Ultrasound Velocimetry in the Healthy Femoral Bifurcation: A Comparative Study Against 4-D Flow Magnetic Resonance Imaging
Majorie van Helvert, Janna Ruisch, Joosje M.K. de Bakker, Anne E.C.M. Saris, Chris L. de Korte, Michel Versluis, Erik Groot Jebbink, Michel M.P.J. Reijnen
Abstract
Objective
Local flow dynamics impact atherosclerosis yet are difficult to quantify with conventional ultrasound techniques. This study investigates the performance of ultrasound vector flow imaging (US-VFI) with and without ultrasound contrast agents in the healthy femoral bifurcation.
Methods
High-frame-rate ultrasound data with incremental acoustic outputs were acquired in the femoral bifurcations of 20 healthy subjects before (50V) and after contrast injection (2V, 5V and 10V). 2-D blood-velocity profiles were obtained through native blood speckle tracking (BST) and contrast tracking (echo particle image velocimetry [echoPIV]). As a reference, 4-D flow magnetic resonance imaging (4-D flow MRI) was acquired. Contrast-to-background ratio and vector correlation were used to assess the quality of the US-VFI acquisitions. Spatiotemporal velocity profiles were extracted, from which peak velocities (PSV) were compared between the modalities. Furthermore, root-mean-square error analysis was performed.
Results
US-VFI was successful in 99% of the cases and optimal VFI quality was established with the 10V echoPIV and BST settings. A good correspondence between 10V echoPIV and BST was found, with a mean PSV difference of -0.5 cm/s (limits of agreement: -14.1–13.2). Both US-VFI techniques compared well with 4-D flow MRI, with a mean PSV difference of 1.4 cm/s (-18.7–21.6) between 10V echoPIV and MRI, and 0.3 cm/s (-23.8–24.4) between BST and MRI. Similar complex flow patterns among all modalities were observed.
Conclusion
2-D blood-flow quantification of femoral bifurcation is feasible with echoPIV and BST. Both modalities showed good agreement compared to 4-D flow MRI. For the femoral tract the administration of contrast was not needed to increase the echogenicity of the blood for optimal image quality.
Keywords
Ultrasound vector flow imaging
Blood speckle tracking
Echo particle image velocimetry
High-frame-rate ultrasound
4-D flow MRI
Blood-flow imaging
Femoral artery
Introduction
Local hemodynamics play an important role in the onset and progression of atherosclerotic disease. Flow disturbances lead to regions with low and oscillating wall shear stresses, which in turn could contribute to local endothelial cell disfunction [1,2]. The bifurcation of the common femoral artery (CFA), into the superficial and deep femoral arteries (SFA and DFA, respectively), is known for atherosclerotic plaque formation [3]. The bifurcating anatomy with associated flow disturbances, as well as the typical triphasic flow profile, are potentially important contributors to the high incidence of atherosclerotic lesions in this area [4,5]. This relationship between atherosclerosis and local hemodynamics makes it of clinical importance to adequately visualize and quantify blood-flow velocity profiles.
Traditionally, duplex ultrasound (DUS) has been the technique of choice to examine the peripheral vascular system. DUS employs focused line-by-line B-mode ultrasound images combined with pulsed-wave Doppler measurements to assess both the anatomy and local blood velocities. While DUS is readily accessible and cost effective, the velocity estimate reliability is affected by its inherent operator- and machine dependency [[6], [7], [8]]. For example, a manual beam-to-flow angle correction is necessary to recover the 1-D cross-beam velocity. Proper alignment with the blood flow is however difficult, particularly in branched anatomies or in diseased vessels, where complex flow patterns arise with varying flow directions in space and time [6]. As such, DUS is unable to accurately estimate blood-flow velocities and cannot unravel the full complexity of the local hemodynamics.
4-D flow magnetic resonance imaging (MRI) can be used to study peripheral flow characteristics over a full cardiac cycle [9,10]. Although MRI offers velocity information along all three spatial directions, it suffers from limited resolution, the need for interleaved sampling, long scan durations and high costs.
Multiple angle-independent ultrasound vector flow imaging (US-VFI) techniques have emerged that enable 2-D or 3-D blood-flow quantification [11,12]. For example, native blood-speckle tracking (BST) tracks the speckle patterns that arise from moving red blood cells through block-matching between subsequent frames [13]. Accurate tracking of fast-moving blood requires high temporal resolution to minimize the inter-frame displacement. While conventional DUS utilizes frame rates of up to 100 Hz, high-frame-rate plane wave ultrasound enables frame rates of up to 10 kHz [14]. However, high-frame-rate ultrasound suffers from a poor signal-to-noise (SNR) ratio due to lack of focusing and decrease in the acoustic pressure field with propagation distance.
High-frame-rate BST operates well in superficial vessels such as the carotid artery [15], but its use is limited in vessels located at larger depths. Contrast-enhanced echo particle image velocimetry (echoPIV) uses ultrasound contrast agents (UCAs) to increase the echogenicity of the blood pool, which thereby overcomes the SNR limitations of BST [16]. Previous work has shown that high-frame-rate echoPIV is even feasible in the aorto-iliac tract and stented femoral arteries of patients with peripheral arterial disease [[17], [18], [19]]. This earlier work did not investigate the potential of BST in the femoral artery, where it could offer a viable and, above all, non-invasive alternative to echoPIV in these rather superficially located vessels. Two examples of BST in the femoral bifurcation are showcased by Hansen et al. [20]. However, studies concerning the in vivo application and validation of echoPIV and BST are sparse, particularly in the peripheral arteries.
This study aims to evaluate the performance of BST and echoPIV to quantify spatiotemporal blood-flow velocity profiles in the femoral trajectory of healthy subjects against 4-D flow MRI as a reference.
Materials and methods
This single-center, prospective, observational study was approved by the authorized regional Medical Ethics Committee (NL80478.091.22) and the institutional review board. The study was registered at clinicaltrials.gov (NCT05451485) and performed in accordance with the principles of the Declaration of Helsinki and Good Clinical Practice guidelines. Twenty healthy subjects (10 men, median age: 49 years; range: 24–75 years) were enrolled after providing written informed consent. Subjects were included when there was no history of cardiorespiratory disease. Subjects were excluded when pregnant or when the use of UCA (SonoVue; Bracco, Milan, Italy) or the execution of an MRI scan was contraindicated. All measurements were performed between November 2022 and March 2023.
US-VFI acquisitions
Each subject underwent the following ultrasound examination of the left femoral bifurcation. First, a conventional DUS evaluation of the distal CFA, the femoral bifurcation and the proximal SFA was performed with a clinical ultrasound system connected to an L3-7 transducer (Hitachi ARIETTA 850; FUJIFILM Healthcare Corporation, Tokyo, Japan). The femoral bifurcation was located first and used as an anatomical landmark. The transducer was then positioned such that the field of view only contained the distal CFA, femoral bifurcation (division of the CFA into SFA and DFA in the center) or proximal SFA and that the centerline of the desired vessel was captured. B-mode images served as anatomical references, and an equivalent view was located for subsequent high-frame-rate acquisitions. High-frame-rate BST data were acquired of the same three locations with a Vantage 256 ultrasound system connected to a L11-4v transducer (Verasonics, Kirkland, WA, USA; see Table 1 for system and acquisition specifics). A B-mode live view of the anatomy was provided through a three-angled plane wave imaging scheme at 100 Hz. Once the vessel of interest was located, the system was switched to high-frame-rate acquisition mode and 3 s of BST data were obtained. A clutter-filtered preview based on singular value decomposition (SVD) was visually inspected to ensure that sufficient signal from the blood pool was present. Finally, echoPIV data were acquired by repeating the high-frame-rate ultrasound measurements after an intravenous injection of 0.75 mL UCA. Contrast levels were monitored using a three-angled pulse inversion sequence next to the B-mode live view. After bolus passage, three consecutive echoPIV measurements were acquired with incremental transmit voltages. Lower voltages compared with BST were used to prevent UCA destruction [17].
Table 1. High-frame-rate ultrasound and particle image velocimetry settings
| Ultrasound property | Setting |
|---|---|
| Ultrasound scanner | Verasonics Vantage 256 |
| Transducer | L11-4v |
| No. of transducer elements | 128 |
| Pitch | 0.3 mm |
| Elevational focus | 20 mm |
| Center frequency | 4 MHz |
| Pulse repetition frequency | 9 kHz |
| Pulse duration (no. of cycles) | 1 |
| No. of steering angles | 3 (-18⁰, 0⁰, 18⁰) |
| Transmission voltage BST EchoPIV |
50V 2, 5, 10V |
| Apodization in transmit | Rectangular |
| Apodization in receive | Rectangular |
| Acquisition duration | 3 s |
| Field of view (axial × lateral) | 50 × 40 mm2 |
| Pixel size (beamforming grid) | 0.19 × 0.19 mm2 |
| PIV property | Setting |
| Similarity metric | Normalized cross-correlation (FFT) |
| No. of iterations | 4 |
| Kernel size Iteration 1 and 2 Iteration 3 and 4 |
16 × 16 pixel (3 × 3 mm2) 8 × 8 pixel (1.5 × 1.5 mm2) |
| Kernel overlap | 75 % |
| Spatial vector resolution | 2 × 2 pixel (0.38 × 0.38 mm2) |
| Correlation averages | 10 |
| Temporal vector resolution | 300 Hz |
| Subpixel estimator | 2 × 3 pixel parabola fitting |
| Final smoothing Spatial (Gaussian) Temporal (moving average) |
3 × 3 vectors (σ = 0.5 × 0.5) 3-ensemble |
BST, blood speckle tracking; PIV, particle image velocimetry.
US-VFI analysis
High-frame-rate ultrasound data were processed offline in MATLAB (version R2022b; The Mathworks, Natick, MA, USA). The Verasonics Vantage System beamformer (v4.6.2) was used to reconstruct the separate angled acquisitions through delay-and-sum beamforming on the 0-degree cartesian grid. Subsequently, the beamformed in-phase quadrature (IQ) data were coherently compounded, resulting in an effective frame rate of 3000 Hz. An SVD-based clutter filter was applied to the IQ data of a whole acquisition to automatically distinguish between static tissue clutter and dynamic blood flow based on the spatial similarity matrix [21].
PIV analysis was conducted on the signals envelope to obtain the 2-D velocity vector fields. A maximum-intensity projection of the ultrasound data over the acquisition duration was used to manually segment the vessel lumen. In addition to its use in the PIV analysis, this rigid vessel contour was saved for later inter-modal image-registration purposes. A modified implementation of PIVlab [22,23] was used to calculate block-wise normalized cross-correlations between image pairs with an iterative interrogation window refinement scheme (see Table 1). Ten consecutive normalized cross-correlation maps were averaged to increase the SNR, and the displacement between two image pairs was estimated via identification of the peak in the correlation map. Finally, 300 velocity vector fields per second were acquired. Identical PIV analysis was applied to the BST and the echoPIV datasets to allow for comparison between both.
Comparison between echoPIV and BST
Differences between the US-VFI modalities were assessed via two metrics: (1) the contrast-to-background ratio (CBR) and (2) the vector correlation. The CBR was computed to compare the signal strength of the blood pool relative to the surrounding background, i.e., tissue clutter or noise that remained after SVD filtering. The vessel lumen and a background region anterior to the vessel were manually segmented. Subsequently, the CBR was calculated per frame as follows: CBR=20log10(RMSblood/RMStissue),where RMS represents the root-mean-square signal power of the blood or the surrounding background.
Vector correlation was retrieved to assess the tracking performance of the PIV algorithm. Given the nonparametric distribution of the normalized cross-correlation values, the median per vector field was taken to retrieve a single value per frame. The vector correlation ranged between 0 and 1, where a higher value indicates that the displacement between two frames was identified more accurately.
Both metrics were calculated for the entire acquisition duration. Moreover, differences within a heartbeat were assessed by identifying the systolic and end-diastolic phase based on the inflection points in the temporal velocity profile.
Based on these metrics, the best echoPIV acquisition was selected to further compare with the BST measurements in terms of velocity profiles. Temporal velocity profiles were acquired along the CFA and SFA centerline, from which peak systolic velocities (PSV) were obtained. Peak systolic spatial velocity profiles were extracted perpendicular to the centerline points. The root-mean-square error (RMSE) between both modalities was computed for both the temporal and spatial velocity profile.
4-D flow MRI
Each subject underwent a free-breathing and retrospective vector cardiographic-gated 4-D flow MRI scan covering the entire femoral trajectory. All images were obtained with a 3.0T MRI system (Ingenia, Philips Healthcare, Best, The Netherlands). Scan settings are listed in Table 2. MRI data were generally acquired on the same day or within 1 week of the ultrasound measurements. Blood pressure and heart rate were measured to assure comparable hemodynamical conditions between the two examinations.
Table 2. 4D flow MRI scan settings
| Scan property | Setting |
|---|---|
| Field of view | 200 × 200 × 68.75 – 85 mm3 |
| Spatial resolution Acquired Reconstructed |
2.5 × 2.5 × 2.5 mm3 1.25 × 1.25 × 1.25 mm3 |
| No. of cardiac phases | 28 |
| Temporal resolution | 24 – 42 msec |
| Repetition time | 3.5–4.1 ms |
| Echo time | 2.0–2.8 ms |
| Flip angle | 8° |
| Velocity encoding | 80–160 cm/s |
| Scan duration | 6–12 min |
MRI data were corrected for phase offset errors and noise masked during the image reconstruction performed by the MRI system. Aliasing artefacts were avoided by setting the velocity encoding sufficiently high based on a 2-D scout measurement. Visual inspection confirmed the absence of aliasing. Further processing was performed using in-house built software in MATLAB. A 3-D magnetic resonance angiography representation was created via temporal maximum-intensity projection of the magnitude images [24]. This volume was then used to automatically segment the femoral trajectory.
US-VFI and 4-D flow MRI comparison
To enable inter-modal comparison, semi-automatic image registration was performed to obtain an oblique slice from the MRI volume that best matched the 2-D ultrasound image (Fig. 1). The user initialized the process by manually selecting a slice in a 4-D flow MRI preview that displayed the anatomical images with labelled segmented vasculature. The user had full freedom to translate and rotate the slice in all three spatial directions, and decided on the best match based on anatomical landmarks. The orientation of the selected slice was then used to align the segmented MRI vasculature with the ultrasound image. A 2-D plane was extracted from the MRI volume and the obtained vessel contour was compared with the vessel contour on the ultrasound image. To this end, the ultrasound image resolution was down-sampled to the MRI resolution, and both images were binarized, i.e., the vessel lumen contained 1s and the area outside the lumen contained 0s. Geometrical similarity of the contours was assessed by calculating the sum of absolute differences between the binarized US and MRI images. Following initialization, the 4-D flow MRI volume was further translated and rotated along all three axes to automatically fine tune the selected slice. The parameter space was defined relative to the vessel diameter (∼10 mm) and the ultrasound acquisition landmarks (desired centerline imaging plane and a set location with respect to the bifurcation point). Hence, a translation window of 5 mm with a step size of 1.25 mm, and a rotation window of 12 degrees with a step size of 2 degrees, was employed in all three directions around the center point of the initialized slice. All combinations of translations and rotations were analyzed, and for each step the geometrical similarity between the vessel contours was evaluated. The contour that gave the highest similarity, i.e., lowest sum of absolute differences, was considered the best matching MRI slice and its corresponding transformation was used to project the 3-D MRI velocity vectors onto the 2-D slice. The registration was performed twice: one time for the best echoPIV acquisition (MRIPIV) and one time for the BST acquisition (MRIBST). Quantitative comparison of the velocity data between modalities was performed similar to the aforementioned ultrasound comparison.
Figure 1. Stepwise inter-model alignment of the 2-D ultrasound image (US-VFI) within the 4-D flow magnetic resonance imaging (MRI) volume. The user initializes the registration by selecting the most suitable slice in the MRI preview based on anatomical landmarks. The transformation matrix of the manually selected slice is used to align the MRI volume with the down-sampled ultrasound image. A set of rotations and translations further fine tunes the MRI slice based on coherent geometrical features in the vessel contours. The lowest sum of absolute differences between the two contours defines the most optimal MRI slice. Two examples of (suboptimal and best) matched contours are given.
Statistical analysis
Statistical tests were performed using IBM SPSS statistics 29 (IBM corporation, Armonk, NY, USA). A Shapiro-Wilk test was used to assess normality. Differences in ultrasound acquisitions were evaluated with a repeated measurement ANOVA test for normally distributed data and a Friedman repeated measures test for non-normally distributed data. When p < 0.05, post hoc analysis between individual groups was conducted using a paired samples t-test (normal distribution) or a Wilcoxon signed-rank test (non-normal distribution) followed by a Bonferroni correction for multiple comparisons. Differences in PSV were assessed using Bland-Altman analysis and a Wilcoxon signed-rank test. Differences in RMSE were also tested via a Wilcoxon signed-rank test. Continuous variables are given as median (interquartile range [IQR]). p < 0.05 was considered statistically significant.
Results
US-VFI was successful in 99% of the cases (238 out of 240 recordings). Two BST recordings (1 CFA and 1 SFA) were excluded from further analysis because only part of the vessel was captured due to transducer displacement prior to high-frame-rate acquisition. 4-D flow MRI was successful in 80% of the cases (16 out of 20 scans). The remaining four MRI scans only imaged the SFA, limiting the comparison with ultrasound to only this location. Moreover, this restricted volume hampered image registration in two out of four datasets due to the lack of anatomical landmarks. An overview of the final datasets used for comparison is provided in Figure 2.
Figure 2. Flowchart representation of the complete datasets used for the inter-modal comparison. The ultrasound vector flow imaging (US-VFI) data of the femoral bifurcation was only used for contrast-to-background and vector correlation comparisons. MRI, magnetic resonance imaging; CFA, common femoral artery; SFA, superficial femoral artery; echoPIV, echo particle image velocimetry; BST, blood speckle tracking.
Comparison between echoPIV and BST
Signal strengths visually improved with increasing echoPIV transmit voltages and were highest for the BST acquisition (Fig. 3). The 2V echoPIV recording shows significantly lower CBR and vector correlation compared with the other acquisitions (p < 0.001; Fig. 4 and Fig. SA.1). Averaged cycle and systolic CBR was also significantly lower in the 5V echoPIV measurement compared with the 10V echoPIV and BST recordings (p < 0.001, except averaged cycle CBR 5V vs. 10V: p = 0.04), while no significant difference was found in vector correlation. No significant differences were found between 10V echoPIV and BST, except during diastole where both CBR and vector correlation was significantly higher in the BST acquisition (p = 0.001 and p = 0.008, respectively).
Figure 3. The effect of different ultrasound settings on clutter-filtered high-frame-rate ultrasound images acquired in the presence (echo particle image velocimetry [echoPIV]) and absence (blood speckle tracking [BST]) of ultrasound contrast agents. The femoral bifurcation of one of the subjects is shown.
Figure 4. Differences in contrast-to-background ratio (left) and vector correlation (right) over the cardiac cycle, during systole and diastole, between the ultrasound acquisitions. Data of all three measured locations are pooled and presented as boxplots (n = 58). Edges of the boxes represent the 25th and 75th percentiles, while the whiskers extent to the minimum and maximum values, excluding outliers (diamond shape). Black lines denote significant differences between groups (p < 0.05). Results per individual location are provided in Fig. SA 1. EchoPIV, echo particle image velocimetry; BST, blood speckle tracking.
Peak velocities showed no significant differences between the US-VFI modalities, with a PSV of 64.0 cm/s (IQR: 9.4 cm/s) obtained with 10V echoPIV compared to 64.0 cm/s (IQR: 19.4 cm/s) obtained with BST in the CFA, and 65.6 cm/s (IQR: 8.0 cm/s) compared to 65.5 cm/s (IQR: 15.5 cm/s) in the SFA. Bland-Altman analysis underpinned these findings, showing a small underestimation in BST PSVs compared to echoPIV, with a bias of -0.5 cm/s and relatively small limits of agreements (-14.1; 13.2 cm/s; Fig. 5). Moreover, good agreement was found among temporal velocity profiles, with an RMSE of 4.4 cm/s (IQR: 1.5 cm/s) and 4.8 cm/s (IQR: 2.4 cm/s) for CFA and SFA, respectively. Spatial velocity profiles showed an RMSE of 7.1 cm/s (IQR: 3.5 cm/s) in the CFA and 6.9 cm/s (IQR: 5.8 cm/s) in the SFA.
Figure 5. Bland-Altman analysis of discrepancies in peak systolic velocity obtained with 10V echo particle image velocimetry (echoPIV) and blood speckle tracking (BST) in the common femoral artery (CFA, n = 19) and superficial femoral artery (SFA, n = 19). SD, standard deviation. CFA, common femoral artery; SFA, superficial femoral artery; echoPIV, echo particle image velocimetry; BST, blood speckle tracking.
Comparison of ultrasound VFI with 4-D flow MRI
2-D velocity vector fields showed equivalent patterns for 10V echoPIV and BST, as well as their corresponding MRI plane. For example, in the CFA, blood flow is seen to move towards the outer curve during the deceleration phase among all modalities (Fig. 6, Video S1–S4). Additionally, a recirculation zone in the distal CFA was visualized by echoPIV and its corresponding MRIPIV, while this was not captured in the BST field of view nor its corresponding MRIBST plane. An example of end-systolic 2-D velocity patterns in the SFA presents a clear recirculation zone imaged by all modalities (Fig. 7, Video S5–S8).
Figure 6. 2-D velocity streamline profiles during the deceleration phase in the common femoral artery of one of the subjects. Similar patterns can be appreciated between the modalities, including higher velocities at the outer curve of the vessel. The 10V echo particle image velocimetry (echoPIV) acquisition and its matched 4-D flow magnatic resonance imaging (MRIPIV) show a clockwise recirculation zone at the distal common femoral artery. Blood speckle tracking (BST) captured a slightly different plane in which this recirculation is not present. White lines delineate the vessel lumen. Imaging width of 0 mm represents the middle of the transducer array.
Figure 7. 2-D velocity streamline profiles at end systole in the superficial femoral artery of one of the subjects. Similar patterns can be appreciated between the modalities, including a counter-clockwise recirculation zone. Note the misalignment in the recirculation zone, indicating that the transducer placement changed between the echo particle image velocimetry (echoPIV) and blood speckle tracking (BST) measurement. White lines delineate the vessel lumen. Imaging width of 0 mm represents the middle of the transducer array.
MRI PSVs were not significantly different compared with the corresponding US-VFI technique. MRIPIV showed PSVs of 69.4 cm/s (IQR: 17.7 cm/s) and 65.8 cm/s (IQR: 7.3 cm/s) in the CFA and SFA, respectively, and MRIBST presents PSVs of 70.7 cm/s (IQR: 21.0 cm/s) and 64.8 cm/s (IQR: 7.7 cm/s). Bland-Altman analysis substantiated these findings, with a positive bias of 1.4 cm/s towards echoPIV compared with MRIPIV (Fig. 8). A positive bias of 0.3 cm/s towards BST was found when comparing with MRIBST, but with slightly increased limits of agreement. Temporal and spatial velocity profiles corresponded well between the modalities, as can be appreciated from the four examples given in Fig. 9 and the RMSEs listed in Table 3. The RMSEs were not significantly different.
Figure 8. Bland-Altman plots with peak systolic velocities. (left) Discrepancies between 10V echo particle image velocimetry (echoPIV) and its corresponding 4-D flow magnetic resonance imaging plane (MRIPIV) in the common femoral artery (CFA, n = 16) and superficial femoral artery (SFA, n = 18). (right) Discrepancies between blood speckle tracking (BST) and its corresponding 4-D flow MRI plane (MRIBST) in the CFA (n = 15) and SFA (n = 17). SD, standard deviation.
Figure 9. Examples of temporal (upper row) and spatial (lower row) velocity profiles measured with all modalities in the common femoral artery (left) and superficial femoral artery (right). Shading represents the standard deviation over the centerline points. RMSEPIV, root-mean-square error between 10V echoPIV and its corresponding 4-D flow MRI plane (MRIPIV) in cm/s; RMSEBST, root-mean-square error between BST and its corresponding 4-D flow MRI plane (MRIBST) in cm/s.
Table 3. Root-mean-square error.
| Inter-model comparison | Temporal velocity profile | Spatial velocity profile | ||
|---|---|---|---|---|
| CFA | SFA | CFA | SFA | |
| MRIPIV – 10V EchoPIV | 6.2 (2.7) | 5.2 (2.0) | 12.3 (7.5) | 13.1 (4.8) |
| MRIBST – BST | 6.8 (4.9) | 5.6 (3.1) | 16.0 (6.9) | 13.3 (12.3) |
Root-mean-square error given as median (interquartile range) both in cm/s. CFA, common femoral artery; SFA, superficial femoral artery; MRI, magnetic resonance imaging; EchoPIV, ultrasound particle image velocimetry; BST, blood speckle tracking.
Discussion
This work investigated the performance of two novel US-VFI modalities to quantify 2-D local hemodynamics in the healthy femoral bifurcation. Assessment of different ultrasound acquisition parameters, i.e., voltage transmission and the addition of UCA, revealed comparable image quality and tracking performance between the 10V echoPIV and BST recordings. Furthermore, both US-VFI techniques showed good agreement with 4-D flow MRI in terms of 2-D velocity patterns as well as temporal and spatial velocity profiles in the CFA and proximal SFA. Only minor differences in PSV were found between the ultrasound modalities, suggesting that both techniques could be used to quantify local hemodynamics in the femoral trajectory, and that the addition of UCA is not necessary to enhance the echogenicity of the blood in these relatively superficial arteries.
EchoPIV recordings at several voltages, i.e., 2, 5 and 10V, were successfully acquired at the three different locations along the femoral tract. BST recordings were also successful in all cases, except for two measurements in which probe displacement caused misalignment between the ultrasound plane and the vessel. For echoPIV, increased transmission voltages resulted in increased CBR for the full cardiac cycle and during systole. CBR generally decreased during diastole, with the largest decrease found for 10V echoPIV; however, a comparable CBR to that measured at 5V remained. Similar trends between CBR and transmission voltage have been described and were attributed to UCA destruction caused by higher transmission voltages and increased exposure time to ultrasound [17,23,25]. A transmission voltage of 2V was therefore included in the sequence to evaluate, and possibly overcome, the issue of microbubble destruction, but data quality turned out to be too poor for a quantitative flow assessment. Another reason for the decrease in CBR can be related to insufficient clutter filtering during diastole, as slow or stagnant blood flow causes the singular values of blood to coincide with that of tissue [21]. This could also explain the decrease in BST CBR at diastole, where no UCA was used. Currently, the SVD cut-off is based on the entire acquisition duration, while tissue and blood characteristics vary across the cardiac cycle. Thus, the use of an adaptive SVD cut-off could improve the clutter filter process [21].
Vector correlation showed comparable outcomes between 5V echoPIV, 10V echoPIV and BST, indicating similar performance in speckle tracking. During diastole lower correlations were found for 10V echoPIV compared with BST, but with large variations in both datasets. The absence of speckle probably led to the lower correlations, whereas the relatively small speckle displacements between consecutive frames led to larger correlations.
Comparisons of 10V echoPIV with 4-D flow MRI, and BST with 4-D flow MRI, demonstrated an overall good agreement between the spatiotemporal velocity profiles in the CFA and SFA. A physiological triphasic velocity waveform, including a clear backflow phase, was captured by all modalities, with peak velocities in a range of 50–90 cm/s. Bland-Altman analysis revealed negligibly small mean PSV differences, with limits of agreement equivalent to earlier work where echoPIV and 4-D flow MRI were compared in the aorto-iliac trajectory of healthy volunteers [26] and in the left ventricle of cardiac patients [27]. An extensive evaluation of velocities obtained in the carotid artery with conventional DUS and 4-D flow MRI showed a negative bias in systolic velocities measured with 4-D flow MRI [28]. Higher DUS peak velocity estimates were also observed in the current study, namely 102 cm/s (IQR: 35.5 cm/s) in the CFA and 95.5 cm/s (IQR: 18 cm/s) in the SFA. This suggests that both US-VFI and 4-D flow MRI have a similar trend that differs from conventional DUS velocities. This difference likely stems from the intrinsic limitations of the modalities and seems to be twofold: The relatively coarse spatial resolution of 4-D flow MRI and the kernel dimensions together with the spatial smoothing used in the PIV algorithm provide a more averaged, underestimated, velocity per grid point. On the other hand, a small sample volume and its effect on spectral broadening causes DUS to overestimate maximum velocities [7].
2-D velocity vector data allows for detailed evaluation of local blood-flow patterns, which offers an attractive benefit over conventional DUS. Similar patterns, including complex flow disturbances such as recirculation zones, were appreciated in all modalities. US-VFI has the advantage over 4-D flow MRI of, first, a higher temporal resolution (300 vs. 28 vector fields per second at 60 bpm) and, second, a higher spatial resolution (0.38 × 0.38 mm vs. 1.25 × 1.25 mm). This can potentially lead to more accurate quantification of small-scale and short-lived events. Moreover, US-VFI allows for the assessment of inter-beat variations, whereas 4-D flow MRI provides an interleaved cardiac cycle [27]. As such, US-VFI could provide novel insights into local flow disturbances that might be related to the onset and progression of atherosclerosis. The computational load of the current application and limited 2-D visualization of 3-D flow phenomena hamper the translation to clinical practice to date.
The present study has several limitations. First, this study was limited by its small sample size and only included healthy subjects. Second, inter-modal plane selection was undertaken with great care and finetuned with the automated registration. However, the latter did not result in lower limits of agreement between US-VFI and 4-D flow MRI-derived velocities compared with the work of others [26], [27], where only manual selection was performed. Manual segmentation of the ultrasound images and lack of distinctive geometrical features possibly contributed to discrepancies between modalities. Third, the absence of real-time feedback on the high-frame-rate US-VFI acquisitions prohibited the sonographer from assessing the image quality and correcting the anatomical position. Last, aside from transmission voltage, US-VFI acquisition and processing was kept identical between echoPIV and BST to enable the best comparison. A transmission frequency at the lower end of the transducers’ bandwidth was used to allow for contrast-specific imaging schemes. The non-contrast BST measurements could have been acquired with a higher transmission frequency, which would improve image resolution. Moreover, the current analysis was based on earlier in vivo echoPIV work, and thus was potentially better suited to these type of recordings. Fundamentally, however, the cross-correlation principles described in the literature are very similar [15], [17].
Future developments of US-VFI should focus on 3-D acquisitions and flow characterizations, and, in parallel, on reducing the computational load. Graphical processing units could be employed to overcome extensive reconstruction times and facilitate real-time feedback. Direct assessment of the acquisitions allows for adjustments in transducer placement and optimization of ultrasound settings, and the judgement if UCA administration is needed, as this will probably vary between individuals and imaging locations.
Conclusion
10V EchoPIV and BST showed overall good agreement in spatiotemporal velocity profiles compared with the reference 4-D flow MRI. The agreement in image quality, tracking performance and acquired velocity profiles between the US-VFI modalities suggest that the administration of UCA is not required to increase the echogenicity of the blood in these relatively superficial femoral arteries. In these cases, no intravenous cannula needs to be inserted, making it an attractive non-invasive alternative to echoPIV.
Conflict of interest
This work was supported in part by the Dutch Research Council VORTECS program (17219) and in part by Rijnstate Vriendenfonds (VF19-a10).
Acknowledgements
The authors thank Juanne Koucher, Jochem Noordzij, Frans Tak and Kristel Dulos for their assistance with the ultrasound examinations, and Marieke Stam, Wessel Berends and Marit Bles-Hartman for their assistance with the 4-D flow MRI scans. The authors thank Jason Voorneveld for sharing the PIV algorithm that was used to analyze the high-frame-rate ultrasound images.
Data availability
The data underlying the findings presented in this article are available from the authors on reasonable request.