OpenBNCT

English | 日本語

OpenBNCT is an open-source research workbench for boron neutron capture therapy (BNCT). Calculate component dose, estimate spatial and time-varying boron uptake, compare biological models, optimize multi-field research plans, and investigate dose uncertainty and prompt-gamma measurements.

It combines a deterministic neutron/photon transport solver with OpenMC integration and MCNP/PHITS interchange. The Rust implementation powers the CLI, Python bindings, desktop app, and browser workbench.

Handbook · First study · Benchmarks explained

Research software — not a medical device, not commissioned for any treatment facility, not for clinical decisions. See DISCLAIMER.md.

What researchers can compute

The capabilities below describe the current source tree; packaged releases can lag development. The usage reference documents command options and supported input formats.

Research task	Implemented capabilities
Component dose and dose-volume analysis	Separate boron, nitrogen, hydrogen, and photon fields; physical totals and available statistical uncertainties; DVHs, D95, Vx, EUD, and region summaries.
Neutron and photon transport	In-house multigroup discrete-ordinates (S_N) transport, P0–P5 neutron scattering, coupled photon transport, forward/adjoint solves, and OpenMC execution. CADIS/FW-CADIS weight windows support Monte Carlo variance reduction. The deterministic solver works without an external transport code.
Imaging and heterogeneous anatomy	DICOM CT, RT Structure Set, MR, and PET import; NIfTI volumes and labelmaps; rigid registration and resampling; HU-to-material calibration; RT Dose export and static-beam RT Plan import/export.
PET-derived boron fields	Voxelwise B-10 estimates from registered SUV images using ratio, calibrated-linear, or uniform uptake models; input-uncertainty propagation, washout correction, and material assignment for transport, or post-hoc application of blood concentration × tissue:blood ratios (or a PET field) to a unit-concentration boron dose under a trace-B-10 approximation. These are model-derived estimates.
Boron kinetics and irradiation timing	Fit mono-/biexponential concentration curves to samples, model changing tissue:blood ratios, compare irradiation windows under declared organ-dose limits, and emit time-integrated component-dose maps.
Multi-field and robust plan research	Beam-direction ranking by tissue path or adjoint importance; objective-driven field selection and iterative weight optimization; spectrum/aperture candidate sweeps and beamlet shaping. Evaluate named uptake, output, and positioning scenarios and optimize weights against their worst-case penalty.
Uncertainty and sensitivity	Systematic boron, positioning, and component uncertainties; first-order plan-metric uncertainty; supported ENDF MF33 nuclear-data covariances propagated to integrated dose-response budgets; Morris/Sobol screening of declared transport inputs.
Biological dose and response models	CBE/RBE component weighting; González–Santa Cruz photon-isoeffective, microdosimetric-kinetic (MKM), and stochastic-MK models; parameter sweeps and model comparisons; fractionation, BED/EQD2, combined-treatment analysis, and declared TCP/NTCP/UTCP models.
Cell-level boron microdosimetry	Compartmental boron localization, cell-to-cell uptake heterogeneity, stochastic captures and alpha/lithium tracks; nucleus specific-energy distributions, untouched-cell fractions, lineal-energy spectra, and population-survival comparisons.
Prompt-gamma reconstruction research	478 keV emission maps, adjoint detector responses with aperture acceptance, expected counts under a declared calibration, and regularized non-negative reconstruction of emission fields.
Beam and beam-shaping studies	Published or user-defined beam spectra, beam-quality metrics, a lithium-target accelerator-source model, beam-shaping assembly definitions, and thickness-sweep generation.
Independent comparison and interchange	Component-wise dose comparison, gamma analysis, analytic and metamorphic checks, and measurement comparisons; MCNP/PHITS deck export and output import; component NIfTI exchange and CSV/XLSX exposure plans.

The research layers have explicit limits. Scenario positioning currently shifts existing dose fields; it does not re-solve transport through moved anatomy. Optimization's isoeffective objectives use fixed component weights, while full nonlinear biological evaluation is a separate path. Current-source joint ensembles propagate explicitly declared shared/correlated sources and supported PK draws; they do not infer a complete patient uncertainty model. Retrospective delivery and measurement-informed boron workflows also require explicit clocks, calibration, observations and reduced-model assumptions. Prompt-gamma reconstruction remains an imaging research model. See the uncertainty guide and usage reference for current scope and remaining limits.

Cases, models, plans, and result artifacts are versioned and SHA-256-bound to their inputs. CLI, GUI, and Python surfaces reuse the Rust crates; individual features may reach these surfaces at different times.

Benchmarks and recorded evidence

The repository includes analytic checks, published transport problems, Monte Carlo comparisons, measured-phantom comparisons, and model-specific research fixtures. These are committed results for their declared inputs and solver versions, not a claim that every current configuration passes.

Case or suite	What the recorded evidence establishes	Scope and remaining work
NF-BNCT-001	A 600M-history OpenMC candidate passes the frozen case's statistical acceptance gates.	Candidate reference; independent transport reproduction is still required for reference promotion.
NF-BNCT-003 absorber	The recorded deterministic result recovers the analytic boron-dose attenuation slope of −0.2308 cm⁻¹ within its declared tolerance.	One-group, near-pure B-10 absorber; a check against a closed-form solution.
Reed, Azmy, and Kobayashi	Canonical tests of source deposition, heterogeneity, spatial/angular discretization, and void streaming. Reed and Azmy pass their case-specific checks; Kobayashi passes the graded near-field probes.	Kobayashi also records severe deep-field ray effects; its near-field pass does not qualify the full field.
Kobayashi MC comparison and layered-head MC comparison	Shared-case OpenMC comparisons separate transport-method differences from input differences.	The layered-head multigroup/continuous-energy discrepancy investigation remains open, including R17-04 group-boundary work.
FiR 1 K63 water, PMMA, and borated liquid	Comparisons with published/digitized phantom profiles under specified beam and material assumptions.	Normalized-profile agreement in selected fixtures does not establish absolute-dose agreement; multigroup results retain documented discrepancies.
Scenario optimizer oracle and FiR 1 scenario study	A known-answer robust-weight problem and a study of declared uptake/output/position perturbations.	The scenario set is not a calibrated probability distribution or a clinical robustness certificate.
PK timing study and cell microdosimetry study	Dose accumulation and cellular-model behavior under declared inputs drawn from published research.	Model checks using declared curves/distributions; not patient validation or independent measurements.
Prompt-gamma geometry study	A transported forward/inverse experiment showing the effect of collimation on localization.	Uncollimated inversion fails to localize; the collimated example still merges the two vial sources.

Conformance suites separately check interchange, adapter parsing, biological models, and endpoint calculations. Parser fixtures do not establish real-engine MCNP/PHITS agreement. NF-BNCT-002 supplies a frozen heterogeneous deep-penetration case; its transport execution remains pending. See validation and the roadmap for the evidence and remaining milestones.

The web build

The same Rust compiles to WebAssembly at openbnct.avilalabs.org. Drop a dose bundle, plan, NIfTI volume, or uncertainty budget onto the page; it is validated and routed to the right workspace. Nothing leaves the browser — there is no upload. The UI runs in English, 日本語, Italiano, 中文, and Español; requires WebGL2 or WebGPU (Chrome, Edge, Firefox work; LibreWolf needs WebGL enabled per-site). Case folders and process execution remain desktop-only.

Install

cargo install openbnct-cli               # CLI from crates.io
pip install openbnct                     # Python bindings from PyPI

Desktop builds are on the releases page; from source: cargo build --workspace, cargo run --bin openbnct-gui.

Try it

From a CT plus RT Structure Set to component dose, boron-scaled dose, DVHs and a report in two commands, no external codes needed. This runs the synthetic NF-BNCT-001 study (a research demonstration, not a patient case):

openbnct benchmark generate study/
openbnct project init --dicom study --output p001 --target CORE --spacing-mm 8
openbnct project run p001

The solve takes several minutes on a laptop with a release build. The report lands at p001/out/report.md (and report.json), with per-structure DVH curves in p001/out/dvh/*.csv. Every step is a hash-bound artifact under p001/out/; the report lists the exact command line of each, and rerunning skips steps whose inputs are unchanged. See docs/USAGE.md. The desktop app's Project tab runs the same workflow (DICOM folder or NIfTI CT volume in, report and Monte Carlo check out):

With OpenMC 0.16.0 and the ENDF/B-VIII.1 library installed, openbnct project verify p001 adds an independent continuous-energy Monte Carlo check of the same study and reports the agreement in the same report.

Accuracy status (2026-09-30): on the synthetic layered-head benchmark the default project workflow (beam histogram bins spread uniformly per eV as in OpenMC/MCNP, capture photons transported with sn photon-solve) gives structure-mean boron, fast-neutron ("hydrogen") and photon dose within about 15% of continuous-energy OpenMC run with the same S(α,β): whole phantom 0.86 / 0.98 / 0.93, target 1.06 / 0.90 / 1.03 (S8, 5e6 histories; the phantom has no nitrogen dose). With photon_transport = false the multigroup data deposits capture-γ energy locally and the photon component is ~3–4× high. One synthetic geometry is not a validation of your study; openbnct project verify runs this Monte Carlo check on any project. See the validation note.

Under the hood, the deterministic solver on its own — the shipped layered-head benchmark — then open the result in the workbench:

git clone https://github.com/AvilaLabs/OpenBNCT && cd OpenBNCT
openbnct sn solve --case benchmarks/synthetic/layered-head-phantom/case.json \
    --data benchmarks/synthetic/layered-head-phantom/multigroup-data-28g-v5-tsl.json \
    --assignment benchmarks/synthetic/layered-head-phantom/assignment.json \
    --source-weighting uniform_in_bin --dose dose.json --output flux.json

With the defaults (S4, P1, CMFD acceleration) this converges in 14 outer iterations — under a minute on 2 cores with a release build.

Or skip the terminal entirely: open openbnct.avilalabs.org and press Load example bundle in the dose workspace — the bundled benchmark artifact is built into the app.

To bring your own phantom: a segmented NIfTI labelmap plus a label→material table becomes a transport case in one command (import labelmap); a digitized facility spectrum becomes a beam-description with beam build. See docs/BYOC.md.

Status

Active research development. External reproduction, BNCT-physicist review, broader measured-data comparisons, and cross-institution execution remain milestones. Current capabilities and acceptance evidence are tracked in docs/ROADMAP.md; the research work handoff details proposed extensions for joint uncertainty, delivery replay, and measurement-informed boron estimation.

Documentation

Start with the OpenBNCT Handbook for installation, first studies, transport, boron/biology, planning, uncertainty, Python and benchmark interpretation. Its source is docs/guide/. It describes current-source capabilities and distinguishes them from older packaged releases and frozen evidence.

Detailed research records and reference material:

• docs/USAGE.md — command and workflow reference
• benchmarks/synthetic/nf-bnct-001/SPECIFICATION.md — the frozen case and its predeclared gates
• conformance/ — public fixture suites
• validation/ — transport and measured-phantom comparisons
• docs/RESEARCH_WORK_HANDOFF.md — scoped implementation packages and acceptance criteria for the next research work
• docs/adr/ — architecture decision records
• docs/research/ — technical baseline, cross-code recipe

Citing

CITATION.cff at the repo root gives the citation; GitHub's "Cite this repository" sidebar renders it.

Questions

Issues and discussions are open.

License

MIT. The repository must not implement Avify Dose patent subject matter without an IP review — see docs/IP_BOUNDARY.md.