Install¶
A platform wheel ships the compiled engine — no build step, no system DICOM library, no codec plugins. Wheels are published for macOS (arm64, x86_64) and Linux (aarch64, x86_64; manylinux_2_28 — glibc 2.28+, i.e. RHEL 8+ / Ubuntu 20.04+ / Debian 11+). One wheel per platform covers CPython 3.12+ (stable ABI).
numpy is the only hard runtime dependency. Optional integrations are
imported lazily, so you install only what you use:
| Feature | Optional package |
|---|---|
DICOMDataset(to_torch=True) / training loops |
torch |
| NIfTI output consumers | none — NIfTI is written natively |
SimpleITK image in/out in radiomics / seg |
SimpleITK |
| Waveform analysis hand-off | mne, neurokit2 |
Note
pydcm distributes as wheels only (no sdist): the high-performance engine
ships as a compiled binary inside the extension, and parts of it are not
open source. Licensing: Apache-2.0 for pydcm; bundled third-party
components are listed in THIRD-PARTY-LICENSES (FFmpeg under LGPL-2.1
with a relink offer).