DICOM metadata validation, collection-consistent pseudonymization, UID remapping and verified binary rewriting in MoonBit.

dicom

de-identification

medical-imaging

moonbit

Download zip

Author

001-Elsa

Version

0.5.0

License

Apache-2.0

Repository

github.com/001-Elsa/MedSeal-MBT

Last updated

8 hours ago

Downloads

Dependencies

#MedSeal-MBT

MoonBit libraries and native tools for DICOM metadata conversion, linked-object consistency and independent output verification. 中文指南.

#0.5.0 data-processing libraries and tools

Add public api/query (typed paths, filters, grouping, CSV/JSONL and anonymous inventory), api/quality (object/collection diagnostics and caller-required fields), api/validation named configuration and immutable conversion plans, and Native api/native metadata scans and streaming closed-collection jobs without Python. The parser additionally accepts a pinned-dictionary Implicit VR Little Endian subset; binary output is normalized to Explicit VR Little Endian.

moon run --target native examples/data_pipeline
moon run --target native cmd/data -- summary INPUT_DIR
moon run --target native cmd/data -- quality INPUT_DIR
moon run --target native cmd/data -- export INPUT_DIR --output NEW.csv --format csv --fields 00080060,00280010,00280011
moon run --target native cmd/data -- convert INPUT_DIR --output NEW_DIR

Read the data-processing guide, 0.5 migration and boundaries, and implementation/acceptance ledger. The installation example below targets 0.5.0. Verify registry availability and the frozen source snapshot using the external release readback report described in the synchronization protocol. The following sections describe capabilities retained in the current candidate; historical release and measurement references keep their original versions.

#Community upgrade in source / 社区升级源码

C1 adds a complete portable public collection SDK: strict closed-batch conversion, bounded inputs, revalidated relationships and privacy-safe reports. The source consumer runs on Native, JS, wasm and wasm-gc; an independent pydicom oracle checks its actual output. These additions are included in the 0.4.1 community release. Run moon run --target js examples/consumer from the source checkout. See SDK usage and verification and 六阶段社区升级方案.

C2 adds an offline rule lab powered by the compiled MoonBit core. Build with python3.12 tools/build_community_lab.py, then open tmp/community-lab/index.html in a regular browser. It uses bundled synthetic objects and provides rule/structure views, failure experiments and public JSON export. 使用与验收边界: Node/DOM contracts pass; V5 also verifies actual Chromium rendering, CSP enforcement, downloads, keyboard activation and a 390px layout. 0.4.1 adds local read-only file inspection with an isolated Worker, explicit private text reveal, anonymous export and cancellation; no image conversion or release authority.

C3a adds public DICOM JSON interchange for 25 VRs, exact decimal token preservation, bounded parsing, and Part 10 reconstruction. A bidirectional pydicom oracle checks actual output. The full consumer now passes through JSON before strict collection conversion. 支持范围与示例. The 实际应用升级计划 records the implementation of dataset preflight, independent reception, measured workloads and usability. V3–V5 now include actual loopback Orthanc reception, licensed phantom fidelity, measured application costs and rendered-browser checks; see the linked plan. V2 adds a bounded dataset preflight with actionable, privacy-safe diagnostics: python3.12 tools/dataset_preflight.py INPUT. 操作说明. These additions are included in the 0.4.1 community release.

#Current scope: 0.5.0 source release

Entry	Implemented behavior	Platform / boundary
Public library	DICOM/JSON exchange, policy, query/filter/group/export, quality, immutable collection plans and privacy-safe explanation	Portable core; directory streaming and authorization providers require Native
Native data tools	Metadata summary/quality, private CSV/JSONL and strict streaming conversion	No Python required; per-job identity, new output destinations, cooperative cancellation
Collection CLI	Patient pseudonyms, identity UID mapping, full-date intervals, references and SHA-256 manifests	Per-job mappings; no persisted secrets in this compatible entry
Private host	Persistent project/source identities, incremental cache/dedup, delivered-history references, precise time handling, resumable sealed jobs, object review, Ed25519 packages and idempotent directory receipts	Python 3.12 + MoonBit Native, controlled POSIX deployment
Classic CT research	Single-frame uncompressed 16-bit MONOCHROME2, independent voxel/HU/geometry checks	Narrow ct-classic-research-v1; synthetic series tested
Local workbench	Loopback job/status/explanation and cancellation interface	Single-host capability/CSRF protection; no multi-user network service

Publication state and the contest remediation ledger are recorded in live competition remediation record. W0–W7 evidence separates implementation, synthetic verification and remaining field evidence. Earlier P/iteration audits are historical snapshots, not the current capability list.

#Install the 0.5.0 library

Install MoonBit using the official instructions. Verified baseline: moon 0.1.20260920 / moonc 0.10.14+7d59c7ec9. In an existing independent MoonBit module:

moon add 001-Elsa/medseal@0.5.0

Create an executable package app/moon.pkg:

import {
  "001-Elsa/medseal/api/policy",
}
pkgtype(kind: "executable")
supported_targets = "+native"

Create app/main.mbt:

///|
fn main {
  try {
    let policy = @policy.strict()
    ignore(policy)
    println("public-policy-ok")
  } catch {
    _ => println("public-policy-rejected")
  }
}

moon run --target native app

Expected output: public-policy-ok. Compiling a policy does not authorize privacy, pixel or release approval. Public import paths are api/dicom, api/dicom_json, api/policy, api/validation, api/explain, api/query, api/quality, and Native api/authorization / api/native in current 0.5.0 source. Versions before 0.5.0 do not provide the new data-processing packages. See API contracts; existing src/ imports remain compatible. This package install supplies the library, not a certified enterprise deployment.

#Run the source CLI

cd MedSeal-MBT
moon update
moon run --target native examples/collection

Use this local 0.5.0 candidate checkout or its source archive for the new APIs. The public repository URL alone does not establish that this candidate was pushed. The demo creates two synthetic patients, four files and two references, converts them and verifies the manifest. It prints a new tmp/demo-... directory. No real patient data is included. In the supplied workspace, tools/moon-local replaces moon using the sibling isolated toolchain without changing shell configuration.

moon run --target native cmd/main -- plan INPUT_DIR --policy examples/research-policy.json
moon run --target native cmd/main -- batch INPUT_DIR --output NEW_JOB_DIR --policy examples/research-policy.json
moon run --target native cmd/main -- verify NEW_JOB_DIR
moon run --target native cmd/main -- inspect samples/input.dcm --json
moon run --target native cmd/main -- validate samples/input.dcm --json
mkdir -p tmp
moon run --target native cmd/main -- anonymize samples/input.dcm --output tmp/single.dcm --report tmp/single.json --remove-private

plan checks the candidate collection without publishing outputs. batch stages anonymous DICOM names, per-object audits and a manifest, then publishes a new non-nested output directory. Source byte changes, identity conflicts and broken unauthorized references fail. verify checks manifest integrity and supported structure, not complete privacy. Basic anonymize keeps UIDs and fixed aliases; its separate output/report cannot be committed as a single directory transaction. See policies and support matrix.

#Persistent workflows and realistic examples

Use operations and offline installation for Python 3.12 private stores, create/run/resume, review/publish/deliver, backup and rotation. Incremental ledger explains source-domain boundaries, exact-repeat reuse and historical-reference checks. The workbench reuses these operations. Four complete scenarios describe batch integration, longitudinal research and embedding the public library, with executable evidence and explicit failure boundaries. They are intended uses supported by synthetic checks, not claims of hospital adoption.

#Verification and contribution

moon info
moon fmt --check
moon check --target all --deny-warn
moon test --target all --deny-warn
moon build --target native
moon package
python3.12 -m pip install -r requirements-qa.txt -r requirements-enterprise.txt
python3.12 tools/check_documentation.py
python3.12 tools/verify_data_processing.py --record tmp/data-processing-ci.json
python3.12 tools/verify_data_consumer.py --record tmp/data-consumer-ci.json
python3.12 tools/verify_with_pydicom.py
python3.12 tools/verify_ct_research.py --record tmp/ct-research.json
python3.12 tools/verify_release_readiness.py

Core CI covers Linux/macOS/Windows. POSIX host acceptance runs on Linux/macOS; Windows is not an enterprise deployment claim. Fixed corpus, generated-rule drift, negative cases and offline distribution verification are in CI. Records state their source revision; a configured workflow alone is not proof of success. See CONTRIBUTING, CHANGELOG and contribution boundaries.

#Privacy and limits

Part 10 Explicit VR Little Endian and pinned-dictionary Implicit VR Little Endian inputs, ASCII / ISO_IR 6 / ISO_IR 192 and documented Sequence/metadata rules are supported. Unknown/private implicit tags and ambiguous VRs are rejected. Binary writes normalize to Explicit VR LE; raw dataset parsing remains Explicit VR only. Classic CT has a limited research profile; MR/SC metadata checks do not establish full IOD conformance. Unsupported transfer syntax, public UID role, encoding and conflicting identities fail explicitly. Pixel bytes are preserved without OCR, redaction or diagnosis. Unknown public metadata and burned-in identifiers require review. Full Annex E anonymization, compressed pixels, PACS/DICOMweb and real field acceptance are not claimed. Security, streaming limits, field evidence gates.

Apache-2.0. Third-party sources and licenses. GitHub · GitLink · Mooncakes.

#Documentation and submission

全部文档索引 / document audit · 项目提交指南 · 申报事实与答辩提纲。姓名：韩桢。正式表单由本人提交；项目发布不代表已经取得报名回执。

文档核对：本文说明已与当前维护代码和实际入口核对；不扩大医疗/生产保证。见文档账本；正式申报由韩桢本人完成。

Current candidate maintenance: API compatibility, 0.5 migration, doctor, and clean-checkout release gate.

Site source Report issue Packages Build queue Skills Statistics