Deep learning-derived magnetic resonance imaging masseter muscle volume is associated with oral impairment and cognitive frailty in older Japanese adults.
Authors
Affiliations (9)
Affiliations (9)
- Department of Aging Research and Geriatric Medicine, Institute of Development, Aging and Cancer, Tohoku University, Sendai, Japan. Electronic address: [email protected].
- Department of Aging Research and Geriatric Medicine, Institute of Development, Aging and Cancer, Tohoku University, Sendai, Japan.
- Department of Aging Research and Geriatric Medicine, Institute of Development, Aging and Cancer, Tohoku University, Sendai, Japan; Department of Geriatric Medicine and Neuroimaging, Tohoku University Hospital, Sendai, Japan. Electronic address: [email protected].
- Department of Aging Research and Geriatric Medicine, Institute of Development, Aging and Cancer, Tohoku University, Sendai, Japan; Department of Pediatrics, Tohoku University Hospital, Sendai, Japan.
- Department of Aging Research and Geriatric Medicine, Institute of Development, Aging and Cancer, Tohoku University, Sendai, Japan; Department of Geriatric Medicine and Neuroimaging, Tohoku University Hospital, Sendai, Japan.
- Department of Preemptive Medicine, Innovation Center for Health Promotion, Graduate School of Medicine, Hirosaki University, Hirosaki, Japan.
- Department of Oral and Maxillofacial Surgery, Graduate School of Medicine, Hirosaki University, Hirosaki, Japan.
- Research Institute of Health Innovation, Hirosaki University, Hirosaki, Japan.
- Department of Epidemiology and Public Health, Graduate School of Medical Sciences, Kyushu University, Fukuoka, Japan.
Abstract
The masseter muscle is associated with oral impairment and age-related functional decline, yet scalable and objective indicators remain limited in epidemiological studies. We developed a convolutional neural network (CNN) to segment the masseter muscle on T1-weighted magnetic resonance images (T1WI) and examined whether CNN-derived masseter muscle volume, as an imaging-based structural phenotype, is associated with oral impairment and cognitive frailty in older Japanese adults. A CNN was developed using an iterative human-in-the-loop procedure based on T1WI acquired at Tohoku University Hospital. The trained model was then applied to available 3D T1WI from the Hirosaki site of the Japan Prospective Studies Collaboration for Aging and Dementia, and 2077 participants (mean age, 69.9 ± 4.1 years) were included in the final analyses after image quality control and application of the study exclusion criteria. Associations among masseter volume, oral impairment, and cognitive frailty were analyzed. Greater masseter volume was associated with a lower composite oral impairment score [odds ratio (OR) 0.67; 95% confidence interval: 0.60-0.74; p < 0.001]. Supplementary analyses showed the strongest association with structural abnormalities (OR 0.60), followed by periodontal inflammation (OR 0.81) and oral functional decline (OR 0.86). Masseter muscle volume showed an inverse association with cognitive frailty; however, this association was not significant after full adjustment. The pretrained CNN is available at https://github.com/bthyreau/masseter_mri. CNN-derived masseter muscle parameters from magnetic resonance imaging were inversely correlated with oral impairment in older adults. This approach enables scalable and objective oral phenotyping from imaging data and highlights its potential application in large-scale epidemiological studies of aging and functional decline.