fastdicom 0.1.0

High-performance DICOM reader for ML pipelines

fastdicom

fastdicom is a Rust-backed Python extension module for high-throughput DICOM reads in ML and data processing pipelines.

It is built with PyO3 + maturin and uses dicom-rs for metadata parsing and pixel decoding.

Quickstart

python3 -m maturin develop

import fastdicom

ds = fastdicom.dcmread("study.dcm")

rows = ds.Rows
cols = ds[(0x0028, 0x0011)]
spacing = ds.get("PixelSpacing", None)
ts_uid = ds.file_meta.TransferSyntaxUID
pixels = ds.pixel_array

Implemented API surface

• fastdicom.dcmread(path, *, stop_before_pixels=False) -> Dataset
• Dataset access patterns:
  • ds.Rows
  • ds[(0x0028, 0x0010)]
  • ds["Rows"]
• Dict-like operations:
  • get, in, iteration, keys, values, items
• Properties:
  • pixel_array
  • file_meta
  • file_path
• Exception classes:
  • DicomError
  • DicomReadError
  • TagNotFoundError
  • PixelDataError
  • UnsupportedTransferSyntaxError

Compatibility notes

• Tag access returns direct values, not pydicom DataElement wrappers.
• Date/time VR values remain strings (no automatic datetime conversion).
• PixelData is not directly exposed through tag lookup; use .pixel_array.

Explicit non-goals (v1)

• No write/mutation APIs (save_as, Dataset mutation)
• No DICOMDIR support
• No file-like object input (path-only read)
• No DIMSE/DICOMweb networking APIs
• No deferred reads / lazy file handles
• No overlay/curve/waveform features

Development checks

cargo check
python3 -m maturin develop
pytest -q