Development of an acoustic respiratory monitor

Patients sometimes suffer apnea during sedation procedures or after general anesthesia. Apnea presents itself in two forms: respiratory depression (RD) and respiratory obstruction (RO). During RD the patients' airway is open but they lose the drive to breathe. During RO the patients' airway is occluded while they try to breathe. Patients' respiration is rarely monitored directly, but in a few cases is monitored with a capnometer. This dissertation explores the feasibility of monitoring respiration indirectly using an acoustic sensor. In addition to detecting apnea in general, this technique has the possibility of differentiating between RD and RO. Data were recorded on 24 subjects as they underwent sedation. During the sedation, subjects experienced RD or RO. The first part of this dissertation involved detecting periods of apnea from the recorded acoustic data. A method using a parameter estimation algorithm to determine the variance of the noise of the audio signal was developed, and the envelope of the audio data was used to determine when the subject had stopped breathing. Periods of apnea detected by the acoustic method were compared to the periods of apnea detected by the direct flow measurement. This succeeded with 91.8% sensitivity and 92.8% specificity in the training set and 100% sensitivity and 98% specificity in the testing set. The second part of this dissertation used the periods during which apnea was detected to determine if the subject was experiencing RD or RO. The classifications determined from the acoustic signal were compared to the classifications based on the flow measurement in conjunction with the chest and abdomen movements. This did not succeed with a 86.9% sensitivity and 52.6% specificity in the training set, and 100% sensitivity and 0% specificity in the testing set. The third part of this project developed a method to reduce the background sounds that were commonly recorded on the microphone. Additive noise was created to simulate noise generated in typical settings and the noise was removed via an adaptive filter. This succeeded in improving or maintaining apnea detection given the different types of sounds added to the breathing data.