Skip to content

v0.2.0

Choose a tag to compare

@Aakash-Tripathi Aakash-Tripathi released this 23 Feb 05:22
· 10 commits to main since this release

honeybee-ml v0.2.0

Highlights

  • Published in Nature Digital Medicine -- HoneyBee paper now published at nature.com/articles/s41746-025-02003-4
  • 313 tests with comprehensive mocked test suite (no GPU or model weights required)

New Features

WSI Pipeline (Pathology)

  • New Slide class with backend abstraction (auto-detects CuCIM or OpenSlide)
  • PatchExtractor for grid-based patch extraction with tissue filtering
  • Patches container with quality scoring, stain normalization, and immutable filtering
  • TissueDetector deep learning model for tissue segmentation
  • Backend abstraction layer (_backend.py) normalizing CuCIM vs OpenSlide APIs

Clinical NLP

  • Fully rewritten modular clinical processor (honeybee/processors/clinical/)
  • Document ingestion pipeline with PDF/OCR support
  • NER engine with pluggable backends (scispaCy, MedSpaCy, MedCAT, Transformers)
  • Ontology mapping via UMLS, BioPortal, and SNOMED CT (Snowstorm) clients
  • FHIR converter and HL7 parser for clinical interoperability
  • Temporal timeline extraction
  • Embedding engine with local and API backends

Model Registry

  • New model registry system (honeybee/models/registry/) with protocol-based provider architecture

Radiology

  • Overhauled preprocessing, segmentation, and processor modules
  • Removed legacy radiology loaders (functionality moved to processors)

Bug Fixes

  • Pathology processor bug fixes for embedding generation and aggregation
  • Clinical processor implementation fixes
  • Stain normalization cleanup (removed _fixed and _working variants)

Documentation

  • Simplified README
  • Expanded docs for clinical, pathology, and radiology processing
  • New pathology and radiology visualization assets on website

Breaking Changes

  • honeybee/processors/clinical_processor.py removed -- replaced by honeybee/processors/clinical/ package
  • honeybee/loaders/Radiology/ removed -- use honeybee/processors/radiology/ instead
  • honeybee/loaders/Scans/ removed
  • Legacy PathologyProcessor methods (load_wsi, detect_tissue, extract_patches, etc.) now emit DeprecationWarning and delegate to new classes

Install

pip install honeybee-ml==0.2.0