Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.1.1] - 2026-04-09

New preset mammography/learning_raw_uint16: same pipeline as learning_raw but exports uint16 npy for ~50% smaller training files
CLI help lists the new learning_raw_uint16 profile

ExportDtype.UINT16 cast now preserves raw pixel values via clip-to-[0, 65535] instead of per-image min-max rescaling. This keeps cross-image calibration intact (critical for training on raw DICOM). Out-of-range values are clipped with a warning.

Initial release of medip
Modality-extensible pipeline architecture: BasePipeline + MammographyPipeline
Profile-driven pipeline with modality-scoped presets: mammography/learning_raw, mammography/learning_presentation, mammography/clinical_display
DICOM-standard transform chain: padding -> modality LUT -> photometric -> VOI -> geometry -> mask -> crop -> export
DicomMetadata dataclass with 20+ DICOM tags for pixel interpretation
pydicom-based authoritative pixel decoding
8 export formats: npy, npz, png8, png16, tiff16, jpeg, dicom (Secondary Capture), nifti
MONOCHROME1 auto-inversion
Pixel Padding Value/Range masking
FOR PROCESSING / FOR PRESENTATION auto-detection (mammography)
Multi-window DICOM support (window index selection)
SpacingPolicy.AUTO: resample to target spacing (default 0.07mm) when PixelSpacing exists in DICOM header
SimpleITK-based resampling with configurable interpolators
Otsu + morphology mask extraction (research/experimental)
Bounding box crop with mm/px margin support
Metadata JSON sidecar export
CLI commands: run, inspect, dump-metadata, validate, presets
CLI --modality flag for pipeline selection
YAML config file support with profile + override pattern
Parallel processing with configurable worker count
Optional dependencies: [cli], [jpeg], [itk], [nifti], [all], [dev]