US20260256363
2026-09-03
Human necessities
A61B5/0205
A novel system for non-invasive physiological monitoring uses a monocular RGB camera to capture facial video, enabling the estimation of cardiovascular parameters and generating wellness indications. The system identifies facial skin regions of interest by segmenting facial landmarks and computes a metric to select stable pulsatile regions. It extracts imaging photoplethysmography (iPPG) signals and temporal features, and employs a convolutional neural network to infer OCT-variation feature maps, all without physical OCT hardware. These elements are combined to generate volumetric tensors for further analysis.
The system employs a trained model to predict pulse rate and other physiological parameters from the fused feature vector, which includes iPPG signals, temporal features, and OCT-variation feature maps. Motion-derived imaging ballistocardiography (iBCG) and facial micro-motion features are also extracted. The model, incorporating convolutional and transformer layers, processes volumetric tensors to capture spatiotemporal dependencies. It is trained using facial video sequences correlated with ground-truth physiological data, ensuring accurate predictions.
The system estimates additional physiological parameters like heart-rate variability, respiration rate, blood oxygen saturation, and blood pressure. It also outputs an uncertainty metric alongside predicted values, enhancing reliability. The model provides interpretability cues by identifying contributing regions, offering insights into the inference process. This approach supports low-cost RGB cameras operating under ambient lighting, making it accessible and practical for various applications.
Latent variables from the spatiotemporal model are transformed into physiological wellness indications. The system generates textual contexts describing the subject's physiological state, which are used to retrieve evidence from a curated knowledge base. A medical or wellness language model then provides grounded indications and recommendations on stress, fatigue, anxiety, and more, offering valuable insights for users and clinicians.
The disclosed system includes a non-contact, non-invasive apparatus comprising a monocular RGB camera, processors, and a memory storing executable instructions. It can be implemented on a non-transitory computer-readable medium, ensuring that the system's operations are efficiently performed by the processors. This setup facilitates seamless integration into various environments, enhancing the scope and utility of physiological monitoring.