The grid of instanced neurons is a simplified slice of primary motor cortex (M1) around the hand-knob area. Each live neuron i carries a synaptic weight wi ∈ [0,1] for the hand motor representation, initialised as a Gaussian bump around the hand centroid:
w_i(0) = exp(−d_i² / 2σ²)
An infarct zeroes every neuron's weight within its radius and marks it permanently dead (grey) — real ischemic core tissue does not repair itself. Training applies a Hebbian-style, distance-weighted update to surviving peri-infarct neurons each simulated day:
Δw_i = η · I · a_i · exp(−r_i / λ) − δ·w_i
a_i = local co-activation with already-recruited neighbours
r_i = distance to the nearest infarct border
I = training intensity (reps/day), δ = passive decay
This mirrors the real neuroscience: Nudo's motor-map microstimulation studies in primates and human TMS/fMRI mapping both show that peri-infarct cortex can take over part of a lost representation, but only with sufficient use-dependent practice — more reps per day (higher I) recruit surviving tissue faster, low intensity lets passive decay (δ) win and the map stays shrunken, and nothing recruits neurons that were inside the infarct itself. This is the mechanistic basis for why rehab dosing (reps, not just calendar time) predicts motor outcome after stroke.
- Infarct radius / Trigger infarct — sets and detonates the ischemic core; dead neurons never return.
- Training intensity — none / low / moderate / high reps-per-day multiplier on the Hebbian term.
- Plasticity rate η — how efficiently each rep converts into synaptic strengthening (roughly: patient age, comorbidities, time-since-stroke window).
- Map recovered — surviving weight mass vs. the pre-stroke total; Centroid shift — how far the functional hand map has migrated from its original cortical location.