# Automatic pain recognition from Blood Volume Pulse (BVP) signal using machine learning techniques

**Authors:** Fatemeh Pouromran, Yingzi Lin, Sagar Kamarthi  
**arXiv:** [2303.10607](https://arxiv.org/abs/2303.10607)  
**Date:** 2023-03-19  
**Summary page:** https://www.summarizepaper.com/en/paper/2303.10607v1/

## Key points

- Physiological responses to pain are a focus for developing automated pain recognition systems
- Blood Volume Pulse (BVP) is considered a potential measure for objective pain assessment
- Machine learning techniques were applied to BVP signals in a recent study for non-invasive pain sensing
- Features extracted from BVP signals and Inter-beat Intervals (IBIs) were used to build machine learning models for pain detection and intensity estimation
- XGBoost model showed superior performance in both pain classification and intensity estimation tasks
- XGBoost achieved high ROC-AUC values for detecting different levels of pain
- Statistical analysis revealed significant variations in BVP characteristics across different pain levels during the cold pressor test
- Feature engineering directly from BVP signals was explored for comprehensive pain assessment
- Ensemble methods like Random Forest and XGBoost showed promising results in reducing variance while maintaining bias

## Summary

Physiological responses to pain have become a focal point for researchers aiming to develop automated pain recognition systems. Among these responses, Blood Volume Pulse (BVP) has emerged as a potential measure for objective pain assessment. In a recent study, machine learning techniques were applied to BVP signals in order to create a non-invasive modality for pain sensing. Thirty-two healthy subjects participated in the study, where novel time-domain, frequency-domain, and nonlinear dynamics features were extracted from BVP signals and Inter-beat Intervals (IBIs). These features were used to build machine learning models for detecting the presence of pain and estimating its intensity. Various machine learning algorithms were explored, including Logistic Regression, Random Forest, Support Vector Machines, Adaptive Boosting (AdaBoost), and Extreme Gradient Boosting (XGBoost). The XGBoost model demonstrated superior performance in both pain classification and intensity estimation tasks. The XGBoost classifier achieved ROC-AUC values of 80.06%, 85.81%, and 90.05% for detecting low, medium, and high levels of pain respectively. Additionally, it distinguished between medium and high levels of pain with an ROC-AUC of 91%. For multi-class classification among three pain levels, the XGBoost model yielded an average F1-score of 80.03%. Furthermore, statistical analysis was conducted to evaluate differences in BVP characteristics between different pain levels during the cold pressor test. Dunn's test was utilized due to the non-normal distribution of data revealing significant variations in features across pain levels. The study also delved into feature engineering directly from BVP signals and implemented a model architecture for comprehensive pain assessment. Different machine learning algorithms were compared with ensemble methods like Random Forest and XGBoost showing promising results due to their ability to reduce variance while maintaining bias. Overall, this research highlights the potential of utilizing BVP signals in conjunction with machine learning algorithms for automated pain assessment. The findings have implications for accurate pain management strategies that could help address issues such as drug-seeking behavior among patients and contribute towards mitigating the national opioid crisis.