Each of the 14 nodes is a schematic brain region wired into a small connectome (thalamo-cortical hubs, the fronto-parietal attention network, the hippocampal memory loop, and the Broca-Wernicke language pathway), laid out here as a 2D node-link graph you can pan and zoom. Lesioning a region sets its health h ∈ [0,1] to 1 − severity; every edge touching it loses strength accordingly:
w_eff(i,j) = w(i,j) · h(i) · h(j)
d(i,j) = 1 / w_eff(i,j) (edge removed if w_eff < threshold)
Distances are then the shortest weighted paths (Floyd-Warshall) through the whole graph — an intact detour can partially route around damage, exactly as collateral white-matter pathways do. Task performance uses global efficiency (Latora & Marchiori, 1994) over the task's critical node set S:
E(S) = 1/(|S|(|S|-1)) · Σ_{i≠j∈S} 1/d(i,j)
Performance = 100 · E(S) / E₀(S) (E₀ = healthy baseline at the same threshold)
- Click a region (in the sidebar or on the graph) to lesion/restore it; the severity slider sets how damaged every lesioned region is.
- Drag on empty graph space to pan, scroll to zoom — the layout is fixed, only your view moves.
- Edge dropout threshold is the w_eff cutoff below which an edge is treated as severed; raising it makes the network more brittle even before any lesion.
- Attention / Memory / Language switches which node set S drives the performance readout — highlighted on the graph.
- Stimulate fires a signal pulse from that task's seed region and lets it propagate hop-by-hop up to the hop-budget slider; pulses die out at low-health nodes or missing edges.
- The distance-matrix panel shows every pairwise shortest path live (bright = short/healthy, dark = far/disconnected); the performance-vs-severity panel resamples the current lesion pattern across the full severity range so you can see the whole deficit curve, not just today's slider position.
This is the same lesion-deficit logic used in real network neuroscience: ablate a node in a connectome model, recompute graph efficiency, and read off the predicted behavioral deficit — the approach behind "virtual lesion" studies that estimate cognitive impact of stroke or trauma before it can be observed clinically.