What slide formats are supported?

SVS, NDPI, and pyramidal TIFF open via histolab/OpenSlide; standard PNG/JPG images open via PIL as a fallback. The slide must live in workspace storage and is referenced by its s3:// URI.

What workflows can I run?

Tile extraction (the default), tissue detection, and feature extraction are the operational workflows. Classification and cell segmentation appear in the schema but are not implemented; jobs that request them fail with an "Unsupported workflow" error.

Why did tile extraction return fewer tiles than the grid?

Every tile is filtered on tissue coverage and blur. Tiles below tissue_threshold (mostly background) or below blur_threshold (too blurry) are discarded, so the retained set is often much smaller than the raw grid. Lower those thresholds to keep more.

What does feature extraction produce?

A per-tile deep-learning feature matrix from a pretrained ResNet-50, 2048 dimensions per tile, saved as features.npy along with the mean feature vector. Use it for clustering, similarity, or classification.

What is stain normalization for?

Histology stains vary between scanners and labs. Normalization (reinhard, macenko, or vahadane) harmonizes color so downstream models see consistent input. It is on by default.

How long does a run take, and does it need a GPU?

Runs are asynchronous and use GPU (A10G). Depending on slide size and workflow, expect 10–60 minutes. Large slides are processed tile by tile.

Can guests use pathology analysis?

No — it is an authenticated, professional-tier tool. Sign in with an account that has access.

Pathology faq

Ivy Biosciences.bloom* docs

login .bloom*

Loading...

Pathology faq

Frequently asked questions

Read next

Pathology outputs

Pathology troubleshooting

On this page