Design of an Inertial Multi-Sensor Network for the Monitoring of the Heart Rate During Sleep using Ballistocardiographic Signals

Diseases of the circulatory system are among the chronic degenerative pathologies with the highest morbidity, representing one of the main causes of disability. The general definition of cardiovascular diseases (CVDs) includes all the diseases of the heart and blood vessels. These represent the first cause of access to first aid, as well as death. In order to contain the number of hospitalizations and healthcare costs, innovative prevention and monitoring techniques for patient’s health are needed. To date, conventional techniques to monitor the patient’s cardiovascular activity are intrusive and invasive. The main goal of this work is to overcome these limitations, by introducing the design of a non-invasive and non-intrusive system for heart rate (HR) detection. This vital sign is obtained starting from the processing of the ballistocardiographic (BCG) signal which is acquired by a network of four IIS2ICLX inclinometers connected to the STM32L4-R9IJ6 microcontroller, produced by STMicroelectronics. The main advantage of this work is to use a multi-sensor network to detect the BCG signal during a sleep cycle, regardless of the subjects positions during the night, suitable for both home and hospital environments. In order to acquire the signal with the highest quality, several tests were performed by placing the sensors in certain regions of the bed, according to the literature. Subsequently, a time-based algorithm was developed in order to compute the heart rate from the ballistocardiographic signal and to manage the most frequent sources of noise, such as motion artifacts. In order to calculate the HR using the signal with the highest SNR among the four acquired, two prediction models were developed: a Multi-Parameter Model (MPM) and a Single-Parameter Model (SPM). The first one identifies, on 10 seconds windows of the ballistocardiogram, the best signal among the four provided by the inertial multi-sensor network, by calculating three statistical parameters (the standard deviation, the kurtosis, and the auto-correlation function). In the same way, the second method identifies, on 30 seconds windows, the best signal by calculating the mean of the cross-correlation coefficients from heart beats found on the ballistocardiogram of the four sensors. To validate the results achieved, the heart rate provided by the proposed system was compared with the HI device, a three-lead certified electrocardiograph designed by STMicroelectronics. Considering the developed predictive models, the designed inertial multi-sensor network provides the heart rate as follows. The Multi-Parameter Model estimates the HR every 10 and 30 seconds, with an average coverage of 87.95 % (MAE of 4.56 bpm ± 7.09 bpm) and 93.76 % (MAE of 4.26 bpm ± 6.55 bpm) of the entire sleep respectively. The Single-Parameter Model estimates the HR every 10 and 30 seconds, with an average coverage of 93.57 % (MAE of 4.25 bpm ± 6.51 bpm) and 97.66 % (MAE of 3.86 bpm ± 5.81 bpm) of the entire sleep respectively. Finally, taking full advantage of the characteristics of the network, the Ideal Model provides the heart rate every 10 or 30 seconds, estimating the vital sign for an average coverage of 94.69 % (MAE of 2.30 bpm ± 4.40 bpm) and 97.68 % (MAE of 1.80 bpm ± 3.22 bpm) of the entire sleep respectively.