---
title: Pathology quickstart
description: Run a whole slide image through tile extraction or tissue detection and download the results.
---

You want tile-level and tissue-level readouts from a whole slide image — a grid of quality-filtered tiles, a tissue mask and coverage percentage, or a matrix of deep-learning features. This runs digital [pathology](/docs/pathology) analysis over your histology data end to end: open the slide, run the workflow you pick, and hand back downloadable artifacts.

## Before you start

Have your slide in the platform's workspace storage as an S3 URI — the tool reads it from there. Supported formats are SVS, TIFF, NDPI, and standard images (PNG/JPG). Decide which workflow you need (see step 2).

## Step-by-step

### 1. Upload or locate your slide

Make sure the whole slide image is in workspace storage. You will reference it by its `slide_url` (an `s3://…` URI).

### 2. Choose a workflow

- **tile_extraction** (default) — cut the slide into tiles and keep the ones that pass tissue and blur filters.
- **tissue_detection** — segment tissue from background and measure how much of the slide is covered.
- **feature_extraction** — extract ResNet-50 features from each retained tile.

### 3. Submit the job

Ask for the analysis in chat, naming the slide and workflow. For example:

> "Run tile extraction on my slide at s3://workspace/slides/case01.svs at 20x, tile size 256."

The orchestrator selects pathology analysis and fills in the parameters (see [inputs](/docs/pathology/inputs)).

### 4. Wait for processing

The job runs on GPU asynchronously and can take **10–60 minutes** depending on slide size and workflow. A progress phase and heartbeat appear while it runs; large slides are processed tile by tile.

### 5. Read and download results

When the job completes, an intelligence card appears on the canvas with the headline numbers, and the artifacts land in workspace storage for download — a `tiles.zip` archive, a `tissue_mask.png`, or a `features.npy` file depending on the workflow. See [outputs](/docs/pathology/outputs) for the full result shape.

## A worked example

You have an H&E slide and want a clean tile set for a downstream classifier:

1. Upload `case01.svs` to workspace storage.
2. Prompt: *"Extract tiles from s3://workspace/slides/case01.svs at 20x magnification, tile size 256, tissue threshold 0.5."*
3. The tool grids the slide, drops tiles that are mostly background (`tissue_percentage` below 0.5) or blurry (Laplacian variance below the blur threshold), and zips the survivors.
4. The card reports the tile count and slide dimensions; `tiles.zip` is available to download and feed into feature extraction next.

## Read next