4D flow MRI · acquisition and quantitative biomarkers
Can a faster 4D flow MRI scan change the pulse wave velocity (PWV) we report? Wang and colleagues investigated whether compressed sensing (CS) acceleration alters aortic PWV estimates. Their study highlights an important distinction: apparent PWV differences can reflect imaging and analysis choices rather than physiological differences alone.
This is an independent study by Wang et al., not a study from our research group. Below, the findings are also connected to our group’s separate work on turbulent kinetic energy (TKE), which examines a different flow-derived quantity.
One aorta, multiple PWV estimates
The researchers acquired thoracic-aortic 4D flow MRI at 3 T using conventional GRAPPA (R = 2) and five CS acceleration factors: R = 4.4, 7.7, 11.5, 15.3, and 20.5. They evaluated five transit-time algorithms: time-to-foot, time-to-median, cross-correlation, fitting plane, and maximum likelihood estimation. They also assessed scan-rescan repeatability.
Although reconstructed temporal resolutions were broadly similar (approximately 27–35 ms), acquisition times varied markedly. Faster acquisition can facilitate clinical use, but waveform characteristics may change enough to affect derived measurements.
What happened to PWV as acceleration increased?
Across the evaluated algorithms, CS-derived PWV estimates were higher than GRAPPA-derived estimates. Importantly, the magnitude of the difference depended on the algorithm used to estimate pulse transit time.
PWV across acceleration levels
| PWV algorithm | GRAPPA R = 2 | CS R = 7.7 | CS R = 20.5 |
|---|---|---|---|
| Time-to-foot | 4.40 ± 0.64 | 5.87 ± 1.12 | 6.62 ± 2.17 |
| Time-to-median | 4.53 ± 0.42 | 5.46 ± 0.74 | 5.60 ± 0.90 |
| Cross-correlation | 4.91 ± 0.63 | 5.43 ± 0.86 | 5.57 ± 0.90 |
| Fitting plane | 4.49 ± 0.42 | 5.57 ± 0.89 | 5.77 ± 1.01 |
| Maximum likelihood | 4.92 ± 0.67 | 5.40 ± 0.85 | 5.57 ± 0.91 |
Table: Mean ± SD PWV (m/s) for three selected acceleration settings. Consult the original publication for all methods, acceleration settings, and participant-level comparisons.
Why does the timing algorithm matter?
Aortic PWV is commonly estimated as propagation distance divided by pulse-wave transit time. The distance can remain unchanged while the estimated time delay changes. Time-to-foot identifies a point on the leading edge of each waveform, whereas cross-correlation makes use of the waveform shape over a wider interval.
The investigators found CS-to-GRAPPA relative biases of +29.3% to +50.5% with time-to-foot, versus +8.6% to +13.4% with cross-correlation. Cross-correlation also had the highest scan-rescan reliability within the tested conditions (ICC up to 0.94). This supports consistency under this particular protocol, not universal superiority across scanners or clinical populations.
A related study from our group: turbulent kinetic energy
Our group’s separate study compared GRAPPA R = 2 and CS R = 7.7 for turbulent kinetic energy (TKE). It included 13 healthy participants and 22 patients with moderate aortic stenosis. Unlike PWV, MRI-derived TKE characterizes unresolved intravoxel velocity fluctuations. There is no reason to assume identical sensitivity to reconstruction, noise, or analysis conditions.
GRAPPA vs CS: TKEmax
Healthy volunteers
Whole-aorta TKEmax · n = 13 · mean ± SD (mJ)P = 0.006Higher with CS; statistically significant
Moderate aortic stenosis
Ascending-aorta TKEmax · n = 22 · mean ± SD (mJ)P = 0.129No statistically significant difference; not an equivalence test
Healthy whole-aorta maximum TKE was higher with CS (1.13 ± 0.73 versus 2.13 ± 0.95 mJ; P = 0.006). Among moderate-AS participants, ascending-aorta TKEmax did not differ significantly (14.0 ± 4.6 versus 14.8 ± 5.5 mJ; P = 0.129). Other analyses were not identical: for example, whole-aorta TKEmax in AS was 16.3 ± 5.5 versus 17.6 ± 6.5 mJ (P = 0.050).
These studies do not directly compare PWV against TKE, nor do they use equivalent cohorts or protocols. Instead, they illustrate the broader methodological lesson: accelerated 4D flow MRI needs parameter-, region-, and population-specific validation.
Implications for quantitative 4D flow MRI
Differences in reconstructed PWV are not necessarily changes in arterial stiffness. To compare published PWV results meaningfully, it helps to report the acceleration and reconstruction method, temporal resolution, vessel path and distance calculation, waveform preprocessing or interpolation, and the algorithm used to define transit time.
The independent Wang study involved young healthy adults and used GRAPPA as the acquisition comparator rather than an external invasive reference standard. Its bias estimates should not be interpreted as correction factors for all populations. The practical lesson is that faster 4D flow MRI may improve feasibility, but every derived biomarker should be validated under the exact imaging and analysis settings used.
Continue reading
References
- Wang B, Sun A, Li H, et al. Highly Accelerated Aortic 4D Flow MRI: Implications for Pulse Wave Velocity Measurements. Journal of Magnetic Resonance Imaging. 2026;63:549–560. doi:10.1002/jmri.70153.
- Park S, Yi J-E, Jin N, et al. Comparison of compressed sensing and conventional 4D flow MRI for turbulent kinetic energy assessment in healthy participants and patients with aortic stenosis. Scientific Reports. 2026;16:26434. doi:10.1038/s41598-026-55843-9.