Using Inertial Measurement Units and Machine Learning to Classify Body Positions of Adults in a Hospital Bed.

Area of Science:

• Biomedical Engineering
• Clinical Monitoring
• Machine Learning in Healthcare

Background:

• Timely clinical interventions are crucial for preventing patient deterioration in hospitals.
• Early recognition of patient decline is vital for effective treatment.
• Patient positioning and movement patterns can indicate the need for further medical assessment.

Purpose of the Study:

• To investigate the efficacy of inertial measurement units (IMUs) combined with machine learning for continuous patient position monitoring in hospital beds.
• To develop a system for capturing, classifying, and visualizing patient positions to aid in early detection of clinical deterioration.

Main Methods:

• Utilized five IMU Xsens DOT sensors placed on participants' forehead, wrists, and ankles in a simulated hospital setting.
• Trained Support Vector Machine (SVM) and K-Nearest Neighbours classifiers using data from various sensor combinations to classify nine distinct body positions.
• Analyzed data from 30 participants to evaluate classification accuracy.

Main Results:

• The highest accuracy of 87.7% was achieved using SVM with forehead and wrist sensors.
• Incorporating data from ankle sensors resulted in a decrease in classification accuracy.
• A privacy-preserving 3D visualization system was developed using Unity.

Conclusions:

• IMUs and machine learning offer a promising approach for continuous patient monitoring in hospital settings.
• Forehead and wrist sensor placement demonstrated optimal accuracy for classifying patient positions.
• The developed visualization system can support clinicians in identifying positional changes indicative of potential clinical deterioration.