NEWS
dicomtk.ai 0.2.0
- IBSI Phase-1 Radiomics (
R/radiomics-ibsi.R): Added ibsi_digital_phantom() and ai_extract_radiomics() computing morphology, first-order intensity, 3D GLCM, 3D GLRLM, and 3D GLSZM features.
- Train/Test-Safe ComBat Harmonisation (
R/combat.R): Added combat_fit(), combat_apply(), and ai_combat_harmonize() for Empirical Bayes scanner batch harmonisation without test-set leakage.
- MONAI MAISI Synthetic Data (
R/maisi-synthetic.R): Added ai_generate_maisi() for 3D synthetic CT DICOM series + multi-organ label masks with DERIVED\\SECONDARY\\SYNTHETIC provenance tagging.
dicomtk.ai 0.1.0
- Initial release of
dicomtk.ai, the AI, MCP, and clinical NLP satellite package for dicomtk:
ellmer-based chat wrapper (ai_chat(), as_ellmer_chat()) and 5-stage clinical agent loop (ai_agent_plan(), ai_agent_dry_run(), ai_agent_approve(), ai_agent_execute(), ai_agent_verify()) with unbypassable monotonic de-identification guardrails.
- Native Model Context Protocol server (
dicomtk_mcp_tools(), start_mcp_server()) powered by mcptools.
- Pluggable AI backend registry (
register_ai_backend(), list_ai_backends()) with segmentation + voxel-wise uncertainty estimation (ai_segment()) and IHE AIR-aligned DICOM SEG + TID 1500 export (ai_export_results()).
- Clinical report de-identification (
ai_deid_report()), weak-label extraction (ai_extract_labels()), and RadLex / LOINC-RSNA protocol harmonisation (ai_harmonize_protocol(), radlex_playbook).
- Foundation study embeddings and
DuckDB similar-case retrieval (ai_embed(), ai_embedding_index(), ai_index_add(), ai_search_similar()).