Machine Learning and ECG‐Derived Biomarkers for Objective Pain Assessment: An Explainable AI Approach Toward Precision Pain Management

Figures

• 

  FIGURE 1. Scheme of the dual‐layer algorithm for pain assessment.

• 

  FIGURE 2. Confusion matrix for the binary pain classification task (CoVAS = 0 vs. CoVAS > 0) using ECG‐derived biomarkers, showing excellent performance with few misclassifications.

• 

  FIGURE 3. Feature importance for the classifier model (CoVAS = 0 vs. CoVAS > 0), showing ECG‐derived biomarkers and heart rate variability measures as the most important predictors.

• 

  FIGURE 4. Feature importance for the regressor model (predicting pain intensity for CoVAS > 0), highlighting the importance of specific ECG signal characteristics as biomarkers for pain intensity estimation.

• 

  FIGURE 5. SHAP summary visualization showing the impact of key ECG biomarkers on model predictions. Features are ordered by importance from top to bottom, with color indicating feature value (red = high, blue = low).

• 

  FIGURE 6. SHAP interaction plot showing the joint impact of heart rate (HR) and interbeat interval (IBI) biomarkers on model predictions. The complex patterns indicate important physiological interactions captured by the model.

• 

  FIGURE 7. Biomarker correlation heatmap showing relationships between key ECG measures. Red indicates positive correlation; blue indicates negative correlation.

• 

  FIGURE 8. Bland–Altman plot comparing predicted and actual pain values for samples with CoVAS > 0. The horizontal lines represent the mean difference (solid black) and limits of agreement (dashed red).

• 

  FIGURE 9. Representation of permutation feature importance, clearly showing the hierarchy of ECG biomarkers.

• 

  FIGURE 10. PDP and ICE of the features.

• 

  FIGURE 11. Diagram of the autonomic activation in response to pain.