Related Experiment Video
Updated: Aug 5, 2026
07:08
Estimate the Cognitive Load Using Electrocardiographic Measure: A Human-AI Collaborative Task
Published on: December 5, 2025
ECG-Only Cognitive Workload State Classification in Laparoscopic Training Using Raw and Recurrence-Plot
Kaizhe Jin1,2, Adrian Rubio-Solis1,2, Ravi Naik1,2
1Hamlyn Centre for Robotic Surgery, Institute of Global Health Innovation, Imperial College London, London SW7 2AZ, UK.
Sensors (Basel, Switzerland)
|July 28, 2026
Summary
Electrocardiography (ECG)-only signals can accurately classify cognitive workload states during laparoscopic surgery training. Raw ECG data demonstrated superior performance in distinguishing low and high workload levels, supporting its use in training research.
Area of Science:
- Physiological sensing
- Surgical training research
- Cognitive workload assessment
Background:
- Multimodal sensing is burdensome for laparoscopic training research.
- Electrocardiography (ECG) offers a lower-burden alternative for physiological sensing.
- Classifying cognitive workload states is crucial for optimizing surgical training.
Purpose of the Study:
- To evaluate ECG-only representations for classifying cognitive workload states during a controlled laparoscopic peg transfer task.
- To compare the performance of different ECG-derived models (raw ECG, recurrence plots, hybrid, HRV-RF) in workload classification.
- To determine the feasibility of ECG-only sensing for workload assessment in surgical training.
Main Methods:
- Twenty surgical trainees performed a laparoscopic peg transfer task under varying cognitive load conditions (Control, N0, N1, N2).
- ECG data were collected and processed into raw windows, recurrence plots, and heart rate variability (HRV) features.
- Four Random Forest (RF) models (raw ECG, RP-derived, hybrid, HRV-RF) were trained and evaluated using a leave-one-round-out cross-validation strategy.
- Primary endpoint: low/high workload classification (Control+N0 vs N1+N2); secondary endpoints: four-class and three-level classification.
Main Results:
- Raw ECG models achieved the highest performance on the primary low/high workload endpoint (macro-F1/balanced accuracy: 0.865/0.865).
- Hybrid models also showed strong performance (0.847/0.847) and were statistically supported over RP-derived and HRV-RF models.
- Recurrence plot (RP)-derived features showed potential complementarity on secondary endpoints, suggesting endpoint-dependent advantages.
Conclusions:
- ECG-only signals are effective for block-level workload-state classification in controlled laparoscopic training settings.
- Raw ECG data provides a robust and accessible method for assessing cognitive workload during surgical tasks.
- These findings support the development of lower-burden physiological sensing solutions for surgical education and research.
