Using the Training View

The Training view is where you label samples, train classifier models, import existing models, and review classifier output for one staining at a time.

Choose the staining and slice

Focus on one staining at a time because each staining has its own classifier model selection.

Move through slices to collect visual variety when staining pattern, intensity, or background changes.

Using 3 to 5 visually different slices is often a good starting point.

Understand the viewer

In cell counting and classification projects, you may be able to switch between the marker image and the Cell Detection marker image, such as DAPI.

In independent-detection projects, the viewer focuses on the selected marker staining. The DAPI toggle is hidden when there is no nuclei-source image to show.

Unclassified cells or candidates appear blue.

Classified examples appear green for positive and red for negative.

Label single cells or candidates

When you label a cell or candidate, it is added as a training sample.

You can relabel or remove samples when needed.

For independent-detection projects, remember that a negative example means rejected candidate or false positive, not a biological marker-negative cell.

Select and label many cells

Rectangle selection is the main workflow for collecting samples quickly.

When you draw a selection, NeuroQP randomly samples cells or candidates from that region instead of adding every visible object.

A practical workflow is to select an interesting region with around 20 to 50 cells or candidates and label the sampled objects as one class, then correct the sampled objects that belong to the other class.

Usually, you should not try to label every visible object in an area.

Use shortcuts to work faster

• 1 for positive
• 2 for negative
• Delete for remove
• arrow keys for slice navigation
• Esc to clear selection

The D shortcut appears only when a DAPI or nuclei-source image toggle is available.

Train or import a model

Training uses the current training samples for the selected staining.

You need at least 20 samples before NeuroQP allows training.

In practice, around 200 samples is a useful first target before the first model.

You can also import an existing classifier model instead of starting from training samples. Imported models can come from the pre-trained catalog or from another project in the same organization.

Review the selected model

After training or importing a model, review its output on real slices.

Switch between available models to compare behavior before choosing which model to use for Results and Statistics.