Longitudinal Data Acquisition for AI Services in Long-Term Care Facilities for Older Adults

Rinu Elizabeth Paul, Pantea Kock, Yale Hartmann, Eckhard Ball, Kathrin Seibert, Hui Liu, Tanja Schultz

Abstract: Data is essential for analysis, processing, feature extraction, and machine learning models, serving as a cornerstone for developing patient-centered digital health technologies for older adults. Most datasets in older adult applications are collected in controlled laboratories, with fewer from natural environments. Data collection and processing in natural settings is challenging, often yielding both usable and unusable data. This paper focuses on collecting data from older residents in long-term care facilities using sensor boxes installed in resident rooms. The sensor box, equipped with a depth sensor, captures depth images around the clock. We collected continuous 24-hour depth images from 45 older residents in nursing homes over 15 months. We describe the ethical, social, and technical conditions for collecting on-site data from depth sensors in nursing homes. We propose a pipeline to process depth images and classify them into different room states and corrupted frames using machine learning models, achieving 93\% accuracy in occupied room classification. Using this dataset, we aim to develop AI services such as fall detection, activity monitoring, gait analysis, sleep position monitoring, and bed exits in long-term care facilities. These insights advance digitally enabled care solutions for older adults, paving the way for innovative, sustainable strategies.