2.1.

Design of Imaging Photoplethysmography Hardware

The custom-made iPPG prototype device has been elaborated. To acquire blood volume changes in the deep cutaneous tissue, the reflection mode remote PPG is combined with orthogonal polarization spectral imaging and bispectral light sources, exploiting the principle of different wavelength penetration into different depths of the skin.^25,27

The custom-made iPPG device has been designed using CAD software (Solidworks, Dassault Systems). The plastic parts were printed with a three-dimensional (3-D) printer (Prusa i3, Custom made, Latvia). The device is composed of a plastic enclosure filled with an adjustable vacuum pillow ($40\times 20$ AB Germa, Sweden) as palm support, four bispectral light sources, monochromatic video camera, and microcontroller board (Arduino Nano, Arduino) (Fig. 1).

Fig. 1

(a) Hardware components of imaging PPG device: bottom view of imaging system, camera, and light sources and (b) the device fastened to the basement, including the body with vacuum pillow supporting the hand.

The illuminating system is composed of four bispectral light sources, each consisting of two high-power light-emitting diode (LED) emitters (Roithner LaserTechnik GmbH, green: $\lambda =530\mathrm{nm}$, 3 W; infrared (IR): $\lambda =810\mathrm{nm}$, 1 W) [Fig. 1(a)]. The designed system was intended mostly for recording signals from the curved surface of the hand (dorsal or ventral aspect), and so achieving a uniform illumination can be difficult. To achieve uniform illumination of the skin surface, an adjustable LED intensity control was introduced via PC-based custom developed software. The wavelengths of LEDs were chosen by taking into account the light-tissue penetration depth and isosbestic wavelength of oxyhemoglobin and deoxyhemoglobin to avoid the effect from saturation changes (${\mathrm{SpO}}_2$) as described by Sandberg et al.²⁸ The microcontroller board provides switching of green and IR LEDs intermittently and sequential triggering of video frames captured by a camera. The camera control was performed by uEye software using the manual trigger mode, fixed exposure time, $2\times 2\text{pixel}$ binning, and triggered at $60\text{frames}/\mathrm{s}$. The camera (8 bit CMOS “IDS-uEye UI-1221LE”) was equipped with “S-mount” 1/2 in. $F=4\mathrm{mm}$ low distortion, wide-field lens (Lensagon). For the full view of an adult palm ($20\times 15\mathrm{cm}$ field of view), the camera lens was placed at 15 cm distance from the skin surface. In order to reduce skin specular reflectance, orthogonally oriented polarizers were placed behind the camera and light sources.

2.2.

Software

In order to extract hemodynamic information from two different depths of the skin, the sequence of video frames was stored as a video format file and further analysis was carried out offline using custom developed MATLAB-based software (University of Latvia, Uldis Rubins) [Fig. 2(a)]. The following algorithm has been implemented in the software [Fig. 2(b)]: (1) the video frames were downsampled, by a factor of 4, by averaging neighboring pixels, for processing speed; (2) PPG signals were calculated from the averaged pixels in the selected region of interest (RoI) of video frames; (3) fast varying PPG (AC) component regarding cardiac activity was calculated [using bandpass (0.7 to 1.5 Hz) second-order Butterworth filter] and slow (DC) component was calculated [by lowpass (0 to 0.1 Hz) second-order Butterworth filter]; and (4) the hemodynamic parameters [pulse rate, PPG amplitude, and perfusion index (PI)] were calculated from the PPG AC and DC signals. The last step involved storage of calculated hemodynamic data accessible for third-party software, such as SigmaPlot.

Fig. 2

(a) MATLAB-based image processing custom developed software and (b) the algorithm for calculation of hemodynamic parameters. The video frames captured at green (530 nm) and near infrared (810 nm) light are processed.

The software allows manual selection of the RoI and time window of video. The hemodynamic parameters are calculated in every heartbeat cycle. The sequence of hemodynamic parameter calculation can be described as follows:

2.3.

Statistical Analysis

The perfusion characterizing indices from different modalities (iPPG and LDI) are expressed as percentage changes from the baseline. A comparison of perfusion data from three signals was accomplished using analysis of variance, while two signal data (${\mathrm{iPPG}}_{530\mathrm{nm}}$ and ${\mathrm{iPPG}}_{810\mathrm{nm}}$) were tested with a $t$-test following normality and equal variance analysis. The relationship between different modality time courses was assessed using the Pearson correlation. The significance level was set at $p<0.05$. The data are expressed as $\text{means}\pm \text{standard}$ error of the mean, unless otherwise stated. The statistical analysis of the data was performed using SigmaPlot 12.5 software (Systat Software Inc.).

2.4.

In Vivo Validation of Imaging Photoplethysmography System

The in vivo validation study has been approved by the University of Latvia Research Ethics Committee and is in accordance with the guidelines of the Declaration of Helsinki. After being explained all the experimental procedures and the possible risks associated with participation in the study, the subjects gave their written informed consent. The selected subjects were nonsmokers and previously asked to restrain from consuming caffeine or alcohol-containing beverages. In order to assess in vivo capability of imaging PPG, four provocation protocols were accomplished. The tests focused on the predominant effect on the skin superficial subpapillary plexus: local skin heating and topical application of liniment, and with generalized effect on both superficial subpapillary plexus and a deeper plexus: postocclusive reactive hyperemia (PORH) and local anesthesia. All protocols were performed under constant conditions (room temperature $23\pm 2°\mathrm{C}$, relative humidity $50\pm 5\%$) to ensure adequate microcirculation. Before every test, the subject was acclimatized for 10 min to the room, comfortably sitting, and the skin surface temperature was measured by infrared thermometer (Raynger MX, Messtechnik, Berlin, Germany).

3.1.

Topical Application of Liniment

The topical application of liniment is a decisive test in the present study as it is able to induce selective vasodilation of superficial microvessels.³¹ Within 3 to 4 min after application of the liniment, reddening of skin (flushing) appeared, indicating hyperemia. The response began with the rapid vasodilation 3 to 4 min after application and reached its plateau 6 to 7 min later, as described elsewhere.^31,32 All subjects individually demonstrated a statistically significant increase of all three signals (${\mathrm{iPPG}}_{530\mathrm{nm}}$, ${\mathrm{iPPG}}_{810\mathrm{nm}}$, LDI) from the baseline. The representative illustration of relative cutaneous perfusion changes during the test is depicted in Fig. 3(a). The subject group relative peak perfusion values are shown in Fig. 3(b) and are ${\mathrm{iPPG}}_{530\mathrm{nm}}=659\pm 297\%$; ${\mathrm{iPPG}}_{810\mathrm{nm}}=99\pm 52\%$; $\mathrm{LDI}=611\pm 228\%$.

Fig. 3

(a) Perfusion changes during application of liniment registered with iPPG and LDI, representative example from one subject; arrows indicate the onset of provocation. (b) Peak perfusion changes from the baseline, group mean $\text{values}\pm \text{standard}$ error of the mean; statistically significant difference at $p<0.05$ are indicated by asterisks. (c) iPPG signal waveforms during baseline recording and liniment evoked hyperemia representative example from one subject.

Although an increase in the ${\mathrm{iPPG}}_{810\mathrm{nm}}$ signal was negligible in comparison to ${\mathrm{iPPG}}_{530\mathrm{nm}}$, as depicted in Fig. 3(a), it has been noticed that the ${\mathrm{iPPG}}_{810\mathrm{nm}}$ waveform during peak perfusion (hyperemia) was noisy at low amplitude in contrast to that of ${\mathrm{iPPG}}_{530\mathrm{nm}}$, see Fig. 3(c). The observed similarity between ${\mathrm{PPG}}_{530\mathrm{nm}}$ and LDI perfusion changes has been confirmed by their high correlation (group mean $r=0.98\pm 0.01$, $p<0.001$), as depicted in Fig. 4.

Fig. 4

Relationship between perfusion obtained from LDI and ${\mathrm{iPPG}}_{530\mathrm{nm}}$ during liniment application test, representative example from one subject.

3.2.

Local Heating Test

The local heating of nonglabrous skin produced a significant increase of cutaneous perfusion from the baseline in all subjects for ${\mathrm{iPPG}}_{530\mathrm{nm}}$, ${\mathrm{iPPG}}_{810\mathrm{nm}}$, and LDI $\sim 6\min$ ($6.0\pm 1.5\min$) following the onset of warming. The largest increase of perfusion was observed for ${\mathrm{iPPG}}_{530\mathrm{nm}}$, and LDI signals, with the same trend forming initial peak with the subsequent nadir, as depicted in Fig. 5(a) (13 to 16 min). A similar response was noted for the ${\mathrm{iPPG}}_{530\mathrm{nm}}$ waveform-signal amplitude as it increased during the heating-evoked hyperemia [see Fig. 5(c)]. The mean relative peak perfusion is depicted in Fig. 5(b) and values were ${\mathrm{iPPG}}_{530\mathrm{nm}}=370\pm 136\%$; ${\mathrm{iPPG}}_{810\mathrm{nm}}=123\pm 29\%$; $\mathrm{LDI}=530\pm 327\%$. There was no statistically significant difference between ${\mathrm{iPPG}}_{530\mathrm{nm}}$ and LDI.

Fig. 5

(a) Perfusion changes during local heating test registered with iPPG and LDI, representative example from one subject, arrows indicate the onset of provocation. (b) Peak perfusion changes from the baseline, group mean $\text{values}\pm \text{standard}$ error of the mean, statistically significant difference at $p<0.05$ is indicated by asterisks. (c) iPPG signal waveforms during baseline recording and local heating induced hyperemia, representative date from one subject.

Correlation analyses revealed a close relationship ($r=0.93\pm 0.01$; $p<0.001$) between LDI and ${\mathrm{iPPG}}_{530\mathrm{nm}}$ PI. The representative data for one subject are displayed in Fig. 6.

Fig. 6

Relationship between perfusion registered with LDI and ${\mathrm{iPPG}}_{530\mathrm{nm}}$, representative example from one subject.

3.3.

Postocclusive Reactive Hyperemia Test

The occlusion test has been selected for the present study to evoke an increase of perfusion in both superficial and deep circulation of glabrous skin. Subjects showed a typical perfusion trend for both iPPG signals, comprising relatively flat baseline, rapid decline at the onset of occlusion, and PORH peak (reactive hyperemia), as shown in Fig. 7(a). The iPPG waveform amplitudes differed in all three conditions, as depicted in Fig. 7(c).

Fig. 7

(a) Perfusion changes during occlusion test registered with iPPG and LDI, representative example from one subject; dotted lines indicate various stages (baseline, occlusion, postocclusion). (b) Peak perfusion changes from the baseline during reactive hyperemia test, group mean $\text{values}\pm \text{standard}$ error of the mean; statistically significant difference at $p<0.05$ is indicated by asterisks. (c) iPPG signal waveforms during baseline, arterial occlusion and postocclusion period.

There was no significant difference between perfusion obtained from ${\mathrm{iPPG}}_{530\mathrm{nm}}$ and ${\mathrm{iPPG}}_{810\mathrm{nm}}$; hence, for volunteers with higher sympathetic tone and subsequently lower baseline, the so-called “biological zero” during occlusion period was higher, exhibiting minor difference between iPPG signals. The group mean values for the relative PORH peak did not significantly differ with the slightly higher ${\mathrm{iPPG}}_{530\mathrm{nm}}=94\pm 24\%$ and lower ${\mathrm{iPPG}}_{810\mathrm{nm}}=71\pm 23\%$, see Fig. 7(b).

3.4.

Regional Anesthesia Test

The recording of glabrous skin perfusion during rest and brachial plexus block is important as an appraisal for the potential clinical application of our iPPG system. The expected effect is a decrease in sympathetic vascular tone resulting in vasodilation and subsequent elevation of perfusion in superficial and deep microvasculature.

The gradual increase in perfusion in both iPPG channels was observed 5 to 7 min following successfully administrated local anesthetics, where the plateau phase was reached in $\sim 15$ to 16 min after administration [see Fig. 8(a)]. The perfusion response timescale slightly differs across the patients, varying depending on heterogeneity of patient group and variance of administration procedure. The patient group mean relative peak perfusion did not differ (${\mathrm{iPPG}}_{530\mathrm{nm}}=413\pm 208\%$; ${\mathrm{iPPG}}_{810\mathrm{nm}}=359\pm 191\%$), indicating similar responses in superficial and deeper vasculature situated in dermal and hypodermal layers [Fig. 8(b)]. The same tendency is seen in iPPG waveform, as depicted in Fig. 8(c).

Fig. 8

(a) Perfusion changes during RA test obtained from iPPG and LDI, representative example from one subject, arrow indicates the onset of provocation. (b) Peak perfusion changes from the baseline, group mean $\text{values}\pm \text{standard}$ error of the mean; statistically significant difference at $p<0.05$ is indicated by asterisks. (c) iPPG signal waveforms during baseline recording and anesthetic evoked hyperemia.

4.1.

Topical Application of Liniment

The skin flushing phenomenon (reddening) observed in the present study after topical application of vasodilatory liniment is characterized by a dramatic increase in perfusion and number of functional capillaries in superficial subpapillary plexus within the upper dermis, where absorption of hemoglobin at 530 nm is high due to the very dense vascular network. During resting conditions and comfortable temperatures, this plexus maintains relatively low in perfusion as it serves mostly for heat transfer purposes. When the liniment is applied on the skin, the active ingredients, nicotinic acid esters (ethyl nicotinate, hexyl nicotinate), are converted to niacin (nicotinic acid) by cutaneous esterases after penetration into the skin.³⁵ At least two different physiological mechanisms account for the vascular component of the flushing, one involving activation of niacin receptors (GPR109A) in dermal Langerhans cells and keratinocytes,^36–39 leading to secretion of prostaglandins, the other activates the Capsaicin receptor (TRPV1) pathway.^40,41 In the end, both pathways produce vasodilation of arterioles and precapillary sphincters and subsequently increased capillary perfusion. The protocols utilizing topical application of rubefacients, such as nicotinic acid, have been used for decades in pharmacological, dermatological, and biophotonic studies.^{28,30–32,42,43} Hence, contactless cutaneous perfusion measurement at different depth is still the exclusive option due to the lack of appropriate instrumentation, and only few studies are available employing two-bandwidth LDI. For example, Jacobi et al.^13,35 utilized a two-bandwidth laser Doppler (0.5 to 5 kHz and 5 to 30 kHz) instrument for measurement of cutaneous microcirculation at two different depths.^13,35 Much like our results, the authors reported higher perfusion in superficial capillaries compared to the deeper vasculature following liniment application. A high-amplitude perfusion signal in the ${\mathrm{iPPG}}_{530\mathrm{nm}}$ channel during liniment test is produced by substantial vasodilation in the relatively dense capillary network. The same perfusion trend was observed in our data from a different modality, LDI signal [Fig. 3(a)], with the relatively wide measurement range (signal acquisition ranges from 0 to 1.5 mm), and hence, a different operating principle.

Conversely, the weak ${\mathrm{iPPG}}_{810\mathrm{nm}}$ signal during the liniment test indicates minor involvement of deeper cutaneous vessels in the flushing response, reported by Jacobi et al. The results obtained in the liniment test indicate the capacity of our iPPG system to register superficial subpapillary blood perfusion utilizing the 530-nm light source.

4.2.

Local Heating Test

The response of skin blood flow to local heating includes complex interactions between neural components and locally produced chemical messengers.⁴⁴ The typical response to local heating of nonglabrous skin is an initial peak (following 10 min of heating), a subsequent drop to nadir, and a further vasodilation to a sustained plateau (after 20 min).⁴⁵ In our protocol, after a relatively short heating time, only the first peak and part of the nadir were observed, while lacking a sustained plateau. The possible cause is the different vasoreactivity of glabrous and nonglabrous skin, which is mentioned in the study of Metzler-Wilson et al.⁴⁶ The similarities between the LDI and ${\mathrm{iPPG}}_{530\mathrm{nm}}$ traces provide evidence that subjects exhibited consistent vascular response, predominantly in superficial vascular plexus situated close to dermal epidermal junction, where densities of capillaries are high. While the blunted response of ${\mathrm{iPPG}}_{810\mathrm{nm}}$ reflect the perfusion of dermal–hypodermal junction, comprising less dense, relatively large caliber vessels. Thee perfusion changes occurring closer to the skin surface will be masked at longer wavelengths by the perfusion changes at deeper levels. This may partly explain the wavelength dependence of the iPPG signal. To the best of our knowledge, the simultaneous blood perfusion registering in superficial and deep plexuses has never been demonstrated by means of the iPPG technique, which makes it difficult to compare our findings. However, the feasibility of single spectral (green light) iPPG during thelocal heating test has been previously reported in our earlier study.⁴⁷ The results of th local heating protocol largely support the findings obtained in the liniment protocol, similarly indicating the feasibility of the iPPG system to selectively acquire perfusion from the superficial vascular plexus utilizing a 530-nm light source.

4.3.

Postocclusive Reactive Hyperemia Test

The occlusion test comprising PORH is often used as a provocation test due to its convenience and strong regional vascular response.⁴⁸ The postocclusive hyperemia phenomenon occurs after the restoration of blood supply due to marked dilatation, in some circumstances even reaching maximal vasodilatory capacity of the vascular bed (usually following 15 min of occlusion). Two major physiological mechanisms are equally contributing to PORH: direct relaxation of vascular smooth muscle by endothelial hyperpolarizing factor and reflectory vasodilation due to stimulation of sensory nerve endings by accumulated metabolites.⁴⁹ In the present protocol, the occlusion time was much shorter (3 min instead of 15 min) to minimize subject discomfort and solely reveal differences between perfusion in deep and superficial vascular plexuses. Previous studies confirmed the strong vasodilation of the entire occluded vascular bed, comprising small arteries and arterioles, supplying skeletal muscle and skin.^30,50,51 This explains the observed negligible difference in relative peak perfusion for ${\mathrm{iPPG}}_{530\mathrm{nm}}$ and ${\mathrm{iPPG}}_{810\mathrm{nm}}$, as superficial dermal and deep cutaneous vessels almost equally respond during PORH. Nevertheless, the slightly larger perfusion for superficial microvasculature during PORH may be attributed partly to different vasoreactivity in dermal end subcutaneous microvascular plexuses, partly due to higher absorption of dense superficial capillaries. Unfortunately, the results were inconclusive, as no studies were identified regarding responses in superficial and deep cutaneous perfusion registered by means of the iPPG technique. However, in studies utilizing single spectral iPPG, the same trend of perfusion changes was shown during baseline, occlusion, and PORH.⁵¹ Summarizing the present findings, we consistently demonstrate that the occlusion test, as expected, induced detectable generalized responses of perfusion in both superficial and deep vascular layers in the glabrous skin.

4.4.

Regional Anesthesia Test

The local anesthesia (plexus brachialis block) protocol was included in evaluation of the iPPG system for two reasons—to assess the sensitivity of the iPPG system during generalized skin sympathetic blockage and to asses our system in real clinical conditions. Despite the fact that the brachial plexus blockade for upper-extremity surgery is a very common clinical procedure,⁵² there are quite a few papers describing iPPG for anesthesia monitoring.^53,54 Our findings are consistent with results obtained in several studies utilizing LDI^55,56 and a thermal camera,⁵⁷ indicating increased perfusion during the brachial plexus block in both superficial dermal (${\mathrm{iPPG}}_{530\mathrm{nm}}$) and deeper vascular layers (${\mathrm{iPPG}}_{810\mathrm{nm}}$). The data are consistent with early findings, which state that a block of sympathetic nerves is associated with a sudden drop of peripheral sympathetic (vasoconstrictor) nerve activity,⁵⁸ thereby leading to vasodilation, increased blood flow, and increase in tissue/skin temperature.^56,57 The results obtained in this protocol are similar to that of PORH because of the generalized response of superficial and deep cutaneous vascular plexuses, confirming that our designed iPPG system could be used in clinical settings.