Detection of Dementia-Related Abnormal Behaviour Using Recursive Auto-Encoders.

Area of Science:

• Gerontology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Global life expectancy is rising, leading to an increase in age-related health issues like cognitive impairment.
• Early detection of cognitive impairment is crucial for timely medical intervention and patient management.
• Current methods often overlook the structural relationships within daily activities, limiting their effectiveness in assessing cognitive status.

Purpose of the Study:

• To investigate the structural relationships between sub-activities for improved cognitive status assessment in the elderly.
• To develop a method for detecting abnormal activities indicative of cognitive impairment using hierarchical activity modeling.
• To evaluate the efficacy of recursive auto-encoders (RAE) in identifying early indicators of dementia.

Main Methods:

• Activities were modeled hierarchically from their sub-activities using recursive auto-encoders (RAE).
• A novel sensor representation, raw sensor measurement (RSM), was introduced to capture intrinsic activity structures.
• Abnormal activities were detected by analyzing RAE's reconstruction error on simulated data reflecting cognitive impairment.

Main Results:

• Recursive auto-encoders demonstrated effectiveness in modeling hierarchical activity structures.
• The RAE approach successfully detected simulated abnormal activities linked to cognitive impairment.
• The method proved particularly useful in semi-supervised and unsupervised learning scenarios with limited labeled data.

Conclusions:

• Hierarchical activity modeling using RAEs offers a promising approach for detecting cognitive impairment in seniors.
• RAEs can serve as valuable decision-support tools for identifying early signs of dementia.
• The proposed RSM representation enhances the ability to capture nuanced activity patterns relevant to cognitive health.