Exporting results

12.1 Cell table export

Menu: Extensions → SP Classify → Export → Cell Table...

For each selected image, writes <ImageName>.csv to the folder you choose, with one row per detection:

Column	Notes
Image	Source image name
CellID	QuPath cell UUID
CentroidX_um / CentroidY_um	Centroid in microns, 2 decimals (pixel × calibration if no micron value)
Area_um2	Cell area in µm², 2 decimals
Classification	Current class (empty if unclassified)
ParentAnnotations	All ancestor annotations, joined with ;
ContainingAnnotations	Every annotation whose outline contains the cell centroid, joined with ;. Includes overlapping annotations that the hierarchy does not record as parents
Geometry_um / Geometry_px	(optional) WKT POLYGON of the cell outline, in microns or pixels. Written only when Export cell polygons (geometry) is ticked
feature columns	One column per measurement or text field you tick in the export dialog

Before exporting, the Select Columns for Cell Table Export dialog opens. It works like Select Features: search box, prefix dropdown, Select Prefix / Clear Prefix, Select All / Clear All. The whole-cell mean measurements and any distance measurements are ticked by default. Numeric measurements are listed first, followed by text fields such as CN Class. To add cell outlines, tick Export cell polygons (geometry) and choose Microns (µm) or Pixels under Units. Any value a cell does not have is written as NA.

12.2 Ground truth export & import

Ground-truth files hold your labelled cells and their feature values, so you can reuse labels in other projects or on other computers.

Export

Menu: Extensions → SP Classify → Export → Ground Truth...

Header (commented):

# SP Classify Ground Truth Export
# Image: my_image.ome.tiff
# Exported: 2026-06-02T14:30:45
Image,Label,CentroidX,CentroidY,Feature1,Feature2,...

Exports raw feature values for labelled cells only. Batch correction is not applied, even when the classifier uses it.

In multi-class mode the export includes labels from the current image and all other project images. In binary mode use Export → Active Binary Ground Truth... instead. It exports only the active marker and includes training rows imported from other projects, so the file can be moved between projects without losing rows.

Import

Menu: Extensions → SP Classify → Import → Ground Truth...

After choosing the CSV, pick one of two modes:

1. Spatial Match (per image) — each imported row is matched to the nearest detection by centroid distance, up to a maximum distance you set (default 20 px). Rows with no detection within that distance are skipped. Use this to re-import labels onto the same image they were exported from.
2. Training Data Only (cross-project) — imports the feature values and labels without matching them to cells. Use this when the source image is not in the current project. The rows are used in the next training run in the same way as labels on cells. Labels such as X-mergedInto(Y) are trained as Y. The sidebar shows the count as Imported rows: N.

For binary mode use Import → Active Binary Ground Truth.... It has the same two modes but applies only to the active marker.

Ground truth can only be exported and imported as single CSV files. There is no ZIP bundle option.

---