Accelerometer-Based Human Activity Recognition for Patient Monitoring Using a Deep Neural Network

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Wearable Sensor Technology

Background:

• Continuous monitoring of hospitalized patients' physical activity is crucial for assessing recovery.
• Traditional methods for activity recognition are often labor-intensive and subjective.
• Developing automated, accurate, and non-invasive activity recognition systems is a significant challenge in healthcare.

Discussion:

• A novel Deep Neural Network (DNN) architecture, combining convolutional and long short-term memory layers, was employed.
• The DNN was trained and validated using data from a single tri-axial accelerometer placed on the trunk.
• Performance was benchmarked against a Support Vector Machine (SVM) classifier, highlighting the DNN's superior accuracy.

Key Insights:

• The DNN achieved 94.52% accuracy in classifying six distinct activities (lying, upright, walking, wheelchair, stair ascent/descent) in a simulated hospital setting.
• This accuracy significantly surpasses the 83.35% achieved by the SVM classifier.
• The findings demonstrate the feasibility of using a single accelerometer and DNN for reliable patient activity recognition.

Outlook:

• This DNN-based approach holds potential for unobtrusive, continuous patient monitoring during hospitalization.
• It can provide objective data to supplement clinical assessments and personalize rehabilitation strategies.
• Future research could explore multi-sensor fusion and real-world clinical validation for broader adoption.