The open-source Acoustic Digital Human Workbench.
Full-body ultrasound computed tomography simulation, 3D reconstruction, multimodal medical signal fusion, and autonomous harness evolution — in your browser.
→ Open Sonic Chamber in your browser
No install. Full USCT reconstruction running in WebAssembly (~130 ms).
Sonic Chamber is an open-source research workbench for Ultrasound Computed Tomography (USCT) — a non-ionising medical imaging modality that reconstructs speed-of-sound and attenuation maps through full-body tissue cross-sections.
It is the open-source reference implementation for the type of acoustic digital human modelling pioneered by emerging platforms like Midjourney Medical. Where those platforms are closed, Sonic Chamber is fully auditable — every algorithm, every weight, every ADR is public.
Midjourney Medical applies generative AI to medical imaging — synthesising, enhancing, and reconstructing diagnostic images from multimodal inputs (CT, MRI, ultrasound, labs). Sonic Chamber is the acoustic physics layer of that stack, built open and reproducible:
| Midjourney Medical | Sonic Chamber |
|---|---|
| Proprietary cloud platform | MIT open source — runs locally + in-browser |
| Black-box reconstruction | Every algorithm documented, ADR-indexed |
| No audit trail | Cryptographic run ledgers (SHA-256 chain) |
| Closed training data | Synthetic phantom + DICOM/FHIR adapters |
| Subscription model | Free forever |
Sonic Chamber works as pure software simulation or as the software stack for a real USCT rig.
| Component | Spec | Cost |
|---|---|---|
| Any modern PC / Mac / Linux | 4-core CPU, 8 GB RAM | — |
| Browser with WASM support | Chrome / Firefox / Safari | Free |
| Rust toolchain | stable 1.78+ | Free |
| Node.js | v22+ | Free |
| Component | Part | Qty | Est. Cost |
|---|---|---|---|
| Controller SBC | Raspberry Pi 5 (8 GB) | 1 | ~$80 |
| Ultrasound transducer array | 128-element ring, 2–5 MHz | 1 | $800–$3,000 |
| Pulser/receiver board | Ultrasonix SonixDAQ or equivalent | 1 | $2,000–$8,000 |
| Water tank (coupling bath) | 30 cm acrylic cylinder, food-grade | 1 | ~$150 |
| Stepper motor (axial drive) | NEMA 17, 1.8°/step + A4988 driver | 1 | ~$25 |
| ADC / FPGA interface | 12-bit, 100 MSPS min | 1 | $200–$500 |
| Coaxial cables + SMA connectors | RG-58, SMA-M | — | ~$50 |
| Total | ~$3,300–$12,000 |
| Component | Part | Qty | Est. Cost |
|---|---|---|---|
| Ultrasound research platform | Verasonics Vantage 256 or equivalent | 1 | $80K–$150K |
| 256-element ring transducer | Imasonic custom, 2–7 MHz | 1 | $15K–$40K |
| High-perf workstation | 32-core, 128 GB RAM, RTX 4090 | 1 | ~$8,000 |
| Degassed water recirculation | Temperature-controlled DI water system | 1 | ~$2,000 |
| Precision stepper + rail | 5-axis stage, sub-mm positioning | 1 | ~$3,000 |
| Total | ~$108,000–$203,000 |
| Platform | Channels | Open HW | Est. Cost |
|---|---|---|---|
| OpenSonics | 64 | Yes | ~$500 |
| ULA-OP | 64 | Partial | ~$15,000 |
| Kwave (simulation only) | N/A | Yes | Free |
| FIELD II (simulation only) | N/A | Yes | Free |
Design first, buy later: Sonic Chamber's simulation produces the same reconstruction behaviour as hardware. Design and validate your acquisition protocol in software before committing to hardware spend.
→ ruvnet.github.io/SonicChamber/demo/
# Install Rust + wasm-pack
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
curl https://rustwasm.github.io/wasm-pack/installer/init.sh -sSf | sh
# Clone and run
git clone https://github.com/ruvnet/SonicChamber
cd SonicChamber
bash scripts/build-sonic-ct-wasm.sh
cd examples/sonic-ct && npm install && npm run dev
# → http://localhost:5184cd SonicChamber
# One reconstruction → PGM images + metrics to stdout
cargo run --release --bin sonic_ct_demo /tmp/out
# Train segmentation model on 24 synthetic phantoms
cargo run --release --bin sonic_ct_train 24 /tmp/out
# Benchmark reconstruction methods
cargo run --release --bin sonic_ct_methods
# Frozen-engine JSON stdio (for ML / harness integration)
echo '{"n":96,"elements":256,"iters":8,"seed":42}' \
| cargo run --release --bin sonic_ct_servedocker run --rm -p 5184:5184 ghcr.io/ruvnet/sonic-chamber:latest
# → http://localhost:5184npm install sonic-chamberimport { buildReconstructionPrior, scoreMultimodalRun } from 'sonic-chamber';
const prior = buildReconstructionPrior([
{ modality: 'mri', uncertainty: 0.1, consentScope: 'research' },
{ modality: 'ekg', uncertainty: 0.05, consentScope: 'research' },
]);
const score = scoreMultimodalRun({ acoustic: engineResult, prior });
// score.acousticResidual → immutable physics output (never modified)
// score.reconstructionStability → +10–12% from multimodal prior fusion┌──────────────────────────────────────────────────────────────────────┐
│ FROZEN PHYSICS ENGINE (crates/sonic-ct) │
│ SART · Backprojection · Landweber · FWI · Acoustic Memory · Organs │
│ Zero external Rust deps · Deterministic · 31 KB WASM binary │
├──────────────────────────────────────────────────────────────────────┤
│ WASM BINDING LAYER (crates/sonic-ct-wasm) │
│ C ABI · Zero-copy F32/U8 buffer access · 2-D + 3-D volume API │
├──────────────────────────────────────────────────────────────────────┤
│ FUSION HARNESS (packages/metabiohacker) │
│ Multimodal ingest · Patient state graph · Prior builder │
│ Darwin evolution · Evidence gate · Immutable run ledger │
└──────────────────────────────────────────────────────────────────────┘
The physics engine is frozen — its acoustic residual is never altered by downstream context.
| Method | Description | WASM time |
|---|---|---|
Backprojection |
Delay-and-sum (1 SART iteration) | ~16 ms |
SART |
Algebraic row-action, 8 iterations default | ~130 ms |
Landweber |
Gradient descent on normal equations | ~200 ms |
FWI |
Full-Waveform Inversion, adjoint-state gradient | ~500 ms/src |
| Tissue | Speed | Attenuation | Dice (tuned model) |
|---|---|---|---|
| Water (coupling bath) | 1480 m/s | 0.5 Np/m | — |
| Fat | 1450 m/s | 8 Np/m | ~0.72 |
| Muscle | 1580 m/s | 15 Np/m | ~0.65 |
| Organ (soft parenchyma) | 1570 m/s | 12 Np/m | ~0.41 |
| Bone (cortical) | 3000 m/s | 120 Np/m | ~0.00 ¹ |
¹ Bone Dice ≈ 0 with straight-ray SART (expected — acoustic shadow ignored). Full-Waveform Inversion (ADR-0026) is the path to sub-mm bone resolution.
- 3-D volume sweep — progressive axial slicing, body composition fractions
- Organ inference — hypothesis from shape + position, never speed alone
- Acoustic memory — NSW graph for longitudinal tracking and FWI warm-start
- Full-Waveform Inversion — adjoint-state, frequency continuation, CI gradient-checked
- Darwin harness evolution — Pareto-front multi-objective parameter search
- Run ledger — SHA-256 chain hash for every pipeline execution
- Evidence gate — Grade A/B-only claims; Grade C/D blocked
- Multimodal fusion — MRI · CT · EKG · EEG · labs · pathology priors
| Gate | What it enforces |
|---|---|
| Acoustic residual immutable | Physics output never altered by downstream fusion |
| Pathology → human review | Pathology/biopsy always sets humanReviewRequired = true |
| Evidence grading | Grade C/D or no-citation claims blocked |
| Honesty gate | Results below Dice 0.4 not surfaced as headline numbers |
| Mode | Scope |
|---|---|
research |
Dataset construction, physics benchmarking, model evaluation |
wellness |
Non-diagnostic self-baseline monitoring |
clinicalReview |
Decision support — licensed professional decides |
Clinical use requires a regulatory pathway (FDA GMLP / PCCP, PHIPA).
No real patient data is included in this repository.
| ADR | Decision |
|---|---|
| ADR-0001 | Zero-dependency Rust engine |
| ADR-0002 | C ABI WASM, 31 KB |
| ADR-0003 | Frozen physics, evolved harness |
| ADR-0004 | SART as default reconstruction |
| ADR-0007 | Uncertainty-first segmentation |
| ADR-0009 | Five acoustic classes only |
| ADR-0014 | Darwin multi-objective evolution |
| ADR-0018 | SaMD regulatory boundary |
| ADR-0019 | Medical Signal OS architecture |
| ADR-0026 | FWI roadmap for sub-mm resolution |
- Fork → branch → PR against
main - All four CI governance gates must pass
- New reconstruction methods require Dice / MAE regression tests
- New harness changes require a governance test in
packages/metabiohacker/test/ - Significant architectural decisions → new ADR in
docs/adr/
MIT — see LICENSE.
Sonic Chamber · Open-source acoustic digital human workbench · ruvnet/SonicChamber
Born from ruvnet/ruvector PR #595 · June 2026