Any-Organ H&E Slide Analysis

What it does

Any-Organ H&E Slide Analysis reads a single H&E whole-slide image from any organ. It is the general-purpose slide read in the library: use it when no organ-specific model fits, or when you need slide-level measurements rather than a diagnosis.

The run checks the scan first: tissue mask, artefacts (blur, folds, pen marks, bubbles, tears) and an adequacy verdict. It then maps tumour probability across the slide, divides the tissue into compartments (tumour, stroma, inflammation, necrosis, other), segments and classifies every nucleus, and finds mitotic figures with a count in the 2 mm² hot spot. For a metastasis or a cancer of unknown primary it ranks the likely tumour type and primary site. Tile and slide embeddings come with the result for developers.

Intended use

Decision support and measurement. The result describes what is on the slide, and the primary-site ranking guides the choice of immunohistochemistry for a metastasis. Rare entities and lymphomas are the hard cases for an organ-agnostic tumour map.

Who it is for

Pathologists working up a metastasis of unknown origin, labs that want automatic scan QC before sign-out, and researchers and developers who need nuclei, mitosis and embedding outputs at slide scale through an API.

Accepted input

• One H&E slide per run: biopsy or resection, any organ.
• Formats: DICOM VL Whole Slide Microscopy Image (tiled multi-frame pyramid) or a vendor file (SVS, NDPI, MRXS, iSyntax, SCN, BIF, CZI, TIFF/OME-TIFF) converted on upload.
• Size: gigapixel slides of 0.5–5 GB are expected.

Requirements

Requirement	Why
H&E stain	IHC, cytology and haematology slides belong to other models
Pixel spacing in µm/px	Areas in mm², densities per mm² and the 2 mm² mitosis field depend on it
20× or 40× scan	Nuclei and mitoses are read at native magnification
Imaged area and pixel matrix in the metadata	Needed to place maps and annotations on the pyramid

Optional context

The organ, specimen type and clinical question can go in clinical_context. A known organ narrows the description; it does not turn the result into an organ-specific diagnosis.

The result

Section	Content
Slide quality	Tissue mask, artefact map and scan-adequacy verdict
Tumour detection	Tumour probability heatmap and tumour regions
Primary site	Ranked tumour types and likely primary sites for a metastasis or cancer of unknown primary
Tissue compartments	Map of tumour, stroma, inflammation, necrosis and other tissue, with area fractions
Nuclei	Instance masks classed as neoplastic, inflammatory, connective, dead (apoptotic or necrotic) or epithelial, with densities per mm²
Mitotic count	Typical and atypical mitotic figures and the count in the 2 mm² hot spot
Embeddings	Tile embeddings with coordinates and a slide embedding, as safetensors or HDF5 arrays
Draft report	Opt-in (options.report): an English narrative that interprets the result for the reader (findings, impression, limitations); every number is checked against the findings

Standards

• DICOM: tumour heatmap as Parametric Map; tissue and tumour regions as Segmentation, including Label Map Segmentation; nuclei and mitoses as Microscopy Bulk Simple Annotations; measurements in SR TID 1500; Encapsulated PDF.
• Counting: mitoses per 2 mm² following the WHO Classification of Tumours, 5th edition convention.
• Coding: SNOMED CT and ICD-O-3 for morphology and topography; FHIR R4 DiagnosticReport with Observation entries.

Embeddings for developers

The embedding arrays support slide search, cohort analytics and training your own slide classifiers without re-reading the image. Each array carries the tile coordinates and the encoder version it was computed with.