasing. Physiological measures are objective and real-time, free of some main limitations in subjective measures. Eye tracker as a minimally-intrusive and continuous measuring device can provide both physiological and behavioral metrics. These metrics have been found sensitive to changes in workload in various domains. Meanwhile, electroencephalography (EEG) signals capture electrical activity in the cerebral cortex and can reflect cognitive processes that are difficult to assess with other objective measures. Both techniques have the potential to help address some of the challenges in RAS.
In this study, eight RAS trainees participated in a 3-month long experiment. In total, they completed 26 robotic skills simulation sessions. In each session, participants performed up to 12 simulated RAS exercises with varying levels of difficulty. For Research Question I, correlation and mixed effect analyses were conducted to explore the relationships between eye tracking metrics and workload. Machine learning classifiers were used to determine the sensitivity of differentiating low and high workload with eye tracking metrics. For Research Question II, two eye tracking metrics and one EEG metric were used to explain participants’ performance changes between consecutive sessions. Correlation and ANOVA analyses were conducted to examine whether variations in performance had significant relationships with variations in objective metrics. Classification models were built to examine the capability of objective metrics in predicting improvement during RAS training.
In Research Question I, pupil diameter and gaze entropy distinguished between different task difficulty levels, and both metrics increased as the level of difficulty increased. Yet only gaze entropy was correlated with subjective workload measurement. The classification model achieved an average accuracy of 89.3% in predicting workload levels. In Research Question II, variations in gaze entropy and engagement index were negatively correlated with variations in task performance. Both metrics tended to decrease when performance increased. The classification model achieved an average accuracy of 68.5% in predicting improvements.
Eye tracking metrics can measure both task workload and perceived workload during simulated RAS training. It can potentially be used for real-time monitoring of workload in RAS procedure to identify task contributors to high workload and provide insights for training. When combined with EEG, the objective metrics can explain the performance changes during RAS training, and help estimate room for improvements.