Classifier Model Management

Classifier models decide which detected cells or candidates should be counted as marker-positive or marker-negative for a staining.

Each staining has its own model selection. The selected model is used when you review output in Training, Results, Statistics, and classification exports.

Model sources

A staining can use different kinds of classifier models:

• Trained models: models trained from samples labeled in the current project
• Pre-trained models: ready-to-use models imported from the NeuroQP catalog
• Project models: models copied from another project in the same organization

Imported models are copied into the current staining. After import, they appear alongside models trained in the project.

When to import a model

Import a model when you want a faster starting point or when a similar project already has a reliable classifier.

A pre-trained model can be useful when the staining pattern matches a common workflow.

A project model can be useful when another project in your organization used similar tissue, staining, imaging settings, and labeling rules.

Always review imported model output on your own slices before using it for final analysis.

When to train a model

Train a model when an imported model does not match your images well enough or needs improvement for your specific dataset.

Training uses the current labeled samples for the selected staining.

The usual workflow is:

1. Label representative positive and negative examples.
2. Train a model.
3. Review the output on real slices.
4. Add targeted samples where the model makes mistakes.
5. Train a new model version.

Managing models

Open the model selector for a staining to manage its classifier models.

You can:

• Select which model is active for the staining
• Import a pre-trained model
• Import a model from another project in the same organization
• Rename trained or project-imported models
• Add comments to trained or project-imported models
• Mark useful models as favorites
• Delete models that are no longer needed

Pre-trained catalog imports keep their catalog name and description.

Deleting models

Deleting a model removes the model and its related classification output for that staining.

If Results, Statistics, or co-expression processing is currently running for that model, wait until processing finishes before deleting it.

Use deletion carefully once analysis has started, especially if you need to preserve the exact model used for a previous export or review.

Good practice

Use clear names and comments so the intended model is easy to recognize later.

For example, describe what changed between model versions, which project a copied model came from, or which labeling rule the model follows.

Before calculating Statistics or exporting results, confirm that the selected model is the one you want to use.