Data from 15 prospectively included neonates and 71 adolescents who underwent cardiovascular magnetic resonance imaging at Skåne University Hospital between 2014 and 2020 were used. An additional adolescent was included for computer phantom construction and was not further incorporated in analyses. The regional Ethics Review Board (Lund, Sweden) approved the study. All participants, and their guardians when appropriate, provided written informed consent before participation in the study.

The pulse wave velocity computer phantom methodology from Dorniak et al. [6] was used to establish the required temporal resolution needed for accurate PWV measurements. Phantom vessel lengths were set to approximately the minimum vessel lengths of the current populations, i.e., 25 mm for neonates’ aortic arch, 60 mm for neonates’ thoracic aorta and adolescents’ aortic arch, and 150 mm for adolescents’ thoracic aorta. Post-acquisition analysis of the study population revealed sufficient timeframes per cardiac cycle for construction of a high-resolution computer phantom for neonates, but not for adolescents. This was due to a limitation in reconstructed number of timeframes. As a result, one additional adolescent (17-year-old female, 162 cm, 55 kg) was included and used for the phantom construction only. Computer phantoms with a reference pulse wave velocity between 2 and 10 m/s were generated to cover the normal pulse wave velocity interval for neonates and adolescents [11], while also including large margins for potential confounders. A representative high-resolution ascending flow curve from each group was upsampled to 10,000 timeframes per cardiac cycle. A simulated corresponding descending flow curve was created by shifting the ascending flow curve in time with a delay computed as the vessel length (aortic arch or thoracic aorta) divided by the reference PWV (i.e., 2 m/s, 4 m/s, 6 m/s, 8 m/s, 10 m/s). Thereafter the descending flow curve was scaled in amplitude with 60% to mimic the blood diversion to the aortic arch branches and finally smoothed with a Gaussian smoothing filter with a width of 2,000 timeframes. The flow profile was then downsampled to 20–60 timeframes per cardiac cycle to mimic image acquisition with different temporal resolutions. For each combination of reference PWV and timeframes per cardiac cycle, the PWV was calculated using the TTF method on the downsampled ascending and descending flow curves and compared to the reference PWV. Differences were presented as % error. The full code for the phantom can be found as part of the open source code distribution for the software Segment.

Participants were imaged in supine position using a 1.5 T MR scanner (Philips Achieva, Best, the Netherlands, or Siemens Aera, Erlangen, Germany). For adolescents a 32-channel phased-array cardiac coil (Philips) or 18-channel and spine coil combination (Siemens) was used according to clinical routine. For neonates a small flexible coil was chosen. Further, neonates were positioned in a vacuum infant immobilizer and either imaged using feed-and-sleep or if this was not possible, using chloral hydrate 25 mg/kg administered rectally. A 2D phase-contrast gradient recalled echo sequence with retrospective ECG gating was used for quantitative flow measurements. For both neonates and adolescents, two phase-contrast flow planes were acquired; one positioned in the ascending aorta as part of the clinical standard protocol which also includes the descending aorta immediately after the aortic arch, albeit potentially angulated to the vessel, and one at the level of the diaphragm. For aortic length measurements in adolescents, 3D oblique sagittal slices covering the thoracic aorta (non-contrast enhanced 3D angiography) and coronal slices covering the entire thorax (coronal overview) were acquired. A subset of adolescents (n = 49) had both 3D non-contrast-enhanced angiography and thoracic coronal overview images acquired for comparison of aortic length measurements. The coronal overview images were based on a balanced SSFP sequence with inplane resolution 1.66 mm and slice thickness 8 mm and slice gap 2.64 mm, whereas the 3D angiography was based on a clinical routine T2-prepared balanced steady-state free precession sequence with isotropic resolution 0.88 mm (Philips) or 0.55 mm (Siemens). In neonates a clinical routine 3D black-blood T1-weighed non-contrast-enhanced angiography sequence was used with isotropic resolution 1.04 mm (Siemens). Typical image parameters for the quantitative flow phase-contrast sequence were TR/TE 4.92/2.67 ms, flip angle 20°, 1.5 × 1.5 × 5 mm³ (Siemens) for neonates. Corresponding parameters for adolescents were TR/TE 9.15/5.54 ms, flip angle 15°,1.2 × 1.2 × 6 mm³ (Philips); and TR/TE 4.92/2.67 ms, flip angle 20°, 1.5 × 1.5 × 5 mm³ (Siemens). Reconstruction was set to a maximum of 50 timeframes per cardiac cycle for neonates and 35 for adolescents. In the one adolescent scanned for construction of the computer phantom, flow data were collected using the Siemens scan protocol but with the reconstruction limit set at 70 timeframes per cardiac cycle to avoid assessment errors. Velocity encoding was 150–250 cm/s to optimize velocity resolution and avoid aliasing. Temporal resolution (ms) was calculated as heartrate duration divided by timeframes per cardiac cycle.

Image analysis was performed using the software Segment version 3.2 R9405 (http://​segment.​heiberg.​se, Medviso AB, Lund, Sweden) [12]. Three observers assessed velocity and volumetric flow curves in the ascending aorta, descending aorta after the aortic arch and at the level of the diaphragm, and length measurements in the aortic arch and thoracic aorta. The observers had 2 years’ (observer 1, assessed coronal overview data; SL), 5 years’ (observer 2, assessed 3D angiography data and flow data; JL), and 21 years’ (confirmed data from observer 1 and observer 2; EHed) CMR experience. A subset of 15 data sets (5 neonates and 10 adolescents) were randomly chosen and delineated by observers 1 and 2 for interobserver assessment. Aortic arch and thoracic aorta lengths by 3D angiography were used for length comparison between observers and the PWV method with least bias and variability was used for transit time comparisons.

Vessel length (Δd), i.e., blood traveling distance between flow planes, was measured in the reference standard 3D angiography images and in 2D coronal overview images. For both image types a manual frame-by-frame centerline vessel tracking method was used (Fig. 1). Lengths reported are from the reference standard 3D angiography unless otherwise specified.

Velocity and flow curves were assessed using automatic segmentation with manual correction where needed [12] with vessel circumference delineated throughout the cardiac cycle in magnitude images supported by phase-contrast images as appropriate and analyzed according to Dorniak et al. [6] for PWV (Fig. 2).

Three transit time (Δt) methods were compared as shown in Fig. 3; namely (1) time-to-foot (TTF), (2) maximum upslope, and (3) time-to-peak (TTP). The algorithms were based on the baseline intersection of an automatically generated tangent superimposed at the point of maximum upslope (TTF); the maximum upslope incline (maximum upslope), and the maximum curve magnitude (TTP). For TTF, which is baseline dependent, baseline correction was applied both automatically and manually. For automatic baseline correction, a modified version of the algorithm previously described by Dorniak et al. [6] was applied, using instead the mean of the 80th–95th % segment of the flow curves (Fig. 4). For manual correction, this study instead proposed to use the segment prior to the systolic upslope, to avoid errors related to the diastolic phase of the cardiac cycle such as backward flow related to insufficiency. Gaussian smoothing was performed on all curves to optimize signal-to-noise ratio with sigma = 0.025. Pulse wave velocity was calculated as PWV = traveling distance (Δd) / transit time (Δt).

Statistical analyses were performed in GraphPad 9.2.0 for Windows (GraphPad Software, La Jolla, California, USA). Group characteristics and vessel lengths were not normally distributed and presented as median (range), whereas PWV comparisons were normally distributed and therefore presented as mean ± SD. Bland–Altman plots assessed agreement between methods with data presented as bias ± SD in text and bias and 95% limits of agreement (LoA; i.e., ± 1.96 SD) in graphs. Intra-individual variations between coronal overview images and the reference standard 3D angiography are presented as median and 2.5th–97.5th percentile. Interobserver variabilities are presented as mean ± SD. Paired t-test or repeated-measures ANOVA with Tukey correction post hoc assessed group differences. P values < 0.05 were considered to indicate statistically significant differences.

Figure 5 shows the computer phantom results as timeframes per cardiac cycle, independent of heart rate. For a neonatal flow curve profile and an aortic length > 25 mm (i.e., shortest neonatal aortic arch), 42 timeframes per cardiac cycle resulted in a PWV error range of − 8–2%, and for an aortic length > 60 mm (i.e., shortest neonatal thoracic aorta) 41 timeframes per cardiac cycle resulted in a PWV error range of − 6–1%. Corresponding number of timeframes for an adolescent flow profile and aortic length > 60 mm (i.e., shortest adolescent aortic arch) and for an aortic length > 150 mm (i.e., shortest adolescent thoracic aorta) were 39 timeframes per cardiac cycle (PWV error range − 8–3%) and 32 timeframes per cardiac cycle (PWV error range − 4–6%), respectively. Fewer timeframes per cardiac cycle and particularly so combined with higher PWV and shorter pulse wave traveling distance yielded larger errors, with neonatal aortic arch showing the largest PWV error of approximately 30% at 33 timeframes per cardiac cycle and a PWV of 8 m/s.

Table 1 shows group characteristics for neonates and adolescents. Reconstructed timeframes per cardiac cycle was 44 in neonates, corresponding to a temporal resolution of 10 ms for an average heart rate of 138 bpm, and 35 in adolescents corresponding to a temporal resolution of 22 ms for an average heart rate of 75 bpm.

Figure 6 shows vessel length measurements and agreement. In neonates the median aortic centerline length by 3D angiography was 30 (23–59) mm for the aortic arch and 68 (55–83) mm for the thoracic aorta. In adolescents, the median aortic arch length was 86 (62–124) mm by the reference standard 3D angiography and 88 (64–113) mm by overview images (p = 0.09), and median thoracic aortic length was 188 (149–258) mm by the reference standard and 195 (153–245) mm by overview images (p < 0.0001). Variations between the reference standard 3D angiography and coronal overview images were − 16–18 mm for aortic arch length and − 25–30 mm for thoracic aortic length. Interobserver variability for 3D angiography length measurements for neonates and adolescents combined were – 2 ± 2 mm.

Figure 7A shows agreement between baseline correction methods for thoracic aorta PWV in neonates. Pulse wave velocity bias between automatic and manual baseline correction was – 0.2 ± 0.2 m/s for velocity curves and − 0.2 ± 0.3 m/s for flow curves. Corresponding bias was 0.5 ± 0.7 m/s and 0.5 ± 0.6 m/s for manual and no baseline correction and 0.6 ± 0.7 m/s and 0.7 ± 0.5 m/s for automatic and no baseline correction.

Figure 7B shows agreement between baseline correction methods for thoracic aorta PWV in adolescents. Pulse wave velocity bias between automatic and manual baseline correction was 0.0 ± 0.2 m/s for velocity curves and 0.0 ± 0.1 m/s for flow curves. Corresponding bias was 0.5 ± 0.3 m/s and 0.4 ± 0.2 m/s for manual and no baseline correction and 0.5 ± 0.3 m/s and 0.4 ± 0.2 m/s for automatic and no baseline correction.

Table 2 shows thoracic aorta PWV differences related to baseline correction methods. In neonates, velocity-based PWV by TTF with no baseline correction was higher (4.8 ± 1.6 m/s) than with automatic baseline correction (4.2 ± 1.0 m/s) (p = 0.01). In adolescents, PWV was higher using both velocity and flow curves for TTF with no baseline correction (4.4 ± 0.6 m/s and 4.2 ± 0.5 m/s) compared with automatic (3.9 ± 0.5 m/s and 3.8 ± 0.5 m/s) and manual baseline correction (3.9 ± 0.5 m/s and 3.8 ± 0.4 m/s; all p < 0.0001). Further, velocity-based PWV by TTF with no baseline correction was higher (4.4 ± 0.6 m/s) than the corresponding flow-based PWV (4.2 ± 0.5 m/s) (p < 0.001).

Figure 8 shows transit time method comparisons. In neonates, thoracic aorta PWV based on velocity curves were 4.2 ± 1.0 m/s for TTF, 6.1 ± 2.4 m/s for maximum upslope, and 1.8 ± 5.4 m/s for TTP. Corresponding PWV based on flow curves were 4.1 ± 0.9 m/s, 6.4 ± 3.5 m/s, and 5.3 ± 5.6 m/s.

In adolescents, thoracic aorta PWV based on velocity curves were 3.9 ± 0.5 m/s for TTF, 3.8 ± 0.6 m/s for maximum upslope, and 2.6 ± 0.8 m/s for TTP. Corresponding PWV based on flow curves were 3.8 ± 0.5 m/s, 3.7 ± 0.6 m/s, and 2.7 ± 1.0 m/s. Time-to-peak thus yielded lower PWV than maximum upslope (p < 0.0001) and TTF (p < 0.0001) for both velocity and flow curves in adolescents.

Figure 9 shows agreement between thoracic aorta velocity and flow curves for each transit time method. Time-to-foot yielded PWV 0.0 ± 0.4 m/s for neonates and − 0.1 ± 0.2 m/s for adolescents. Corresponding values were for maximum upslope 0.3 ± 1.8 m/s and − 0.1 ± 0.4 m/s, and for TTP 3.5 ± 8.2 m/s and 0.1 ± 0.7 m/s.

Figure 10 shows neonatal and adolescent PWV based on TTF with automatic baseline correction. Neonatal PWV based on centerline distance by 3D angiography was 2.8 ± 1.0 m/s for the aortic arch and 4.1 ± 0.9 m/s for the thoracic aorta. Adolescents’ aortic arch PWV was 4.3 ± 1.1 m/s based on centerline distance by 3D angiography and 4.4 ± 1.2 m/s when centerline distance was based on coronal overview images (p = 0.08). Corresponding values for adolescents’ thoracic aorta PWV were 3.8 ± 0.4 m/s and 3.9 ± 0.5 m/s (p < 0.0001). Variation in adolescents’ PWV derived from coronal overview and 3D angiography length measurements was more prominent (− 0.8–1.1 m/s) for the aortic arch than for the thoracic aorta (− 0.3–0.6 m/s). Interobserver variability for flow PWV by TTF with automatic baseline correction was 0.0 ± 0.1 m/s, corresponding to an error of − 1 ± 2%.

This study showed that both shorter aortic lengths as related to age [13] or flow plane acquisition location, and increased PWV as related to age, grade of hypertension, or other cardiovascular diseases [14] risks yielding larger errors at inadequate temporal resolution in both neonates and adolescents. All these factors are thus important to consider in order to ensure sufficient temporal resolution in the individual patient. Pulse wave velocity of the aortic arch can be derived from a single flow plane covering both the ascending aorta and descending aorta immediately after the aortic arch, but requires increased temporal resolution to compensate for the short distance between flow planes. The applicability of low and high temporal resolution acquisitions have been investigated [15, 16] but only Dorniak et al. [6] has of yet dissected the impact of a continuum of timeframes per cardiac cycle and temporal resolutions, despite being addressed as an important aspect already in 2014 [1]. This study presented temporal resolution as the heartrate independent measurement timeframes per cardiac cycle. For comparison between studies and for standardization of clinical application an increased transparency regarding acquired timeframes per cardiac cycle is beneficial.

The current study tested the hypothesis that standard coronal overview images could be used interchangeably with the reference method non-contrast enhanced 3D angiography for measuring accurate blood traveling distance and to calculate PWV in the aortic arch and thoracic aorta. On a group level, a small and likely clinically insignificant difference between 3D angiography and coronal overview was observed. However, when comparing the intra-individual length and corresponding PWV variability for 3D angiography and coronal overview, the variation is both large and unpredictable (Figs. 6 and 10). Differences in slice thickness, slice gap, and number of slices covering the aorta likely explain differences in measured aortic lengths and correspondingly PWV. Therefore, as non-contrast-enhanced 3D angiography can be acquired in only a few minutes it is preferred for accurate PWV calculations also in the young and healthy. The same logic can be applied for the aortic arch where, due to its shorter length and its curvature, accurate centerline length measurements are even more critical. This was also shown in the current study with larger PWV variation in the aortic arch. Retrospective analysis of clinical data sets with only coronal overview images and a single flow acquisition including both the ascending and descending aorta is therefore not recommended, but if used, should only be applied on a group level and with caution.

The current study showed no difference in PWV based on velocity or flow curves. Flow curves are commonly used for PWV by CMR but are affected by diversion of blood. For PWV measured in the aortic arch or thoracic aorta, blood mainly diverge into aortic arch vessels leading to a decline in blood volume of approximately 40% between the ascending and descending aortic flow measurement planes. This can be noted as the difference between the descending and ascending flow curves as shown in Fig. 2. Velocity curves are on the other hand routinely acquired by Doppler ultrasound for applanation tonometry-based PWV. They are less affected by diversion of blood but are instead susceptible to local variations in velocity [17]. Velocity curves are rarely used for PWV by CMR but may be advantageous in more challenging populations due to the more homogeneous curve shapes for transit time estimations. This however remains to be tested.

For baseline correction, automatic correction based on the 80th–95th % segment of the flow curve, i.e., late diastole, agreed with manual correction based on the more variable pre-systolic segment in both neonates and adolescents and for both velocity and flow curves, implying an adequate automatic algorithm. The previously proposed automatic method used the mean of the 62.5th–87.5th % segment of the flow curve [6], i.e., early diastole. Using the diastolic phase for correction may yield false results if extended to populations including patients with diastolic retrograde flow. By instead using a timepoint closer to the systolic upslope, as proposed in the current study, the influence of retrograde diastolic flow may be reduced. This hypothesis was not tested in the current healthy population, and a separate study including patients with retrograde diastolic flow is needed to confirm this hypothesis.

The commonly used transit time methods TTF, maximum upslope, and TTP [1, 6‐10] were compared in the current study, showing that TTF with automatic baseline correction had the narrowest limits of agreement in both neonates and adolescents. This agrees with previous studies showing TTF to be a reproducible method also in other populations [1]. Importantly, neonatal PWV did not differ between methods, whereas in adolescents PWV by TTP was lower than when using TTF and maximum upslope. Time-to-peak not only underestimated PWV, but also risks reporting false negative flow as shown in Fig. 2. This also explains the considerably lower, albeit not statistically significant, PWV value for velocity-based TTP in neonates (1.8 ± 5.4 m/s) as compared to TTF (4.2 ± 1.0 m/s) and maximum upslope (6.1 ± 2.4 m/s). In addition, the current study assessed each method’s robustness, suggesting the use of TTF with automatic baseline correction as the method of choice in neonates and adolescents as it is more robust and less variable than the other tested methods. This will also increase comparability between hospitals and research studies, and use of reference values.

Reference values are available for aortic arch PWV by CMR in adolescents [9, 18]. In both these studies, a sufficient number of timeframes per cardiac cycle was used for flow assessment in the aortic arch, assuming their populations are comparable to the current population in terms of heart rate and aortic arch centerline distance. However, these previous studies used maximum upslope [9] and time-to-half peak [18] as transit time methods, which is important to consider when applying these reference values in relation to the method used locally, as it may lead to wrongly interpreted results.

Whether this suggested standardized method also is accurate compared to invasive measurements remains to be answered. Compared to the current methodological situation, however, the current study presents a reproducible and precise transit time method which is a step forward in increasing comparability and applicability both in clinics and research.

The current study did by design not include elderly and patients with cardiovascular disease and the results shown may not be directly transferable to these populations. It may be hypothesized that the differences shown between methods in the current young population are even larger in elderly and in patients with cardiovascular disease. Further, the current study had a reconstruction limit of 35 timeframes per cardiac cycle for flow data acquired in adolescents, which is slightly fewer than that shown to be needed in the adolescent aortic arch, but not thoracic aorta, by the computer phantom. Therefore, the presented adolescent aortic arch PWV values should be taken with caution. However, the method comparisons should not be affected as aortic arch PWV measurements was used only for comparison between use of 3D angiography versus coronal overview images for centerline distance.