Machine Learning Approach to Identifying Hypercapnia Through Breath Sounds in Mask Worn Breathing Systems
Abstract
pnia is thought to be caused by malfunctions in a selfcontained underwater breathing apparatus (SCUBA), especially in systems with rebreathers, that result in a failure to remove adequate proportions of CO2 from operator breathing gas. Hypercapnia can also occur due to physical changes that happen as depth increases such as reduced minute ventilation and alveolar hypoventilation. Hypercapnia causes increased heart and breath rate and vasodilation in the cerebral cortex that results in tissue hypoxia in neurons, and at extreme levels, operator incapacitation. When hypercapnia occurs in operational and training settings, it places military divers at a high risk of significant physiological stress, panic, incapacitation, drowning, and death. To mitigate hypercapnia associated risks in diving, methods to identify the early signs of the condition in-situ are required.Hypercapnia is typically measured through end-tidal CO2, which measures the amount of CO2 expired at the end of each breath. Efforts to develop the capability to accurately measure end-tidal CO2 in dynamic operational environments like diving have been unsuccessful due to several issues, including the mixing of expired and supplied gas in the breathing line, the humidity of expired breath, and the rapid changes in pressure inside the breathing loop and mask. Since respiratory sounds are found to convey invaluable information, an alternative method is to employ a small non-invasive microphone and monitor respiration in-situ using the sound dynamics in an operators breath. The reliable measurement of respiration rate and volume using a simple microphone in operational settings would be a significant increase in current physiol will develop a machine learning based algorithe, unobtrusive way to measure important breath based indicators of physiological stress associated with hypercapnia. Specific breathing gestures (e.g., inhalation and exhalation, cough, sniff etc) will be analyzed through Praat and openSMILE software to identify acoustic (energy and spectrum related) and duration-related features. Features can be used to identify inhalation and exhalation frequency, intensity, and duration. This information will be utilized to provide a moving average of breath rate, breath duration, and inhalation and exhalation dynamics. Trends can be tracked over time to significantly improve our situational awareness of operator physiologic status.A study will be conducted to demonstrate the systems ability to correctly identify hypercapnia. Breathing sounds will be recorded in separate sessions for the non-stressed condition as well as four stressed conditions (described below) while wearing a breathing mask. Two types of breathing sounds will be collected: (1) ten iterations of previously identified breathing gestures (e.g., a sniff or a cough) which participants will specifically be asked to produce, and (2) their normal breathing in each condition for 10 minutes. Participants will breathe four different gas mixtures for a period of 15 minutes per mixture (stressed condition). A mixed gas system will deliver gas mixtures of 1.0% CO2, 2.5% CO2, 4.0% CO2, and 5.5% CO2. In year two a second study will be performed at Duke University to confirm that the system works underwater and with various dive configurations.The development of a simple, reliable, and field expedient measure of Respiration Rate would be a significant advancement in real-time operator monitoring. This development can be expanded to identify other physiological stressors like hypoxia and breathing resistance, and sounds associated with other life support system malfunctions.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 05, 2021
- Source ID
- N000142112667
Entities
People
- Jeff Phillips
Organizations
- Florida Institute for Human and Machine Cognition
- Office of Naval Research
- United States Navy