A team led by Dr. Hsin-Hsu Chou from Chiayi Christian Hospital, in collaboration with Melten, Inc., has developed the "Bao-Ming Fall Prevention - Smart Pediatric Bed Fall Alert System." This system combines deep learning algorithms, IoT technology, and ICT to accurately predict bed fall risks and issue alerts, significantly enhancing the safety of hospitalized children.
Serious Issue of Pediatric Bed Falls, Limited Effectiveness of Traditional Prevention
According to the Ministry of Health and Welfare, 51.2% of pediatric ward fall incidents involve children falling from beds, often due to caregivers' lack of fall prevention awareness, insufficient vigilance, or improper use of bed rails. Traditional prevention methods like educational outreach and regular monitoring have limited effectiveness, and commercially available bed exit alarms often generate too many false alarms, leading to poor user experience.
Thermal Imaging Combined with AI for Real-Time Bed Fall Risk Alerts
The alert system uses thermal imaging cameras to capture real-time images of children and caregivers. By integrating pre-trained AI models, it recognizes the positions and postures of individuals, such as sitting, lying, approaching, or moving away from bed rails. The system also uses bed rail sensors to transmit the height and position of the bed rails via low-power Bluetooth to the processing module. Through comprehensive analysis based on a set of compound judgment rules, the system categorizes bed fall risk into high, medium, and low levels. When the risk reaches medium or high levels, the system immediately provides auditory and visual alerts to both the family members and children, and sends real-time alerts to the nursing station and designated nursing staff's mobile phones. Additionally, the nursing station's server visually displays each child's bed fall risk level, allowing healthcare providers to intervene promptly and prevent accidents.
Clinical Simulation Achieves 99.8% Accuracy
To ensure the AI model's accuracy, the team simulated various potential fall scenarios in actual clinical settings. They used thermal imaging to capture raw image data, defined bed location information, applied masks to filter out irrelevant heat sources, and employed K-means clustering to segment temperature data. The highest temperature clusters were used to segment images and identify all objects within them. By analyzing the distance and relative position between bed rails and objects, they identified the relevant subjects, distinguishing between children and caregivers using feature selection techniques. Installed and tested at Chiayi Christian Hospital, the system achieved a 99.8% accuracy in interaction status detection and a 99.5% accuracy in bed fall risk alert categorization.
Expanding to Home Care for Comprehensive Healthcare Quality and Safety Improvement
The alert system not only effectively reduces the incidence of bed falls in pediatric wards but also alleviates the burden on healthcare providers and improves the doctor-patient relationship. In the future, the system is expected to be extended to elderly wards, nursing homes, and even home care settings, providing safety assurances to more care-dependent populations and enhancing overall healthcare quality.
Resource (mandarin): 預防病童床跌受傷！熱成像偵測即時預警，同步聲光提醒照護者，準確率達99.5%！
