Pilot clinical study of quantitative ultrasound spectroscopy measurements of erythrocyte aggregation within superficial veins

2.1 Outlines

For this pilot clinical evaluation, healthy volunteers were selected and measurements were repeated to evaluate the intra-subject variability. A homemade bracelet to control the blood flow was designed (Fig. 1), and a computer platform to monitor measurements was developed. The monitoring system allows a user to visualize the B-mode ultrasound image of the vein and to record radiofrequency (RF) data. RF images are then processed using a particle image velocimetry algorithm to measure the maximum blood velocity so as to provide feedback to the user for adjusting the pressure applied by the bracelet over the vein to reduce the blood flow. Acquired RF data at a predetermined low shear rate and under natural flow were used to measure RBC aggregate structural parameters with the SFSAE algorithm. Measurements were performed on two vascular sites that would be accessible for patient monitoring.

2.2 Subject recruitment and protocol

After approval by the Ethics Committee of the Centre Hospitalier de l’Université de Montréal (CHUM) and having obtained signed informed consents, fifty participants (25 men and 25 women) were enrolled in this pilot study. Men and women were aged 43.9±19.4 and 46.7±18.6 years old, and their body mass indices (BMI) were 24.3±4.4 and 25.1±3.1, respectively. This cohort involved European (43), Asian (4) and North-African ancestries (3) participants. Some of them self-reported having at least one of the following chronic disease conditions: type 2 diabetes (3), lupus (1), aortic aneurysm (1), high cholesterol (9) and hypertension (5). Participants were lying down on an exam bed and ultrasound data were acquired on the palmer side of the forearm on the antebrachial vein, and on the tibial vein on the front of the leg. Ultrasound acquisitions were performed with a high-frequency scanner (Vevo770, 35 MHz central frequency probe #RMV-703, Visualsonics, Toronto, Canada). For each measurement site, veins with a diameter larger than 1 mm were quested to facilitate ultrasound post-processing. After the ultrasound probe was set in place with a rigid stand, a B-mode image was acquired to measure the vein depth, which corresponds to the span between the skin surface and the closer vessel wall. The internal vein diameter was also measured in this manner before each assessment of RBC aggregation.

2.3 Low shear rate acquisition procedure using the designed bracelet

As introduced earlier, a homemade pneumatic compression bracelet (Fig. 1) was designed to apply a smooth pressure on the skin over the vein for regulating the flow and reducing the shear rate into the blood vessel. This was based on Garcia-Duitama et al. experiments [14] that have shown that keeping a pre-measuring low residual venous shear rate enables reproducible in vivo RBC aggregation measurements. The bracelet was positioned 10 mm downstream to the ultrasound acquisition site. After positioning the ultrasound probe, the blood flow was initially stopped by inflating the pneumatic device and then released to validate that the bracelet could control the blood flow. The compression bracelet included an embedded load cell (mini40, ATI Industrial Automation, Apex NC., USA) allowing real-time measurement of the force applied on the skin. The load cell controller was calibrated and sent the applied force [in Newton] through a RS232 communication directly to a Matlab program (version 2014a, Natick, Ma, USA) every second. Five ultrasound measurements were made with and without the flow controlling bracelet to measure the intra-subject variability, so as to reach low shear rate and natural shear rate conditions, respectively. An infrared instrument was used prior to ultrasound probe and bracelet positioning to facilitate the identification of the venous architecture (model AV400, AccuVein, Huntington, NY, USA). With a non-permanent pen, the vasculature was mapped on the surface of the skin and a straight venous segment without major bifurcations within a distance of at least 40 mm was selected. When such a vein segment was not available (17 cases for the leg and one for the forearm), only measurements without bracelet were performed (i.e., natural flow).

OPEN IN VIEWER

Fig.1

Picture showing the load cell mounted on the pneumatic bracelet device, the venous architecture of the forearm and leg marked with ink, and the Visualsonics ultrasound probe RMV-703 positioned over the region of the measure.

Acquired RF echoes were amplified (model 5900PR, Panametrics inc, Waltham, Ma, USA) and recorded using a 12-bit 250 MHz digitizer (GageScope 12501, Dynamic Signals LLC, Lockport, IL, USA) at a frame rate of 26 per second. A computer platform interface (Fig. 2), developed in Matlab, allowed monitoring the blood flow velocity within the vein (Fig. 2D), and the pressure applied by the bracelet (Fig. 2E). The blood flow velocity was computed using a particle image velocimetry (PIV) algorithm [15] applied on RF blood signals. The maximum velocity in the central portion of the vein (horizontal lines in Fig. 2A) was determined every second and used to manually control the pressure into the bracelet. The RF data quality control (Fig. 2B) displayed an A-line at the center of the B-mode image to validate that no saturation occurred in the blood signal (data clipping would affect the computation of the backscatter coefficient). The B-mode quality control (Fig. 2A) displayed the central position of the vein that was adjusted to match the transducer focus depth at 10 mm, thus ensuring an optimum signal-to-noise ratio.

OPEN IN VIEWER

Fig.2

Illustration of the control panel used during data acquisition. Each panel was automatically refreshed over time. (A) Longitudinal B-mode image of the vein; (B) acquired radiofrequency ultrasound data (depth versus normalized amplitude); (C) displacement map of the blood flow, the length of each vector corresponds to the instant displacement. The blood speed map region corresponds to the region delimited by the 2 white horizontal lines within the flow stream of panel A. (D) Maximum flow velocity within the blood speed map over time (in mm/s versus s), the green line corresponds to the threshold limit according to the vein diameter; (E) force applied by the bracelet and monitored by the force sensor over time (in Newton versus s).

During bracelet inflated pressure adjustment, the maximum blood velocity was monitored instead of the mean shear rate for convenience purpose. Indeed, instead of computing the shear rate from PIV data using the spatial derivative of the velocity field, which would require a real-time vessel wall segmentation to determine the vein boundary, it was determined experimentally that a low residual shear rate condition could be fulfilled if the maximum speed remained close to 1 mm/sec. This method was suitable for most participants, even those with very small venous flow. For every measurement, the bracelet pressure was gradually increased until the maximum blood velocity into the vein reached 1 mm/s (Fig. 2D). When the optimum pressure allowing the targeted maximum speed was attained (Fig. 2E), one minute was allowed prior to RF ultrasound data acquisition to ensure a stabilized RBC aggregation condition. Then, the bracelet pressure was released and one minute was left between every repeated acquisition.

Natural shear rate measurements were made after controlled flow acquisitions; whenever applicable, the ultrasound probe was left in place to maintain the same recording site within the vein. The protocol for natural flow acquisitions was the same as described above. The bracelet was deflated and left in place. A one minute waiting period was left between every repeated acquisition. Because the natural blood flow velocity produced PIV aliased measures (velocity too high considering the frame rate), the maximum blood velocity was assessed using pulsed Doppler ultrasound after the last acquisition of repeated measurements.

2.5 Backscatter coefficient assessment and SFSAE modeling

Details on QUS SFSAE estimation can be found elsewhere [16]. A fourth-order Butterworth bandpass filter (15–37 MHz) was first applied on RF signals. Then, a region of interest (ROI) was manually delineated within the central region of the vein lumen (Fig. 3A). To avoid computing SFSAE values with RF data corresponding to the vessel wall, the segmentation was made to reject signals too close (less than 64 samples) to the vessel boundary. The ROI was then partitioned into resolution cells of 5 scanlines laterally, each comprising 128 samples axially with 50% overlap.

OPEN IN VIEWER

Fig.3

B-mode (A) and parametric images of D and W (B and C, respectively) computed for the segmented region (red boundaries in A) of the vessel lumen of patient ID number 29. D is the mean RBC aggregate diameter and W is the packing factor. Regions close to the vessel wall are intentionally not selected to compute the backscatter coefficient only for blood flow signals. Black areas in B and C correspond to missing information due to a poor correlation between the backscatter coefficient and the SFSAE spectral model.

The mean experimental backscatter coefficient as a function of frequency was computed for every resolution cell, and averaged in the time-domain over 20 consecutive images prior to processing the SFSAE method. In order to compensate for instrument dependent signal conditioning, the reference backscatter coefficient method was used [17]. Two parameters of the SFSAE method were computed: D and W (Fig. 3B, C). D represents the mean diameter of aggregates (expressed in RBCs diameter assuming isotropic clustering). W corresponds to the packing factor (dimensionless), which describes the compactness of aggregates. Both D and W parameters increase with the enhancement of RBC aggregation. Based on simulations [18], it was demonstrated that W = C₁ × D² + C₂, where constants C₁ and C₂ are close to 1.0 and 0.0, respectively. Values of these constants are influenced by the hematocrit that is an input parameter of the SFSAE model. The blood hematocrit of each participant was measured by micro centrifugation.

To filter outliers in parametric images of D (Fig. 3B) and W (Fig. 3C), SFSAE spectral models from resolution cells providing poor fits with experimental backscatter coefficients were rejected (i.e., if the correlation R² within the frequency bandwidth of 15–37 MHz was below 0.5, see [19]). To extract a representative value of the aggregation state, median values of D and W were computed within the segmented venous ROI. The median was then averaged over 5 repeated measurements, prior to computation of correlations with other parameters (see below).

A 4 mL blood sample was taken less than 2 hours before or after US examination for all participants. Blood hematocrit was adjusted to 40% with autologous plasma and processed in an erythroaggregometer (Regulest, Florange, France). The aggregation index at 10 sec (S₁₀) was computed, which is a dimensionless parameter proportional to the RBC aggregation kinetics [20]. This measure corresponds to the relative surface under the reflected laser backscatter curve during the first 10 sec after applying a high shear rate to disaggregate RBCs. This value was used as a gold standard reference to compare with the proposed ultrasound method for assessing the systemic RBC aggregation level of flowing blood.

2.7 Data reporting and statistical analysis

The SFSAE D and W parameters are presented as mean±standard deviation (SD) computed over all measurements for a given configuration. The reliability of the SFSAE method was assessed by computing the intraclass correlation coefficient (ICC) between 5 repetitive measurements. The ICC and its 95% confidence interval were computed using R statistical software (version 3.2.5, R Foundation, Vienna, Austria). An ICC between 0.75 and 0.9 corresponds to a good reliability, whereas a value between 0.5 and 0.75 indicates moderate reliability [21]. Spearman correlation was computed between D, W, and ex vivo S₁₀ measurements, since these parameters vary non-linearly as a function of erythrocyte aggregation. To verify the potential influence of gender on values of D, W and S₁₀, Student t-tests were used. The mean intra-subject variability was calculated as the variance divided by the mean of the 5 repeated measurements and averaged over all participants. Repeated ANOVA (RANOVA) tests were performed to evaluate any relationship between the 5 repeated SFSAE measures and the measurement order, and any relationship between the force applied and the measurement order. The median shear rate within the selected vein was estimated by substitution of the measured maximum velocity and vessel diameter into the Ostwald-De Waele power law model [22]. See Appendix 1 for a mathematical description of this model.

The force applied by the pneumatic occluder to reduce the flow within the brachial vein of the forearm was quite small at 1.5±1.2 N (Table 1). Over the leg, the required force to adjust the flow was slightly higher at 2.7±1.9 N. Figure 4 reports the applied force on the forearm for every subject. For a single participant, the required force was similar along the 5 measurements (mean intra-subject variability of 16% and 17% on the forearm and leg, respectively). Between participants, the mean applied force varied from 0.3 to 5.2 N (inter-subject variability of 76% and 70% on the forearm and leg, respectively). For both venous sites, the patient age, vein’s depth, and BMI were not correlated with the applied force required to control the flow. Measurements showed no trend between the force applied and the acquisition order (p-value >0.05) for both forearm and leg (i.e., no memory effect).

OPEN IN VIEWER

Fig.4

Applied force required to reduce the blood flow within the vein to the targeted maximal blood velocity of 1 mm/sec for 5 repeated ultrasound measurements on the forearm of every participant. Participants #28 and #43 had no data for this configuration. The color of data points corresponds to the order in which measurements were taken. There was no association between the calibrated force applied on the skin and the measurement order (p-value >0.05 among 5 measurements).

The mean value, standard deviation and intraclass correlation coefficient (ICC) of D and W for all measurement configurations are shown in Table 2. When looking at averaged values of D or W, it can be noticed that they are similar for measurements on the arm or the leg. The same observation holds for the standard deviation and the intraclass correlation coefficient. As expected, both D and W showed higher values at low shear rate. In general, ICCs reveal better reliability for D than W. With ICC values higher than 0.75, measurements presented a good reliability except for W at natural shear rates, which presented moderate reliability. According to reported ICC values, the leg site seemed a better location for quantitative ultrasound aggregation measurements.

In the remaining of this manuscript, to simplify figure displays, only the SFSAE parameter D is reported (similar results were obtained for W). To illustrate all data points used to compute a single value in Table 2, all individual measurements of D at low shear rate on the forearm are displayed in Fig. 5. The mean value of D for this configuration was 1.46, with an intra-subject variability of 26.3%, ranging from 5.7% (participant #26) to 91.5% (participant #4). The color of data points corresponds to the order in which measurements were taken. Measurements showed no trend between data point values and their acquisition order in the sequence of 5 measures (p-values >0.05 for both forearm and leg configurations).

OPEN IN VIEWER

Fig.5

Plot of 5 measured values of the ultrasound aggregation diameter (parameter D) on the forearm at a low shear rate for every participant. Participants #28 and #43 had no data for this configuration. The color of data points corresponds to the order in which measurements were taken. There was no association between the aggregate diameter and the measurement order (p-value >0.05 among 5 measurements).

Measurements of D on the leg and on the forearm were correlated as displayed in Fig. 6 for both shear rate conditions. Low shear rate measurements spread well along the regression line (round blue dataset over the blue dotted line) with a coefficient of determination of 0.77. This strong relation is confirmed by the Spearman’s correlation coefficient ρ= 0.84 (p-value <0.01). Results were similar for W at low shear rate (ρ= 0.78, p-value <0.01). At natural shear rate, less variation in RBC aggregation between individuals was noticed; measurements of D in Fig. 6 were packed close to the origin (orange diamond dataset). Correlation coefficients for D and W between the leg and the forearm at natural flow were the same at ρ= 0.52 (p-values <0.01).

OPEN IN VIEWER

Fig.6

Aggregate diameter (D parameter) on the forearm and leg are proportional for measurements at low shear (blue) and natural shear rate (orange) conditions. SR: shear rate.

Figure 7 compares D values that were measured for each participant at natural and low shear rate conditions. Both forearm (black dot dataset) and leg (red diamond dataset) measurements are displayed on this figure and present similar trends. Similarities of measurements at both shear rates are confirmed by Spearman’s correlation coefficients ρ = 0.74 (p-value <0.01) for the forearm, and ρ = 0.56 (p-value = 0.025) for the leg. With W, correlations were ρ = 0.76% (p-value <0.01) for the forearm, and ρ = 0.44 (p-value = 0.090, non-significant) for the leg. As noticed in Fig. 7, some measurements were close to the abscissa, as highlighted by the green ellipse. For these participants, the natural blood velocity measured by Doppler ultrasound was higher than for the other participants, on average 38 mm/s compared to 23 mm/s. For these individuals, the natural shear rate was too high to promote aggregation. In these particular cases, D values given by the SFSAE method at natural shear was close to 0.1 compared to 1.0 for measurements at low shear rate on the same participant. Above the regression line are two participants presenting high D values at natural shear and lower D at low shear rate, as highlighted by the blue box. The natural blood flow speed for these two participants was too small to be measured by Doppler ultrasound, whereas at reduced flow controlled by the bracelet the mean velocity measured by PIV was around 0.5 mm/s.

OPEN IN VIEWER

Fig.7

SFSAE D parameter on both forearm and leg at natural shear rate versus low shear rate conditions. The low shear obtained by using the compression bracelet increased the range of D values for participants with naturally high venous flow (green ellipse), and contributed to standardise measurements for those with very low venous flow (blue box).

RBC aggregation measurements made on blood samples using the erythroaggregometer ranged from 14.2 to 28.9 (21.6±3.6). Since S₁₀ and in vivo SFSAE parameters describe laser and ultrasound RBC aggregate backscattering properties, respectively, it was interesting to explore the relation between both approaches. Figure 8 presents D at low shear rate as a function of the ex vivo parameter S₁₀. One can notice that D tends to increase proportionally with S₁₀. Similar relations were noticed for D at natural flow and for W at both flow conditions (not shown). Significant correlation coefficients were measured between D or W assessed on the forearm at both shear rates and S₁₀ (0.64 < ρ <0.72, p-values <0.01). SFSAE measurements on the leg at low shear rate condition were also correlated with S₁₀ (for D: ρ= 0.81, p-value <0.01; and for W: ρ= 0.76, p-value <0.01). However, SFSAE measurements at natural flow on the leg were weakly correlated with S₁₀ (for D: ρ= 0.35, p-value = 0.09, non-significant; and for W: ρ= 0.34, p-value = 0.04).

OPEN IN VIEWER

Fig.8

SFSAE D parameter on both forearm and leg at low shear rates versus the ex vivo erythroaggregometer parameter S₁₀. D values increased proportionally with S₁₀, as shown by the linear fitting curve.

4.1 Adjustment of the force applied to reduce the blood flow

As seen in Fig. 4, a wide range of applied pressure was required among participants to reduce the venous flow to the targeted value. Indeed, Garcia-Duitama et al. [14] proposed keeping the shear rate below 3 s^–1 for obtaining measurements under stable aggregation conditions. However, as reported in Table 1, the shear rate at natural flow was on average 37 s^–1, approximately. Consequently, this means that without any flow control, the shear rate in the vein would be well above the proposed threshold for optimum measures. Also, as noticed in Fig. 7, two participants had natural flow close to stasis (blue box) and values of D (and W) higher than under the controlled flow condition. For these participants, the use of the bracelet was not helpful to control the blood flow. Thus, according to results of Figs 4 and 7, a standard pressure that would apply to all participants to reduce the blood flow could not be established. Also, as demonstrated by Garcia-Duitama et al. [14], assessing RBC aggregation under stasis condition could not be a solution. Indeed, this approach is not reproducible because the natural flow before stoppage would provide variable RBC aggregation states affecting the aggregation kinetics and ultrasound measurements.

As indicated in Table 1, standard deviations of the shear rate at natural flow were quite large, especially in lower limbs. Coefficients of variation of 59% and 102% at natural flow were reduced to 30% and 38% under controlled flow. Therefore, the bracelet was useful to avoid variable measurements under natural flow due to the different states of RBC aggregation; see data in the green ellipse of Fig. 7. By decreasing the shear rate, RBC aggregation was enhanced and better controlled, and a wider range of D and W could be observed, reflecting different levels of aggregation. However, even if coefficients of variation on reported shear rates were lower when the bracelet was used, corresponding ICCs with the bracelet were 0.45 in the forearm and 0.49 in the leg, showing poor reliability of the shear rate control. This suggests that the blood flow adjustment technique and device could be further improved. An automatic control of the blood pressure applied by the bracelet through a feedback loop would undoubtedly provide greater stability of the blood flow and better repeatability of the aggregation measurement. In addition, because only 19 out of 50 measurements could be performed on the leg site at low shear rate, this also supports the need of redesigning a new compression device. A rotating bracelet with a different footprint coupled with a miniaturised transducer might be more adapted for measurements over the leg (ankle).

4.2 Ultrasound SFSAE aggregation measurements

4.3 Ex vivo S₁₀ vs SFSAE aggregation measurements

When the blood flow was controlled with the bracelet, strong correlations were observed between D and W parameters and the ex vivo S₁₀ reference standard. This finding remained valid for measurements on forearm’s and leg’s sites (Fig. 8). Consequently, as discussed above, ultrasonic measurements of RBC aggregation can be seen as a local measure reflecting a systemic property. At the natural shear rate, the same trend was observed on the forearm’s site but not on the leg, where data showed weak and non-significant correlations with S₁₀. In our cohort, S₁₀= 21.6±3.6, which is close to the value of 23±4 reported for healthy subjects [5].

4.4 Limitations

The sample size of this cohort was too small to study the influence of diseases, age, drugs, native hematocrit or ethnicity on aggregation parameters D, W or S₁₀. However, Student t-tests showed significant differences in D and W values between men and women. This difference in RBC aggregation according to sex was already reported using ex vivo laser backscatter measurements [34]. On the other hand, a relation between obesity and erythrocyte aggregation has also been described [35–37]. The current cohort was composed of healthy participants having a BMI around 25 and did not include obese individuals. However, a significant correlation was nevertheless noticed between the BMI and S₁₀ (ρ= 0.54, p-value <0.01), but not between BMI and SFSAE parameters.

Some limitations still hamper the immediate use of this technology for clinical diagnosis. The current bulky ultrasound system may not be optimum for use in critical care or surgery units. A compact ultrasound device would be more appropriate for these applications. Furthermore, acquisitions were performed at 35 MHz, which is not a typical frequency used by clinical scanners that operate typically between 3 to 15 MHz. Consequently, a dedicated portable miniaturized bed-site high-frequency ultrasound system would need to be developed. Such system would ideally include a monitor displaying time-varying D and W parameters. Indeed, for monitoring RBC aggregation and thus systemic inflammation, the specific value of a given parameter might be less crucial than its evolution over time under similar flow conditions. From a clinical point of view, it is difficult to predict whether the venous flow will remain constant throughout the aggregation monitoring procedure. This is why it seemed more appropriate to control the venous flow and to adjust it at a very low shear for which RBC aggregation is promoted in order to maintain a margin if the patient’s condition deteriorates. It would eventually be important to assess the day-to-day or week-to-week reproducibility of SFSAE measures. Thus to summarize, challenges remain important but feasible.