G. Bilen-Rosas, I. Ong, H. Rosas
University of Wisconsin School of Medicine and Public Health Madison,
United States
Keywords: ultrasound, respiration, monitoring, airflow
Summary:
Delayed detection of respiratory depression and diminished airflow to the lungs, both of which can arise during sedation, and possibly lead to lack of oxygen, resulting in brain or cardiac damage and even death. The ability to continuously monitor respiratory parameters in sedated patients is imperative to identifying subtle changes in respiratory effort or airflow to allow for early lifesaving interventions and avoid life-altering complications. Current monitoring systems measure secondary or downstream effects well after respiratory complications occur, giving healthcare providers at times mere seconds to initiate lifesaving airway maneuvers. Anesthesia closed claims studies indicate that 21% of adverse effects are due to respiratory depression, which are considered preventable with better monitoring. 97% of postoperative complications were due to delayed detection of opioid induced respiratory depression. One review looking solely at the economic burden of minor sedation related complications reported a 2.2 billion dollar loss in the U.S. alone. This work presents a novel method using ultrasound signaling to continuously and quantitatively measure respiration. We conducted a feasibility study with 38 non-sedated human participants and demonstrated a strong correlation (cross correlation > 0.9 and Spearman correlation for doppler envelope of > 0.8) between ultrasound and respiratory flow measurements obtained by spirometry, the current gold standard for measuring airflow. We employed machine learning methods to predict respiratory flow from ultrasound data, achieving high accuracy. Our machine learning model, trained to predict flow based on the Doppler signals currently achieves a root mean-squared error of < 0.1 and high adjusted R2 of 0.7. We further trained a deep learning model to distinguish exhalation from inhalation, which was able to achieve an area under the ROC curve of 85% using 5-fold cross validation. Our preliminary study in patients receiving conscious sedation for eye surgery shows promise in detecting respiratory compromise earlier than standard conventional monitoring methods by detecting obstructive changes in real time. This monitoring method has potential to be developed into a non-invasive, self-adhesive monitoring device that detects and predicts respiratory compromise in real time, affording the ability to act expeditiously and efficiently avoid complications resulting from prolonged periods of oxygen deprivation to the brain and heart and ultimately save lives. Healthcare providers can potentially utilize our system to recognize respiratory depression and airway obstruction early and be alerted to initiate corrective and resuscitative measures.