Multimodal Non-Contact Sensing of Respiration and Movement in Neonates

Area of Science:

• Biomedical Engineering
• Neonatal Care
• Sensor Technology

Background:

• Traditional contact-based monitoring devices are suboptimal for preterm neonates due to their size and physiology.
• Adhesive sensors pose risks of infection, discomfort, and clinical care interference.
• Remote health monitoring presents a significant benefit for fragile neonates.

Purpose of the Study:

• To develop and evaluate a noncontact multimodal monitoring device for neonates in the Neonatal Intensive Care Unit (NICU).
• To integrate video and radar technologies for estimating respiratory rate (RR) and detecting movement.
• To compare unimodal (video or radar only) and multimodal fusion approaches.

Main Methods:

• Utilized a Video Vision Transformer (ViViT) architecture with early, intermediate, and late fusion techniques.
• Integrated data from a video camera and a radar system.
• Conducted preliminary experiments on neonatal simulator mannequins and one neonate.

Main Results:

• Early fusion multimodal approach improved respiratory rate estimation by 14% and 20% in mean absolute error compared to unimodal radar and video, respectively.
• Fusion techniques achieved >97% accuracy and precision for movement detection against human labeling.
• Multimodal analysis demonstrated superior performance over unimodal approaches in challenging conditions.

Conclusions:

• Multimodal analysis using video and radar shows potential to outperform unimodal methods for neonatal monitoring.
• This noncontact sensing technology offers a viable alternative to current gold standard adhesive sensors in the NICU.
• Remote sensing of vital signs can improve care for vulnerable preterm infants.