Qupath set perameters

Sorry, it’s trouble again.
I want to know in the qupath, in the add intensity feature parameter setting, region choose roi, square tiles, Circular tiles, cell nucleus, under what circumstances to choose different regions, such as whether I should choose in the process of cell detection cell nucleus, and other options such as square tiles and circular tiles, are they the choices for detecting superpixel segmentation, and as to whether the ROI is the entire annatation that I marked, I don’t know if I understand it this way, right?
Regarding the setting of another parameter, in run compute intensity features, process regions will pop up,will it be different to choose different regions? I don’t quite understand the difference between cell and detections.
I look forward to your answer and thanks a lot.

Hi @day0,

I’m not sure I understood correctly your questions. But what I can tell you is this:

  • Cells are a type of Detection. E.g. after running Cell detection, you will get cell objects, which are also detections. On the other hand, if you use say the Pixel classifier, you can choose to get detections but not cells objects, as the objects created from it do not represent cells. You can check the documentation for more info about this.
  • If you have cells and want to add some intensity feature to them, you can choose the Cell nucleus option in the Compute intensity features window, then choose to process all Cells.

I tend to use the Circular tiles with an area slightly larger than my objects if I need to do some sort of background subtraction using the local environment (occasionally to remove subcellular detections in messy ISH staining).

For SLICs, I have used it to look at a specific area surrounding the centroid of the SLIC, since smoothing for SLICs can sometimes have… interesting results when the shapes are allowed to be flexible. Note that, in the case of long thin SLICs, a small circular area may not actually cover the entire SLIC, and centroids are weighted by the object (not always the center of the bounding box).

I don’t think I have ever used a square tile for a measurement, but maybe someone else will have ideas for that :slight_smile:

In the Process regions dialog, you are choosing the objects to which measurements will be added.

In the Compute intensity features dialog you are choosing where the measurements are made relative to each object.

The right choices really depend upon what exactly you want to measure.

In the example below, I can choose to measure the selected object, in which case I use only pixels inside the yellow boundary (the ROI option under Compute intensity features).

Alternatively, I can choose to make measurements in a circle or square surrounding the ROI – which may be bigger or smaller than the ROI itself. These are the ‘tiles’ options. No new tile objects are explicitly created, but QuPath uses them internally to figure out which pixels to include in the measurements.

If I want to measure the staining/texture within the annotated nucleus (for example), then I’d choose ROI. But if I want to create texture features that include the surrounding area (e.g. to help classify the cell as tumor or non-tumor) then I might choose a larger square or circle – because this will give some extra information about the local neighborhood to the classifier.

The ability to separate the size of the region used to make measurements from the size of the objects in QuPath helps mostly with classifying small regions. For example, in this image below I created some superpixels at high resolution. If I calculate textures only within the superpixels then I probably don’t have enough information for each superpixel to classify it very accurately according to the kind of tissue it is part of. If can help if each superpixel has texture information that extends beyond its own ROI.

If you have cell objects, you have two ROIs to choose from (one for the nucleus and one for the full cell), but the idea remains the same. The ROI option refers to the whole cell, the Cell nucleus option to the pixels inside the nucleus only.

1 Like