Electrodermal Activity Based Pre-surgery Stress Detection Using a Wrist Wearable.

Area of Science:

• Biomedical Engineering
• Psychophysiology
• Machine Learning in Healthcare

Background:

• Preoperative stress significantly impacts patient well-being and surgical outcomes.
• Current methods for assessing stress are often invasive or subjective.
• Objective and non-invasive stress monitoring is crucial for surgical patients.

Purpose of the Study:

• To develop an automatic stress detection scheme for surgical patients using electrodermal activity (EDA).
• To create a non-invasive, wrist-wearable system for continuous EDA monitoring.
• To implement a machine learning approach for accurate stress level classification.

Main Methods:

• Collected continuous EDA data from 41 surgical patients using a wrist wearable device.
• Developed a supervised machine learning algorithm to detect and remove motion artifacts from EDA data (97.83% accuracy).
• Employed a novel localized supervised learning scheme with adaptive partitioning for stress classification, mitigating interindividual variability.

Main Results:

• The motion artifact detection algorithm achieved high accuracy.
• The localized learning scheme successfully classified stress levels (low, moderate, high) with 85.06% accuracy on new users.
• This approach proved more effective than general supervised classification models by accounting for person-specific EDA variations.

Conclusions:

• An accurate and non-invasive method for detecting pre-surgery stress using EDA is feasible.
• The developed localized learning scheme effectively addresses interindividual variability in EDA measurements.
• This technology holds potential for improved preoperative patient assessment and management.