< Back to QuPath article list

Putting it all together

Take all of the steps and put them together into a workflow!

Finally, we are taking everything we have learned and putting it to use!

This is currently an incomplete skeleton, but provides a framework for the use of all of the information provided in the previous pages. If you have questions about making this work, please post on the image.sc forum! If there are enough requests, we might fill this out a bit more.

Putting it all together

  1. Start with a project

  2. Generate cells throughout the project, duplicate the folder/project

  3. Train a pixel classifier using annotation objects for Tissue/Ignore

    1. Use orion4 5 and 6

    2. RT w/variables

    3. Moderate

    4. All channels, all resolutions

    5. First 4 measurements => Gaussian+weighted

    6. balanced classes

    7. load training

  4. Create a thresholder for CD31 areas

    1. Very high

    2. Gaussian 1

    3. Thresh 600=>1000

  5. Train a cell classifier

    1. Add measurement smoothing

    2. B-cells

    3. T-cells

    4. macrophages

    5. Endothelial/Epithelial cells

    6. Create thresholds for PD-L1 and add that into a composite

  6. Transfer the classifiers back over to the main project.

  7. Assemble a script

    1. Add in a distance to annotations2D

  8. Run the script and export the data

  9. Export rendered images

  10. Take a quick look over the exported data

    1. Include some extra scripts for things like the average distance per cell type for the tissue annotation in the text? (See the image.sc forum for examples)

Object visualization

The user interface has a decent selection of visualization options, including turning the three main types of objects on or off, toggling the pixel classifier overlay, toggling names, and a slider that controls the transparency of most on screen objects, though not the currently selected object.

  1. Measurement Maps: All detections have some set of measurements associated with them, and measurement maps can be a nice way of visualizing those measurements. As you gain more comfort with the interface and scripting, this can become a powerful tool as you can create new measurements related to things like the overlap or correlation between multiple channels, which might otherwise be difficult to visualize with the channel viewer across your entire sample.

  2. Density Maps: See the main QuPath docs

  3. Export images using options discussed here.

  4. Export data as described in the main documentation. As there are many different ways to use/parse the final data that tend to be individual to any given project, I suggest asking on the forum if you have specific questions not addressed in the documentation.