Research Paper Volume 14, Issue 3 pp 1280—1291
Predicting neuropsychiatric symptoms of persons with dementia in a day care center using a facial expression recognition system
- 1 Aging and Health Research Center, Taipei, Taiwan
- 2 Institute of Public Health, National Yang-Ming Chiao-Tung University, Taipei, Taiwan
- 3 Center for Geriatrics and Gerontology, Taipei, Taiwan
- 4 uAge Day Care Center, Taipei Veterans General Hospital, Taipei, Taiwan
- 5 Value Lab, Acer Incorporated, New Taipei City, Taiwan
- 6 Graduate Institute of Clinical Pharmacy, National Taiwan University, Taipei, Taiwan
- 7 School of Pharmacy, National Taiwan University, Taipei, Taiwan
- 8 Department of Pharmacy, National Taiwan University Hospital, Taipei, Taiwan
- 9 Taipei Municipal Gan-Dau Hospital, Taipei, Taiwan
Received: October 12, 2021 Accepted: January 17, 2022 Published: February 3, 2022https://doi.org/10.18632/aging.203869
How to Cite
Copyright: © 2022 Chen et al. This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Background: Behavioral and psychological symptoms of dementia (BPSD) affect 90% of persons with dementia (PwD), resulting in various adverse outcomes and aggravating care burdens among their caretakers. This study aimed to explore the potential of artificial intelligence-based facial expression recognition systems (FERS) in predicting BPSDs among PwD.
Methods: A hybrid of human labeling and a preconstructed deep learning model was used to differentiate basic facial expressions of individuals to predict the results of Neuropsychiatric Inventory (NPI) assessments by stepwise linear regression (LR), random forest (RF) with importance ranking, and ensemble method (EM) of equal importance, while the accuracy was determined by mean absolute error (MAE) and root-mean-square error (RMSE) methods.
Results: Twenty-three PwD from an adult day care center were enrolled with ≥ 11,500 FERS data series and 38 comparative NPI scores. The overall accuracy was 86% on facial expression recognition. Negative facial expressions and variance in emotional switches were important features of BPSDs. A strong positive correlation was identified in each model (EM: r = 0.834, LR: r = 0.821, RF: r = 0.798 by the patientwise method; EM: r = 0.891, LR: r = 0.870, RF: r = 0.886 by the MinimPy method), and EM exhibited the lowest MAE and RMSE.
Conclusions: FERS successfully predicted the BPSD of PwD by negative emotions and the variance in emotional switches. This finding enables early detection and management of BPSDs, thus improving the quality of dementia care.