Interactive companion to From Clocks to Coordinates
Primary research artifact: 54-page full paper
Interactive atlas · 6-page judge brief · Research repository
HumanVector Atlas is the reviewer-facing demo for From Clocks to Coordinates. It turns the research repository into a guided, interactive story: begin with one real ECG, split the heartbeat into electrical subsystem clocks, follow the null-to-mechanism-to-transport discovery arc, and then explore how the same coordinate question was tested across brain MRI, skeleton CT, organ systems, movement, and regulatory genomics.
The product framing is deliberately precise: this is a cross-cohort evidence atlas plus a future whole-person blueprint, not a multimodal patient digital twin.
- On
/, use the unified Atlas Viewer to switch among a real five-surface heart, complete LEMON MRI volumes, a 63-part skeleton, and a 115-structure whole-body CT. - In Brain MRI, choose a released LEMON record, scrub all 256 axial/coronal/sagittal slices, and switch among native T1, six-tissue Atropos, and DKT groups.
- In Heart MRI, click a chamber surface or choose a named chamber; phase buttons project an explicitly explanatory localization map onto the real MRI segmentation.
- In Skeleton CT or Whole-body CT, rotate, zoom, click, and filter the named TotalSegmentator masks and their population-level volume–age associations.
- Open
/participantto filter one of three real, de-identified PTB-XL ECG examples inside the same workstation-style shell and inspect the measured trace, subsystem clocks, A/D geometry, and signed q coordinates. - Open
/resultswith the participant selection preserved in the URL. Present the six-slide evidence deck with its numbered rail, Previous/Next controls, or the arrow keys; every slide keeps the result and its scientific boundary together.
- The ECG, MRI, CT, NHANES, motion, and genomics panels come from independent datasets. They are not measurements from one person.
- The three selectable profiles contain real ECG inputs and frozen HeartVector outputs only. Other modalities are clearly labeled as separate-cohort evidence.
- The 3D heart contains four chamber-cavity surfaces and an epicardial surface manually segmented from a separate 1.5 T MRI. The phase colors are explanatory overlays, not voltage measurements, causal saliency, or an ECG reconstruction.
- The LEMON viewer loads complete 256³ released MRI records and their processed segmentations. The displayed chronological values are source age bins; model-derived brain age is labeled separately. Records are different participants, not before/after scans.
- The whole-body and skeleton explorers use scan-derived meshes from representative TotalSegmentator CT subject
s1045. Their colors are cross-sectional population correlations from 1,227 CT exams; they are not that subject's disease map or longitudinal aging trajectory. - The external QT-label association is conditional. The marginal association was null, and the demo makes no diagnostic claim.
- Motion includes a prespecified null, and Skeletome remains a prospective hypothesis screen.
| Surface | Source in the research project | Demo treatment |
|---|---|---|
| ECG profiles | PTB-XL examples 00233_hr, 00420_hr, 00176_hr |
Real 500 Hz lead-II waveform, phase ages, A, D, and q geometry |
| Heart anatomy | Gerach et al. Four-Chamber Human Heart Model, Zenodo 10.5281/zenodo.5573921 |
Five separately named MRI-segmented surfaces with interactive, explanatory phase coloring |
| Brain | LEMON skull-stripped MP2RAGE T1, ANTsPyNet Atropos tissue masks, DKT parcellation, and lemon_disagreement_result.json |
Four complete, interactive 256³ volumes plus aggregate cohort result |
| CT anatomy | TotalSegmentator v2 subject s1045, 117-class mask, and 1,227-exam cohort associations |
Interactive 63-mesh skeleton / 115-mesh whole body, exact structure filters, voxel counts, rho values, and raw CT/mask proof |
| Organ systems | nhanes_cox_results.csv, canonical survey Cox table, CV ladder, and leave-one-system-out table |
Filterable aggregate dashboard; original available-case per-system estimates are kept separate from canonical survey-weighted A/D results |
| Perturbation | Randomized IKr blockade analysis | Frozen signed-direction figure and summary |
| External transport | Chapman–Shaoxing / Ningbo analysis | Conditional frozen-direction result |
| Motion | NHANES MotionVector | Prespecified null boundary result |
| Genomics | Skeletome / Borzoi screen | Prospective hypothesis-screen result |
- Confirmatory disagreement-radius result: HR 1.010, 95% CI 0.983–1.039, LRT p = .478; n = 74,715, 2,583 deaths.
- IKr direction: bootstrap median cosine 0.941 with 99.99% sign stability; nominal permutation p = .033.
- External conditional association: OR 1.225/SD, 95% CI 1.10–1.364, p = 1.92 × 10⁻⁴; n = 44,550, 386 cases.
- Brain: older D 0.517 versus young D 0.340; difference +0.177, p = 4.3 × 10⁻⁸, d = 0.844; n = 220. Balanced 69-versus-69 resampling kept the contrast positive in all 5,000 resamples (median d = 0.60), supporting an age-group shift rather than universal dysregulation.
- CT atlas: a multi-region volume-only model achieved age MAE 8.56 years and R² = .432 in 1,227 exams. Skeleton panels show cross-sectional volume associations, not bone density or structure-specific aging rates.
- NHANES: A_body HR 1.206, D_body HR 1.213; cross-validated C-index 0.8127 → 0.8262 → 0.8304.
- Motion primary reserve test: OR 1.10, 95% CI 0.99–1.23, p = .070.
- Skeletome: 1,955 HAR substitutions screened, 722 candidates retained; matched enrichment null p = .24.
Requires Node.js 22.13 or newer.
npm ci
npm run dev
npm run buildThe GitHub Pages build is a static Next.js export. All scientific figures, ECG samples, named 3D anatomy, and aggregate association metadata are committed under public/ or app/data/; the demo does not depend on a research-data API. .github/workflows/pages.yml builds and publishes out/ whenever main changes.
app/AtlasDashboard.tsx— unified text-first system selector and cross-cohort boundary.app/HeartExplorer.tsx— clickable Three.js viewer for five named MRI-segmented heart surfaces.app/BrainMRIExplorer.tsx— direct browser rendering of complete T1, tissue, and DKT NIfTI volumes.app/HumanVectorAtlas.tsx— participant explorer, real ECG filters, synchronized phase views, and reference heart.app/ResultsDeck.tsx— presenter-first, keyboard-accessible six-slide evidence deck.app/ResultsStory.tsx— retained long-form evidence narrative and scientific boundaries.app/AnatomyExplorer.tsx— Three.js viewer for clickable scan-derived skeleton and whole-body meshes.app/OrganExplorer.tsx— filterable aggregate NHANES organ-system dashboard.app/data/ecgProfiles.json— real example ECG traces and frozen outputs.app/data/organSystems.json— committed aggregate per-system values transcribed from the project result tables.app/data/ct*Associations.json— exact CT structure groups, voxel counts, colors, and cohort-level rho values.app/globals.css— responsive visual system.public/evidence/— supporting cohort anatomy and retained source figures.public/models/heart-four-chamber-*.glb— MRI-segmented four-chamber reference heart with neutral and phase-highlight variants, CC BY-NC 4.0.public/models/ct-skeleton.glb,public/models/ct-whole-body.glb— real TotalSegmentator CT-derived meshes, CC BY 4.0.public/mri/lemon/— canonical RAS LEMON T1/Atropos/DKT NIfTI volumes and their manifest.public/og.png— typography-only social-preview artwork with no synthetic anatomy or linked-patient implication.PRESENTER_SCRIPT_LAST_TWO_MINUTES.md— committed 1:00–3:00 click sequence, spoken copy, timings, and claim guardrails.
The project code follows the parent repository's MIT license. The three ECG examples come from PTB-XL v1.0.3 by Wagner et al. and are redistributed under CC BY 4.0. TotalSegmentator v2 assets are also CC BY 4.0. The MRI files were obtained from the public MPI-LEMON release on the canonical INDI portal, which lists PDDL as its data-sharing license; Atropos and DKT volumes are locally generated derivatives and are identified as such. The CardioMechanics heart is CC BY-NC 4.0 and is suitable here only for attributed, noncommercial hackathon display. Each redistributed scientific asset retains the source citation, license note, and scientific boundary appropriate to that source.
The authoritative asset-by-asset record is ASSET_PROVENANCE.md. The full paper committed under public/papers/ is the primary write-up for judging; the shorter brief is a navigation aid, not a replacement for the paper.