Development Of Wearable Apnea Recognition System

Respiration-related diseases like sleep apnea-hypopnea syndrome(SAHS)pose a great threat to human health nowadays,which features high incidence rate and great hazards.Accurate detection and pattern recognition of oronasal airflow signals are conducive to the diagnosis and treatment of SAHS.Though currently used both at home and abroad,the clinical application of gold standard polysomnography(PSG)is being restricted due to its high costing and operating environmental limitations.Therefore the development of an apnea recognition algorithm and a wearable apnea recognition system has great significance and application value.In this paper,a set of apnea recognition algorithm is first proposed.By means of frequency reduction,adaptive normalization and segmentation,the algorithm preprocesses oronasal airflow signals to eliminate the interference and guarantee the signal integrity.Then two different ways are proposed to generate sufficient and efficient features.The first approach is based on pattern analysis,calculating permutation entropy and time-frequency domain statistical parameters-as eigenvectors;the second approach is based on convolutional neural networks,extracting dense layer neurons as eigenvectors.The extracted features are fed into several classifiers for testing,combined with various performance evaluation indices.The experimental results based on the authority Apnea-ECG database show that both approaches can describe the respiratory signals effectively and achieve more than 95%recognition accuracy.In contrast,the convolutional neural network scheme features higher identification accuracy and the pattern analysis scheme features less resource consuming.Compared with existing solutions in literature,our algorithm possesses excellent comprehensive performance and great practical value.Based on the algorithm,a wearable apnea recognition system is further developed.The system adopts modularization design,achieving human respiration signals acquisition,transmission,storage and analysis.The thermal flow sensor is calibrated to improve the measurement accuracy and meet the respiratory measurement requirements.The system airtightness simulation analysis and experimental verification are carried out to prove the positive correlation between the measured and the actual airflow signals.The combination of this peculiarity and the algorithm is further tested to prove that the pattern analysis scheme possesses immune ability against airflow distortion,which can be further generalized and maintain superb recognition performance.Finally,a GUI interface is developed to render algorithm results in real-time,which is convenient for subsequent experimental studies.Preliminary clinical validation shows that the apnea recognition system constructed in this paper features good stability,robust algorithm,high recognition accuracy and acceptable portability,possessing promising application prospects and tremendous commercial potentials.The system can monitor human respiratory status under the family conditions,which has significant effects on the diagnosis and rehabilitation of SAHS patients.