Location
Orbital ATK Conference Center
Start Date
5-7-2018 11:15 AM
Description
Hypoventilation remains a cause of unexpected hospital death due to limitations in detecting the problem and the associated high false positive alerts, alarm fatigue, and delayed diagnosis. Pulse oximetry, while affordable, is a delayed and indirect indicator of hypoventilation, especially for patients receiving supplemental oxygen. Capnometry is a direct measure of ventilation, but is too expensive and cumbersome for use in general ward patients, for whom the risk of mortality remains high during postoperative pain control via opioids. We hypothesize that we can identify changes in the pattern of the red photoplethysmography waveform and distinguish among periods of apnea, one minute preceding apnea, and normal breathing. Methods: Using data recorded from volunteers during administration of sedatives and opioids, we explored the feasibility of using machine learning to classify three types of events observed in the red photoplethysmography waveform including one minute preceding apnea, apnea and normal breathing. Nine feature parameters calculated from the red photoplethysmography waveform and a line fitted to the bottom of the photoplethysmography were used as inputs for a neural network pattern recognition algorithm which classified the three types of waveform- normal breathing, apnea and one minute before apnea. Results: We observed a high rate of success for classifying three types of events including one minute preceding apnea, apnea and normal breathing in the red photoplethysmography waveform. The neural network correctly classified 100% of the evaluation database as a measure of the three types of waveform accuracy.
Detecting Apnea Events Based on Pulse Oximetry’s Red Photoplethysmography Waveform
Orbital ATK Conference Center
Hypoventilation remains a cause of unexpected hospital death due to limitations in detecting the problem and the associated high false positive alerts, alarm fatigue, and delayed diagnosis. Pulse oximetry, while affordable, is a delayed and indirect indicator of hypoventilation, especially for patients receiving supplemental oxygen. Capnometry is a direct measure of ventilation, but is too expensive and cumbersome for use in general ward patients, for whom the risk of mortality remains high during postoperative pain control via opioids. We hypothesize that we can identify changes in the pattern of the red photoplethysmography waveform and distinguish among periods of apnea, one minute preceding apnea, and normal breathing. Methods: Using data recorded from volunteers during administration of sedatives and opioids, we explored the feasibility of using machine learning to classify three types of events observed in the red photoplethysmography waveform including one minute preceding apnea, apnea and normal breathing. Nine feature parameters calculated from the red photoplethysmography waveform and a line fitted to the bottom of the photoplethysmography were used as inputs for a neural network pattern recognition algorithm which classified the three types of waveform- normal breathing, apnea and one minute before apnea. Results: We observed a high rate of success for classifying three types of events including one minute preceding apnea, apnea and normal breathing in the red photoplethysmography waveform. The neural network correctly classified 100% of the evaluation database as a measure of the three types of waveform accuracy.
Comments
Session 3