Project Summary/Abstract Chronic and acute pain conditions increase with aging and with age-associated conditions, including Alzheimer's disease and similar dementias, causing suffering, exacerbating diminished quality of life, and likely leading to further morbidity and increased mortality. Understanding and treatment of pain is particularly challenging among populations whose ability to communicate verbally may be impaired, among which patients with Alzheimer's disease and related dementias are of particular importance if only because of the increased aging population. A substantial body of research suggests that methods based on nonverbal indicators of pain yield reliable and valid measurement of pain in patient with dementias. Some of this work has indicated that patients with dementia may in fact be hyper-reactive to pain, rendering the pursuit of improved techniques for assessing pain in the dementia population even more important. Existing techniques for assessing nonverbal indicators of pain have several disadvantages which limit their utility. In particular, they require human observers, are dependent on specialty training, can be laborious, and cannot be applied on a continuous basis. Advances in computer vision and machine learning have the potential to overcome some of these shortcomings. Our “parent proposal” project aims to develop advanced, automated computer-vision and machine-learning models for assessing nonverbal aspects of pain in a longitudinal cohort study of people with low back pain. This supplement proposal seeks to extend our work on the “parent grant”, which aims to “build a fully automatic, multimodal (face, head, and body movement) system to measure the occurrence and intensity of low back pain from video”. In this supplement project, these aims will be extended to older adults with dementia. We will refine the principles discovered for pain assessment in the low back pain population, extend, and evaluate their application in a sample of older adults with dementias in extended-care facilities. Participants, long-term care residents with dementia, were video-recorded during a baseline state as they were lying still on a bed or examination table and then as they underwent a standardized protocol of movements designed to identify painful areas. Participants' face, head, and body movement will be used for the development of automatic measures of the occurrence and intensity of pain. To do so, face, head, and body movement will be automatically tracked using fully- automatic methods. The tracking results will be used to train end-to-end deep- leaning based classifiers to automatically measure the occurrence and intensity of pain in older adults with dementia. To investigate the validity of the proposed classifiers, we will compare automated measurement of pain intensity to reliable and objective pain intensity coding. MANOVA will be used to quantify the relationship between the individual modalities and their combination f...