A. Wallace, T. Heintz, A. Badathala
Atapir,
United States
Keywords: Machine Vision, Artificial Intelligence, Medical Monitoring, Software as a Service
Summary:
Introduction: The AVD-M monitor uses a camera array with an RGB (red-green-blue), IR (Infrared), and a depth camera to identify patients in a hospital bed and monitor heart rate, respiratory rate, pulse oximetry, mood, pain, and other parameters using machine vision and artificial intelligence. Methods: Neural network algorithms identify and track the patient in the room. Eulerian magnification algorithms are used to monitor the pulsatile signal from the skin. Depth imaging allows measurement of chest wall motion. Explicit time domain and frequency domain algorithms are used for heart rate, respiratory, and SpO2 measurements. 48 facial action units are tracked on the face to monitor mood and pain. Neural network algorithms are used for mood and pain parameters. Results: Data from thirty (30) volunteers were used for the HR, RR, and SpO2 testing. Seventy-seven (77) patients had measurements of pain and mood prior to, after surgery in the recovery room, and then daily in the hospital. The correlation between AVD-M monitored heart rate (HR), respiratory rate (RR), and oxygen saturation (SpO2) was measured in 30 volunteers. AVD-M accurately measured respiratory rate (RR) (r2 = 0.88). AVD-M was able to accurately measure heart rate (HR) (r2 = 0.98). AVD-M accurately measures SpO2 with r2 = 0.87 and RMSE 1.55. The receiver operator curve for pain measurements from patients having surgery depends on the model. Random Forest AUROC 0.88. Discussion: The AVD-M monitor represents a new platform for medical monitoring relying on machine vision and artificial intelligence to create software as a monitor.