Accurate classification of pain experiences using wearable electroencephalography in adolescents with and without chronic musculoskeletal pain

Figures

• 

  Figure 1. Subjective pain ratings before and after the CPT. Bar charts displaying self-reported pain scores (numeric rating scale 0–10) at baseline and during the CPT (average score throughout the condition). All healthy participants reported no pain (0) at baseline. Both healthy and chronic pain participants had significantly higher pain scores during the CPT compared to baseline. Chronic pain participants had significantly higher scores than healthy participants at baseline, but average pain scores throughout the CPT were not significantly different between groups. ***denotes significance at p < 0.001 level.

• 

  Figure 2. Baseline vs. CPT: feature importance. Most important features (top 10%) for classifying between pain and no pain conditions based on logistic regression model weights for chronic pain (A) and healthy (B) participants. Note: Model coefficients cannot be generated when using a radial basis function kernel for SVM models; thus, overall model feature importance was only explored using the logistic regression models. Solid boxes represent permutation entropy features, while stripped boxes indicate directed phase lag index features. No graph theory or peak frequency features were in the top 10% based on model coefficients. Positive model weights are predictive of tonic cold pain conditions, while negative model weights are predictive of no-pain conditions.

• 

  Figure 3. Individual electrode classification accuracy for chronic pain patients permutation entropy features (theta frequency band). Pain vs. no pain classification accuracy for each individual electrode and across all electrodes for PE (theta band) features only in chronic musculoskeletal pain participants. Topographic maps and bar charts are displayed for SVM models (A / B) and logistic regression models (C / D).

• 

  Figure 4. Individual electrode classification accuracy for health controls permutation entropy features (theta frequency band). Pain vs. no pain classification accuracy for each individual electrode and across all electrodes for PE (theta band) features only in healthy participants. Topographic maps and bar charts are displayed for SVM models (A / B) and logistic regression models (C / D). White circles designate electrodes where models in healthy participants had significantly higher classification accuracy than chronic pain participants.