# MedSeal-MBT MedSeal-MBT is a small DICOM metadata validation and basic de-identification tool written in MoonBit. It parses DICOM Part 10 files using Explicit VR Little Endian, validates basic file structures, removes or replaces common direct identifiers, optionally removes private elements, preserves pixel data, and writes a new DICOM file with a JSON audit report. ## Features - DICOM Part 10 file parsing - Explicit VR Little Endian support - Common text and integer VR decoding - Defined-length and common undefined-length nested Sequence support - Structural and basic image-metadata validation - Basic metadata de-identification - Optional private tag removal - Pixel Data byte preservation - Staged DICOM rewriting with parse-after-write verification - JSON audit reports that never include original sensitive values - Native command-line interface ## Build and test MoonBit version `0.1.20260724` or newer is recommended. ```bash moon fmt --check moon check --target all --deny-warn moon test --target all --deny-warn moon build --target native moon info moon package ``` ## Usage Inspect a file without printing patient values: ```bash moon run cmd/main inspect samples/input.dcm moon run cmd/main inspect samples/input.dcm --json ``` Validate a file: ```bash moon run cmd/main validate samples/input.dcm moon run cmd/main validate samples/input.dcm --json ``` Perform basic metadata de-identification: ```bash moon run cmd/main anonymize samples/input.dcm \ --output samples/output.dcm \ --report samples/report.json \ --remove-private ``` An existing output is rejected unless `--force` is supplied. The input path can never be used as the output path. The writer stages the result, reparses and validates it, then atomically replaces the destination. Other commands: ```bash moon run cmd/main --help moon run cmd/main inspect --help moon run cmd/main validate --help moon run cmd/main anonymize --help moon run cmd/main --version ``` Exit codes are `0` for success, `1` for validation failure, `2` for command errors, `3` for unsupported DICOM formats, `4` for file I/O errors, and `5` for de-identification or write-back failures. ## Built-in policy Version 0.1.1 replaces Patient Name with `ANONYMOUS`, replaces Patient ID with `MEDSEAL-ID`, and empties the configured birth date, institution, address, physician, operator, device serial number, and related direct identity fields. Patient Sex is retained. Study Description, Series Description, and Protocol Name are retained with warnings. UIDs are retained to avoid breaking object relationships. Private elements are retained with a risk warning by default. Use `--remove-private` to remove elements whose group number is odd. Original private values are never written to the report. ## Important limitations MedSeal-MBT does not inspect pixel data for burned-in personal information. It does not provide medical diagnosis and does not guarantee complete anonymization. Version 0.1.1 supports only Explicit VR Little Endian files. It does not implement Implicit VR, Explicit VR Big Endian, compressed image decoding, DICOM networking, PACS, DICOMweb, OCR, pixel-region redaction, batch directory processing, or the complete DICOM standard. Unsupported Transfer Syntax UIDs are rejected explicitly; the parser never guesses. The limits are 256 MiB per file, 128 MiB per element, 100,000 elements, and 32 Sequence nesting levels. ## Privacy and safety The audit report records only tag, field name, and action. It never records the original field value. The CLI does not print patient names or identifiers. Only synthetic files generated by this repository are committed under `samples/`. See [SECURITY.md](SECURITY.md) and [samples/README.md](samples/README.md). ## Mooncakes The module is `001-Elsa/medseal`, version `0.1.1`, with `native` as the preferred target. It is published at [mooncakes.io/docs/001-Elsa/medseal](https://mooncakes.io/docs/001-Elsa/medseal). The CLI entry package is `cmd/main`. Release acceptance evidence is documented in [ACCEPTANCE.md](ACCEPTANCE.md). Competition submission metadata is in [PROJECT_APPLICATION.md](PROJECT_APPLICATION.md). Architecture and provenance are documented in [ARCHITECTURE.md](ARCHITECTURE.md) and [THIRD_PARTY.md](THIRD_PARTY.md). ## License Apache-2.0. See [LICENSE](LICENSE).