Development of Invisible Sensors and a Machine-Learning-Based Recognition System Used for Early Prediction of...

Area of Science:

• Biomedical Engineering
• Machine Learning
• Human-Computer Interaction

Background:

• Accurate monitoring of patient mobility is crucial for healthcare, particularly for fall prevention.
• Existing methods for detecting bed-leaving behavior often lack real-time accuracy and comprehensive pattern recognition.

Purpose of the Study:

• To develop and evaluate a novel bed-leaving sensor system for real-time recognition of diverse patient behavior patterns.
• To assess the effectiveness of machine learning algorithms in classifying specific actions such as sleeping, sitting, and leaving the bed.

Main Methods:

• A sensor system comprising five bed pad sensors and one safety rail sensor was developed.
• Load tests were conducted to characterize sensor linearity with respect to load and speed.
• Machine learning algorithms were employed to recognize five distinct behavior patterns from benchmark datasets.

Main Results:

• The proposed sensor system demonstrated improved recognition accuracy for both continuous and discontinuous behavior datasets.
• Sensor output exhibited a linear relationship with applied load and speed, validating sensor characteristics.
• Integration of learning datasets further enhanced the overall recognition accuracy of the system.

Conclusions:

• The novel bed-leaving sensor system effectively achieves real-time recognition of patient behavior patterns.
• Machine learning integration significantly improves the accuracy of distinguishing between various mobility states.
• This system offers a promising solution for enhanced patient monitoring and safety in healthcare settings.