Non-contact monitoring of ventilation function parameters using optical flow

Abstract

Respiration rate is important parameter of a patient’s health. The majority of patient monitoring devices estimate vital parameters on contact-based principle. These solutions require attachment of sensing device to the body. Such methods typically involve contact to patient’s body, and require preliminary patient preparation which might not be convenient for patients and medical personnel in various clinical situations. Therefore, non-contact systems have potential to complement or replace existing contact-based solutions. In this paper, the approach for estimation of extended lung ventilation parameters during non-contact monitoring using video camera was proposed. This method could be used in non-contact single-camera systems for vital parameters monitoring.The advantage of this using video cameras in comparison to other non-contact methods (radar methods, the map of depth reconstruction using laser projectors, etc.)is the passiveness of the technology. This approach does not have any influence on the patient’s body such as radar or laser radiation. The most widespread technique that implements object movement estimation in video sequences is based on assessment of optical flow coming from the patient. The proposed approach allows to estimate duration of inhale, exhale using video stream from the upper torso. Extraction of respiration-related signal was done using Horn-Schunk optical flow algorithm. Obtained respiration signal was filtered and processed using algorithm for local minima extraction. Durations of inhale and exhale were estimated based on respiration curve minima information. Validation of obtained results was done using processed capnography data that was collected at the same time as video records. For the reference, medical device was used (Mainstream Capnography EtCO2 Sensor Respironics Capnostat 5). Advanced parameters allow to extract additional information for diagnostic of patient state during non-contact monitoring. The proposed approach was tested on video records of five healthy volunteers. The average relative error for duration of inhale was 9.7 percent and for duration of exhale was 18.3 percent. Developed during the experiment software allows to analyze video, plot the respiration signal and the values or estimated respiration parameters, as duration of inhale and duration of exhale. The respiration data could be stored for further analysis. The value of error for the developed non-contact method in comparison to the contact-based medical capnography sensor confirmed its consistency. The current approach could be developed to decrease number of errors and make it more robust. This approach could be implemented on various devices and platforms that have a single camera and are able to perform processing of the real-time video sequences: select region of interest, estimate optical flow, find local minima on respiration curve, calculate respirations parameters as respiration rate and durations of inhale and exhale, etc. Ref. 11, fig. 4, tabl. 1.