Sternal vibrations during head-out immersion: A preliminary demonstration of underwater wearable ballistocardiography

Ballistocardiography (BCG) is a method to measure cardiovascular (CV) health from vibrations of the human body originating from the beating heart and can be measured by wearable accelerometers or sensors in beds, chairs, or weighing scales.^1–4 Although electrocardiography (ECG) is used to measure the electrical aspects of cardiovascular function in the clinic, and ultrasonic technologies such as echocardiography are also widely used by clinicians to image local tissues and fluids, research interest in BCG has increased rapidly in the last few years to develop measurements of cardiovascular health from infrasonic mechanical waves that propagate throughout the whole body. This study is a step toward our goal to enable a new class of biomedical devices based on mechanical sensors for use outside of clinical settings.^5–7 With an improved understanding and application of the acoustics of the human body at low frequencies, these devices promise to measure and trend cardiovascular metrics unobtrusively and at lower cost than existing techniques.

Although BCG has been studied in air, to our knowledge it has never been investigated as a means of monitoring cardiovascular physiology underwater. A mechanical sensor can be completely sealed against moisture and affixed to the body conveniently with a wrap or sleeve. This electrode-less approach would greatly simplify cardiac monitoring in aquatic environments.

Immersion up to the neck induces a series of physiological responses in the body that have been well-studied in previous work. Two key mechanisms by which the cardiovascular physiology is acutely affected following immersion are (1) a displacement of blood volume from the periphery to the thorax, and (2) peripheral vasoconstriction or vasodilation associated with cold or warm water, respectively. Rapid displacement of blood to the thorax causes a 32%–62% increase in cardiac output (CO) and stroke volume (SV) with little change in heart rate (HR) or blood pressure (BP). Vasodilation in warm water causes an increase in peripheral blood flow and corresponding decrease in total peripheral resistance (TPR); this effect is opposite in cold water.⁸ 

Three healthy male adults (age, 22–32 years; height, 178–190 cm; weight, 67–88 kg; systolic BP, 102–121 mmHg; diastolic BP, 64–82 mmHg) volunteered for this preliminary study. The study was approved by the Georgia Institute of Technology Institutional Review Board (IRB), and the subjects provided written consent.

First, a waterproof single-axis ultra-low noise accelerometer (5958-A, Brüel & Kjær, Nærum, Denmark) was placed on each subject's sternum with the measurement axis oriented head-to-foot. This particular accelerometer was selected for its low noise (400 μg RMS for a bandwidth of 1–14 000 Hz), wide bandwidth (1–8000 Hz, ±2 dB), and reasonably small size (44 g, 19 mm diameter × 55 mm length). The direction of the positive axis was toward the head, and the sensor was secured with an elastic bandage (ACE, 3M, St. Paul, MN) wrapped around the body. The sensor's cable was also secured to the waist with some slack to minimize undesirable signal artifacts from the cable's movement in the water. The accelerometer signal was amplified with a custom circuit (G = 60 dB, BW = 125 Hz) and captured with a 1 kHz sampling rate using a USB oscilloscope (Analog Discovery, Digilent, Pullman, WA). The waveforms were saved for offline processing.

Each trial consisted of 30 s of rest, a 15-s Valsalva maneuver, and 30 s of recovery; however, only the resting period was examined in this study. In addition to the natural hemodynamic changes expected due to immersion, the subjects' physiologic states were further modulated by modifying water, or skin, temperature (T_SK); trials in both warm water (T_SK = 36.7 °C) and cold water (T_SK = 27.8 °C) were performed and compared with trials in air (T_ambient = 22 °C). The resting period of a recording in warm water while kneeling appears in Fig. 1. Each subject was given at least 2 min to acclimate before each trial, and separate recordings were taken with the subjects in several different configurations as shown in Fig. 2. Trials in air were recorded first followed by immersion in cold water and finally warm water. The subjects breathed freely in all trials, and the subjects' heads were fully above the water in the standing and kneeling cases. The subjects' feet were placed on the floor, and the subjects' arms hung loose at their sides.

The underwater (UW) BCG signals were first bandpass filtered (BW = 2–50 Hz) in matlab (Mathworks, Natick, MA). While previous work has often used cutoff frequencies lower than 2 Hz, this value was chosen to remove low-frequency artifacts from the movement of the body due to the water. This effect was particularly pronounced when the body was partially afloat during the supine trials. The upper cutoff frequency of 50 Hz was chosen after the FFT revealed that most of the energy was below this frequency, consistent with the existing literature.⁹ 

Next, ensemble averages (EA) of each recording were computed. Ensemble averaging is employed to improve the signal-to-noise ratio (SNR) of a repeating event—in this case, the heartbeats in the UW BCG signal—by computing the sample-by-sample average over multiple instances. A fiducial point is chosen in the waveform, or a separate simultaneous waveform such as an ECG, and used as a reference point to align the instances. The EA is then calculated as shown in Eq. (1) where n is the current sample index, i is the index of the current heartbeat in the set, and k_i is the offset of each heartbeat as determined with the fiducial points,

A certain number of low-noise heartbeats were hand-picked via examination since motion noise corrupted several heartbeats in every data set, and the location of the maximum in each squared UW BCG heartbeat signal was used as the fiducial point k_i as shown in Eq. (2),

Finally, two scalar measurements were computed from the EAs as shown in Eqs. (3) and (4): the time-domain root mean square (RMS) and the median frequency (MF), respectively. MF is a technique commonly applied to study muscle fatigue with surface electromyography (EMG) and is the frequency f_MED where half of the total power in the power spectral density (PSD) lies below and above f_MED.¹⁰ MF was chosen to better understand the frequency domain effects of immersion and temperature on the UW BCG. In Eq. (4), P(f) is the PSD of the EA and f_S is the sample rate,

Figure 2 shows the EA for each trial, and Fig. 3 shows the PSDs of the EAs of the standing trials. Because of motion of the water during the supine trials, the signals captured for this posture were corrupted by motion artifacts and yielded poor results while the standing and kneeling trials provide much better insight. The mean increase in the RMS UW BCG from air to cold water immersion while standing was 360 μg [standard deviation (SD), 49], or 36%. The average increase in the MF for these trials was −8.8 Hz (SD, 2.8), or −54%. Thus, a large shift toward lower frequencies also occurred during water immersion. The differences between cold water and warm water immersion while kneeling were less clear. The mean increase in the RMS UW BCG was −450 μg (SD ± 466), and the mean increase in the median frequency (MF) was −0.83 Hz (SD ± 1.61).

The mean air-to-cold-water change in the RMS of the UW BCG of 36% is particularly notable for two reasons. First, a previous work demonstrated that changes in the RMS BCG measured with a weighing scale trended with changes in CO.¹¹ Although UW BCG, as a measure of acceleration, is related primarily by double integration to the weighing scale BCG,^6,7 relative changes in RMS amplitude can be compared due to linearity of integration. Second, existing physiology literature shows that the change in CO from open air to total body immersion up to the neck is between 32% and 62%.^12,13 The average change in the RMS of the UW BCG in this study thus appears to be within the expected change in CO during immersion suggesting that this measurement may track changes in CO during water immersion. However, the change in the RMS signal of −27% from cold to warm water immersion while kneeling is interesting because little change in CO is normally expected in the case of cold-to-thermoneutral water immersion. This may be attributed to inter-subject variability and small sample size, as well as open research questions regarding cardiovascular physiology effects of warm water immersion.

The large reduction in the median frequency from air to cold water while standing compared to the negligible change from cold to warm water was also particularly interesting. We speculate that the cause is rapid translocation of ∼700 mL of blood to the thorax that occurs upon immersion since such a change would likely affect the mechanical system of the body and therefore the BCG signal. This raises questions to us about whether median frequency could be used to trend changes in blood volume or distribution.

An important limitation of this study that must be noted is the small sample size of only three subjects. The purpose of the work was to demonstrate, using a preliminary proof-of-concept study, that BCG measurements are feasible with subjects immersed in water and the initial features of these BCG signals are consistent with physiological expectations.

In this paper, we presented and analyzed preliminary results of what we believe is the first study comparing vibrations of the human body to physiologic changes during water immersion. A new signal we call the UW BCG measures vibrations of the sternum in the head-to-foot direction with an ultra-low noise accelerometer. Although the sample size was small, the results demonstrated clear changes in the various temperatures, and especially in immersion, versus recordings in air. These exciting preliminary results support the idea that it may be possible to measure hemodynamic changes underwater without electrodes using body vibrations and warrant further investigation.