User guide, Part II — from mesh to simulated heartbeat

Each workflow is launched from the Runtime panel and runs on the cluster, not in your browser. You may close the tab and come back.

The arrows are the whole story. Unload is the root: everything that needs a resting heart reuses one rather than recomputing it, and ED/ES quietly builds one first if none exists. Optimize is the exception — it never chains, and asks you to pick the Calibrate run it should refine.

Every workflow begins by staging the mesh (staging_mesh) and ends by publishing its archive. Between those, the named stages differ.

A heart imaged in a living patient is already stretched by the blood inside it, so it is the wrong starting shape for a simulation. Unload runs the inflation backwards: it searches for the geometry that, pressurised to the assumed end-diastolic pressure, returns the shape you actually imaged. Along the way it fits an empirical filling curve to your ventricle and reports the passive stiffness implied by it.

Two quasi-static solves bracket the heartbeat. The first fills the ventricle passively to its measured end-diastolic volume. The second searches for the peak active tension that squeezes it down to the measured end-systolic volume against arterial pressure. Both solves must converge, both volumes must land within volume_match_frac of their targets, the end-systolic pressure within es_pressure_tol_mmhg millimetres of mercury. The tension itself must sit strictly inside its plausible band rather than resting against a limit — a value pinned at tmax_lo_pa means the search never truly found one.

Calibrate fits the passive filling behaviour first, then the active twitch, and writes the parameter set that every later run on this mesh reads. It also leaves behind a fast approximate model of the ventricle, which is what makes Optimize cheap.

Optimize searches the fast approximate model — not the full heart — against your pressure–volume targets, so it explores thousands of parameter sets in the time one simulation would take. It refuses to start without a finished Calibrate run to draw that model from.

Simulate solves the beating ventricle in three dimensions, coupled to a circulation model that supplies realistic filling and afterload, and returns a full pressure–volume loop with regional strain. It always starts from the canonical parameter set, inheriting this case's current upstream results where they exist. A mesh with no upstream results may still be submitted, and may fail to get off the ground. If the solver cannot complete its first pressurisation, the job stops with LEGACY_FE_NLOAD_NON_CONVERGENCE rather than stalling; calibrate the mesh and resubmit.

Results opens on the pressure–volume loop, live while the job runs. Strain and the ED/ES summary get their own tabs, and the Artifacts tab holds one archive per job — never a list of loose files.

Each workflow runs either on this platform's own compute or, when your account is granted access, on the university's high-performance cluster. Unload and ED/ES prefer the cluster. The choice changes nothing about what you submit or what you get back.

The status box names the reason; the raw log under Advanced holds the detail. These are the failures you can act on yourself.

Thresholds, units and confidence levels for each of these are recorded alongside the code, together with the ranges the platform treats as physiologically plausible.