LeaFlow turns clinical scans into patient-specific simulations, and an agentic pipeline runs them across whole cohorts instead of one idealised model. The first application, validated and running today, is airflow and aerosol transport in the airways.
Why it matters
An idealised model is one geometry. Real patients differ in calibre, branching and angle, and those differences change the flow a drug or a device works in. Our cohort showed a thirty-fold spread where a single model shows one number.
The move of pressurised inhalers to low-global-warming propellants changes the spray. Each reformulation raises the same question: does the drug still reach the same airways?
Casts and in-vitro rigs cover a handful of geometries. Simulation can cover dozens before a prototype is built, and points the physical tests to the cases that matter.
Agentic, at cohort scale
An AI agent carries each case from the CT to the result without manual meshing: segmentation, geometry, set-up on a regular lattice-Boltzmann grid, simulation and report. In our internal study the unmodified pipeline ran 43 airway trees from routine clinical CT, at resting and exertional flow, in roughly one day on a GPU cluster.
Internal study, manuscript in preparation; not yet peer reviewed.
Who it is for
Compare formulations and devices on the same set of real anatomies.
Place the device in front of a reconstructed airway and test design variants.
Stents, tubes and endobronchial valves in the anatomy they are designed for.
Where we are
Today
Lattice-Boltzmann solver on GPU, validated against the SimInhale benchmark: measured velocity profiles and the flow split among the main bronchi.
The airway surface from the CT is voxelised automatically onto the solver grid; inlet and outlets are proposed by the platform.
Inhaler, nebuliser and spacer presets emit particles that the flow carries through the airways.
Each attested run carries a signed record of software, hardware and data, built with EQTY Lab.
The solver records where each particle touches the airway wall, with identical results on CPU and GPU. Validation against the SimInhale deposition data is under way, then the same 43-patient cohort; we would like to do this with industry partners.
Particles from 1 to 10 µm in the same simulation, each followed through the airways.
The same engine and pipeline on vessels and other anatomies, as research and development.
How to work with us
One device or formulation question, a defined set of anatomies, a written report. Weeks, not months.
Build and validate regional deposition together, with your in-vitro or clinical data as the reference.
Run LeaFlow yourself, in the cloud or on your own GPUs, with the same agentic pipeline.
Tell us the product, the formulation or the patient population you care about. We will propose a pilot on real anatomies.
LeaFlow is a research and engineering tool. It is not part of the CE-marked LeaRes medical device and it is not intended for clinical decisions on individual patients.
