Time delay neural network based apnea detection and comparative analysis

Sleep apnea is described as a cessation of breath for at least 10 seconds during the sleeping. These apneas can occur in hundreds depending on the severity of the disease during the sleep. Busy schedules and high costs of sleep laboratories in hospitals make apnea diagnosis in the society difficult task. In order decrease loads of sleep laboratories and increase the number diagnostic attempts with a low cost, there is a need for using portable apnea devices which can trace possible apnea patients out of hospitals. The purpose of this study is to analyze and reveal the proper combination sets of physiological signals for detecting apnea episodes in order to decide whether the standard but more complicated polysomnography test stage might be required or not for possible apnea patients by examining physiological signals recorded by portable recording devices used for prescreening purposes. Thus, air flow, oxygen saturation and ECG signals are used separately for apnea detections. For this reason, neural networks are trained and tested by extracted dynamic features associated with each signal. For a sample case, the neural network trained with hybrid (including three channels) data generated from the recordings obtained by the portable recording device has shown 89.6% high detection rate based on the expert scores. In the last phase of the study, a software interface was also developed in order to examine obtained physiological signals for possible apnea patients and the data analysis useful for apnea scoring and labeling. With the tools provided by the interface using third level portable apnea device definition, three channels, namely air flow, oxygen saturation and ECG signal, can be examined comprehensively for general apnea analysis. In order to have detailed analysis, embedded neural network topologies can be chosen through the designed interface and associated features that are derived for each corresponding channels can be used in this platform for implementing training and testing phases.