Automated Detection Of SAS Based On BP Neural Network

Sleep Apnea Syndrome(SAS) has been called "sleep killer". It has kept increasingly attention of public since it usually happens sudden without omen, and it could cause lots of seriously complications. Nowadays, the using of PSG to help clinical diagnosis is the most reliably method to diagnose SAS. However, PSG could not be popularized because of the complicated diagnose process and the expensive cost. Most patients come from areas where health level is less advanced cannot receive diagnosis and therapy promptly, so it has significant practical meanings in developing an alternative diagnose method which is characterized by costless and operation easily.A kind of automatic SAS diagnose model which based on BP neural network is provided in this paper. The mainly works as follows:1.The model adopts wavelet decomposition algorithm and wavelet threshold diagnosing algorithm to eliminate baseline drift, electrical interference and power frequency interference.2. We achieved the location of each peak of R wave in ECG on the basis of the modulus maximum principle and the Lipschitz constant.3.We designed related algorithm in optimizing false detection and omission detection that we could extract RR intervals accurately. Then we did analyze the HRV based on RR intervals, and got fifteen character parameters of time domain and frequency domain. We’ve taken databases of Apnea-ECG from PhysionNet as training and test samples for our model, and imported the fifteen character parameter as input vector to the BP neural network so that we could get the weight value and threshold value of the model.4.We tested the unknown thirty-five samples, and found that the classification accuracy reached 85.7143%. The work we’ve done closely meet the request anticipated before.