Unobtrusive Pain Monitoring in Older Adults With Dementia Using Pairwise and Contrastive Training

Area of Science:

• Gerontology
• Computer Science
• Artificial Intelligence

Background:

• Older adults, particularly those with dementia, are frequently undertreated for pain due to communication barriers.
• Nursing staff face challenges in regular, validated pain assessment in long-term care due to resource and expertise limitations.
• Existing computer vision pain detection methods lack validation on elderly populations and individuals with dementia.

Purpose of the Study:

• To develop and validate the first fully automated, vision-based pain detection technique for older adults with and without dementia.
• To address the gap in automated pain assessment for non-verbal individuals in long-term care.
• To improve pain recognition and management in a vulnerable patient group.

Main Methods:

• Development of a deep learning-based computer vision system for detecting painful facial expressions.
• Collection of an unobtrusive video dataset from older adult participants with and without dementia.
• Introduction of a pairwise comparative inference method for personalized calibration and efficient training.
• Implementation of a fast contrastive training method to enhance cross-dataset performance.

Main Results:

• The developed pain estimation model significantly outperforms existing baselines.
• The system demonstrates particular effectiveness when evaluated on the facial expressions of individuals with dementia.
• The pairwise comparative inference method shows improved efficiency in using training data.

Conclusions:

• The presented vision-based system offers a promising automated solution for detecting pain in older adults, especially those with dementia.
• This technology has the potential to aid nursing staff in identifying and managing pain more effectively in long-term care.
• The study provides a validated tool and resources for advancing automated pain assessment in geriatric populations.