Skip to content

Core Concepts

Data Hierarchy

Team
└── ImageSet (one or more)
    └── Image (one or more)
        └── Annotation (zero or more)
            └── Verification (zero or more)

Team

A Team is a group of users who share access to a set of ImageSets. Each team has one or more admins who can manage membership. Users can belong to multiple teams.

ImageSet

An ImageSet groups related images — for example, all slides from one study or one batch. Key properties:

Property Meaning
Public Anyone can view (read-only) without logging in
Public collaboration Anyone can annotate, not just team members
Priority Flags the set as high / low labeling priority
Image lock Prevents new uploads when set
Products Which annotation type groups apply to this set

Image

A single file inside an ImageSet. EXACT stores the file on disk and records metadata (width, height, resolution, number of frames, …) at upload time.

Multi-dimensional images are represented as:

  • Z-stacks — multiple focal planes or time points, navigated with a frame slider
  • NIfTI volumes — full 3D volumes with axial / coronal / sagittal MPR views

Annotation

An Annotation marks a region of interest inside an image. It has:

Property Meaning
Annotation Type Shape, label, and color
Vector The coordinates (bounding box corners, polygon vertices, …)
Frame Which frame / plane the annotation lives on
Concealed / Blurred Flags for special rendering
Verified Whether verification has been completed

Annotation Types and Products

An Annotation Type defines the shape and semantic label of a class of annotations. It belongs to a Product.

A Product is a named collection of annotation types for a particular study or labeling task (e.g., "Mitosis detection" with types "Mitotic figure" and "Mitosis-like"). Products are assigned to ImageSets to control which labels are available in the annotator.

Shape types

Type Description
Bounding box Axis-aligned rectangle
Circle Center + radius
Polygon Arbitrary closed or open contour
Line Open polyline
Point Single coordinate
Global Image-level label (no spatial extent)

Verification Workflow

The verification workflow gives a second annotator a chance to confirm or reject each annotation:

  1. Annotator A draws an annotation → status: unverified
  2. Annotator B opens the Verification view
  3. B clicks ✓ → status: verified (or ✗ → rejected)

Rejected annotations can be corrected and re-submitted for verification. This creates a lightweight quality-control loop without a separate review system.

Plugins

Plugins are server-side jobs that process images or annotations automatically. Typical uses:

  • Run inference with a trained model and insert predictions as annotations
  • Compute statistics over existing annotations
  • Generate heatmaps or density maps

Plugins are managed under Administration → Plugins and can be triggered manually from the imageset view or via the REST API.

REST API

EXACT exposes a full REST API at /api/v1/. Authentication uses token-based auth:

# Obtain a token
curl -X POST http://localhost:8000/api/auth/token/login/ \
  -H 'Content-Type: application/json' \
  -d '{"username": "exact", "password": "exact"}'

# Use the token
curl http://localhost:8000/api/v1/images/image_sets/ \
  -H 'Authorization: Token <your-token>'

See the API Reference for the full endpoint list.