Open Access
Add to BibSonomy
Abstract
Non-invasive continuous blood pressure monitoring remains elusive. There has been extensive research using the photoplethysmographic (PPG) waveform for blood pressure estimation, but improvements in accuracy are still needed before clinical use. Here we explored the use of an emerging technique, speckle contrast optical spectroscopy (SCOS), for blood pressure estimation. SCOS provides measurements of both blood volume changes (PPG) and blood flow index (BFi) changes during the cardiac cycle, and thus provides a richer set of parameters compared to traditional PPG. SCOS measurements were taken on the finger and wrists of 13 subjects. We investigated the correlations between features extracted from both the PPG and BFi waveforms with blood pressure. Features from the BFi waveforms were more significantly correlated with blood pressure than PPG features (R = − 0.55, p = 1.1 × 10−4 for the top BFi feature versus R = − 0.53, p = 8.4 × 10−4 for the top PPG feature). Importantly, we also found that features combining BFi and PPG data were highly correlated with changes in blood pressure (R = − 0.59, p = 1.7 × 10−4). These results suggest that the incorporation of BFi measurements should be further explored as a means to improve blood pressure estimation using non-invasive optical techniques.
© 2023 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
Non-invasive, continuous, and reliable blood pressure monitoring is a long sought after goal in the biomedical research community [1]. The current gold standard of blood pressure monitoring is the cuff-based sphygmomanometer, which can only provide intermittent measurements of blood pressure [1]. This limits blood pressure monitoring during daily activities, such as sleep. In addition, it is difficult to monitor acute changes in blood pressure that may be associated with a cardiac event. The Centers for Disease Control and Prevention (CDC) estimates that nearly half of the adults in the United States have hypertension, and one in five adults are unaware of their hypertension and thus remain untreated [2]. The ability to continuously and non-invasively monitor blood pressure could greatly improve diagnosis of hypertension, also known as the “silent killer”.
Photoplethysmography (PPG) is ubiquitous in clinical settings for heart rate monitoring, and when conducted with at least two wavelengths that are sensitive to hemoglobin oxygen saturation, it may be used to perform pulse oximetry. Recently, pulse oximeters have been integrated into smart watches, enabling continuous monitoring of peripheral arterial oxygen saturation for the general population [1]. PPG measurements in adults are typically performed in transmission mode on the finger, or in reflection mode on the wrist. The PPG signal itself arises from changes in blood volume and a consequent modulation in light absorption. For transmission-based PPG measurements, red or near infrared (NIR) light is often used because this wavelength is sensitive to hemoglobin absorption and can penetrate through the finger for detection [3]. Reflection-based PPG measurements used in heart rate monitoring are typically performed using green light, whereas red or NIR light is typically used for oxygen saturation measurements [4]. There has been extensive research utilizing the PPG signal to estimate blood pressure, and various strategies have been applied. For example, physiological models such as the Windkessel and the Moens-Korteweg models have used the PPG signal as an input to estimate blood pressure (BP) [1,5–11]. Alternatively, features extracted from the PPG waveform have been shown to correlate with BP. The most common features include the slope of the systolic upstroke, the temporal location and amplitude of the dichrotic notch, the temporal width of the PPG waveform, and the amplitude of the second derivative of the PPG waveform peaks [1,12–18]. These features have been used as inputs to a wide range of machine learning techniques for BP estimation, including support vector machine learning, adaptive boosting regression, random forest regression, decision trees, ensemble trees, Gaussian process regression artificial neural networks, and autoregressive models [15,19–24]. Despite the extensive work utilizing the PPG signal for BP estimation, there are few examples of continuous non-invasive BP technologies deployed in clinical settings.
Speckle contrast optical spectroscopy (SCOS) utilizes a long coherence length laser and a complementary metal oxide semiconductor (CMOS) or a charge-coupled device (CCD) sensor to measure the time-evolving interference patterns, i.e. speckle pattern, generated from diffuse scattering of the laser light on tissue [25]. The speckle pattern temporal fluctuations depend on the rate of blood flow in the tissue. When speckle images are acquired at a sufficient sampling rate, changes in blood flow within each cardiac pulse can be resolved. SCOS has traditionally been used for monitoring blood flow and perfusion [26,27]. However, there has been minimal literature investigating the morphology of the pulsatile speckle waveform itself. Prior work by Ghijsen et al. suggested that the speckle waveform may be correlated with cardiovascular health parameters [28]. A 785 nm laser diode was used to simultaneously measure speckle contrast and intensity changes (PPG) through finger. They found that the morphology of the PPG waveform and the pulsatile blood flow index (BFi) waveform were distinct. In addition, the time delay between the two waveforms was highly correlated with age. They also found that the pulsatile speckle signal was more robust to changes in perfusion compared with the PPG signal [28]. Dunn et al. similarly measured simultaneous PPG and BFi from the finger [29]. They collected ECG signals and compared estimation of heart rate variability (HRV) using PPG to HRV estimation using BFi. They found that BFi signals were highly correlated with the ECG estimation of HRV and outperformed PPG estimation, and hypothesized that the superior performance of BFi was due to BFi’s lower sensitivity to temperature and motion artifacts [29]. Additional literature suggests that the pulsatile speckle waveform may provide better classification of peripheral artery disease progression than traditional methods [30–33]. In summary, this body of literature thus far suggests that the pulsatile speckle waveform contains abundant, yet unexplored, features highly relevant to cardiovascular health. When combined with the PPG waveform, we hypothesized that the speckle waveform has the potential to improve BP estimation.
Here, we used SCOS to measure 13 subjects and extract both PPG and BFi waveforms. We then investigated the correlations between PPG and BFi pulse waveforms (PWFs) with BP before and after an exercise activity.
2. Methods
2.1 Optical instrumentation
The optical set up is shown in Fig. 1. It consisted of two long coherence length lasers at wavelengths of 532 nm (Crystalaser, model CL532-075-SO) and 808 nm (Crystalaser, model DL808-100-S30) coupled into multimode fibers (0.22 NA, 50 µm core diameter, Thorlabs M42L01). Wavelengths of 532 nm and 808 nm were chosen because they probe different tissue volumes. 532 nm is more strongly absorbed by tissue and penetrates less deeply, and has been used frequently in wrist PPG measurements [4]. In contrast, 808 nm will penetrate more deeply into tissue. In addition, 808 nm is the isosbestic point between oxy and deoxyhemoglobin absorption, allowing us to decouple changes in blood volume from changes in oxy and deoxyhemoglobin concentration. A 50:50 fiber optic splitter was used to split the 808 nm laser light into two fibers. One 808 nm source fiber was placed on the finger, the other on the wrist, and the 532 nm source fiber was placed 1 cm away from the 808 nm source on the wrist. The light was collected using three detector fibers, one for each source fiber. For the finger, the detector was placed on the opposite side of the finger as the source fiber to measure the light transmitted through the finger. Transmissive SCOS measurements taken on the finger were previously shown to correlate with age [28]. In addition, we were interested in exploring the differences in morphology and signal quality between BFi and PPG waveforms collected in transmission and reflection mode. For the wrist measurements, the detection fiber was placed 3.5 mm and 4.5 mm from the source for the 532 nm and 808 nm wrist measurements, respectively. The three detector fiber bundles had 3770 individual fibers with a 37 µm core diameter and an NA of 0.66 (Fiber Optics Technology, Inc, FTIMG25911). The detection fibers were imaged onto two cameras (Basler boA1936-409cmHSP) using two plano convex lenses (Edmund Optics, #67-581 and #67-531). Camera 1 measured the 808 nm light, and camera 2 measured the 532 nm light. A low-pass filter with a cut off wavelength of 700 nm (Thorlabs FELH0700) was placed between the two plano-convex lenses prior to camera 2 to block 808 nm light. Likewise, a high-pass filter with a cut off wavelength of 700 nm (Thorlabs FESH0700) was placed prior to camera 1 to block 532 nm light.
Fig. 1. Optical set up. Two long coherence lasers were coupled into multimode fibers and placed on the finger and wrist. The transmitted and reflected light was collected by detection fibers and imaged onto two cameras using two plano convex lenses. High and low pass filters were placed in between the lenses of camera 1 and 2, respectively. The cameras were triggered simultaneously by a frame grabber, which also collected the data.
2.2 Experimental protocol
For each subject, optical and blood pressure measurements were acquired at four different time points. Prior to data collection, the wrist measurement location was selected by collecting images at a frame rate of 25 Hz while scanning the probe over the volar side of the wrist until cardiac pulses were visible. For all subjects, pulses appeared most clearly near the radial artery, as confirmed by palpation. This is consistent with the literature suggesting that the reflective PPG signal on the wrist is strongest near the radial artery [4,34]. After the wrist measurement location was identified, an initial baseline measurement was taken. Subjects were then asked to walk up and down seven flights of stairs to induce blood pressure changes [28,35]. Immediately after the stair exercise, two more optical and blood pressure measurements were obtained consecutively. As the first measurement post-exercise was typically performed four minutes after the completion of the exercise, this measurement will be referred to as ‘Exercise + 4’ hereafter. Similarly, the second measurement post-exercise was typically performed eight minutes after exercise was completed, and will be referred to as ‘Exercise + 8’. Fifteen minutes after the completion of the stair exercise, a final optical and blood pressure measurement was performed, termed ‘Exercise + 15’. Participants were asked to sit and relax until all post-exercise measurements were completed. The measurement protocol is shown in Fig. 2. Each optical measurement consisted of one minute long 532 nm and 808 nm reflective SCOS measurements on the wrist, and an 808 nm transmissive measurement on the finger. The images were collected at 390 Hz and a 2.5 ms exposure time, and each image was 400 × 400 pixels. Blood pressure was simultaneously collected from the contralateral arm using the Finapres Nova (Finapres Medical Systems) continuous blood pressure monitor [10,24,36]. The Finapres measures blood pressure at a sampling rate of 200 Hz, and this data was collected by a National Instruments DAQ (NI USB 6361) with an acquisition rate of 1000 Hz. The blood pressure and optical data were then aligned in time with the optical data using an exposure active transistor- transistor logic (TTL) signal from the camera.
Fig. 2. Left: measurement protocol. Right: representative subject with Finapres blood pressure measurements being taken. Inset: subject’s left arm with optical probes attached. Exercise + 4 refers to the first optical and blood pressure measurements taken after the stair exercise, typically taken 4 minutes after the stair exercise. Likewise, Exercise + 8 and Exercise + 15 refer to the subsequent measurements taken 8 and 15 minutes post exercise. SCOS: speckle contrast optical spectroscopy.
2.3 Subject pool
A total of 13 healthy volunteers were recruited. The subject’s age, height, weight, and Fitzpatrick score were collected. The Fitzpatrick score was used to quantify skin tone [37]. Each subject was asked to select a number between one and six depending on how easily their skin burns when exposed to sun, with 6 corresponding to never burning and the darkest skin tone, and 1 corresponding to always burning and the lightest skin tone. The mean +/- standard deviation (SD) age of the subjects was 24.3 +/- 1.6 years, and the Fitzpatrick number was 2.9 +/- 0.8.
2.4 Data processing
The data processing pipeline is as shown in Fig. 3. The PPG signal is obtained from the intensity fluctuations relative to baseline. Optical blood flow measurements are calculated from the squared value of the speckle contrast, which is the spatial variance of the speckle image divided by squared spatial mean intensity [38]. The blood flow index (BFi), a unitless value proportional to the Brownian motion of blood, is inversely proportional to the squared speckle contrast [39]. However, spatial variance may be caused by other noise sources, including shot, read, and quantization noise. Their contribution is estimated and subtracted to isolate the fundamental squared contrast arising from the speckle patterns.
Fig. 3. Data processing pipeline. K and I were calculated for each 7 × 7 pixel sliding window (represented by the yellow square). The average K and I for each image were then calculated.
To obtain the PPG time series, a 400 × 400 pixel region of interest (ROI) imaging the fiber bundle was selected on each camera. Intensity values in the ROI pixels and across frames were averaged to obtain a baseline intensity, ${\bar{I}_{baseline}} = \; {\langle\langle I\rangle_{pix}}\rangle_t$, for the ROI. Then, the PPG signal was calculated as the natural log of the baseline intensity divided by spatially averaged intensity in each frame, $PPG = \textrm{ln}({{\raise0.7ex\hbox{${{{\bar{I}}_{baseline}}}$} \!\mathord{\left/ {\vphantom {{{{\bar{I}}_{baseline}}} {{\langle I\rangle_{pix}}}}} \right.}\!\lower0.7ex\hbox{${{\langle I\rangle_{pix}}}$}}} )$.
To obtain the BFi time series, speckle contrast and intensity were calculated for each image. The ROI was divided into 7 × 7 pixel windows with maximum overlap along both x and y axis to reduce the effects of variance introduced by illumination heterogeneity. For each window, spatial variance and squared mean intensity were calculated to determine the measured squared contrast, $K_{all}^2 = \; \; {\sigma ^2}/{\bar{I}^2}$. The measured spatial variance contains contributions from shot, read, and quantization noise [40].
Shot noise contribution was estimated from mean intensity and camera gain. Shot noise follows a Poisson distribution with a variance equal to the number of photoelectrons generated at each pixel. The camera amplifies and digitizes the photoelectrons to camera counts. Thus, shot noise variance scales proportionally to this gain as $\sigma _{shot}^2 = gI$, where $g$ = gain (analog-digital conversion units (ADU)/photoelectrons) and $I\; $= intensity (in camera counts). Camera gain was estimated prior to the experiment by measuring the slope of the relationship between variance and mean intensity from incoherent illumination (e.g. LED) at different exposure times. Read noise is assumed to follow a normal distribution, with zero mean and a constant standard deviation. Read noise was measured by capturing images with no illumination (cap on aperture) and constant offset added to ensure that the read noise induced a normal distribution that is within the camera’s dynamic range. Read noise induced variance was calculated as the temporal variance averaged across pixels, $\sigma _r^2 = \frac{1}{{n - 1}}\sum {({{\langle I\rangle_t} - I} )^2}_{pix}$. Quantization noise induced variance is 1/12 with less than $\mathrm{1\ \times 1}{\textrm{0}^{\textrm{ - 3}}}$ deviation when the signal’s true variance is 0.42 and quantization step size is 1 ([41–43]). For each noise contribution, the corresponding squared contrast was calculated by dividing the noise variance by the squared mean intensity. Then the BFi was calculated as ${\raise0.7ex\hbox{$1$} \!\mathord{\left/ {\vphantom {1 {K_F^2}}} \right.}\!\lower0.7ex\hbox{${K_F^2}$}}$, where ${K_F}$ is the fundamental speckle contrast.
Although PPG and BFi PWFs can be obtained from all three measurement locations, the BFi PWFs from the 532 nm wrist measurement and the PPG PWFs from the 808 nm wrist measurement were too noisy in the majority of subjects to reliably extract features. Therefore all data analysis in this paper was performed using the PPG and BFi PWFs from the finger, the PPG signal from the 532 nm wrist measurement, and the BFi signal from the 808 nm wrist measurement.
The PPG and BFi pulses from each measurement were averaged to obtain an average pulse from which features were extracted. To find the average pulse waveform, we aligned the pulse peaks and averaged each pulse. The pulse peaks were identified by calculating the derivative of the data between 0.5 to 1.25 periods of the expected pulse to pulse duration. The resulting time series was used to find the first time point the derivative crosses zero after the peak positive value. That time point is defined as the time of the pulse peak. The process repeats, identifying the next peak between 0.5 and 1.25 periods after the end of the previous peak. After the pulse peaks were found, data 0.3 periods before and 0.7 periods after each pulse peak were found. Each peak was then averaged. For features that are dependent on the arrival times of pulses, the signals were first filtered with a Chebyshev 2 low pass filter with a cut off frequency of 18 Hz to remove high frequency noise (Chebyshev2, Matlab 2021b). Examples of raw data and average waveforms for each measurement are shown in Fig. 4. All data processing and analysis was performed in Matlab R2021b (Mathworks, inc).
Fig. 4. Examples of raw data and average waveforms for each measurement location and type. The average plots show the average pulse waveform of all pulses within the one minute measurement (red) overlaid on each indivivual pulse (black). The average pulse is plotted over one heart rate period. The 532 nm wrist blood flow index (BFi) measurement and the wrist 808 nm photoplethysmography (PPG) measurements were too noisy in the majority of subjects for reliable feature extraction.
3. Analysis
3.1 Feature extraction
In total, 16 features were extracted from the PPG and BFi PWFs (Table 1).
Table 1. List of features extracted from PWFs. Features highlighted in blue were extracted from the BFi or PPG waveforms alone. Features highlighted in yellow were extracted from a combination of BFi and PPG features.
Figure 5 shows a visual representation of these features. All features extracted from the average waveforms were dependent on the identification of the base and maximum of the systolic peak (Fig. 5). To identify the peak max, the maximum value of the average waveform was selected. All waveforms were visually inspected to confirm accurate selection of these two points. Eight additional features were calculated from these points. Previous work has shown that the time delay between the BFi (earlier) and PPG (later) PWF peaks contains information related to cardiovascular health parameters such as age and vascular resistance [28]. We were interested in investigating the correlations between different time delays in our PWFs and blood pressure. To extract time delays, peaks were identified for each signal by first roughly identifying each peak using Matlab’s findpeaks function. In addition, the period of the cardiac cycle (Tc) was determined based on the heart rate found as the maximum of the Fourier transform of the signal within the range of 0.2 to 3.0 Hz. For each initially identified peak, the region 0.5*Tc before and after the peak was selected and the maximum value within that region was selected as the peak. The average time delays between PPG and BFi PWFs on the finger were determined, as well as time delays between the 532 nm PPG and 808 nm BFi signals on the wrist, and the time delays between those features and the PPG and BFi PWFs from the finger (Fig. 5). In some cases, noise in measurements prevented reliable feature extraction and that subject’s data was removed. Noise was likely due to motion during the measurement or shifting of the probe location between measurement time points.
Fig. 5. A) and B) examples of features extracted from average waveforms. The base and max of the systolic peak are represented by yellow circles. C) Example of time delays extracted form the time series data. The time delay is defined as the time between the two identified peaks within a single cardiac cycle.
3.2 Correlation with blood pressure
The blood pressure data for each subject consisted of four minute long blood pressure waveforms acquired with the Finapres device at each measurement time point. To identify the average systolic blood pressure for each measurement, the peaks were identified using Matlab’s findpeaks function and the values were averaged. The same function was then applied on the inverted signal to identify the diastolic troughs, which were averaged to obtain the diastolic blood pressure.
For each extracted feature, the correlation between systolic and diastolic blood pressure for that feature was determined using the Pearson’s linear correlation coefficient R, and a p-value was computed. In addition, we identified the correlation between feature changes relative to the baseline condition with blood pressure changes relative to baseline. Linear regression was performed to determine the relationship between each feature and blood pressure. Based on that relationship, an estimate for blood pressure from that feature was determined and the mean absolute error between the estimated blood pressure and the actual blood pressure was found.
4. Results
4.1 Blood pressure changes
Fig. 6. Systolic (left) and diastolic (right) blood pressure changes over each measurement condition. The middle line represents the median. The bottom and top of the box and whisker plots represent the 25th and 75th percentiles, respectively. The whiskers represent +/- 2.7 standard deviations.
The changes in systolic blood pressure (SBP) and diastolic blood pressure (DBP) at each measurement time point are shown in Fig. 6. The mean increase in systolic blood pressure between baseline and Exercise + 4 was 12.3 +/- 13.3 mmHg, and the mean increase in diastolic blood pressure was 4.7 +/- 8.3mmHg.
4.2 Correlation with blood pressure
Tables 2 and 3 show the correlation between extracted features and blood pressure for the ten most signigicant features. Statistical significance of the linear relationship is indicated by an asterisk, and the threshold of significance was determined using the Benjamini-Hochberg procedure, which takes into account multiple comparisons [44].
Table 2. Ten most significantly correlated features with SBP. Asterisk indicates statistical significance. N is the number of subjects for each measurement. Symbol Δ indicates correlation between the change in feature value relative to baseline and the change in blood pressure relative to baseline.
Table 3. Ten most significantly correlated features with DBP. Asterisk indicates statistical significance. N is the number of subjects for each measurement.
Of the features, diastolic time and $\textrm{T}{\textrm{d}_{\textrm{BFi}}}\mathrm{\ast T}{\textrm{s}_{\textrm{PPG}}}$ extracted from the finger BFi data had the strongest correlations with SBP. $\textrm{T}{\textrm{d}_{\textrm{BFi}}}\mathrm{\ast T}{\textrm{s}_{\textrm{PPG}}}$ can be interpreted as the diastolic time from the BFi PWF scaled by the systolic time from the PPG PWFs, combining temporal dynamics between the PPG and BFi waveforms. In addition, the changes in time delays between PPG and BFi peaks calculated from the finger data were highly correlated with changes in SBP across measurement conditions. Overall, features extracted from the BFi PWFs or BFi and PPG PWFs combined were more significantly correlated with systolic blood pressure than those extracted from solely the PPG PWFs, indicating that the BFi PWFs may contain additional information related to systolic blood pressure compared with PPG alone. Similarly, the diastolic time extracted from the finger BFi waveforms was most significantly correlated with diastolic blood pressure, suggesting the BFi PWFs may also be used to estimate diastolic blood pressure. Correlation plots for the top two most highly correlated features with SBP are shown in Fig. 7(A) and (B). In addition, a correlation plot for the $\Delta $ Time delay between finger BFi and PPG feature and the most significantly correlated feature with DBP are shown in Fig. 7(C) and (D).
Fig. 7. A: Correlation plot for the most significantly correlated feature with SBP. B: Correlation plot for the second most significantly correlated feature with SBP. C: Correlation plot for the change in time delay between BFi and PPG relative to baseline. D: Correlation plot for the significantly correlated feature with DBP. Red line represents the linear line of best fit.
5. Discussion
Here we present SCOS as a method to estimate BP. SCOS allows extraction of both the PPG and BFi pulsatile waveforms from a single laser speckle measurement. The two signals are temporally and morphologically distinct, and each may provide useful information about cardiovascular health. We found that features extracted from the BFi waveforms were typically more significantly correlated with blood pressure than features extracted from the PPG waveforms. In addition, features combining information extracted from both the BFi and PPG PWFs had strong correlations with both SBP and DBP.
The changes in time delays between the PPG and BFi peaks on the finger were strongly correlated with changes in SBP. This feature has previously been correlated with age and it was hypothesized that the time delays may be influenced by arterial stiffness and compliance [23]. When blood pressure was elevated following exercise, the time delay between the BFi and PPG waveforms shortened. This could be due to decreases in arterial resistance following exercise. Since blood pressure is affected, in part, by arterial resistance, the ability to optically probe this parameter could help to improve blood pressure estimation.
Features related to diastolic and systolic time were highly correlated with both SBP and DBP. For example, the combined feature of $\textrm{T}{\textrm{d}_{\textrm{BFi}}}\mathrm{\ast T}{\textrm{s}_{\textrm{PPG}}}$ was highly correlated with SBP. Systolic time may be influenced by both arterial resistance and compliance, which could both affect the time it takes for an arterial to reach maximum expansion ($\textrm{T}{\textrm{s}_{\textrm{PPG}}}$) and maximum flow rate ($\textrm{T}{\textrm{s}_{\textrm{BFi}}}$). Diastolic time is related to both systolic time and heart rate. Our results suggest that the combination of time related features from BFi and PPG PWFs may enable the probing of complex cardiovascular dynamics related to compliance, resistance, and heart rate.
BFi signals from the finger were typically less noisy than the PPG signals. Previous work has shown that the BFi signal may be more robust to low perfusion conditions than the PPG signal [45]. The decreased presence of noise in the finger BFi data may improve the reliability of feature extraction and enable accurate feature extraction despite low perfusion conditions, which has been a challenge in previous work utilizing the PPG signal to estimate blood pressure [1]. On some subjects, clear inverted PPG waveforms were observed in the 808 nm reflectance PPG measurement on the wrist, although the BFi signals extracted from the same dataset were not inverted. We hypothesize that the presence of inverted signals in the 808 nm PPG reflectance measurements may have led to increased noise. Conversely, the 532 nm reflectance BFi measurements on the wrist typically had higher noise compared to the PPG signal. The physiological mechanisms underlying these differences in noise level are unclear.
The strongest correlations were typically found in data collected from the finger, which may be due to several factors. The finger measurement may be less sensitive to the location of the probe compared with the wrist measurements. Slight movement of the probe between measurements may be a confounding factor in feature extraction from wrist data. For the 532 nm wrist PPG signals, the peaks are typically flatter than the PPG measured from the finger, which may reduce reliability of systolic peak extraction. In future work, a more robust peak detection algorithm may be utilized to minimize this effect.
One limitation of this study is the small sample size. In addition, the subject pool was mainly comprised of young, healthy subjects with medium skin tones. In order to fully validate the potential for SCOS to estimate blood pressure, a larger and more diverse subject pool must be measured. Using data from a larger subject pool, multiple features extracted from the BFi and PPG waveforms could be input into models or machine learning algorithms for blood pressure estimation. This would enable a more direct comparison of blood pressure estimation between PPG, BFi and PPG + BFi features. Previous work has shown that by using multiple features as inputs to machine learning algorithms and applying feature selection and optimization algorithms, the blood pressure estimation can be greatly improved [15,19–24]. An additional limitation is that we used a stair exercise to induce blood pressure changes. However, this exercise also induces changes in heart rate. Future work will aim to decouple heart rate and blood pressure changes by using a blood pressure perturbation method that has less impact on heart rate.
6. Conclusion
In conclusion, we have investigated the correlation between features extracted from PPG and BFi PWFs on the finger and wrist measured with SCOS. We found that features extracted from the BFi waveforms were highly correlated with both SBP and DBP, and were more significantly correlated with BP than features extracted from the PPG features alone. In addition, the combination of features from the PPG and BFi PWFs yielded high correlations with both SBP and DBP. We also found that the time delays between the PPG and BFi peaks were strongly correlated with changes in SBP. These results indicate that the BFi PWF contains information related to cardiovascular state that could provide a means to improve non-invasive blood pressure estimations either alone or in combination with the PPG PWF.
Funding
Meta.
Acknowledgements
This research was supported by funding provided by Meta Platforms Inc.
Disclosures
DB: Meta Platforms Inc. (F), DR: Meta Platforms Inc (F).
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
References
1. M. Elgendi, R. Fletcher, Y. Liang, N. Howard, N.H. Lovell, D. Abbott, K. Lim, and R. Ward, “The use of photoplethysmography for assessing hypertension,” NPJ Digit. Med. 2(1), 1–11 (2019). [CrossRef]
2. Centers for Disease Control and Prevention: Facts about Hypertension, https://www.cdc.gov/bloodpressure/facts.htm, 2021.
3. A. A. Alian and K. H. Shelley, “Photoplethysmography,” Best Practice & Research Clinical Anaesthesiology 28(4), 395–406 (2014). [CrossRef]
4. A. A. Kamshilin and N.B. Margaryants, “Origin of photoplethysmographic waveform at green light,” Phys. Procedia 86, 72–80 (2017). [CrossRef]
5. J. Liu, B.P.-Y. Yan, W.-X. Dai, X.-R. Ding, Y.-T. Zhang, and N. Zhao, “Multi-wavelength photoplethysmography method for skin arterial pulse extraction,” Biomed. Opt. Express 7(10), 4313–4326 (2016). [CrossRef]
6. J. Liu, Y.-T. Zhang, X.-R. Ding, W.-X. Dai, and N. Zhao, “A preliminary study on multi-wavelength PPG based pulse transit time detection for cuffless blood pressure measurement,” in 2016 38th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) (IEEE, 2016).
7. C.C.Y., Poon and Y.T. Zhang, “Cuff-less and Noninvasive Measurements of Arterial Blood Pressure by Pulse Transit Time,” in 2005 IEEE Engineering in Medicine and Biology 27th Annual Conference (IEEE, 2005).
8. R. Mukkamala, J.-O. Hahn, O.T. Inan, L.K. Mestha, C.-S. Kim, H. Töreyin, and S. Kyal, “Toward ubiquitous blood pressure monitoring via pulse transit time: theory and practice,” IEEE Trans. Biomed. Eng. 62(8), 1879–1901 (2015). [CrossRef]
9. R.C. Block, M. Yavarimanesh, K. Natarajan, A. Carek, A. Mousavi, A. Chandrasekhar, C.-S. Kim, J. Zhu, G. Schifitto, and L.K. Mestha, “Conventional pulse transit times as markers of blood pressure changes in humans,” Sci. Rep. 10(1), 1–9 (2020). [CrossRef]
10. J. Liu, B.P. Yan, Y.-T. Zhang, X.-R. Ding, P. Su, and N. Zhao, “Multi-wavelength photoplethysmography enabling continuous blood pressure measurement with compact wearable electronics,” IEEE Trans. Biomed. Eng. 66(6), 1514–1525 (2019). [CrossRef]
11. A.D. Choudhury, R. Banerjee, A. Sinha, and S. Kundu, “Estimating blood pressure using Windkessel model on photoplethysmogram,” in 2014 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (IEEE, 2014).
12. P.S. Addison, “Slope transit time (STT): A pulse transit time proxy requiring only a single signal fiducial point,” IEEE Trans. Biomed. Eng. 63(11), 2441–2444 (2016). [CrossRef]
13. Y. Chen, S. Cheng, T. Wang, and T. Ma, “Novel blood pressure estimation method using single photoplethysmography feature,” in 2017 39th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) (IEEE, 2017).
14. M. Elgendi, Y. Liang, and R. Ward, “Toward generating more diagnostic features from photoplethysmogram waveforms,” Diseases 6(1), 20 (2018). [CrossRef]
15. X. Xing, Z. Ma, M. Zhang, Y. Zhou, W. Dong, and M. Song, “An unobtrusive and calibration-free blood pressure estimation method using photoplethysmography and biometrics,” Sci. Rep. 9(1), 1–8 (2019). [CrossRef]
16. M. Elgendi, “Detection of c, d, and e waves in the acceleration photoplethysmogram,” Comput. Methods Programs Biomed. 117(2), 125–136 (2014). [CrossRef]
17. M. Elgendi, I. Norton, M. Brearley, D. Abbott, and D. Schuurmans, “Detection of a and b waves in the acceleration photoplethysmogram,” BioMed. Eng. OnLine 13(1), 1–18 (2014). [CrossRef]
18. W.-H. Lin, H. Wang, O.W. Samuel, G. Liu, Z. Huang, and G. Li, “New photoplethysmogram indicators for improving cuffless and continuous blood pressure estimation accuracy,” Physiological Measurement 39(2), 025005 (2018). [CrossRef]
19. M.H. Chowdhury, M.N.I. Shuzan, M.E. Chowdhury, Z.B. Mahbub, M.M. Uddin, A. Khandakar, and M.B.I. Reaz, “Estimating blood pressure from the photoplethysmogram signal and demographic features using machine learning techniques,” Sensors 20(11), 3127 (2020). [CrossRef]
20. K. Duan, Z. Qian, M. Atef, and G. Wang, “A feature exploration methodology for learning based cuffless blood pressure measurement using photoplethysmography,” in 2016 38th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) (IEEE, 2016).
21. Y. Liang, Z. Chen, R. Ward, and M. Elgendi, “Photoplethysmography and deep learning: enhancing hypertension risk stratification,” Biosensors 8(4), 101 (2018). [CrossRef]
22. G. Acciaroli, A. Facchinetti, G. Pillonetto, and G. Sparacino, “Non-invasive continuous-time blood pressure estimation from a single channel PPG signal using regularized ARX models,” in 2018 40th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) (IEEE, 2018).
23. M.S., Tanveer and M.K. Hasan, “Cuffless blood pressure estimation from electrocardiogram and photoplethysmogram using waveform based ANN-LSTM network,” Biomedical Signal Processing and Control 51, 382–392 (2019). [CrossRef]
24. A.S. Zadi, R.M. Alex, R. Zhang, D.E. Watenpaugh, and K. Behbehani, “MathEMATICAL MODELING OF ARTERIAL BLOOD PRESSURE USING PHOTO- PLETHYSMOGRAPHY SIGNAL IN BREATH-HOLD MANEUVER,” in 2018 40th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) (IEEE, 2018).
25. C.P. Valdes, H.M. Varma, A.K. Kristoffersen, T. Dragojevic, J.P. Culver, and T. Durduran, “Speckle contrast optical spectroscopy, a non-invasive, diffuse optical method for measuring microvascular blood flow in tissue,” Biomed. Opt. Express 5(8), 2769–2784 (2014). [CrossRef]
26. M. Draijer, E. Hondebrink, T. van Leeuwen, and W. Steenbergen, “Review of laser speckle contrast techniques for visualizing tissue perfusion,” Lasers Med. Sci. 24(4), 639–651 (2009). [CrossRef]
27. T. Dragojević, J.L. Hollmann, D. Tamborini, D. Portaluppi, M. Buttafava, J.P. Culver, F. Villa, and T. Durduran, “Compact, multi-exposure speckle contrast optical spectroscopy (SCOS) device for measuring deep tissue blood flow,” Biomed. Opt. Express 9(1), 322–334 (2018). [CrossRef]
28. M. Ghijsen, T.B. Rice, B. Yang, S.M. White, and B.J. Tromberg, “Wearable speckle plethysmography (SPG) for characterizing microvascular flow and resistance,” Biomed. Opt. Express 9(8), 3937–3952 (2018). [CrossRef]
29. C.E. Dunn, D.C. Monroe, C. Crouzet, J.W. Hicks, and B. Choi, “Speckleplethysmographic (SPG) estimation of heart rate variability during an orthostatic challenge,” Sci. Rep. 9(1), 1 (2019). [CrossRef]
30. T.B. Rice, B. Yang, and S. White, “Effect of skin optical absorption on speckleplethysmographic (SPG) signals,” Biomed. Opt. Express 11(9), 5352–5361 (2020). [CrossRef]
31. S. Kikuchi, K. Miyake, Y. Tada, D. Uchida, A. Koya, Y. Saito, T. Ohura, and N. Azuma, “Laser speckle flowgraphy can also be used to show dynamic changes in the blood flow of the skin of the foot after surgical revascularization,” Vascular 27(3), 242–251 (2019). [CrossRef]
32. E. Gkaliagkousi, A. Lazaridis, P. Anyfanti, K. Stavropoulos, K. Imprialos, A. Triantafyllou, K. Mastrogiannis, S. Douma, and M. Doumas, “Assessment of skin microcirculation in primary aldosteronism: impaired microvascular responses compared to essential hypertensives and normotensives,” J. Hum. Hypertens. 36(12), 1066–1071 (2022). [CrossRef]
33. M.K. Razavi, D.P.T. Flanigan, S.M. White, and T.B. Rice, “A real-time blood flow measurement device for patients with peripheral artery disease,” Journal of Vascular and Interventional Radiology 32(3), 453–458 (2021). [CrossRef]
34. H. Lee, H. Ko, and J. Lee, “Reflectance pulse oximetry: practical issues and limitations,” ICT Express 2(4), 195–198 (2016). [CrossRef]
35. J. Proença, J. Muehlsteff, X. Aubert, and P. Carvalho, “Is pulse transit time a good indicator of blood pressure changes during short physical exercise in a young population?”in 2010 Annual International Conference of the IEEE Engineering in Medicine and Biology (IEEE, 2010).
36. Soltan Zadi, A. R. Alex, R. Zhang, D.E. Watenpaugh, and K. Behbehani, “Arterial blood pressure feature estimation using photoplethysmography,” Comput. Biol. Med. 102, 104–111 (2018). [CrossRef]
37. S. Sachdeva, “Fitzpatrick skin typing: applications in dermatology,” Indian J. Dermatol. Venereol. Leprol. 75(1), 93 (2009). [CrossRef]
38. D.A., Boas and A.K. Dunn, “Laser speckle contrast imaging in biomedical optics,” J. Biomed. Opt. 15(1), 011109 (2010). [CrossRef]
39. A., Fercher and J.D. Briers, “Flow visualization by means of single-exposure speckle photography,” Opt. Commun. 37(5), 326–330 (1981). [CrossRef]
40. S. Zilpelwar, E.J. Sie, D. Postnov, A.I. Chen, B. Zimmermann, F. Marsili, D.A. Boas, and X. Cheng, “Model of dynamic speckle evolution for evaluating laser speckle contrast measurements of tissue dynamics,” Biomed. Opt. Express 13(12), 6533–6549 (2022). [CrossRef]
41. B. Oliver, J. Pierce, and C.E. Shannon, “The philosophy of PCM,” Proc. IRE 36(11), 1324–1331 (1948). [CrossRef]
42. W.R. Bennett, “Spectra of quantized signals,” The Bell Syst. Tech. J. 27(3), 446–472 (1948). [CrossRef]
43. W.F. Sheppard, “On the calculation of the most probable values of frequency-constants, for data arranged according to equidistant division of a scale,” Proceedings of the London Mathematical Society 1(1), 353–380 (1897).
44. Y., Benjamini and Y. Hochberg, “Controlling the false discovery rate: a practical and powerful approach to multiple testing,” Journal of the Royal Statistical Society: series B (Methodological) 57(1), 289–300 (1995).
45. C.E. Dunn, B. Lertsakdadet, C. Crouzet, A. Bahani, and B. Choi, “Comparison of speckleplethysmographic (SPG) and photoplethysmographic (PPG) imaging by Monte Carlo simulations and in vivo measurements,” Biomed. Opt. Express 9(9), 4306–4316 (2018). [CrossRef]
