Heartbeat Detection Using Multidimensional Cardiac Motion Signals and Dynamic Balancing

Abstract

Ballistocardiography (BCG) is seeing a new renaissance mainly due to access of new miniaturized and sensitive MEMS accelometers and gyroscopes that provides us a new tool for unobstrusive measurement of cardiac signals. These signal, however, suffer from high signal morphology variability and commonly signals are at least partly of low quality. A characteristic of a BCG signal is commonly a brief oscillation associated with each heartbeat which caused by the hearts mechanical movement. We developed an algorithm to detect these wavelets using an envelope enhancement filtering and subsequent dynamic balancing to alleviate the problem of high peak amplitude variability. The beat detection resulted in 0.87 % missed beats and 0.31 % false beats using the gyroY axis of the mobile phone’s integrated motion sensors. Also it is shown, that if the used axis could be chosen optimally for each measurement accuracy of 0.22 % missed beats and 0.21 % false beats could be reached within the used measurements. A photoplethysmography (PPG) signal was used as a verification reference. The data set consisted 2 min recordings from 66 healthy subjects and in total 8870 beats.