Research On Obstructive Sleep Apnea Syndrome Detection Algorithm Based On VMD-CNN-LST

Purpose: OSAS is a common sleep apnea disorder,while polysomnography,the gold standard for diagnosing OSAS,has many limitations in terms of resource constraints,high cost,and sleep disruption.Thus,a simplified and practicable method has become the goal of many researchers.At the same time,various portable health devices are widely used nowadays,and ECG signals have been shown that have a strong correlation with respiratory signals and OSAS.Therefore,the following study was carried out on the basis of ECG signals.Method: In the study of preprocessing ECG signal,firstly,VMD was applied to decompose and reconstruct the ECG signal adaptively,which can retain some key subtle features in the ECG signal while completing the noise suppression and getting a purer ECG signal.Then,based on the amplitude threshold method,an improved algorithm called dynamic differential threshold was designed to effectively detect the QRS waves which is the significant in the ECG signal.Later,according to the influence of respiratory motion on the ECG signal,a reasonable algorithm was designed to extract the EDR signal by using the relevant information of R wave and S wave.In the study of OSAS classification model,unlike many detection algorithms that rely on complex feature engineering,a classification network was constructed for the two source signals separately by making special use of CNN,which is good at automatic feature extraction,and LSTM,which is good at preserving temporal information,with the understanding of the characteristics of ECG and EDR.And a series of experiments were constructed to optimize parameters of network.Finally,we determine a practical and feasible detection model.Results: The signal-to-noise ratio of signals processed by VMD and Butterworth digital filter was compared to confirm that VMD can better suppress the main noise in the ECG signal.The Apnea database of Physio Net and the self-collected signals from a portable device were selected as materials for QRS wave detection,using dynamic differential thresholding method,and extraction of EDR signals.The former achieved 99.85% detection accuracy,and the latter confirmed the effectiveness of the algorithm after waveform comparison and count verification.The ECG signal and corresponding annotations in the database were selected for training and validation of the classification model.Last,92.87% sensitivity,93.75% specificity and 93.41% accuracy were achieved on the test set.Conclusion: The proposed OSAS detection algorithm based on VMD-CNNLSTM in this study not only achieves ideal results in the signal pre-processing stage,but also the designed detection model of OSAS performs well.Overall,we reach the expected goal and provide a reference for achieving convenient and efficient detection of OSAS in portable devices in the future.