Real-Time Postural Disturbance Detection Through Sensor Fusion of EEG and Motion Data Using Machine Learning

Area of Science:

• Biomedical Engineering
• Neuroscience
• Public Health

Background:

• Falls are a major global health issue, necessitating improved real-time detection methods.
• Current fall detection systems often lack accuracy and timely alerts.
• Detecting the brain's reaction to postural changes is key for advanced fall prediction.

Purpose of the Study:

• To develop an accurate and efficient fall detection system using electroencephalogram (EEG) data.
• To analyze EEG signals for recognizing reactions to postural disturbances.
• To compare novel state-space methods with traditional autoregressive (AR) and Shannon entropy (SE) techniques.

Main Methods:

• Utilized a state-space-based system identification approach for EEG feature extraction.
• Compared performance using EEG epochs starting 80 ms post-event versus from onset.
• Developed a real-time algorithm integrating EEG and accelerometer data.

Main Results:

• State-space methods showed improved performance in detecting reactions to postural perturbations.
• EEG-based classification achieved high sensitivity (90.9%), specificity (97.3%), and accuracy (95.2%).
• The integrated real-time system detected falls in under 400 ms with over 99% accuracy for unexpected falls.

Conclusions:

• EEG data, particularly when analyzed with advanced methods, shows significant potential for fall detection.
• Combining EEG with accelerometer data enhances real-time fall detection accuracy and speed.
• This technology can improve safety and quality of life for at-risk populations, including the elderly.