Smart wireless, flexible hybrid electronics for fall risk monitoring among older adults

Thumbnail Image
Issue Date
Al Muslim, Asra Muslim
Lee, Yongkuk

Elderly’s fall is a serious medical risk since it might result in irreversible health injuries such as bone damage due to their normal changes of aging, especially elderly age 65 or older. However, there is still a lack of automatic fall detection systems for the elderly. Therefore, the objectives of this study are to develop 1) a wireless, flexible, skin-wearable electronic device with motion sensors for accurate motion sensing and user comfort and 2) a deep learning-based classification algorithm using training datasets collected from elderly for reliable fall detection. A flexible and ultrathin skin-wearable device for user comfort was designed and fabricated by using traditional MEMS techniques. It included a 6-axis motion sensor and can be directly laminated on the skin for the collection of accurate motion data. It exhibited excellent connectivity by optimizing the RF efficiency at 2.45 GHz, which allowed wireless transmission up to 10 m while operating on the skin. To study accurate fall detection for elderly, the motion data was collected from a total of 20 participants from two different age groups (young adults aged 21-30 and elderly aged 65 or older) while performing different human activities (e.g., falling, running, walking, sitting, and climbing stairs). Various deep-learning models (CNN, LSTM, CNN-LSTM, and Conv-LSTM), including their hyperparameters and different body locations (chest, wrist, and a form of necklace) and input datasets (acceleration, gyroscope, and magnitude of acceleration) were investigated. Our results indicate 1) the CNN model and non-normalized acceleration dataset achieved the highest accuracy, 99.4% (± 0.3), with 40 epoch, 2) the optimal location to place the device is the chest, achieved 99.4% (± 0.3) accuracy, compared to wrist, 55.9% (± 1.2), and necklace, 62.8% (± 2.4). Also, our result suggests a large motion dataset directly from elderly is required to improve the accuracy of fall detection for elderly since it achieved, 74.6% (± 2.0) accuracy, compared to training young adults’ datasets with elderly’s testing datasets that achieved, 43.3% (± 1.5).

Table of Content
Thesis (M.S.)-- Wichita State University, College of Engineering, Dept. of Biomedical Engineering