Quantitative assessment of carotid ultrasound diameter measurements in the operating room: a comparable analysis of long-axis versus rotated and tilted orientation

Objective. Carotid ultrasound (US) has been studied as a non-invasive alternative for hemodynamic monitoring. A long-axis (LA) view is traditionally employed but is difficult to maintain and operator experience may impact the diameter estimates, making it unsuitable for monitoring. Preliminary results show that a new, i.e. rotated and tilted (RT) view is more robust to motion and less operator-dependent. This study aimed to quantitatively assess common carotid diameter estimates obtained in a clinical setting from an RT view and compare those to corresponding estimates obtained using other views. Approach. Carotid US measurements were performed in 30 adult cardiac-surgery patients (26 males, 4 females) with short-axis (SA), LA, and RT probe orientations, the first being used as a reference for measuring the true vessel diameter. Per 30 s acquisition, the median and spread in diameter values were computed, the latter representing a measure of robustness, and were statistically compared between views. Main results. The median (IQR) over all the patients of the median diameter per 30 s acquisition was 7.15 (1.15) mm for the SA view, 7.03 (1.51) mm for the LA view, and 6.99 (1.72) mm for the RT view. The median spread in diameter values was 0.18 mm for the SA view, 0.16 mm for the LA view, and 0.18 mm for the RT view. There were no statistically significant differences between views in the median diameter values (p = 0.088) or spread (p = 0.122). Significance. The RT view results in comparable and equally robust median carotid diameter values compared to the reference. These findings open the path for future studies investigating the use of the RT view in new applications, such as in wearable ultrasound devices.


Introduction
Carotid ultrasound (US) imaging is a commonly used diagnostic tool to assess stenosis and plaque (Aboyans et al 2018).Carotid US has also proven to be a useful tool for hemodynamic monitoring (Suriani et al 2022), e.g. by measuring blood flow (Gassner et al 2015), and has been investigated as a method for estimating blood pressure (Vermeersch et al 2008).Using carotid US for these applications would diminish the need for invasive methods, associated with catheter-related complications.Both for hemodynamic monitoring and blood pressure estimation, the arterial diameter, acquired through US B-mode imaging, is a measurement of interest.Blood flow is computed by multiplying the cross-sectional area p ( ) diameter 1 4 2 with the blood velocity.Blood pressure can be estimated using the fluctuations in the common carotid artery (CCA) diameter, e.g. by scaling peripheral blood pressure acquired from tonometry to the carotid diameter waveform (Vermeersch et al 2008, Zaccari et al 2008).The carotid artery is an attractive location as, owing to its superficial location, it is easy to find and measure by US.However, because this 3D structure is typically visualized in 2D, the probe orientation is a key factor when performing carotid US measurements.
Traditionally, carotid US images are acquired with the US probe oriented parallel to the vessel, also known as the long-axis (LA) view (figure 1(A)).In the calculation of blood flow, the assumption of a circular vessel is typically made, which is an acceptable approximation for larger arteries such as the CCA (Weber et al 2016).Based on this assumption, US measurements should be made along the mid-axis of the vessel.However, an important limitation of the LA view is that the probe may not be correctly aligned along the mid-axis of the artery, resulting in inaccurate (underestimated) diameter values (Stadler et al 1996, van Knippenberg et al 2020, de Boer et al 2023).Such misalignments can be caused by operator hand motion, patient movement, and the absence of visual feedback and make the LA view unsuitable for continuous monitoring purposes.To leverage the potential of carotid US imaging, it is paramount to accurately estimate the diameter.
As opposed to the LA view, the short-axis (SA) view is obtained by 90°rotation of the probe from the LA orientation (figure 1(B)).Although the assumed circular cross-section of the CCA implies that the SA view provides a realistic estimation of the diameter, this view cannot be used to visualize blood velocity.The reason is that as the angle between the blood flow direction and the US beam approaches 90°, the Doppler frequency shift approaches zero.However, a simultaneous assessment of diameter and blood velocity is needed for the calculation of blood flow.Using a matrix probe with xPlane mode, it is possible to simultaneously assess the LA and SA views.While this view improves the chance of proper placement of the probe mid-axially, the B-mode of xPlane duplex images currently yields only a single frame that is not updated over time, rendering the assessment of dynamic diameter changes impossible.Moreover, a matrix probe is larger and more expensive than a linear probe.
Another way of assessing the cross-section of the carotid artery is by rotating and tilting the probe, a view that is easier to visualize and assess for sonographers (figure 1(C)).Aiming at increasing operator independency, previous articles describe dedicated algorithms to analyze cross-sectional carotid US images (Kitabatake et al 1990, Schorer et al 2017, van Knippenberg et al 2020).By computing the diameter over the minor axis of the elliptical-shaped cross-section of the vessel, the most recent algorithm allows for diameter estimation under any angle of rotation and tilt, i.e. free-hand scanning (van Knippenberg et al 2020).These feasibility studies show that rotating and/or tilting the probe results in accurate blood flow measurements in vitro (Kitabatake et al 1990, Schorer et al 2017, Van Knippenberg et al 2020).As blood flow is computed based on diameter estimates, these results suggest that the diameter can in principle be accurately estimated from an elliptical-shaped cross-section.Furthermore, they conclude that for blood velocity estimations a free-hand scan, in which both rotation and tilting are allowed, is more robust to motion, and less operator-dependent compared to LA acquisitions (van Knippenberg et al 2020).The aforementioned studies were performed on simulated and phantom data, as well as on a very small cohort of healthy volunteers (1-5 volunteers), and thus warrant further validation in a realistic clinical population.In addition, none of these studies specifically investigate in vivo diameter measurements.
In this study, we evaluate the potential of carotid US to correctly estimate vessel diameter from images with a rotated and tilted (RT) probe orientation.We do so by comparing the diameter computed from LA and RT images to the diameter computed from an SA view.Due to the aforementioned limitations of the LA view, we hypothesize that with respect to the reference SA view, the LA-derived diameter values will be smaller and display a larger variance than those derived from an RT view.

Methods
For this prospective study, patients were included in the Catharina Hospital in Eindhoven between October 2021 and September 2022.Study approval was obtained by the Medical Research Ethics Committees United (NCT05193760).

Study population
Eligible adult patients presenting for elective cardiac surgery were approached and included upon signing an informed consent form.Exclusion criteria were known stenosis of the carotid artery, open wounds at the measurement area around the CCA, and arrhythmias.A sample size calculation showed that 30 patients were sufficient for this pilot study aiming to assess arterial diameter measurements with an RT probe orientation (appendix A).
After inclusion and during analysis, US data quality was evaluated and if image quality was compromised, a new inclusion was performed.The image quality of the acquired images was quantitatively assessed using the generalized contrast-to-noise ratio (gCNR) (Kempski et al 2020).The gCNR was computed using 256 bins on the first frame of each acquisition, as it was deemed to be a good representation of the acquisition's best image quality and probe location.When necessary, a more appropriate and representative frame was selected.Two rectangular regions of interest were defined in the vessel lumen and on the vessel wall, based on the pulsed-wave Doppler (PWD) sample volume and the detected vessel wall location, respectively.gCNR values vary between 0 and 1, with values close to 1 representing good image contrast.

Ultrasound imaging
US acquisitions were performed using a commercially available EPIQ CVx US machine (Philips Ultrasound LCC, Bothell, USA) with a linear array (3-12 MHz) probe (L12-3 broadband linear array transducer, Philips Ultrasound LCC, Bothell, USA).The images were acquired using duplex mode (B-mode and PWD modalities), with a B-mode frame rate varying between 19 and 39 Hz, with a mean ± SD value of 28 ± 7 Hz, dependent on the depth of the artery.During analysis, all diameter waveforms were resampled to the highest used sample frequency.Table 1 shows the US imaging settings.
In each patient, one set of measurements was obtained, consisting of 30 s duplex images with an SA, LA, and RT probe orientation.For the RT view, a 3D-printed biocompatible tool with 30°rotation and 20°tilting angles was used to standardize the angles at which the acquisitions were made, where the LA view was defined as 0°r otation (figure 2 and appendix B).Measurements were made approximately 2 cm below the carotid bifurcation.Tilting the probe will result in visualizing the CCA at a different location.We adjusted for this by shifting the probe over the skin until the same part of the vessel was visualized as before tilting was applied.

Study protocol
The optimal US settings were determined by one sonographer while patients were waiting to be transferred to the operating room (OR).At the OR, acquisitions were performed after induction of anesthesia by one of two experienced sonographers, who both had >5 years of experience in ultrasonography.Within one patient, the same sonographer performed all of the data acquisition.Figure 3 provides an overview of the workflow.

Carotid diameter estimation
For the goal of this study, the carotid diameter was defined as the distance between the adventitia layers.To derive numerical diameter values from the US DICOM data (figure 4, step 1), a diameter waveform was obtained by segmenting the CCA.The algorithm is used for diameter estimation in all views.Outlier detection and filtering were performed at multiple levels from a coarse to a finer mesh.Data analysis was performed in MATLAB version 2022b (Mathworks, MA, USA) using a Philips proprietary algorithm.
A step-by-step explanation of the algorithm used for diameter estimation is shown in figure 4. Firstly, the B-mode DICOM file was pre-processed, consisting of the remapping of the grey values and pseudo-bilinear filtering (figure 4, step 2).This step was performed with the intent of increasing the contrast on the B-mode image, improving the visibility of the adventitia layer, and assisting in the filling of small openings.For the LA and RT views, remapping the grey values was performed by saturating the top and bottom 20% of all pixel values.For the SA view, the top 20% and bottom 30% of pixel values were saturated.The threshold values used for grey-level remapping were empirically defined and adapted for each view based on a visual assessment.Secondly, per B-mode image, nine seed points were automatically selected based on the PWD sample volume.For the seed points, the center of the PWD sample volume was used including 8 additional points around it at a 1.25 mm distance from each other, selected in a squared fashion (figure 4, step 3).Subsequently, the B-mode image was binarized having as a threshold the grey level information in the blood region.For each seed point, the properties of the black and white ellipsoid structure, e.g.minor and major axes length, were characterized using the Matlab function regionprops (figure 4, step 4).For the subsequent frame, the previously determined ellipse was used to define the region of interest within which the algorithm searched for the vessel.The definition of the region of interest was optimized per view, e.g. the shape of the ROI was circular for the SA view, rectangular for the LA view, and elliptical for the RT view.In all probe orientations, the minor axis of the ellipse corresponds to the vessel's diameter.Hence, the grey level intensity values were appraised in the direction of the minor axis over a width of approximately 1 mm (figure 4, inset of step 4 and step 5) and used to detect the adventitia layers (figure 4, step 6, red crosses).The estimated diameter for each cross-sectional line was evaluated to be within a physiological range of 0.3-1.3cm (Krejza et al 2006), and excluded otherwise.Per frame, a diameter estimate was computed by averaging the diameter values obtained from each cross-sectional line along the 1 mm width (figure 4, inset of step 4).Subsequently, a moving median and Gaussian filter were applied to the temporal domain to assess frames if there was no diameter estimate.As such, a frame-by-frame diameter estimate was obtained per seed point (figure 4, step 7).
Thirdly, per acquisition one of the nine waveforms was selected for further analysis based on criteria established to identify the best diameter waveform.The median (μ 9 ) and SD (σ 9 ) over the nine waveforms and the median of each individual waveform were computed.In case the median of an individual waveform deviated more than two times σ 9 from μ 9 , that individual waveform was discarded.Since in some acquisitions structures other than the adventitia layer were detected, the diameter waveform with the smallest spread, defined as the 95%-5% difference, was selected and used for further analysis.In case the waveform selected based on these criteria resulted from a visually incorrect tracing of the adventitia layer, a waveform was manually selected.
Visual inspection of the obtained diameter waveforms and US acquisitions revealed that some timeframes contained large motion artifacts, where B-mode images displayed extensive motion of the US probe, leading to incorrect detection of the adventitia layer and incorrect diameter estimation.These frames were manually deleted from the waveforms.The diameter waveforms were resampled to the highest frame rate (39 Hz) and low-pass filtered with a cut-off frequency of 3 Hz, determined by spectral analysis, to remove high-frequency noise introduced by the frame-by-frame diameter estimation.Per waveform, outliers were removed by discarding values that deviated more than three times the SD to the mean of that waveform.From the denoised diameter waveform, the median diameter and interquartile range (IQR; 75th-25th percentiles) per 30 s acquisition were computed.

Statistical analysis
Statistical analyses were performed using IBM SPSS Statistics version 28.0 (IBM Corporation, IL, USA) and MATLAB version 2022b (Mathworks, MA, USA).Continuous variables were expressed as mean ± SD or as median and interquartile range (IQR), depending on normality.Normality was visually checked using quantilecomparison plots and numerically assessed using a Shapiro-Wilk test.
Bland-Altman plots were used to visualize the accuracy (using the bias) and precision (using the limits of agreement (LOA)) of the diameter estimates obtained from the RT and LA views, both versus the reference SA view.Clinically acceptable boundaries for the bias and LOA are dependent on the target population and should, therefore, be defined in advance (Montenij et al 2016).Based on previous literature, a bias of ±0.3 mm and LOA of ±0.7 mm were defined to be clinically acceptable (Wendelhag et al 1991, Persson et al 1992).Moreover, boxplots were used to visualize the agreement and precision of the median diameter estimates obtained from the three different views.The robustness of the views was assessed using a measurement of spread (the IQR).Friedman tests were used to test for significant differences in diameter estimates and spread, at a significance level of α = 0.05.

Study population
During the study period, 52 patients were included, of whom one withdrew informed consent.Of the patients suitable for measurements, 15 patients were excluded from the study for the following reasons: in 9 patients measurements were logistically not possible, e.g. because the patient had undergone an emergency surgery; in 4 patients there were anatomical restrictions, e.g. a low bifurcation or a short neck due to which the images could not be made 2 cm below the bifurcation; and in 2 patients the US machine experienced a technical error.Therefore, measurements were performed on 36 patients.When checking the images, a total of 6 sets of images were excluded: in 4 image sets, data quality was not sufficient, e.g.no proper probe positioning; and 2 sets of images were excluded due to anatomical reasons, e.g. the discovery of carotid plaque impeding a correct detection of the vessel wall by the used algorithm.As a result, data from 30 patients were available for analysis (figure 5).The demographics of the analyzed patients can be seen in table 2.
The median (IQR) gCNR was 0.98 (0.05) for the SA view, 0.96 (0.09) for the LA view, and 0.99 (0.02) for the RT view.Due to anatomical restrictions, the area of the ROI on the vessel wall was approximately half of that in the vessel lumen.

Carotid diameter estimation
In the analyzed images, timeframes containing large motion artifacts were manually deleted from the diameter waveform.Both for the SA and RT views, timeframes had to be deleted in 6 patients, with an average length of 4 s per waveform.
Table 3 presents the median (IQR) over all patients of the median diameter value per 30 s acquisition.The percentage difference to the reference was −1.7% for the LA and −2.2% for the RT view (figure 6).Differences in the median of the median diameter estimates per 30 s acquisition were not statistically significant (p = 0.088).The distribution of median diameter estimates per view was comparable between male and female patients.
Bland-Altman analyses revealed a bias of −0.25 mm for the LA and −0.27 mm for the RT views, with respect to the reference SA view (figure 7).This means that compared to the reference, both the LA and RT views tend to slightly underestimate the median diameter.Since we defined a bias of ±0.3 mm to be clinically acceptable, the accuracy of both the LA and RT views seem to be comparable to the reference SA view.The 95% confidence interval (CI) of the mean difference can be used to evaluate whether there is a significant difference between the measurements (Kaur and Stoltzfus 2017).The 95% CI of the mean difference with the reference was −0.52 to 0.014 mm for the LA view and −0.54 to 0.0063 mm for the RT view.As for both the LA and RT views the line representing zero falls within the 95% CI of the mean difference, there is no significant difference with the reference SA view.Compared to the SA reference, the LOA for the LA, ranging from −1.70 to 1.20 mm, and RT views, ranging from −1.76 to 1.23 mm, (figure 7) are wider than the a priori-defined clinically acceptable boundaries of ±0.7 mm, suggesting that the LA and RT views do not result in a clinically acceptable precision.
A detailed visualization of the diameter estimates can be seen in figure 8, where the median and spread are presented per patient and probe orientation.Although in some patients the three views result in comparable median diameter values, in others that is not the case, e.g. in patients 1 and 20.The spread is small in all patients and views except for one (patient 18, SA view).These findings are likely a result of a different probe location between views or drift during an acquisition.As a measure of robustness, we assessed the IQR values per 30 s acquisition (table 3), which were not significantly different between views (p = 0.122).

Discussion
In this study, we investigated how diameter estimates are impacted by different US probe orientations.Our results suggest that the estimated median diameter of the CCA per 30 s acquisition is comparable for the SA, LA, and RT views.Furthermore, to evaluate the robustness of each view, we investigated the spread in diameter  estimates.We have found that the diameter estimates were equally robust for all three views.Together, these results suggest that diameter estimates derived from cardiac surgery patients imaged in the OR seem to be interchangeable between the views, with all three views leading to equally robust measurements.
The diameter values analyzed in this study were obtained using a proprietary algorithm that segments the vessel, resulting in a frame-by-frame diameter estimation.The median CCA diameter values derived in our study are consistent with previously published literature.Krejza et al (2006) reported a luminal CCA diameter of 6.52 ± 0.98 mm in males undergoing carotid endarterectomy with a plaque-free CCA, and Veller et al (1993)  described a mean thickness of the intima-media complex of 0.63 ± 0.16 mm in healthy volunteers (51% female).From these values, it is possible to derive an inter-adventitia distance of 7.78 ± 1.0 mm, in line with the values obtained in this study (median of 6.99 to 7.15 mm depending on the view).
The prevalence of females included in this study (14%) is somewhat lower than what would be expected from the literature (23%-41%) (Koch et al 2003, Health Council of the Netherlands 2007).As one of these studies was performed in a different continent than ours it is possible that different demographics could explain the discrepancy in values.Literature has shown sex-based differences in carotid diameter, with slightly larger values reported in men (Sass et al 1998).We do not expect this to impact the validity of our study results.Firstly, because the distribution of median diameter estimates per view was comparable between males and females, and most importantly, due to our study design which consisted of an intra-patient comparison of the three US views, meaning that the results should in principle be robust to inter-patient variability.
To our knowledge, this is the first study investigating and comparing different probe orientations for the estimation of vessel diameter in a clinical setting.The traditionally used LA view can lead to inaccurate diameter values caused by an incorrect alignment of the US probe along the mid-axis of the vessel (Stadler et al 1996, van Knippenberg et al 2020), a problem that might be solved by using a different probe orientation.Diameter estimations cannot only be used for hemodynamic monitoring but also for blood pressure estimation, and in both cases, a reliable diameter value is paramount (Vermeersch et al 2008, Zaccari et al 2008).The SA view represents the vessel diameter with the highest fidelity but it is impossible to estimate velocity with this specific view.Hence, the RT view emerges as a potential alternative to the LA and SA views, in the context of e.g.hemodynamic monitoring and blood pressure estimation.
We hypothesized that with respect to the reference SA view, the RT view would be evenly robust and accurate, while the LA view would be less robust and underestimate the vessel diameter.While this was true for the RT view, the LA view did not result in significantly less robust and accurate diameter estimates.Several factors could explain this discrepancy.First, prior sonographer training might have substantially impacted our measurements.The sonographers received traditional clinical US training, which focuses on imaging in the LA and SA views, possibly leading to a bias towards these views.Second, the tool used to visualize the RT view might have increased variability in these measurements.While free-hand scanning is the ultimate goal, a tool with known rotation and tilt angles was used to standardize the RT view between patients.However, usage of this tool might have introduced an additional layer of complexity for the sonographer to acquire and maintain the RT view, considering that both hands were needed, resulting in a less ergonomic posture.Third, the length and order of measurements might have played a role in the derived diameter estimates.A duration of 30 s for US image acquisition is quite lengthy, as it is difficult to keep the probe steady for such a long period while ample events occur in clinical practice.Since the sonographers had to obtain at least 3 images with a length of 30 s each, the sonographer might have been less focused when performing the third image.Due to the standardization of the image acquisition protocol, the third acquired image was always the one with an RT probe orientation.Together, all these factors might have contributed to LA diameter estimates that are more robust and comparable to the reference-derived estimates than initially hypothesized.
A strength and simultaneous limitation of the current study is the setting in which the measurements were performed.On one hand, it is important to be time efficient at the OR; therefore, the US measurements had to be performed quickly.Moreover, the sonographer needed to stand at the head of the patient while measuring, which is not very ergonomic when using the tool for RT measurements and might have introduced undesired variability.Furthermore, measuring the three views at the same time was not possible and, therefore, the measurements were performed subsequently.Despite marking on the skin being used to mark the measurement location, the probe location might have differed between views.Towards the bifurcation, the CCA becomes the carotid bulb, which has a larger diameter than the CCA.As a result, a different probe location might result in a slightly different diameter estimate.While the acquisitions included in this study had a high gCNR and, thus, a high image quality, differences in contrast between the lumen and vessel wall will certainly still play a role in the quality of the diameter estimation.Factors such as probe movement, incorrect probe positioning as well as image quality will be the determining factors in deriving accurate and precise diameter estimates.
On the other hand, the OR setting is also a strength of this study as it represents a real-world clinical scenario.Moreover, although the measurements were performed subsequently, no changes in hemodynamics and, hence, CCA diameter, are expected within the measurement period due to the controlled environment at the OR.
This paper introduces a semi-automatic algorithm that can be used for the detection of the vessel wall.The literature describes that manual delineation of the vessel wall, e.g.luminal diameter, is subject to errors and has a large variability within and between experts (Wendelhag et al 1991, Loizou 2014).Automated algorithms may aid in filling this gap and decreasing the variability between experts (Loizou 2014).By using a semi-automatic algorithm, we tried to minimize variability caused by human input.This study focused on the inter-adventitia distance since the intima layer was not visible in all probe orientations while the adventitia layer was.This mitigated any error due to vessel layer tracking, which would bias the comparison between views.However, it should be noted that the luminal diameter is needed in case one is interested in blood flow.Considering that detecting the adventitia layer is likely a less complex task than identifying the intima layer, our results indicate what are possibly the upper boundaries of accuracy for the identification of the luminal diameter.Additional research should identify how many degrees a probe can be tilted while still visualizing the intima layer.
The implications of our results with the RT view go beyond the usage of the LA and SA views in current US clinical practice.Due to their non-invasive nature, wearable US devices are attractive for continuous hemodynamic monitoring (Wang et al 2018, Kenny et al 2021), e.g. to alert physicians of an imminent cardiovascular insufficiency (Pinsky and Payen 2005).As a general remark, it should be noted that carotid US cannot be performed in patients with a short neck or low carotid bifurcation, as measurements should be performed 2 cm below the bifurcation (Suriani et al 2022).In the last decades, a shift has taken place from static to dynamic hemodynamic monitoring, and towards functional hemodynamic monitoring, in which the cardiovascular system is challenged to assess preload responsiveness (Pinsky and Payen 2005).Both diameter and blood velocity are needed to compute CCA blood flow.While only the diameter can be obtained from the SA view, the LA and RT views allow for the measurement of both diameter and velocity.An advantage of the RT over the LA view is that the former allows for patient movement.In the LA view, patient movement, e.g.rotation of the head, will result in the CCA moving out of the US field of view, leading to either incorrect visualization of the CCA mid-axial plane, or no longer visualizing the CCA at all.In the RT view, patient movement will only result in displacement of the CCA within the US field of view.Hence, it would be of great interest to further investigate the RT view with the ultimate goal of implementation in wearable US devices.

Conclusion
The findings presented in this paper suggest that the newly introduced RT probe orientation can be used for an accurate estimation of CCA vessel diameter.This paves the path for future studies on the usage of the RT view in wearable US devices.Ultimately, this might lead to the development of a continuous, non-invasive US method for hemodynamic monitoring.

Figure 1 .
Figure 1.Schematic representation of the used probe orientations: (A) the long-axis (LA), (B) short-axis (SA), and (C) rotated and tilted (RT) views.It displays the common carotid artery (CCA), represented for simplicity as a straight cylinder (pink), the ultrasound imaging plane (brown), and the vessel cross-section (blue).

Figure 2 .
Figure 2. 3D-printed biocompatible tool.By resting the probe on the brackets of the tool, 30°rotation and 20°tilting angles were obtained.

Figure 3 .
Figure 3. Flowchart of the workflow and the performed acquisitions.

Figure 4 .
Figure 4. Step-by-step explanation of the algorithm used for diameter estimation.Note that while rotated and tilted images are used as examples in this figure, the described algorithm is used on all probe orientations.(1): A B-mode DICOM file was given as input to the algorithm.(2): Pre-processing consisted of remapping of the grey values and pseudo-bilinear filtering.(3): Automatic seed point selection based on the center of PWD sample volume.(4): Per seed point, the properties of the binarized vessel were characterized using an elliptical shape (red ellipse).The inset shows the 1 mm width over which the intensity levels were appraised.(5): Grey level intensity profile of one of the cross-sectional lines used for wall detection.(6): Detection of the adventitia layer.(7): Per seed point a diameter waveform was computed.The colors of the diameter waveforms correspond to the colors of the seed points as shown in step 3.

Figure 6 .
Figure 6.Boxplot of the median diameter estimate per 30 s acquisition over all patients, including the percentage difference with respect to the reference view.

Figure 7 .
Figure 7. Bland-Altman plots of (A) the traditional long-axis (LA) view versus the reference short-axis (SA) view; (B) the rotated and tilted (RT) view versus the reference SA view.The blue lines represent the mean difference (solid line) and 95% confidence interval of the mean (dashed lines).The red lines represent the limits of agreement.

Figure 8 .
Figure 8. Median and spread of diameter values per 30 s acquisition per patient per view.The blue square represents the long-axis view, the red circle represents the short-axis view, and the green diamond represents the rotated and tilted view.

Table 1 .
Ultrasound (US) settings used to obtain the images.PWD: pulsed-wave Doppler; CCA: common carotid artery.

Table 2 .
Baseline characteristics of the study population.CABG: coronary artery bypass grafting.

Table 3 .
Median (IQR) over all patients of the median and spread in diameter values per 30 s acquisition.IQR: interquartile range.