Machine learning methods on BioVid heat pain database for pain intensity estimation.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Pain Management Research

Background:

• Traditional pain scales (NRS, VAS, VRS) are subjective and unreliable for certain patient groups.
• Objective pain assessment is crucial for accurate treatment and improved patient quality of life.
• Machine learning offers a promising avenue for developing objective pain measurement tools.

Purpose of the Study:

• To introduce and evaluate an objective pain measurement model using machine learning techniques.
• To compare the performance of Convolutional Neural Networks (CNNs), VGG16, Convolutional Vision Transformer (CvT), and MobileViT in classifying pain intensity from facial expressions.
• To assess the potential of automated pain assessment for clinical applications, especially for non-communicative patients.

Main Methods:

• Utilized facial images from the BioVid Heat Pain Database, capturing peak pain and no-pain moments.
• Trained and evaluated four machine learning models: CNN, VGG16, CvT, and MobileViT.
• Compared model performance based on accuracy in classifying pain intensity.

Main Results:

• The CNN model achieved the highest average accuracy (0.71) in classifying pain intensity.
• The CvT model demonstrated strong performance with an average accuracy of 0.69.
• VGG16 (0.56) and MobileViT50 (0.60) showed significantly lower accuracy, indicating challenges in classifying pain from facial images.

Conclusions:

• Machine learning models, particularly CNNs, show significant potential for objective and consistent pain assessment.
• Automated facial expression analysis can aid in pain evaluation for non-verbal or cognitively impaired patients.
• Future research should focus on multimodal data integration to enhance pain detection system robustness.