Blood pressure monitoring might one day become as easy as taking a video selfie, according to new research in Circulation: Cardiovascular Imaging, an American Heart Association journal.
Transdermal optical imaging measures blood pressure by detecting blood flow changes in smartphone-captured facial videos.
Ambient light penetrates the skin’s outer layer allowing digital optical sensors in smartphones to visualize and extract blood flow patterns, which transdermal optical imaging models can use to predict blood pressure.
Finding an accessible, easy way to monitor blood pressure is important given that nearly half of American adults have high blood pressure and many don’t even know they have it, according to the American Heart Association.
“High blood pressure is a major contributor to cardiovascular disease – a leading cause of death and disability.
To manage and prevent it, regular monitoring of one’s blood pressure is essential,” said study lead author Kang Lee, Ph.D., professor and research chair in developmental neuroscience at the University of Toronto in Canada.
“Cuff-based blood pressure measuring devices, while highly accurate, are inconvenient and uncomfortable.
Users tend not to follow American Heart Association guidelines and device manufacturers’ suggestion to take multiple measurements each time.”
Lee and his colleagues measured the blood flow of 1,328 Canadian and Chinese adults by capturing two-minute videos using an iPhone equipped with transdermal optical imaging software.
The researchers compared systolic, diastolic and pulse pressure measurements captured from smartphone videos to blood pressure readings using a traditional cuff-based continuous blood pressure measurement device.
The researchers used the data they gathered to teach the technology how to accurately determine blood pressure and pulse from facial blood flow patterns.
They found that on average, transdermal optical imaging predicted systolic blood pressure with nearly 95% accuracy and diastolic blood pressure with pulse pressure at nearly 96% accuracy.
The technology’s high accuracy is within international standards for devices used to measure blood pressure, according to Lee.
Researchers videoed faces in a well-controlled environment with fixed lighting, so it’s unclear whether the technology can accurately measure blood pressure in less controlled environments, including homes.
Also, while the study’s participants had a variety of skin tones, the sample lacked subjects with either extremely dark or fair skin tones. Lee and colleagues are also looking into reducing the needed video length from 2 minutes to 30 seconds, in order to make the technology more user-friendly.
People in the study all had normal blood pressure.
“If future studies confirm our results and show this method can be used to measure blood pressures that are clinically high or low, we will have the option of a contactless and non-invasive method to monitor blood pressures conveniently – perhaps anytime and anywhere – for health management purposes,” Lee said.
“This study shows that facial video can contain some information about systolic blood pressure,” said Ramakrishna Mukkamala, Ph.D., Circulation Imaging editorial author and professor in the Department of Electrical and Computer Engineering at Michigan State University in East Lansing.
“If future studies could confirm this exciting result in hypertensive patients and with video camera measurements made during daily life, then obtaining blood pressure information with a click of a camera may become reality.”
Humans encounter various stressful situations everyday at work, home, and school. Such stress when experienced at high degrees and/or for a long duration of time could lead to cardiovascular diseases, cognitive dysfunctions, and psychological disorders (Kofman et al., 2006; Pan and Li, 2007; Crowley et al., 2011).
Currently, the assessment of stress relies on the analysis of psychometric (e.g., self-report questionnaires) and/or biometric (e.g., electrocardiography) data.
While psychometric data can provide a glimpse into an individual’s psychological state and stress level, it is heavily dependent upon a subjective reflection of events and conditions.
On the other hand, biometric data can provide an objective evaluation of physiological activity that has been demonstrated to correlate well with psychological stress (Sharpley and Gordon, 1999; Tavel, 2001).
However, biometric data are often obtained using instruments which require the attachment of electrodes or sensors onto the body by trained individuals.
This use of physiological measurement instruments can be inconvenient.
Thus, to date, we still face difficulties in monitoring stress levels both reliably and conveniently. The present research aimed to address these difficulties directly.
Over the last half century, research has revealed that human physiological changes in response to psychological stress, such as the amplitude of respiratory sinus arrhythmia (RSA), can reflect individual stress (Porges, 1995).
When individuals encounter a stressful situation where a threat is perceived, their autonomic nervous system (ANS) works to adjust the internal state of their body and react to the situation.
The two branches of ANS, the sympathetic and parasympathetic nervous systems, contribute in stress reaction.
The sympathetic nervous system is concerned with challenges from the external environment, triggering the fight-or-flight response in stressful situations.
In contrast, the parasympathetic nervous system is concerned with returning the body to a resting state or the state of homeostasis. When an individual experiences stress, the parasympathetic nervous system struggles to maintain homeostasis (Porges, 1995).
Thus, an assessment of stress can be obtained by examining the level of homeostasis.
As part of the parasympathetic nervous system, the vagus nerve plays an essential role in the regulation of homeostasis because it is responsible for signaling the heart, lungs, and digestive tract to slow down and relax.
The activity of the vagus nerve, otherwise known as vagal tone, would then be indicative of the level of homeostasis within the body.
If individual stress decreases, then vagal tone increases, the heart slows down, and homeostasis is maintained.
If individual stress increases, then vagal tone decreases, the heart quickens, and homeostasis is disrupted.
A recent review of research by Castaldo et al. (2015) showed that parasympathetic vagal activity, as determined by heart rate variability (HRV) time series computed from electrocardiography (ECG) recordings, indeed decreases reliably during sessions involving stress. In addition, irregular increase and decrease of vagal tone would indicate chronic stress.
Although vagal tone can provide insight into an individual’s stress level, the changes in vagal tone cannot be measured directly. Rather, vagal tone and corresponding information involving stress can be measured indirectly but reliably by RSA.
RSA is the rhythmic increase and decrease in the beating of the heart, which occurs in the presence of breathing (Berntson et al., 1997).
The heart rate increases with inhalation and decreases with exhalation.
In order to obtain a measurement of RSA, variations in heartbeat must first be measured. Experimental evidence primarily relies on the use of ECG to observe HRV, analyzing the time period in milliseconds between each R-wave to obtain the R-R Interval (RRI). With information regarding the RRI, inferences can be made about stress. An increasing RRI variation indicates excitation of the vagus nerve as it works to decrease heart rate, and thus we can infer stress level to be low.
A decreasing RRI variation indicates an inhibited vagus nerve, allowing heart rate to increase, and thus we can infer stress level to be high (Castaldo et al., 2015, 2016). However, assessment of RRI is not enough to determine vagal tone.
The issue is that respiration is not the only contributor to variations in heart rate.
There are oscillations at frequencies slower than that of respiration, such as Traube-Hering-Mayer waves, which provides information regarding the sympathetic nervous system rather than the parasympathetic nervous system and stress (Porges, 1986).
Thus, data from ECG recordings must be filtered to obtain various HRV features, including measurement of RSA and in effect an estimate of vagal tone that can provide information regarding individual stress levels.
Based on the evidences of cardiovascular changes in response to stress, we have specifically developed a new imaging technology called transdermal optical imaging (TOI) to assess stress conveniently, contactlessly, and remotely.
This technology uses a conventional digital camera to video record participants’ faces from a distance, analyzing facial blood flow information to obtain participants’ heart rate and HRV.
Our TOI technology is built upon a century of research that has revealed cardiovascular activities to be obtainable via analyses of blood flow changes.
Information regarding blood flow changes reveal cardiovascular changes given that movement of blood from the heart to the rest of the body is part of the cardiovascular system.
These discoveries have lead to the development of various methodologies (e.g., laser Doppler flowmetry, photoplethysmography) that measure cardiovascular activities optically. However, similar to the utilization of electrocardiography, these methodologies require the attachment of sensors to the body, which can be inconvenient.
Transdermal optical imaging overcomes the limitations of current methodologies by utilizing a digital video camera to conveniently, contactlessly, and remotely capture video images of the face for extraction of cardiovascular changes.
This is possible because re-emitted light from underneath the skin is affected by chromophores, primarily hemoglobin and melanin (Nishidate et al., 2004), which have different color signatures.
Given the difference in the color signatures, we can use machine learning to separate images of hemoglobin-rich regions from melanin-rich regions, ultimately obtaining video images of hemoglobin changes under the skin (Figure Figure1; for details, see Lee and Zheng, 2016).
The face is ideal for analysis of blood flow changes because it is rich in vasculature and exposed, allowing us to obtain blood flow information conveniently, contactlessly, and remotely.
In the present study, we examined the validity of TOI in measuring heart rate and HRV, which reflects individual stress.
We measured participants’ cardiovascular activities while they were in a state of rest to assess their basal stress levels.
We used the TOI methodology to obtain facial blood flow data reflecting heart rate, HRV, and basal stress levels. At the same time, in order to validate our TOI methodology, we compared the measurements obtained from TOI with those collected concurrently from an ECG system.
We hypothesized that if there is a high positive correlation between data obtained from TOI and ECG, then cardiovascular changes as assessed by TOI should correspond with those by ECG, which were previously proven to correlate with individual stress. Thus, we would provide evidence to suggest TOI to be a valid methodology for assessing stress conveniently, contactlessly, and remotely.
Transdermal Optical Imaging Analysis
Transdermal optical imaging analysis is a novel imaging method that is capable of isolating hemoglobin concentration (HC) from raw human face images taken from a conventional digital camera.
Light travels beneath the skin, and re-emits after traveling through different skin tissues.
The re-emitted light may then be captured by optical cameras (Anderson, 1991; Stamatas et al., 2004; Demirli et al., 2007). The dominant chromophores affecting the re-emitted light are hemoglobin and melanin (Nishidate et al., 2004).
Since hemoglobin and melanin have different color signatures, it has been found that it is possible to obtain images mainly reflecting HC under the epidermis.
Capitalizing on this, TOI analysis first obtains each captured image, and then performs operations upon the image to generate a corresponding optimized HC image of a participant’s face.
Isolating HC is accomplished by analyzing bitplanes in the video sequence to determine and isolate a set of the bitplanes that provide high signal-to-noise ratio (SNR) with regard to the facial cardiovascular activities.
The determination of high SNR bitplanes is made with reference to a first training set of images constituting the captured video sequence coupled with facial blood flow measurements concurrently taken with FDA approved medical instruments that measure cardiovascular activities on the face (facial blood flows with a laser Doppler machine, and blood pressure waves with a continuous cuff-based oscillatory blood pressure monitor).
With respect to bitplanes, a digital image consists of a certain number of pixels; typically referred to as a configuration of width-times-height. Each pixel has one or more channels associated with it.
Each channel has a dynamic range, typically 8 bits per pixel per channel. For color videos, each image typically has three channels: Red, Green, and Blue (RGB). As such, a bitplane is a view of a single bit of an image across all pixels (i.e., a 1-bit image per bit per channel).
Using the raw images that consist of all bitplanes of all three RGB channels, signals that change over a particular time period (e.g., 120 s) on each of the pixels are extracted. Using the signals from each pixel, machine learning is employed to systematically identify bitplanes that will significantly increase the signal differentiation and bitplanes that will contribute nothing or decrease the signal differentiation.
After discarding the latter, the remaining bitplane images optimally determine the blood flow.
To further improve SNR, the result can be fed back to the machine learning process repeatedly until the SNR reaches an optimal asymptote.
The machine learning process involves manipulating the bitplane vectors using image subtraction and addition to maximize the signal differences in all ROIs over the time period for a portion (e.g., 70, 80, 90%) of the subject data and validate on the remaining subject data. The addition or subtraction is performed in a pixel-wise manner.
The resulting images thus contain information corresponding to HC in each pixel, which were then put together as video images to reflect HC changes in all parts of the face (for details, see Lee and Zheng, 2016).
For the present study, we divided the face into nine regions of interests (ROIs): Forehead Small, Nose Between Eyes, Nose Bridge Full, Nose Tip Small, Right Cheek Narrow, Left Cheek Narrow, Upper Lip, Lower Lip, Chin Small (Figure Figure4). We averaged the data obtained from all pixels in each ROI to further increase SNR. Next, we applied Hilbert-Huang transform to filtered ROI signal (Li et al., 2011).
The transform provided us with the principle frequency component of TOI signal. Using synthesized frequency, peaks of heartbeat were reconstructed to obtain heart rate and the intervals between heartbeats (i.e., RRI) were measured.
Using the above process, each video of participant’s face was analyzed for facial blood flow information that reflects cardiovascular activities that correlate with stress. With a reconstruction of the peaks of heartbeat and a measurement for RRIs, we obtained stress in the same way that we extracted the information from data collected using the BIOPAC ECG. We plotted the RRIs on a Poincaré plot and analyzed for HRV features, specifically SD1/SD2.
To compare the data collected from TOI against those from the BIOPAC ECG, we used MATLAB to assess the agreement and correlation between TOI and BIOPAC measurements, specifically for measures of heart rate and SD1/SD2 (i.e., stress). We assessed agreement by constructing Bland–Altman plots for the measures of heart rate and stress obtained from TOI and BIOPAC.
The Bland–Altman plot is often used to determine the differences between two measurements, with the mean difference signifying bias and the standard deviation of the differences signifying the limits of agreement. We computed 95% limits of agreement for comparison of TOI and BIOPAC measurements. We assessed correlation by calculating for the correlation coefficients between measures of heart rate and stress obtained from TOI and BIOPAC.
More information:Circulation: Cardiovascular Imaging (2019). DOI: 10.1161/CIRCIMAGING.119.008857
Journal information: Circulation: Cardiovascular Imaging
Provided by American Heart Association