Presentation
Real-Time Mental Workload Assessment in Robotic-Assisted Surgery Using a Time-Varying Cox Framework
SessionPoster Session 2
DescriptionRobotic-assisted surgery (RAS) enhances dexterity, visualization, and ergonomics, which have driven the adoption of RAS across diverse surgical domains, including urology, gynecology, and gastroenterology. However, RAS also imposes high mental workload (MWL) on surgeons due to increased technical complexity, novel interfaces, and disconnection with the surgical team. High MWL depletes cognitive resources, slowing responses and impairing decision-making, which may compromise both performance and patient safety. Consequently, MWL management strategies such as task redistribution, cognitive support, and neuroadaptive interventions become crucial to prevent surgeons from entering high MWL states. By implementing these strategies, surgeons are better able to sustain performance and mitigate safety risks, yet the effectiveness of MWL management ultimately depends on accurate and timely MWL assessment. Several approaches have been explored for MWL assessment. Subjective scales such as NASA-TLX and SURG-TLX are widely used due to their convenience and ease of administration. Performance metrics such as completion time, error counts, or motion economy provide behavioral proxies of MWL changes. Nevertheless, these methods are post hoc and tend to recall bias. More recently, physiological data, such as electroencephalogram (EEG), eye movement, heart rate variability (HRV), functional near-infrared spectroscopy (fNIRS), has enabled continuous real-time measurement, and machine learning based methods applied to these signals have achieved promising accuracy in classifying MWL levels. However, current methods remain limited in two critical aspects. First, current research suffers from a lack of large-scale, ecologically valid datasets from RAS scenarios. Existing datasets are often limited to simplified simulator tasks that fail to capture the full spectrum of cognitive demands, especially the patterns associated with high MWL. As a result, models trained on such data cannot effectively learn or generalize to MWL conditions encountered in RAS. To address this limitation, we designed a set of RAS tasks, including peg transfer, continuous suturing, anastomosis, and chicken wing dissection, that progressively induce higher MWL during performance. In addition, we recruited 50 surgeons to ensure the size of our datasets. This experimental design ensures the models developed on this dataset can generalize across different RAS scenarios and achieve reliable performance in detecting mental overload. Second, existing MWL detection models are able to detect whether MWL is high at the present moment but cannot estimate when mental overload will occur. Without a time-to-event formulation, MWL management strategies remain reactive and are triggered only after mental overload has emerged, which forces surgeons to remain in high MWL states longer. A predictive framework that predicts mental overload would provide actionable lead time, enabling proactive MWL management before performance breakdowns occur. To bridge these gaps, we propose a time-varying Cox proportional hazards framework applied to multimodal physiological streams. This model treats MWL as a longitudinal process, providing time-to-overload predictions by estimating and inverting the semi-parametric Cox survival function, which enables timely MWL management in RAS.
In summary, by creating a large-scale and ecologically valid MWL dataset across diverse RAS tasks and introducing a time-to-event modeling framework, the proposed work can enhance the generalizability of MWL detection models, enable proactive workload management, and ultimately improve surgeon performance and safeguard patient safety.
In summary, by creating a large-scale and ecologically valid MWL dataset across diverse RAS tasks and introducing a time-to-event modeling framework, the proposed work can enhance the generalizability of MWL detection models, enable proactive workload management, and ultimately improve surgeon performance and safeguard patient safety.
Event Type
Poster Presentation
TimeTuesday, March 244:45pm - 6:15pm EDT
LocationRhinelander Gallery
Patient Safety Research and Initiatives