Abstract

Ballistocardiography (BCG), a non-invasive technique for measuring micro-body vibrations arising from cardiac contractions. It also contains motion arising from breathing, snoring and body movements. Long-term acquisition of respiratory signal finds relevance in various applications such as sleep analysis as well as monitoring of respiratory disorders. Current methods (such as nasal thermistor and Respiratory Inductance Plethysmography) are costly, inconvenient and require technical expertise to setup and analyse. In this paper we assess how BCG based contact-free methods can allow for an accurate, cost-effective and convenient long-term monitoring from the ease of home environment. We propose a novel algorithm to detect breathing cycles from BCG signal, achieving an accuracy of *95% in determining respiration rate for 30 s epochs with a detection rate of 72.8% compared to current methods. Long-term continuous monitoring of respiratory signals with a high accuracy will allow for detection of abnormalities like respiratory distress and apnea/hypopnea episodes.

Introduction

Continuous monitoring of body vitals is poised to be one of the key elements of future healthcare including monitoring critical illnesses, personalized therapeutics and predictive medicine. Continuous monitoring of patients by a team of healthcare professionals in a hospital setting has proven to be extremely effective over traditional (and sparsely) periodic manual measurements by an attendant [1]. Such manual monitoring, apart from being prone to inefficiency and human error, is also not feasible due to already stressed healthcare systems, especially in developing countries where healthcare needs are already at severe odds with their fast growing aging populations and an increasing shortage of trained professionals – a projected deficit of 14 million healthcare workers globally by 2030 [2].
We can broadly classify applications for continuous monitoring of body vitals into two classes: (a) short term and (b) long term. The short term class involves uses like generating real-time alerts based on abnormal deviations in vitals to prevent a critical event like a cardiac arrest [3]. On the other hand, long term monitoring collects data over longer periods of time (days, months) to detect a gradual shift in vitals of an individual from their healthy or expected baseline to monitor recovery (or relapse) or catch a disorder at its onset. While the short term use cases are almost always appli- cable to point of care establishments like hospitals, long term monitoring enables collecting health data while an individual is at home, work, travelling and so on.
Continuous monitoring explored so far has primarily been for cardiac activity (achieved using a full-sized ECG machine, a holter machine and the likes of these), body temperature and blood pressure. Respiration rate has been a highly neglected body vital [4], and has been shown to indicate health deterioration in critically ill patients, leading to onset of critical events [5]. In a clinical setting, respiration is measured as air flow using a nasal thermistor or by respiratory effort using Respiratory Inductance Plethysmography (RIP) belts for abdomen and/or chest. These are usually costly, require a trained professional to set them up and cause discomfort to the individual, making it infeasible for a regular use over long term. Innovative solutions to overcome these challenges in a non-invasive manner can enable long-term continuous monitoring of respiration in hospitals and at home.
Ballistocardiography (BCG) is one such promising technique. While it was originally developed to measure the micro body motions produced during each cardiac contraction [6], the phenomenon is able to detect any physiological parameter that produces a motion – including breathing, snoring and limb movements [7] and [8]. Moreover it can be made to work in a contact-free and non-wearable manner given a medium that can propagate body vibrations, such as when placed on a solid surface under a mattress, while a subject is lying over it. (For the rest of the paper, we assume this setup.)
However, BCG has its own set of challenges. BCG is prone to undesired noise from the setup environment. This could be mechanical vibrations or movement near and around the setup. Even heavy body movements can overpower the cardiac and respiratory signals resulting in a lower detection rate. Apart from these noises, BCG raw signal is a superimposition of respiratory effort, cardiac contractions and vibrations due to snoring; effective signal conditioning is required to segregate these signals for high fidelity analysis. Additionally, the signal can vary for different people and with different mattresses. It can even vary for the same subject in different postures and body positions.
In this paper, we propose a novel unsupervised clustering algorithm that is effectively able to identify each respiratory cycle from an unconstrained BCG signal. In the next section, we discuss the previous related work. Further in Sect. 3, we describe the algorithm with validation methodology and results in Sect. 4 respectively. Future potential work is highlighted in Sect. 5 with conclusion in Sect. 6.

Background

A normal respiration signal comprises alternate inhales and exhales, and is typically a sinusoidal waveform. The same is true for respiration effort as captured in a BCG setup. Figure 1(a) shows raw BCG signal in comparison with respiratory effort captured from RIP belts on chest & abdomen along with airflow signal from the nasal thermistor acquired simultaneously.
Previous studies have attempted to extract respiration rate from a BCG signal. One study [9] describes an algorithm that uses the amplitudes of the filtered BCG signal as the only feature with thresholding to identify inhales and exhales. While this approach works good for normal sinusoidal respiration pattern, it fails to correctly identify respiration cycles in the presence of movements and posture change events that abruptly change the amplitude of the signal. This method is also inconsistent for disorder respiration patterns like hypopnea events, Biot’s respiration where the respiration signal has a non uniform amplitude. In another study [10], maximas in the respiration signal, derived after removing the heart signal from the raw BCG signal, were identified as the respiration cycles. This was tested in a controlled environment for only 3 min per subject. The subjects were asked to breathe with an external sound indication for a part of the 3 min. This approach, however, wasn’t tested in uncontrolled settings for longer durations such as overnight testing where the subjects are asleep. Another recent study [11] proposes two methods, one to calculate breathing rate using maximas in filtered BCG signal that cross a threshold. However, fixed thresholds don’t yield consistent results across various breathing patterns. The other method proposed only calculates an estimate to breathing rate through the most prominent frequency in the fast fourier transform of the filtered BCG signal. The lack of identification of each breath limits this approach for detailed analysis of breathing signal.

Read the full document here: Unsupervised Extraction of Respiration Cycles Through Ballistocardiography | SpringerLink