The Optimization Of Apnea Detection Using Tracheal Sounds In Patients Recovering From Anesthesia

Objective: The use of tracheal sounds to detect apnea has been used in many occasions,such as post anesthesia care unit,obstructive sleep apnea,etc.Failure to correct apnea in time may lead to bradycardia,tissue hypoxia,and even cardiac arrest.Breath monitoring is very important.During the collection of breathing sounds,the talking,machine alarms,etc.,may occur as tracheal sounds during apnea,and the apnea may be missed.When the log-var method is used to detect apnea,when multiple breathing cycles are linked together,the patient's tracheal sounds,groans and continuous talking can cause misdiagnosis of apnea.The purpose of this study was to explore the feasibility of adaptive filtering algorithms to improve the performance of apnea detection,and to explore the accuracy of hidden Markov models for the detection of apnea with breath sounds.Methods: 1.Tracheal sounds and ambient noise acquired from the subject in quiet and noisy environments,respectively.The nasal flow pressure signal collected with PSG was used as the reference signal for the breathing status of the subject.The adaptive filtering algorithm was used to process(the wavelet domain adaptive filtering algorithm is also used in the noisy environment)the collected tracheal sound and ambient noise,and the performance of the tracheal sound detection apnea algorithm before and after the adaptive filtering was compared.2.The mentioned collected subject data and the patient's tracheal sound and noise data collected in PACU and nasal flow pressure signals were used as data sources.The above data were filtered by adaptive filtering algorithm.The data in quiet environment was trained and tested firstly by hidden markov algorithm.The model obtained was validated with data in noisy environment and data in PACU,and the results of apnea detection with hidden markov model and threshold method were compared.Results: 1.The tracheal sounds and nasal flow pressure signals from 46 healthy subjects in a quiet and noisy environment were collected and the total duration was 26.5 hours.Without adaptive filtering in a quiet environment,the apnea detection algorithm's sensitivity changed from 97.2% to 98.2% and the specificity changed from 99.9 % to 99.8%,in the noisy environment without adaptive filtering the algorithm's sensitivity is 81.1%,the specificity is 96.9%,with adaptive filtering the algorithm's the sensitivity is 91.5%,the specificity is 97.4%,and with wavelet domain adaptive filtering the algorithm's the sensitivity is 93.08% The specificity is 96.27%.2.Comparison of apnea detection with log-var method and hidden markov model method was as follows: In the quiet environment,the algorithm's sensitivity is 97.2% and 98.8%,the specificity is 99.9% and 100%,and in the noisy environment the sensitivity is 91.46% and 94.11%,the specificity is 97.41 and 98.12%,in PACU,the algorithm's sensitivity is 91.95% and 83.1%,and the specificity is 97.54% and 99.02%.Conclusion:Adaptive filtering technology can effectively improve the quality of tracheal sounds in noisy environments,and improve the performance of detecting apnea using tracheal sounds.Hidden markov model provides an effective method for detecting the apnea using tracheal sounds.