Access AortaMix (free, instant)

Access is granted automatically. We only ask who you are so we can understand how the data is used and contact users about updates. Research use only; not for clinical decisions.

Log in or Sign Up to review the conditions and access this dataset content.

AortaMix: VA-ECMO aortic mixing-zone simulations

Quick links: interactive viewer · full medico-technical report (PDF, 9 pages) · 2-page technical report (PDF) · physics checks

Simulated blood flow in an idealized human aorta under peripheral veno-arterial ECMO, with a passive tracer separating ECMO-returned blood (c = 1) from native left-ventricular blood (c = 0). The mixing zone ("watershed") decides which arteries receive oxygenated ECMO blood, which is the mechanism behind differential hypoxia (Harlequin syndrome).

To our knowledge (survey on 2026-11-16) this is the first public VA-ECMO aortic CFD dataset with released 3D fields. It has two fidelity levels that are never mixed:

fidelity what where
3D-nekRS spectral-element Navier–Stokes + tracer, 3D fields over time runs/<run_id>/
0D-lumped lumped network model of the same geometry and boundary conditions, 6011 samples zerod/

These are simulations of an idealized model, not patient data, and not clinical guidance.

3D cases

run_id cohort ECMO site ECMO fraction total flow (L/min) period (s) N cycles planned simulated time reached (s) exit periodic?
a50 adult femoral 0.5 5.0 0.8 5 3.0 2.4 0 False
a50x adult femoral 0.5 5.0 0.8 5 4.5 3.6 0 False
c75 paed_child carotid 0.75 1.8 0.545 5 3.0 1.635 0 False
c75x paed_child carotid 0.75 1.8 0.545 5 4.5 2.453 0 False
a75 adult femoral 0.75 5.0 0.8 5 3.0 2.4 0 False
a75x adult femoral 0.75 5.0 0.8 5 3.75 3.0 0 False
a90 adult femoral 0.9 5.0 0.8 5 3.0 2.4 0 False
a90x adult femoral 0.9 5.0 0.8 5 3.75 3.0 0 False
A1_75 adult femoral 0.75 5.0 0.8 5 2.0 1.6 0 False
mz6 adult_benchmark_mazzoli2025 femoral 0.8 7.5 0.8 5 2.0 0.19 1 False

exit is 0 for a run that reached its end time; runs stopped by the campaign deadline keep every checkpoint written before the stop. periodic? comes from the post-processing check (outlet cycle-mean flows within 1 % and c = 0.5 position within 1 mm between the last two cycles); runs with fewer than 3 cycles are not periodic.

Model

  • Solver: nekRS v26.0 (HIP backend, AMD MI300X), spectral elements, polynomial order N = 5 (one N = 4/6 probe is labelled), OIFS sub-cycling, target CFL 4.
  • Geometry: idealized parametric aorta (geometry/make_aorta.py): ascending aorta, arch with BCA, LCCA and LSA, descending thoracic and abdominal aorta with one lumped visceral outlet (celiac + SMA + renals), iliac bifurcation. Variants: adult baseline, A1 (dilated/older), A2 (smaller), paediatric teen (uniform scale 0.8), paediatric child (scale 0.55). Hex meshes 12.5k–14.7k elements, no inverted elements (mesh/, mesh_all/).
  • Blood: Newtonian, rho = 1.06 g/cm3, mu = 3.5 cP; rigid walls; laminar (no turbulence model).
  • Inflow: LV half-sine pulse, systole fraction ~0.44; ECMO steady flow. ECMO share of total flow is the "ECMO fraction".
  • ECMO site: femoral (adult, teen): ECMO enters through the full left-iliac cross-section. Carotid/neck (child): ECMO enters through the BCA into the arch.
  • Outlets: 3-element Windkessel (RCR) at every outlet, resistance split by assumed resting flow fractions, MAP target 75 mmHg adult / 70 teen / 60 child, total compliance 1.0 / 0.8 / 0.35 mL/mmHg, explicit coupling.
  • Tracer: Sc = 1000 with nekRS AVM regularization; at Pe ~ 1e7 the interface width is numerical.
  • Every assumption is stored per case in runs/<run_id>/case_params.json.

Extensions (*x runs)

a50x, c75x, a75x, a90x continue a50, c75, a75, a90 from their last checkpoint. Checkpoint numbering restarts at f00000 after every restart, so always order files by the simulated time stored in each file. a75x and a90x add only 0.6 s (< 1 period): their summary.json and fields.h5 re-analyse the parent cycle 1.6–2.4 s; their new information is the raw checkpoints and monitor.csv rows with t > 2.4 s. See NOTE_EXTENSION.txt in each folder.

Known limitations

  • Idealized geometry; no patient anatomy, no coronary arteries, no left-leg perfusion.
  • ECMO modelled as a full-vessel inlet, not a 15–19 Fr cannula jet (Khamooshi 2024 shows jet details matter near the tip).
  • Rigid walls, Newtonian blood, laminar flow.
  • Short runs (3 cardiac cycles for most cases) because of the compute window; check periodic?.
  • No experimental validation yet. Closest published benchmark: Mazzoli et al. 2025 (doi:10.3389/fbioe.2025.1620897), reproduced in spirit by case mz6 (ECMO 6 L/min, native output reduced 70 %).

Files

runs/<run_id>/aorta.par, aorta.udf, case_params.json   # exact solver input
runs/<run_id>/aorta0.f*                                 # nekRS checkpoints, 24 per cardiac cycle
runs/<run_id>/monitor.csv                               # every 50 steps: Q and tracer per port, Qsum, umax, outlet pressures
runs/<run_id>/fields.h5, derived.h5, summary.json       # when post-processing finished: 128^3 grid x 24 phases, vortex metrics, watershed
mesh*/                                                  # meshes, centerlines, port definitions, mesh-quality reports
zerod/zerod_samples.parquet, zerod_meta.json            # 0D sweep (one row per sample)
metadata/cases.parquet, cases.csv                       # one row per 3D case
logs/                                                   # complete solver and pipeline logs

Units: CGS (cm, s, g). In monitor.csv pressures are kinematic (p / rho, cm^2/s^2); multiply by 1.06 / 1333.22 for mmHg. Port ids: 1 LV root, 2 left iliac, 3 BCA, 4 LCCA, 5 LSA, 6 visceral, 7 right iliac, 8 wall. Flows are positive out of the domain.

0D companion

Same geometry, inflows and Windkessel outlets as a rigid R–L segment network (BDF integration, 8 cycles, last cycle reported). Randomly sampled over cohort (adult / teen / child), size scale, ECMO fraction 0–1, total flow, heart rate, MAP, compliance, proximal-resistance share and outlet split. It reports outlet pressures and flows, root (LV afterload) pressure, ECMO-blood fraction per outlet (perfect mixing at junctions) and the junction where cycle-mean LV and ECMO flows meet. Rows with anchor set reproduce the 3D case settings for one-to-one comparison. Low fidelity: no 3D mixing physics.

Suggested splits

Case-level splits only (never split time steps of one case across sets). Suggested hold-out: the A1 anatomy and the child case for out-of-distribution testing.

Citation / licence

CC BY 4.0. Please cite this repository. Code: generation scripts are included in each run folder (aorta.udf, aorta.par); the full pipeline will be released with the code repository.

How to cite

If you use AortaMix, please cite:

@misc{aortamix2026,
  title        = {AortaMix: simulated blood mixing in the aorta during peripheral VA-ECMO (3D nekRS CFD + 0D companion)},
  author       = {{AortaMix project}},
  year         = {2026},
  howpublished = {Hugging Face Datasets},
  url          = {https://huggingface.co/datasets/vhmaadhav/aortamix-vaecmo-simdata},
  note         = {CC BY 4.0. DOI to be added.}
}

Tell us how you use it

We would love to hear from you, especially clinicians and device engineers. Please open a thread in the Community / Discussions tab with your use case, questions, corrections or requests (new cannulation sites, flows, anatomies). Feedback directly shapes the next release.

Downloads last month
120

Space using vhmaadhav/aortamix-vaecmo-simdata 1