Automated assessment of mental workload from PPG sensor data using cross-wavelet coherence and transfer learning

Area of Science:

• Physiological computing
• Biomedical signal processing
• Cognitive science

Background:

• Mental workload (MW) assessment is crucial for safety in high-stress jobs.
• Photoplethysmogram (PPG) signals offer a non-invasive method to gauge psychological load.
• Increased sympathetic nervous system activity during high MW alters PPG waveform morphology.

Purpose of the Study:

• To develop a time-frequency analysis framework for automatic mental workload assessment using PPG signals.
• To investigate the efficacy of cross-wavelet coherence (WTC) for extracting distinguishing PPG features.
• To validate the proposed method on a dataset of individuals performing a cognitive task.

Main Methods:

• A cross-wavelet coherence (WTC) approach was employed to analyze time-frequency information of PPG signals during mental workload and rest.
• PPG data from 22 healthy individuals performing an N-back task were analyzed.
• A customized pre-trained Inception-V3 model was used for classifying low and high MW based on WTC-generated images.

Main Results:

• The WTC method successfully captured distinguishing PPG features related to mental workload.
• Classification accuracy for low and high MW reached 93.86% (validation) and 93.07% (test).
• Optimal performance was achieved with a 1200-sample window for WTC image creation.

Conclusions:

• The developed time-frequency analysis framework effectively assesses mental workload using PPG signals.
• This approach shows significant potential for real-time monitoring and improving safety in demanding work environments.
• PPG signal analysis, particularly with WTC, is a promising tool for non-invasive mental workload assessment.