Brain view: how the field map is made

The Brain tab of each program's page (Programs → a program → Brain) shows how strong the coils' magnetic field is across a standard brain, which target regions get how much, and what each region is known (or claimed) to do. It answers "where does the field go?". It does not answer "what does the field do?": no study has shown that fields of a few microtesla switch any of these regions on.

What you see

• Glass or solid brain coloured by field strength in µT on a logarithmic scale from 0.05 µT (dark) to 10 µT (bright). Peak = the largest field during the phase; RMS = the root-mean-square over time, which counts how long a coil is on.
• Deep structures (amygdala, hippocampus, basal ganglia, thalamus) as separate surfaces inside the glass brain.
• Coils at their 10-05 sites, face 3 mm off the scalp, axis toward the centre of the head. Coils not used in the selected phase are dimmed.
• Live: runs the program in the browser (real time, 10× or 50× slower) and shows the instantaneous field and coil drive. It is the same preview as on the program's Overview, and the intensity is the one Start uses.
• What this program aims at: the program's targets, with the estimated field at each.
• Regions by field: every curated region ranked by its estimated field. Tap one for its depth, mainstream function, what electrode or TMS stimulation of it did, and the Persinger-lab or Murphy claims about it; it is marked on the brain.
• Tap any point on the brain for its value and the nearest named region.

Method

1. Coil model. Each 19 × 12 mm, 495-turn coil (winding radius 4.1–9.5 mm, 11 mm long) is split into 4 × 4 current loops, and its ferrite/iron core (4 mm radius, 12 mm, flush with the face) is a uniformly magnetised cylinder represented by 8 loops of its surface current (about 1630 ampere-turns per ampere; demagnetisation limits the apparent susceptibility to about 5, so the core material hardly matters). Each loop's field is exact (complete elliptic integrals); beyond 40 mm a point dipole is used (they agree within 1 %). On the axis this gives about 1.86 µT/mA at 20 mm (the firmware default rounds to 1.8) and 105 µT/mA at the face, about 2.5× the air-core winding; the air-core version reproduces tools/coil_field.py within 2 %. The inductance this model implies (4.1–4.6 mH) matches the seller's 4.1 mH.
2. Calibration. Fields are scaled the way the firmware sets intensity: the program's µT value is the peak on the coil axis 20 mm in front of the face (CALIBRATION.md). The ferrite core changes the coil's strength, which calibration absorbs, but hardly the shape of the field at these distances.
3. Program. For every phase the engine's own spatial gains (rotation, cycling, alternation, weights) are sampled for 12 s at 1 ms. Coils that are on together add as vectors. The waveform's peak and RMS scale the result; with two tracks the peaks add (an upper bound) and the RMS values add in quadrature.
4. Brain. The surfaces are built from the Brainnetome atlas (maximum-probability map, 1 mm, MNI152 space): surface nets on a 2-mm grid, holes filled, Taubin-smoothed. Brainnetome has no cerebellum or brainstem; those come from the CerebrA atlas (MNI ICBM 2009c symmetric space, so a few millimetres off the rest).
5. Sites. 10-05 positions from MNE-Python's fsaverage montage, shifted by a translation fitted so that scalp-to-brain distances match published values (34 sites, 1.6 mm RMS error after the fit).
6. Regions. About 40 targets (cortical areas, deep nuclei, cerebellum and brainstem) with MNI centres, depths, functions and claims from the project research (RESEARCH.md), plus anatomy under 43 scalp sites.
7. What stimulation did. For each region, a short note on what direct electrical stimulation (epilepsy surgery, milliamps) or TMS (about 100 V/m) produced, with sources, and neutral Neurosynth meta-analysis terms. This is context for what a region can do when driven that hard, not a prediction for this helmet. Most stimulated cortical sites produce nothing reportable (Fox 2020).

Accuracy

• Template head: real heads and brains differ from MNI152 by 5–10 mm; MNI templates are also somewhat larger than average brains.
• Placement by tape is repeatable to about 5–10 mm, comparable to the radius of the field spot a coil makes on the cortex.
• Absolute values depend on calibration; the map is exact only for the stated µT at 20 mm.
• Tissue does not change a low-frequency magnetic field, so the µT values need no head model.
• The induced electric field depends on dB/dt. Smooth signals give only microvolts per metre (a 7 Hz sine at 8 µT: about 1.7 µV/m at 20 mm). The steps of the 2 kHz output add brief spikes: about 0.6 mV/m for a 1-ms staircase, and at most about 11 mV/m at 20 mm for a full-swing edge lasting a few microseconds. Compare 0.1–1 V/m for the brain's own fields, about 1 V/m to affect spiking in animals, and about 100 V/m for TMS.

Data and licences

Rebuilding the assets

# download the two inputs into an empty folder, then:
node tools/brain_assets.mjs --bna BNA-maxprob-thr25-1mm.nii.gz --sites fsaverage_1005.tsv \
  --cerebra CerebrA.nii --regions regions-with-depths.json --out webapp/src/assets/brain

--regions is optional; it only enables the site-alignment fit. --grid 3 gives a coarser mesh at 40 % of the size.

The code: webapp/src/engine/field.js (coil model and exposure), webapp/src/lib/brain.js (assets, caching, colours), webapp/src/components/BrainScene.vue (three.js scene), webapp/src/views/BrainView.vue.