Voice-Based Pain Level Classification for Sensor-Assisted Intelligent Care.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Signal Processing

Background:

• Intelligent healthcare systems face challenges in pain assessment due to staff shortages.
• Traditional pain detection methods have limitations in cost, accessibility, and professional support requirements.
• Acoustic sensor-based pain detection offers a promising alternative for real-time monitoring.

Purpose of the Study:

• To develop a lightweight, sensor-assisted system for real-time pain level classification.
• To investigate the effectiveness of voice spectral features for pain detection.
• To validate the system using simulations and a hardware prototype.

Main Methods:

• Utilized acoustic sensors to capture voice signals.
• Employed Convolutional Neural Network (CNN) models trained on voice spectral features.
• Developed a three-level pain classification approach.
• Validated the system through Jupiter Notebook simulations and a Raspberry Pi hardware prototype.

Main Results:

• Achieved an average accuracy of 72.74% for three-level pain classification.
• Outperformed existing methods by 18.94-26.74% for the same pain-level granularity.
• Demonstrated real-time processing speeds (6-22s) on a low-cost hardware prototype (<100 USD).
• Showcased the applicability of various machine learning algorithms (ANN, XGBoost, Random Forests, Decision Trees).

Conclusions:

• The proposed sensor-assisted system provides an effective and affordable solution for real-time pain detection.
• Voice spectral analysis combined with machine learning, particularly CNNs, is a viable method for pain assessment.
• The system has the potential to enhance intelligent healthcare and assisted-living environments.