Real-time measures of context to improve fall-detection models

Area of Science:

• Gerontology
• Biomedical Engineering
• Computer Science

Background:

• Real-time fall detection is crucial for elder care but faces challenges due to rare true events and frequent confounding factors in data.
• Existing automated fall detection algorithms struggle with accuracy, hindering commercial viability.

Purpose of the Study:

• To develop a robust fall detection system using a decision theoretic approach.
• To improve the accuracy of fall classification and alerting by incorporating contextual information.

Main Methods:

• Implemented a decision theoretic framework for classification and alerting.
• Utilized contextual data (location, activities) to enhance probability and utility estimates.
• Developed methods for real-time patient state assessment using monitored context to improve training datasets.

Main Results:

• The context-aware approach significantly improves the probability and utility estimates for detecting true falls, near falls, and confounding events.
• Real-time patient state assessment using context enhances the quality of training data.
• Improved classification, detection, and alerting capabilities were demonstrated.

Conclusions:

• A decision theoretic approach integrating context offers a promising solution for reliable real-time fall detection.
• Contextual information is vital for overcoming limitations of current fall detection algorithms.
• This method enhances the practical application of fall detection technology in healthcare settings.