Frontiers in neurology · 2026
Real-world feasibility of privacy-preserving, non-wearable AI for real-time fall detection with disease-specific video classification in parkinsonian syndromes: a proof-of-concept clinical study.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Kamo Hikaru, Oyama Genko, Nagaki Koichi, Tsuboi Yoshio, Hatano Taku, Hattori Nobutaka
Аннотация
BACKGROUND: Falls in Parkinson's disease (PD) and progressive supranuclear palsy (PSP) are frequent yet often missed in clinical care. We evaluated a privacy-preserving, non-wearable AI system for real-time fall detection and exploratory video analysis in Parkinsonian syndromes.
METHODS: Residents of a PD-specialized facility were continuously monitored for 22-63 days using a ceiling-mounted depth camera with in-sensor edge AI. All video-confirmed falls were reviewed by two blinded raters and classified into 10 predefined operational categories. Detection performance was summarized descriptively, and fall characteristics were explored using chi-square tests with false-discovery-rate control.
RESULTS: Across 6,272 h of monitoring, 74 real-world falls were confirmed. AI sensitivity was 52.7% across the full study and 67.6% during the post-update evaluation phase, whereas conventional facility reporting identified 28.4% and 29.4% of video-confirmed falls, respectively. The corresponding positive predictive values of the AI system were 45.3% and 41.8%. Comprehensive video review established the reference count of confirmed fall events within the analyzable recordings. Exploratory video analysis suggested that falls in PD were more often classified as involving observable contexts compatible with attentional, executive, or visuospatial contributions, whereas falls in PSP were more often classified as involving inadequate weight shift or transfers. Most falls in PD and PSP occurred during routine movements without obvious external perturbations.
CONCLUSION: Continuous, privacy-preserving depth-camera monitoring identified falls that were not captured by routine facility reporting. The observed disease-related fall patterns are exploratory and require validation in larger, independent cohorts.
METHODS: Residents of a PD-specialized facility were continuously monitored for 22-63 days using a ceiling-mounted depth camera with in-sensor edge AI. All video-confirmed falls were reviewed by two blinded raters and classified into 10 predefined operational categories. Detection performance was summarized descriptively, and fall characteristics were explored using chi-square tests with false-discovery-rate control.
RESULTS: Across 6,272 h of monitoring, 74 real-world falls were confirmed. AI sensitivity was 52.7% across the full study and 67.6% during the post-update evaluation phase, whereas conventional facility reporting identified 28.4% and 29.4% of video-confirmed falls, respectively. The corresponding positive predictive values of the AI system were 45.3% and 41.8%. Comprehensive video review established the reference count of confirmed fall events within the analyzable recordings. Exploratory video analysis suggested that falls in PD were more often classified as involving observable contexts compatible with attentional, executive, or visuospatial contributions, whereas falls in PSP were more often classified as involving inadequate weight shift or transfers. Most falls in PD and PSP occurred during routine movements without obvious external perturbations.
CONCLUSION: Continuous, privacy-preserving depth-camera monitoring identified falls that were not captured by routine facility reporting. The observed disease-related fall patterns are exploratory and require validation in larger, independent cohorts.