pyXenium.multimodal.run_histoseg_lazyslide_structure_workflow#

run_histoseg_lazyslide_structure_workflow(sdata_or_path, *, output_dir=None, contour_key='histoseg_structures', contour_geojson=None, contour_id_key='polygon_id', contour_coordinate_space='xenium_pixel', contour_pixel_size_um=None, he_image_key='he', he_source_path=None, wsi_reader=None, slide_mpp=None, model='plip', text_model=None, text_terms=None, prompt_set_name='breast_histology_v1', prompt_source='manual exploratory prompt set', prompt_review_status='not pathologist-confirmed', relative_prompt_axes=None, tile_px=224, mpp=0.5, device='cuda', amp=True, batch_size=64, max_tiles=None, table_format='csv', include_rna=True, include_wta_programs=True, include_boundary_programs=True, include_prediction_benchmark=True, wta_program_library='breast_tme_wta_v1', program_library='tumor_boundary_v1', rna_markers=None, lazy_backend=None, precomputed_tile_features=None, precomputed_feature_table=None, precomputed_program_scores=None)#

Run a HistoSeg structure-to-H&E feature workflow with optional LazySlide.

HistoSeg owns segmentation and structure proposals. LazySlide owns WSI tile extraction and image-model inference when the optional backend is used. pyXenium owns coordinate alignment, structure-level aggregation, and RNA/image interpretation artifacts.

Parameters:

  • sdata_or_path (XeniumSlide | str | Path)

  • output_dir (str | Path | None)

  • contour_key (str)

  • contour_geojson (str | Path | None)

  • contour_id_key (str)

  • contour_coordinate_space (str)

  • contour_pixel_size_um (float | None)

  • he_image_key (str)

  • he_source_path (str | Path | None)

  • wsi_reader (str | None)

  • slide_mpp (float | None)

  • model (str)

  • text_model (str | None)

  • text_terms (Sequence[str] | None)

  • prompt_set_name (str)

  • prompt_source (str)

  • prompt_review_status (str)

  • relative_prompt_axes (Sequence[Sequence[str]] | None)

  • tile_px (int)

  • mpp (float)

  • device (str)

  • amp (bool)

  • batch_size (int)

  • max_tiles (int | None)

  • table_format (Literal['csv', 'parquet'])

  • include_rna (bool)

  • include_wta_programs (bool)

  • include_boundary_programs (bool)

  • include_prediction_benchmark (bool)

  • wta_program_library (str)

  • program_library (str)

  • rna_markers (Sequence[str] | None)

  • lazy_backend (Any)

  • precomputed_tile_features (DataFrame | str | Path | None)

  • precomputed_feature_table (Mapping[str, Any] | None)

  • precomputed_program_scores (DataFrame | str | Path | None)

Return type:

dict[str, Any]