Skip to content

SP Classify for QuPath

A cell classifier for QuPath 0.7, for highly multiplexed images, that learns from cells you label. It trains two models on the same labels (XGBoost and LightGBM by default) and lists the cells where the two models predict different classes, so you can check and correct those cells first. You then retrain and repeat. It runs inside QuPath and does not need Python.

Citing SP Classify

If you use this tool, please cite both the software and the CellTune paper it derives from, plus any method papers for the features you use. See How to cite.

What it does

  • Dual-model disagreement detection — two different models (default XGBoost + LightGBM; Random Forest also available) train on the same labels; the cells where they disagree are flagged for review. Training runs on the CPU.
  • Weighted uncertainty sampling — a 5-tier strategy (FOV balance → most-confused classes → rare types → user-specified confusions → random fill) puts review effort where it matters.
  • Interactive review mode — step through sampled disagreements and accept or correct a prediction in one click, with per-cell spatial context and optional auto channel-switching.
  • Beyond classification — marker-based gating, project-wide clustering (k-means & Leiden), cellular-neighborhood analysis, intensity heatmaps, batch normalisation (UniFORM), spatial distance measurements, and whole-image QC.

The active-learning loop

① Seed labels via Manual label mode
        │
        ▼
② Train dual classifiers (XGBoost + LightGBM)
        │
        ▼
③ View inter-model confusion matrix → per-class agreement rates
        │
        ▼
④ Enter review mode → step through disagreement cells → assign/correct labels
        │
        ▼
⟳ Merge new labels → retrain → repeat until satisfied

Getting started

  1. Install & launch the extension (QuPath 0.7, Java 25).
  2. Follow the multi-class quick start or the binary + composite quick start.
  3. Dig into the full multi-class workflow and review mode.

Relationship to CellTune

SP Classify provides similar functionality to parts of CellTune by the Keren Lab, reimplemented as a QuPath extension. It is an independent tool, not the official CellTune software.