Summary:
UCLA researchers in the Department of Anesthesiology have developed a novel pulse oximetry methodology utilizing deep neural networks for non-invasive monitoring.
Background:
In the US alone, over 5 million patients are admitted to the ICU for oxygen saturation monitoring. They, as well as the more than 15 million patients undergoing surgery, require continuous blood oxygen saturation monitoring to ensure positive outcomes. Pulse oximetry is the medical standard for continuous, noninvasive measurement of peripheral arterial oxygen saturation (SpO2). Recently, concerns about the accuracy of SpO2 measurement techniques have been raised. More specifically, the CDC has expressed concerns about the inaccuracy of pulse oximetry in patients with respiratory deterioration conditions. A more accurate reading of arterial oxygen saturation is possible, but the standard of care requires intermittent blood samplings, proving costly and inconvenient. There remains a need in the healthcare space for non-invasive, continuous and accurate monitoring of blood oxygen saturation.
Innovation:
Current pulse oximetry manufacturers use proprietary calibration curves for their devices based on limited prior population samples, which ultimately limit their accuracy on an individual patient basis. In order to account for these inaccuracies, researchers at UCLA have applied a deep neural network to the development of a novel oxygen monitoring protocol. Their algorithm leverages the patient’s individual features and statistics into the measurements, incorporating precision medicine techniques into SpO2 monitoring. These personal factors include static identifiers like age and ethnicity as well as temporal signals like hemoglobin levels and blood pressure, increasing measurement accuracy compared to conventional methods. Notably, this technique is more accurate than gold-standard methods when measuring the SpO2 of non-Caucasian patients, a particular weakness of standard oximeters.
Potential Applications:
• Pulse oximetry
• Screening patients for hypoxia probability
• ICU and surgical monitoring
• Hemodynamic monitoring
• COVID-19 screening
Advantages:
• More accurate than current pulse oximeters
• Non-invasive
• Continuous
• Racially agnostic
State of Development:
The inventors have developed the neural network and analyzed over 70 thousand patients using established datasets. The described algorithm has shown higher accuracy than current pulse oximeters in both Caucasian and African American patients. Successful demonstration of the invention has been completed as of 4/19/2024.
Reference:
UCLA Case No. 2024-227