Dicom
Pure Elixir DICOM toolkit focused on DICOM Part 10 files. Zero runtime dependencies.
Built on Elixir's binary pattern matching for fast, correct parsing of DICOM medical imaging files.
Features
- P10 file parsing -- read DICOM Part 10 files into structured data sets
- P10 file writing -- serialize data sets back to conformant P10 files
- Streaming parser -- lazy, event-based parsing for large files and pipelines
- Data dictionary -- comprehensive PS3.6 tag registry (5,035 entries) with VR, VM, keyword lookup, and retired flags
- DICOM JSON -- encode/decode DataSets to/from the DICOM JSON model (PS3.18 Annex F.2) for DICOMweb, with strict decode errors and explicit
BulkDataURIresolution - Pixel data frames -- extract individual frames from native and encapsulated pixel data (PS3.5 Section A.4)
- De-identification -- best-effort PS3.15 Basic Profile helpers with supported-tag cleaning, consistent UID replacement, and an explicit
retain_private_tagsswitch for retaining all private tags - Character set support -- decode text values for supported single-byte Specific Character Set repertoires plus UTF-8; ISO 2022 escape-sequence switching is not implemented
- Value decoding -- automatic VR-aware decoding (numeric, string, date, UID, etc.)
- SOP Class registry -- 232 SOP Classes (183 storage + service/Q-R/print/worklist) with modality mapping, retired flags, and O(1) lookup
- Transfer syntaxes -- 49 transfer syntaxes tracked by the library (34 active + 15 retired); strict rejection of unknown UIDs with opt-in lenient mode
- Sequences -- defined-length and undefined-length SQ with nested items
- Encapsulated pixel data -- fragments with Basic Offset Table
- Validation -- File Meta Information validation per PS3.10 Section 7.1
- Zero dependencies -- pure Elixir, no NIFs, no external tools
Scope
This library is strongest in DICOM file and data-set workflows:
- PS3.10 read/write for Part 10 files
- PS3.5/PS3.6 value, VR, transfer syntax, dictionary, sequence, and pixel data helpers
-
PS3.18 Annex F.2 DICOM JSON conversion for DataSets, including resolver-based
BulkDataURIdecode
It is not a full DICOM stack. In particular:
- It does not implement DIMSE networking or provide a DICOMweb server
- It preserves encapsulated pixel payloads and frame boundaries, but it does not decode JPEG/JPEG 2000/JPEG-LS/MPEG/HEVC codec bitstreams
- De-identification support is a best-effort helper over the library's supported tag/action set, not a regulatory or standards-conformance guarantee
retain_private_tagsretains all private tags; this library does not claim PS3.15 safe-private evaluation
Installation
Add dicom to your mix.exs dependencies:
def deps do
[
{:dicom, "~> 0.5.0"}
]
endQuick Start
# Parse a DICOM file
{:ok, data_set} = Dicom.parse_file("/path/to/image.dcm")
# Access attributes by tag
patient_name = Dicom.DataSet.get(data_set, Dicom.Tag.patient_name())
study_date = Dicom.DataSet.get(data_set, Dicom.Tag.study_date())
modality = Dicom.DataSet.get(data_set, Dicom.Tag.modality())
# Decode values with VR awareness
raw_element = Dicom.DataSet.get_element(data_set, Dicom.Tag.rows())
rows = Dicom.Value.decode(raw_element.value, raw_element.vr)
# Build a data set from scratch
ds = Dicom.DataSet.new()
|> Dicom.DataSet.put({0x0002, 0x0002}, :UI, "1.2.840.10008.5.1.4.1.1.2")
|> Dicom.DataSet.put({0x0002, 0x0003}, :UI, Dicom.UID.generate())
|> Dicom.DataSet.put({0x0002, 0x0010}, :UI, Dicom.UID.explicit_vr_little_endian())
|> Dicom.DataSet.put({0x0010, 0x0010}, :PN, "DOE^JOHN")
|> Dicom.DataSet.put({0x0010, 0x0020}, :LO, "PAT001")
# Serialize to binary and write
{:ok, binary} = Dicom.write(ds)
:ok = Dicom.write_file(ds, "/path/to/output.dcm")
# Parse from binary
{:ok, parsed} = Dicom.parse(binary)
# DataSet bracket access and Enumerable
patient = data_set[Dicom.Tag.patient_name()]
tags = Enum.map(data_set, fn {tag, _elem} -> tag end)
# Tag parsing and date/time conversion
{:ok, tag} = Dicom.Tag.parse("(0010,0010)")
{:ok, date} = Dicom.Value.to_date("20240115")
# Inspect for quick debugging
IO.inspect(data_set)Streaming
# Stream events lazily from a file (constant memory)
events = Dicom.stream_parse_file("/path/to/large_image.dcm")
# Tune file read-ahead when needed
events = Dicom.stream_parse_file("/path/to/large_image.dcm", read_ahead: 8_192)
# Filter for specific tags without loading the entire file
patient_tags =
events
|> Stream.filter(&match?({:element, %{tag: {0x0010, _}}}, &1))
|> Enum.map(fn {:element, elem} -> {elem.tag, elem.value} end)
# Or materialize back into a DataSet
{:ok, data_set} =
Dicom.stream_parse(binary)
|> Dicom.P10.Stream.to_data_set()Architecture
lib/dicom/
dicom.ex -- Public API: parse, write, stream_parse, stream_parse_file
data_set.ex -- DataSet struct (elements + file meta)
data_element.ex -- DataElement struct (tag + VR + value + length)
tag.ex -- Tag constants and utilities
vr.ex -- Value Representation types and padding
uid.ex -- UID constants, generation, and validation
value.ex -- VR-aware value encoding and decoding
transfer_syntax.ex -- Transfer syntax registry (49 TSes) and encoding dispatch
sop_class.ex -- Dicom.SOPClass registry (232 classes) with modality mapping
character_set.ex -- Specific Character Set decoding for supported single-byte repertoires and UTF-8
character_set/
tables.ex -- ISO 8859-{2..9} and JIS X 0201 lookup tables
json.ex -- DICOM JSON model encoder/decoder (PS3.18 Annex F.2)
pixel_data.ex -- Pixel data frame extraction (PS3.5 Section A.4)
de_identification.ex -- De-identification / anonymization (PS3.15 Table E.1-1)
de_identification/
profile.ex -- Profile options struct (10 boolean columns)
p10/
reader.ex -- P10 binary parser (preamble, file meta, data set)
writer.ex -- P10 binary serializer (iodata pipeline)
file_meta.ex -- Preamble validation and File Meta Information
stream.ex -- Streaming API: parse/1, parse_file/2, to_data_set/1
stream/
event.ex -- Event type definitions
source.ex -- Data source abstraction (binary + file I/O)
parser.ex -- State machine: preamble -> file_meta -> data_set -> done
dictionary/
registry.ex -- PS3.6 tag -> {name, VR, VM} lookup (5,035 entries)DICOM Standard Coverage
| Part | Title | Coverage |
|---|---|---|
| PS3.4 | Service Class Specifications | 232 SOP Classes (storage, Q/R, print, worklist, etc.) with modality mapping |
| PS3.5 | Data Structures and Encoding | VR types, transfer syntax handling, data encoding, sequences, pixel data frame extraction |
| PS3.6 | Data Dictionary | Comprehensive tag registry (5,035 entries), keyword lookup, retired flags |
| PS3.10 | Media Storage and File Format | P10 read/write, File Meta Information, preamble |
| PS3.15 | Security and System Management | Best-effort Basic Application Level Confidentiality Profile helpers for the supported tag/action set |
| PS3.18 | Web Services | DICOM JSON model encoding/decoding for DataSets (Annex F.2) |
Transfer Syntaxes
| Transfer Syntax | Read | Write |
|---|---|---|
| Implicit VR Little Endian (1.2.840.10008.1.2) | Yes | Yes |
| Explicit VR Little Endian (1.2.840.10008.1.2.1) | Yes | Yes |
| Deflated Explicit VR Little Endian (1.2.840.10008.1.2.1.99) | Yes | Yes |
| Explicit VR Big Endian (1.2.840.10008.1.2.2, retired) | Yes | Yes |
| Other registered compressed and video transfer syntaxes | Metadata only | Metadata only |
Unknown transfer syntaxes are rejected by default. Use TransferSyntax.encoding(uid, lenient: true)
to fall back to Explicit VR Little Endian for unrecognized UIDs.
Performance
Benchmarked on Apple Silicon (Elixir 1.18, OTP 27):
| Operation | Throughput |
|---|---|
| Parse 50-element data set | ~10 us |
| Parse 200-element data set | ~50 us |
| Stream parse 50 elements | ~20 us |
| Stream parse 200 elements | ~80 us |
| Stream enumerate 200 elements | ~55 us |
| Write 50-element data set | ~13 us |
| Write 200-element data set | ~55 us |
| Roundtrip 100 elements | ~37 us |
| Parse 1 MB pixel data | ~1 us |
Run benchmarks with mix test test/dicom/benchmark_test.exs.
Testing
mix test # Run all tests (1000+ tests)
mix test --cover # Run with coverage report
mix format --check-formattedProperty-based tests using StreamData verify encode/decode roundtrips across all VR types and streaming parser equivalence.
Project Positioning
dicom is aimed at file-centric DICOM workflows in Elixir: parse, inspect,
transform, write, stream, and validate Part 10 objects without native code or
external tooling.
That means the library is a strong fit for ingestion pipelines, metadata processing, archive tooling, DICOM JSON conversion, and controlled de-identification passes over known data. If you need DIMSE networking, a full codec stack for compressed pixel payloads, or formal privacy/compliance validation, those concerns should sit alongside this library rather than inside it.
For DICOM JSON specifically, BulkDataURI entries are not treated as raw bytes.
Use Dicom.Json.from_map/2 with bulk_data_resolver: when you want to resolve
external bulk data during decode.
AI-Assisted Development
This project welcomes AI-assisted contributions. See AGENTS.md for instructions that AI coding assistants can use to work with this codebase, and CONTRIBUTING.md for our AI contribution policy.
Contributing
Contributions are welcome. Please read our Contributing Guide and Code of Conduct before opening a PR.
License
MIT -- see LICENSE for details.