Design metrics as precise as your question

Measure any property, per object

Every detected object comes with a default set of measurements — but you're not limited to them. Add or remove metrics per model without retraining:

• Common metrics: Position (X/Y), area, perimeter, average intensity per channel, solidity, eccentricity, major/minor axis length, curved length, model confidence score.
• Advanced metrics: Equivalent diameter, Euler number, extent, Feret diameter, orientation, area (bounding box, convex, filled), colocalization (Pearson, Manders), and area threshold — quantify the signal above or below a specific brightness/contrast cutoff on any channel.

Adjust the confidence threshold per model to predict more or fewer objects, directly from a slider — no need to retrain to tune sensitivity.

Build hierarchies between object classes

Class hierarchy — define parent-child relationships to answer questions like "how many of class X are inside class Y":

• Organoid → cell (cells per organoid)
• Cell → puncta (puncta per cell)
• Tissue region → positive cell (positive cells per region)
• Tumor → immune cell (immune infiltration)

Size group ROI hierarchy — constrain where a child class is even segmented, so it only runs inside its parent region (e.g., cells only segmented within an organoid, positive cells only within a tissue region). This isn't just cleaner data — it saves analysis time and credits by skipping segmentation outside the region that matters.

Grow, shrink, and transform objects

Expand or contract detected objects by a set number of pixels to answer questions segmentation alone can't:

• Grow — associate cytoplasmic or extracellular signal with a nucleus, or connect nearby puncta to their nearest cell.
• Shrink — remove autofluorescence or edge artifacts near a boundary.
• Hollow or filled — turn a grown/shrunk region into a ring or a solid shape, then rename and recolor the resulting class.

See some examples