A Study Of Physiological Information Detection Based On Ballistocardiography

In recent years,with the rapid development of the Internet and the increasing awareness of health among people,the research and application of the unobtrusive monitoring of body has been greatly developed.Ballistocardiography(BCG)can record the physiological information of cardiovascular system and respiratory system in a noninvasive and noncontact way.The main research of this paper is to detect the physiological information of cardiovascular system and respiratory system from the BCG acquisition equipment embedded in the mattress.Nowadays,most research used single-channel sensor system to collect BCG data.If the subject's body does not cover the sensor,it may cause the problems of the decrease of data quality and the loss of data.However,this paper used a multi-sensor system to collect BCG data,which improved the accuracy and the coverage of detection.Besides,in the field of apnea detection based on BCG,most studies divided the signal into independent windows and convert an apnea detection task into a classification task.Through this way,only whether the apnea happens in the window can be detected.And this article introduced a semantic segmentation model,which can detect apnea in a fine-grained way and obtain the apnea region and frequency,with higher clinical meaning for the diagnosis of apnea.The innovations mainly focus on the following points:(1)This paper proposed a heart rate period detection algorithm from multi-channel BCG signal based on cepstrum.This algorithm can quickly obtain the estimation of heart rate period by the characteristics of cepstrum,select signal channels by spectrum,and fuse multi-sensor information in frequency domain.In the short-term and long-term experiments,the mean absolute error was40 ms and 42.28 ms.This algorithm has a low computation complexity and has the ability to estimate heart rate period in a quick way.Thus,it has high practicability in applications.(2)This paper proposed an Inter-beat Interval estimation algorithm from multi-channel BCG signal based on Kalman filter.This algorithm relied on J-wave detection in BCG signal and used Kalman filter to fuse multi-sensor data.Experimental results have shown that the algorithm achieved profound J-wave recognition ability and low detection error of Inter-beat Interval estimation.In the short-term and long-term experiments,the mean absolute error was 40 ms and 42.28 ms.Compared with the heart rate period detection algorithm based on cepstrum,this method can get a fine-grained detection results,which can be applied in the clinical heart rate variability analysis.(3)This paper proposed an apnea detection algorithm based on convolution neural network and designed “U-Breath” apnea detection model.Compared with the previous research,"U-Breath" model added semantic segmentation network to segment positive apnea samples to identify the region of apnea.The mean dice coefficient of the segmentation network can reach 0.808.The frequency and duration of apnea can be obtained in this model,which has higher clinical application value.In general,this paper has solved some physiological information detection problems in BCG multi-sensor acquisition system and achieved profound experiment results.In addition,this article provided fine-grained detection idea and method in the field of apnea detection.