Recruitment Sweeps and Selectivity

Recruitment Sweeps & Selectivity

This is the analysis layer that turns a solved model into the numbers you cite: recruitment curves, per-fiber thresholds, selectivity between target and off-target branches, and side-by-side design comparison. It runs on the cached fiber field, so amplitude sweeps are fast.

Source: golgi/pipeline/sweep.py, golgi/pipeline/selectivity.py, golgi/projects/sweep_cache.py, golgi/figures/recruitment.py + selectivity.py. Schemas in golgi/jobs/schemas.py.

---

Sweeps

A sweep is described by a SweepRequest and run via Study.run_sweep (or the GUI Sweep panel). Two modes:

• recruitment — simulate every fiber across an amplitude axis (amplitudes_mA) and record an activated matrix of shape (n_fibers, n_amplitudes).
• threshold — find each fiber's activation threshold by bisection (bisect_lo_mA, bisect_hi_mA, bisect_tol_uA), recording thresholds_uA of shape (n_fibers,).

from golgi.jobs.schemas import SweepRequest

# recruitment across amplitude
req = SweepRequest(mode="recruitment",
                   amplitudes_mA=[0.1, 0.25, 0.5, 1.0, 2.0],
                   backend="pyfibers", model_name="MRG_INTERPOLATION")

# or per-fiber thresholds by bisection
req = SweepRequest(mode="threshold",
                   bisect_lo_mA=0.01, bisect_hi_mA=2.0, bisect_tol_uA=10.0)

result = study.run_sweep(req)

You can restrict the sweep with fiber_indices, branch_filter, or fiber_type_filter, and set pulse_params. The SweepResult carries activated / thresholds_uA plus per-fiber fiber_diameters_um, fiber_branch_idx, and fiber_type_labels, so every metric below can be sliced by branch and type.

The amplitude shortcut

Because the FEM is linear in stimulus current, an I_stim/amplitude axis is computed by rescaling the cached per-fiber Vₑ rather than re-solving the field — only the (cheap) fiber simulation re-runs per amplitude. Axes that touch σ, mesh, electrode geometry, or fiber paths re-enter the appropriate upstream stage instead. See Finite-Element Solver.

Caching

Each result is written to <project>/sweeps/sweep_<sha>.npz (plus .json and CSVs), keyed by a SHA over the request — re-running an identical sweep is instant. Results are also tagged per config so multiple electrode configs coexist. CSV exports cover recruitment, thresholds, and the activation heatmap.

---

Selectivity metrics

pipeline/selectivity.py computes, as pure post-processing on sweep results:

• Branch recruitment — fraction of fibers in a branch activated at each amplitude.

• Veraart selectivity index (Veraart et al. 1993) for a target branch against off-target branches:

  $$\mathrm{SI} = \frac{R_\text{target} - R_\text{off-target}}{R_\text{target} + R_\text{off-target}} \in [-1,,1]$$

  +1 = only the target is recruited; 0 = equal; −1 = only off-target.

• Threshold ratio — median(T_off-target) / median(T_target) (>1 means the target activates at a lower amplitude — desirable).

• Spatial activation map — per cross-section position, the minimum amplitude to activate each fiber.

In the GUI you pick a target branch (e.g. the cardiac branch); selectivity is computed against the union of the others (or custom target/off-target sets). These power the selectivity figures (Figures & Reports).

---

Comparing designs & steering current

A project can hold several electrode designs and configs on the same nerve. The Compare configurations view puts two designs side by side (axis + slice fields) with a selectivity bar chart and threshold-ratio table, so you can answer "which cuff is more selective?" directly.

Because the field is a per-contact lead-field basis, you can also steer current — change which contacts are anodes/cathodes and their current fractions — and re-evaluate recruitment without re-solving the FEM. This makes multi-contact montage exploration (the workhorse of selective VNS) interactive. See Electrodes & Cuff Designer for setting polarities and fractions.

---

References

• Veraart C, Grill WM, Mortimer JT (1993). Selective control of muscle activation with a multipolar nerve cuff electrode. IEEE Trans Biomed Eng 40:640. — the selectivity index.
• Musselman ED, Cariello JE, Grill WM, Pelot NA (2021). ASCENT: a pipeline for sample-specific computational modeling of electrical stimulation of peripheral nerves. PLOS Comput Biol. — recruitment/threshold methodology.

---

See also

Fiber Models & Activation · Finite-Element Solver · Figures & Reports · Python API · Validation