Deep learning

If you want to test a trained model, its application doesn`t differ from the use of other filters.

Example: excluding non-stromal structures

Task: highlight all non-stromal structures inside ROI.

ROI consisting of glands and stroma

For this task, we have previously trained a simple U-net model, named “weights.h5”

The model is then applied as a filter. The steps are as follows.

  • Select filter ( # 6 by default).
  • Expand, to see the properties.
  • Set path to the trained model

Apply the filter inside the annotation.

All non-stromal component are highlighted using a simple U-net trained model

Such a simple model can be used to quantify the precise percentage of a tumor inside ROI.

NOTE: application of deep learning models is done at level 2 by default. You might get a better result if changed to 1 or 0; the higher the resolution, the more resources are needed to make a calculation.

NOTE: the precision of the result depends on many factors, as the quality of the slide, level of scanning, etc. The more different samples are used for training, the better the results are.