Pipeline Overview

Pipeline Overview

golgi models peripheral nerve stimulation as a directed sequence of stages, from a raw image to a fiber-type-selective recruitment result. The same stages run whether you drive them from the GUI, the Python API, or the CLI — they all act on one shared golgi.Study state on disk.

flowchart TD
    A["Image / surface / mask"] -->|segment / import| B["Segmentation + 3-D reconstruction"]
    B -->|geometry| C["Multi-region tetrahedral mesh"]
    M["Electrode / cuff design"] --> C
    S["Conductivities (Cole–Cole / IT'IS)"] --> D
    C --> D["Anisotropic FEM (per-contact lead fields)"]
    B --> E["Fiber populations + curved 3-D trajectories"]
    D -->|Vₑ on fibers| F["Fiber activation (NEURON / AxonML)"]
    E --> F
    F --> G["Recruitment · selectivity · steering"]
    D -.->|reciprocity| R["Recording / CNAP"]
    E -.-> R
    G --> H["Hashed, replayable study bundle"]

Loading

---

The stages

#	Stage	Input → Output	API	Page
1	Geometry	image/mask/surface → 3-D nerve geometry (endo/epi, fascicles)	import_nerve (+ GUI segmentation)	Geometry Import & Segmentation
2	Electrode design	electrode type + placement → cuff geometry, contacts, polarities	set_electrodes / Cuff Designer	Electrodes & Cuff Designer
3	Meshing	geometry + cuff → multi-region tetrahedral mesh	set_mesh + run_mesh	Meshing
4	Materials	tissue conductivities (anisotropic, frequency-aware)	set_mesh(...) / Materials drawer	Conductivity & Tissue Properties
5	FEM	mesh + σ + electrodes → extracellular field Vₑ, lead fields, impedance	run_fem	Finite-Element Solver
6	Fibers	geometry → curved 3-D fiber trajectories, branch-classified	set_fiber_seed + run_fibers	Fiber Populations & Trajectories
7	Activation	Vₑ on fibers + pulse → spikes, thresholds	(within run_sweep / GUI panels)	Fiber Models & Activation
8	Analysis	thresholds → recruitment curves, selectivity, comparison	run_sweep	Recruitment Sweeps & Selectivity
—	Recording	fiber currents + reciprocity lead fields → CNAP	pipeline/recording	Recording & CNAP
9	Export	the whole study → bundle / figures / PDF report	export_bundle	Reproducible Bundles · Figures & Reports

---

Why the stages are ordered this way

• Geometry before meshing — the mesh is built from the reconstructed surfaces (endoneurium, epineurium) plus the cuff and contacts.
• Electrodes feed both meshing and FEM — contacts become tagged facets in the mesh and current boundary conditions in the solve.
• Fibers are independent of the cuff — trajectories are generated on a nerve-only mesh (decoupled from cuff position), so the same fibers can be evaluated against different electrode designs.
• FEM produces lead fields, not just one field — because the quasi-static problem is linear in current, golgi solves a per-contact basis once and combines it for any montage and any amplitude. This is what makes multi-contact current steering and fast amplitude sweeps possible without re-solving.
• Activation reads Vₑ sampled on the fibers — run_fem writes paths_Ve.npz (the field along every fiber), which the fiber models turn into spikes and thresholds.

Caching & reuse

Every stage writes its artifacts under the project directory and is content-hashed, so unchanged inputs are not recomputed:

• FEM lead fields are fingerprinted on (mesh, σ, contact geometry, solver preset).
• Sweeps are cached by a SHA over the request — re-running an identical sweep is instant.
• An I_stim amplitude sweep rescales the cached field instead of re-solving the FEM (linearity).

This same hashing underpins the reproducible study bundle: the bundle's manifest records the stage DAG and the hash of every input and output, so golgi replay can verify the whole chain byte-for-byte.

Multiple designs, one nerve

A project can hold several designs (electrode geometries) and configs (polarity/current montages) on the same nerve. The nerve and far-field bath are canonical and shared; only the cuff moves between designs. This is what powers cuff-position sweeps and side-by-side design comparison.

---

See also

Architecture · Configuration Reference · Getting Started