Vasomotion analysis of speed resolved perfusion, oxygen saturation, red blood cell tissue fraction, and vessel diameter: Novel microvascular perspectives

1 INTRODUCTION

Vasomotion is the spontaneous oscillation in vascular tone in the microcirculation and is believed to be a physiological mechanism facilitating the transport of blood gases and nutrients to and from tissues. Vasomotion can be described as recurrent organized fluctuations in vessel diameter notable over time. By frequency analysis, distinct frequency intervals related to physiological aspects can be presented as heart rate, respiration, myogenic activity, neurogenic, and endothelial dependent activity.1 The higher frequency intervals that can be verified by measurement of pulse and respiration frequency are related to cardiac and pulmonary pressure variation. The myogenic interval has been studied in animal models and by mapping of cellular mechanisms.2-4 The neurogenic interval has been studied by blockage of sympathetic nerves using microdialysis catheters delivering bretylium tosylate.5 The endothelial interval has been studied by iontophoresis of the acetylcholine, an endothelium-dependent vasodilator, and sodium nitroprusside, an endothelium independent vasodilator,6 and by inhibition of endothelial nitric oxide synthase by intradermal microdialysis leading to suppression of the endothelial interval.7

Vasomotion analysis potentially offers measures that are more directly related to physiological mechanisms than the measurement of microvascular perfusion alone and have been utilized to investigate microvascular dysfunction in various disease states. Type 2 diabetes is a well-known risk factor for microvascular dysfunctions including neuropathy.8 Alterations in the neurogenic frequency has been observed in patients with type 2 diabetes10-12, 9 and type 1 diabetes.13 Vasomotion has also been used to study altered microvascular endothelial function during smoking,14 benign and malignant melanocytic skin lesions and has been found to be a specific, sensitive method for the in vivo identification of malignant melanoma,15 in patients with paranoid schizophrenia,16 schizophrenia,17 lepromatous leprosy patients,18 obesity,19 systemic sclerosis patients,20 alcoholic withdrawal syndrome,21 venous insufficiency,22 during general anesthesia23 and compared between different ages5 as well as been used in description of pharmacological properties of drugs such as exenatide.24 Although studied in many patient groups, it has still not reached clinical routine.

Laser Doppler flowmetry (LDF) has constituted the gold standard for in vivo vasomotion analysis. Other methods used have been photoplethysmography,25 correlation mapping optical coherence tomography (cmOCT),26 and high-precision ultrasonic echotracking device.27 The Periflux 6000 EPOS system (Perimed AB, Järfälla, Stockholm, Sweden) combines LDF and diffuse reflectance spectroscopy (DRS) to assess speed resolved perfusion, red blood cell (RBC) tissue fraction, oxygen saturation, and average microvascular vessel diameter, simultaneously from the same measurement volume.28, 29 The related instrument O2C (Lea Medizintechnik, Giessen, Germany) combines the assessment of conventional (not speed-resolved) perfusion, relative blood concentration, and oxygen saturation in a similar manner. These techniques enable further analysis of the physiological aspects involved in vasomotion. Since vasomotion appears from smooth muscle vasodilation and vasoconstriction resulting in variation of vessel lumen and recruitment of capillaries, the Periflux 6000 EPOS that enables assessment of the vessel diameter is hypothesized to be suitable for vasomotion analysis.

Different authors have used different techniques to calculate the frequency intervals. The most commonly used method for analyzing vasomotion has been the wavelet transformation,30, 31 followed by the short-time Fourier transform.32, 33 A multitude of different normalization strategies have also been used, regarding normalization of the signals prior to the transformation calculations, and normalization of the energy in the studied frequency bands. Furthermore, the borders of the frequency intervals vary slightly between authors. The lack of consensus regarding the analysis of vasomotion patterns makes it difficult to compare results between studies. Therefore, a methodological standardization is needed to facilitate future comparisons and reach clinical application.

The aim of this study was to present a framework for vasomotion analysis of signals related to oxygen saturation, RBC tissue fraction, average vessel diameter, and speed resolved perfusion. We analyzed the data using scalograms of individual signals and by correlation between the signals, enabling a deeper understanding of local microvascular regulation.

2 MATERIAL AND METHODS 2.1 Measurement system

Data were recorded using the Periflux 6000 EPOS system (Perimed AB, Järfälla-Stockholm, Sweden).28, 34, 35 The system collects both DRS and LDF data from the same tissue volume. The data analysis is based on artificial neural networks that has been trained on modeled data with added system specific noise, using measured spectra as input and any of the output parameters oxygen saturation, RBC tissue fraction, average vessel diameter, speed resolved perfusion, or sample depth as output.36 In the case of speed-resolved perfusion, both DRS and LDF spectra were used as input, whereas only the DRS spectra were used in the other cases.

Normally, the average vessel diameter is not an output from the EPOS system. However, the shape of the DRS spectra are affected in a characteristic manner when the average vessel diameter is changed, due to the so called vessel packaging effect,37, 38 and we have previously shown that this effect can be used to estimate the average vessel diameter from DRS spectra.28, 39

The sample depth was estimated based on the depths of the photon paths. The sample depth was defined as the depth above which urn:x-wiley:0909752X:media:srt13106:srt13106-math-0001 of the photon paths were situated. The photon paths and thus this sample depth were known in the training data, and a separate artificial neural network was trained to the sample depth. Also compare with the sampling volume description in Fredriksson et al.28

Data were collected at 15 Hz sampling frequency, that is, 15 sets of spectra per second.

2.2 In vivo measurements

Data were collected on five healthy volunteers (two females) aged 21–42 years, BMI 19–28 kg/m2, without regular medication and no known skin disease. The subjects gave their written informed consent before the start of the measurement, and the protocol was approved by the Swedish Ethical Review Authority, d.no. 2019–04713. The measurement probe was attached using double-adhesive tape (PF 105-1, Perimed AB) on the volar side of the right forearm, about 15 cm above the wrist, avoiding any visible vessels. Before the start of the measurement, the subjects rested for 20 min in the same room and in the same position as data later was recorded. After acclimatization, each subject was recorded during a period of 10 min. The room temperature was measured during the experiment and kept within the range of 21–22°C.

2.3 Wavelet analysis

The oscillations in the measured microvascular signals caused by vasomotion as well as respiration and heart beats have period lengths varying from about 1 s (heartbeat) to more than a minute. In this study, the frequency content is analyzed using the Morlet wavelet transformation.30, 31 Similar results can be obtained by using the short-time Fourier transform.32, 33 An advantage with the wavelet analysis is that the time resolution is different for different frequencies, partially overcoming the problem to choose between high time resolution or high frequency resolution. Another advantage is that the frequency resolution can be adapted to a logarithmic frequency scale, which is advantageous for visual representation.

The conventional perfusion signal was normalized with its mean value before the wavelet calculation. From a theoretical fluid mechanics perspective, this normalization adds to two important effects. Firstly, it makes the signals insensitive to the overall perfusion in the sampling volume, and secondly, it makes the signals sensitive to relative variations in vessel diameter. All speed-resolved perfusion signals as well as the total perfusion (sum of the speed-resolved perfusion) were normalized with the average total perfusion for the same reason. The RBC tissue fraction signal was normalized with its mean, and the wavelet was calculated from the square-root of the normalized signal, since it can be argued that variation in the square-root is related to variations in the vessel diameter. For the oxygen saturation and the vessel diameter signals, no normalization was done.

All signals were further pre-processed by subtracting the mean to avoid spectral leakage from the zero frequency. For the signals that were normalized, the subtraction of the mean was done after normalization.

The Morlet wavelet transform was calculated by using the Wavelet Toolbox in Matlab 2021a (The MathWorks, Inc). The calculation results in a scalogram that is a 2D-map of signal magnitude to scale and time, where the scale corresponds to the frequency, sometimes denoted pseudo-frequency. A time-averaged power spectrum was calculated as the time average of the squared scalogram for each frequency. An example is seen in Figure 1. The shaded area in the magnitude spectrogram at low frequencies denotes times where the border effect significantly affects the estimation of the magnitude for those frequencies. Those areas have been excluded when calculating the time-averaged wavelet spectrum.

image

Example of magnitude scalogram and the time-averaged wavelet spectrum for total perfusion. The frequency limits between the five different frequency intervals are marked with the dashed lines. Overtones of the heart frequency (sligtly above 1 Hz) at about 2 and 4 Hz can be observed

The magnitude scalogram and the time-averaged wavelet spectrum were divided into the following frequency intervals: endothelial 0.0095–0.02 Hz, neurogenic 0.02–0.06 Hz, myogenic 0.06–0.15 Hz, respiratory 0.15–0.4 Hz, and cardiac 0.4–1.6 Hz. These intervals are identical to the once used by Stefanovska, Bracic et al,30 and similar to those used by for example Söderström, Stefanovska et al,40 Humeau, Koïtka et al,31 and Rossi, Bazzichi et al.20

2.4 Correlation analysis

In addition to the wavelet analysis, the covariation between the different signals in the different frequency intervals was also analyzed. This was done by first applying zero-phase band-pass filtering with a fourth order Butterworth filter with the frequency limits of each of the five frequency intervals given above. Then, the cross-correlation between each pair of signals, for example between total perfusion and oxygen saturation in the endothelial frequency interval, was calculated. The time-shift corresponding to the maximum absolute correlation, limited to urn:x-wiley:0909752X:media:srt13106:srt13106-math-0002 where fhigh is the upper cut-off frequency of the given frequency interval, was considered. An analysis by the maximum correlation and the time shift that corresponds to, reveals information about potential synchronization in the physiological processes underlying the different microvascular signals.

3 RESULTS

The estimated sample depth was 0.4 mm in the measurements from all five test subjects.

3.1 Examples from one individual

In Figures 2-5, example magnitude scalograms and time-averaged wavelet spectra are shown for one of the five subjects. For the speed resolved perfusion, speeds < 1 and > 1 mm/s are chosen, as well as the total perfusion. For comparison with previous works, results for conventional perfusion are also shown, and these are very similar to the results for total perfusion.

image

Magnitude scalograms and time-averaged wavelet spectra for speed resolved and conventional perfusion

image

Magnitude scalograms and time-averaged wavelet spectra for oxygen saturation

image

Magnitude scalograms and time-averaged wavelet spectra for RBC tissue fraction

image

Magnitude scalograms and time-averaged wavelet spectra for average vessel diameter

There are several differences and similarities in the patterns of the different types of signals. Comparing perfusion for speeds below and above 1 mm/s, respectively, reveals that the frequencies related to cardiac activity is more pronounced for the higher speed region, which is generally the case for all five test subjects. Furthermore, the energy for speeds > 1 mm/s is much higher than for the low speeds indicating that the perfusion variations are larger for higher speeds. It can also be seen that the peaks in the myogenic and neurogenic intervals are more visible in relation to the other frequency intervals for the lower speed, but that is not a general trend for all five test subjects.

The cardiac and respiratory frequency intervals are considerably less pronounced for the non-perfusion signals, that is, oxygen saturation, RBC tissue fraction, and vessel diameter. They are almost indiscernible for the oxygen saturation. That is the case for all five subjects.

The general pattern of the magnitude scalograms of the high-intensity periods of the myogenic, neurogenic, and endothelial frequency intervals is similar in all signals. One example is the high intensity region at about 200 s in the myogenic frequency interval, that is clearly visible in all signals. But there are also differences in how they are pronounced for the different signals. For example, in general for all five test persons, especially the neurogenic activity is considerably more pronounced in relation to the total energy for the oxygen saturation, RBC tissue fraction, and vessel diameter signals, than for the perfusion signals. That can be seen when comparing the neurogenic frequency interval in the time-averaged wavelet spectra in Figure 2, with the corresponding interval in Figures 3-5. In the latter case, there is a clear peak, whereas that is hardly discernible in the spectra from the perfusion signals.

Another interesting aspect to notice in the magnitude scalograms is the long time between periods of high activity. Studying Figure 5, in the myogenic frequency interval, these periods appear with intervals of 100–200 s, whereas in the neurogenic frequency interval, there is only one such high-activity period.

3.2 Summarized results for all test subjects

In Table 1, the average (standard deviation) of the total energy in each frequency interval for each signal, including conventional (conv.) perfusion, is given. The averages and standard deviations are calculated for the five subjects. The total energy is calculated from the time-averaged wavelet spectra.

TABLE 1. Total energy in each frequency interval. Data presented as mean (standard deviation) over the five test subjects Perfusion Frequency interval <1 mm/s Oxygen saturation RBC tissue fraction Vessel diameter Cardiac 0.4–1.6 Hz 0.016 (0.014) 0.20 (0.13) 0.22 (0.16) 0.14 (0.10) 0.63 (0.26) 0.00074 (0.00036) 1.2 (0.2) Respiratory 0.15–0.4 Hz 0.0036 (0.0039) 0.0039 (0.055) 0.053 (0.083) 0.031 (0.041) 0.39 (0.22) 0.00026 (0.00013) 0.62 (0.39) Myogenic 0.06–0.15 Hz 0.0026 (0.0019) 0.029 (0.020) 0.042 (0.030) 0.032 (0.018) 6.7 (7.1) 0.00040 (0.00037) 1.5 (1.8) Neurogenic 0.02–0.06 Hz 0.0027 (0.0017) 0.031 (0.012) 0.045 (0.012) 0.042 (0.015) 25 (15) 0.00083 (0.00064) 3.5 (2.6) Endothelial 0.0095–0.02 Hz 0.0010 (0.0006) 0.014 (0.010) 0.020 (0.011) 0.021 (0.013) 16 (10) 0.00047 (0.00017) 1.6 (0.8) 3.3 Correlation between signals

Results of the correlation analysis for the neurogenic frequency interval (0.02–0.06 Hz) are shown in Tables 2–4. We have chosen to include time shifts only for absolute correlation coefficients exceeding a threshold of 0.5, since the estimation of the time shift is less accurate for smaller correlation coefficients. That also emphasizes the cases when the absolute correlation coefficient is high. It can be seen that the absolute correlation coefficient is above 0.5 in the majority of the test subjects for correlations between: total perfusion and oxygen saturation; total perfusion and vessel diameter; and oxygen saturation and vessel diameter. In all cases where the correlation coefficient exceeded 0.5, the time shift between the total perfusion and the other signals was negative, meaning that the change in the total perfusion preceded the change in the other signal (oxygen saturation, RBC tissue fraction, and vessel diameter). Estimated time shifts between oxygen saturation and RBC tissue fraction, oxygen saturation and vessel diameter, and RBC tissue fraction and vessel diameter were much smaller.

TABLE 2. Maximum correlation coefficient and time shift for the maximum correlation coefficient when it exceeded 0.5 between total perfusion and oxygen saturation, RBC tissue fraction, and vessel diameter, respectively, for the neurogenic frequency interval Oxygen RBC tissue fraction Vessel diameter Subject Time shift Correlation coefficient Time shift Correlation coefficient Time shift Correlation coefficient 1 0.19 0.12 0.38 2 −4.7 0.72 0.37 −3.7 −0.57 3 0.40 −0.19 0.35 4 −2.5 0.62 −1.0 0.54 −1.5 −0.52 5 −4.4 0.80 −2.9 0.68 −4.6 −0.66 TABLE 3. Same as Table  but correlation between oxygen saturation and RBC tissue fraction and vessel diameter, respectively RBC tissue fraction Vessel diameter Subject Time shift Correlation coefficient Time shift Correlation coefficient 1 0.37 −0.49 2 0.19 1.3 −0.80 3 0.20 −1.3 −0.54 4 1.2 0.61 0.8 −0.87 5 −0.1 0.73 0.4 −0.88 TABLE 4. Same as Table  but correlation between RBC tissue fraction and vessel diameter Vessel diameter Subject Time shift Correlation coefficient 1 −0.40 2 0.08 3 0.42 4 −0.5 −0.62 5 0.9 −0.60

It can be observed that the vessel diameter is generally negatively correlated with the other signals. That is true for all cases where the absolute correlation coefficient exceeds 0.5, also for the frequency intervals not shown in Tables 2–4. The correlation between all other signals is positive where the absolute correlation coefficient is above 0.5.

In Table 5, number of subjects having an absolute maximum correlation coefficient exceeding 0.5 between the different signals in the three lowest frequency intervals, it is summarized how many of the five test subjects that had a maximum absolute correlation coefficient above 0.5 for correlations between the different signals for the myogenic, neurogenic, and endothelial frequency intervals. For the respiratory and cardiac frequency intervals, there were no absolute correlation coefficients exceeding 0.5 between: perfusion and oxygen saturation; perfusion and RBC tissue fraction; perfusion and vessel diameter; oxygen saturation and vessel diameter; or RBC tissue fraction and vessel diameter. For the cardiac frequency interval, there was a (weak) negative correlation between oxygen saturation and RBC tissue fraction, between −0.3 and −0.6 for all five test subjects.

TABLE 5. Number of subjects having an absolute maximum correlation coefficient exceeding 0.5 between the different signals in the three lowest frequency intervals Vessel diameter RBC tissue fraction Oxygen saturation Endothelial Neurogenic Myogenic Endothelial Neurogenic Myogenic Endothelial Neurogenic Myogenic Total perfusion 1 3 1 0 2 0 3 3 2 Oxygen saturation 4 4 5 3 2 1 RBC tissue fraction 3 2 1

The correlation coefficient between perfusion for speeds below 1 mm/s and total perfusion was above 0.5 in all cases for the endothelial, neurogenic, and myogenic frequency intervals, but below 0.5 in most cases for the respiratory and cardiac frequency interval. The situation was similar for the correlation of perfusion for speeds below 1 mm/s to above 1 mm/s. The correlation coefficient between perfusion for speeds above 1 mm/s and total perfusion was above 0.5 in all cases for all frequency intervals. The correlation between perfusion and oxygen saturation, RBC tissue fraction, and average vessel diameter was in all cases similar no matter if total perfusion or only perfusion below or above 1 mm/s was considered. Therefore, we choose to present such results only for the total perfusion and not for the perfusion divided into low or high speeds.

An example of the correlation between total perfusion and oxygen saturation is shown in Figure 6, and an example between oxygen saturation and vessel diameter is shown in Figure 7.

image

Total perfusion (blue curve) and oxygen saturation (red curve) as a function of time after band-pass filtering with cut-off frequencies 0.02 and 0.06 Hz, and normalization with their respective standard deviation to facilitate comparison (A). The oxygen saturation signal with no time-shift is represented by the dotted red curve. In (B), the two signals are plotted against each other, where changes are generally found along the positive diagonal giving a positive correlation

image

Oxygen saturation (blue curve) and average vessel diameter (red curve) as a function of time after band-pass filtering with cut-off frequencies 0.02 and 0.06 Hz, and normalization with their respective standard deviation to facilitate comparison (A). In (B), the two signals are plotted against each other, where changes are generally found along the negative diagonal giving a negative correlation. The estimated time-shift giving the maximum correlation was negligible and not included in these plots

4 DISCUSSION

We have introduced new signals possible for vasomotion analysis enabling new tools to investigate the microcirculation. The oxygen saturation and vessel diameter signals represent dimensions of the microcirculation that are intuitively relatively independent from the perfusion. That is manifested by the much less impact of the cardiac and respiratory frequency intervals on those signals. Even if not clearly seen in the data presented in this study, it is also possible that the vasomotion analysis of the perfusion signals originating from blood flow with different speeds can lead to new insights in the underlying physiology of vasomotion, not seen with conventional perfusion. Vasomotion based on conventional perfusion has to date been by far the most common choice, now challenged by these new signals.

Vasomotion can be described as recurrent fluctuations in vessel diameter, and the introduction of the vessel diameter signal is a more direct observation than the conventional perfusion signal. It has been suggested that the oscillatory pattern leads to a more favorable tissue perfusion and oxygenation.41 Negative correlation between average vessel diameter and total perfusion, oxygen saturation and RBC fraction, was generally observed in this study. The variation in the average vessel diameter signal could either be explained by variation in vessel diameter in each active vessel, or in combination with a local capillary recruitment explained by the theory where the lowering of resistance in terminal arterioles increase the perfusion of nutritive capillary beds.42 Capillary recruitment would lead to a smaller average vessel diameter. Increased capillary recruitment would inherently increase RBC fraction and also oxygen saturation in the case of constant metabolism. Thus, negative correlation indicates that variation in the average vessel diameter signal foremost depends on capillary recruitment, and not diameter changes in individual vessels. The sample depth of in the measurements was 0.4 mm, likely foremost including capillary, arteriole, and venule blood vessels with limited smooth muscle control.

Speed resolved perfusion and optically assessed average vessel diameter has for the first time been used in the analysis of vasomotion. Vasomotion has previously been observed in signals related to oxygen saturation43 and RBCs have been proposed as a sensor of local tissue hypoxia.44 RBC tissue fraction can be compared with photoplethysmography that has previously been used to asses vasomotion.25

The origin of the cardiac and pulmonary frequency interval is based on central pressure variation. Although frequency analysis can identify these parameters, they should not be misinterpreted with local regulations of the myogenic, neurogenic, and endothelial intervals.

Traditionally, vasomotion has been measured for 5–30 min. Based on the scalograms in Section 3.1, the chosen 10 min protocol in this study may have been too short to accurately estimate the energy within the lower frequency intervals. Possibly is vasomotion a tool for precision medicine, and we recommend future studies to observe over a longer-time period as well in larger cohorts. To our knowledge, observations of vasomotion over significantly longer time have not been studied.

In this study, we have used the Morlet wavelet transform to calculate vasomotion energy in various frequency intervals. Principally equal results could have been achieved using the Fourier transform, but the wavelet transform was preferred for better visualization of the scalograms. The normalization strategy has a larger impact on the results when comparing different subjects. We have chosen to normalize some of the signals (perfusion and RBC tissue fraction) so that the vasomotion energy in those signals depends on relative variations and not on the mean signal value. We argue that this normalization results in energies that reflect the relative variation in vessel diameter more accurately, than analyzing unnormalized perfusion. For the oxygen saturation and vessel diameter signals, the same argumentation does not apply, and thus we choose not to use normalization for those signals. Furthermore, we do not perform a normalization of the energy in the various frequency intervals to the total energy for all frequencies, as was for example proposed in,30,

留言 (0)

沒有登入
gif