EEG-Based Pain Classification via Sample Selection to Mitigate Subjective Label Bias

Area of Science:

• Neuroscience
• Biomedical Engineering
• Pain Research

Background:

• Accurate pain quantification is crucial for personalized pain management.
• Electroencephalography (EEG)-based pain estimation is promising, especially for non-communicative patients.
• Existing EEG pain models suffer from subjectivity due to self-reported pain labels, limiting reliability.

Purpose of the Study:

• To develop a novel method for EEG-based pain level classification using reliable sample selection during training.
• To enhance the robustness and reliability of EEG pain prediction models by addressing label subjectivity.
• To identify objective EEG biomarkers for pain intensity in non-communicative individuals.

Main Methods:

• Proposed a reliable sample selection technique that quantifies sample informativeness and label reliability.
• Excluded unreliable or uninformative samples to improve model robustness.
• Evaluated the method using EEG data from 41 participants under various thermal stimuli, with pain labels from the Numerical Rating Scale (NRS), employing 5-fold cross-validation.

Main Results:

• The proposed method achieved statistically significant improvements in multi-class EEG pain classification (3, 6, and 10 classes) compared to baseline models.
• Demonstrated generalization to novel thermal stimulation types, indicating potential for objective pain assessment.
• Identified delta-band activity at frontotemporal electrodes (F7, F8) as a key EEG feature strongly associated with perceived pain intensity.

Conclusions:

• Reliable sample selection significantly enhances the accuracy and robustness of EEG-based pain classification.
• The developed method shows promise for objective pain assessment in challenging patient populations, such as those unable to communicate.
• Specific EEG features, particularly delta-band activity, can serve as objective indicators of pain intensity.