Research On Real-Time Detection Method Of Respiratory Events In Positive Pressure Ventilation

Positive pressure ventilation is the most effective treatment for obstructive sleep apnea at present.However,the domestic independently developed treatment machine cannot accurately warn of abnormal breathing in advance,which has a great impact on the efficacy of sleep breathing monitoring and treatment for patients.In order to improve the real-time and accurate warning performance of the treatment machine,the pressure and flow integrated sensors were used to detect the breathing waveform of continuous positive pressure ventilation in this paper.Combined with the signal processing technology,the methods of dynamic real-time prediction of respiratory waveform and the breathing event detection method were studied respectively in order to achieve the realtime and accuracy of early warning.In order to improve the real-time early warning,the respiratory waveform was predicted.Firstly,six prediction methods were compared,and the results show that the prediction errors of these six methods were relatively small,among which the Kalman filter algorithm has obvious advantages in predicting calculation time.In order to realize fast prediction and improve the prediction accuracy,a new prediction model was proposed by combining wavelet transform and Kalman filter.Respiratory waveform was decomposed by wavelet simultaneously with frequency doubling filter bank,state equation and measurement equation of each component were calculated,filter models were established respectively,and then filtered signals were reconstructed to get the predicted respiration waveform.The experimental results show that compared with the traditional Kalman filter,the prediction error of the optimized prediction model in this paper was significantly reduced,and it can realize the rapid and accurate prediction of respiratory waveform.In order to improve the accuracy of early warning,five types of respiratory events such as normal breathing,apnea,hypopnea,coughing and swallowing were detected.It analyzes and extracts features from the time-frequency domain and waveform decomposition respectively,and uses five classification methods to detect respiratory events.Among them,the K-nearest neighbor classification algorithm has the highest classification accuracy and the lowest time complexity.On this basis,a new classification model was constructed by combining Ball-Tree space search and Kernel methods to improve classification efficiency and accuracy.The results show that the total accuracy of the classification of the new model was 96.2%,which was 2.6% higher than the traditional K-nearest neighbor algorithm;the training time was increased by 0.08 s,and the test time was increased by 0.02 s.Experimental results show that the classification model proposed in this paper had better recognition rate and recognition efficiency for respiratory events,and can realize rapid and accurate detection of respiratory events.In this paper,an optimization model based on Kalman filter was proposed to realize dynamic real-time prediction of respiratory wave travel.On the basis of waveform prediction,a fast and accurate classification model based on K-nearest neighbor was proposed to detect respiratory events.The experimental results show that the research work can realize real-time detection of respiratory events.It had important reference value for improving the real time and accurate warning performance of positive pressure ventilation therapeutic apparatus and further pressure feedback control.