When editing/checking segmentation it would be very useful to have a annotation filter on Nessys for roundness to weed out incorrectly segmented gaps between cells which are naturally edgier than real cells.
Ideally, you would rely on your segmentation tool for this. If you are using Nessys, the classifier that you train should get rid of the sort of shapes you are talking about.
But I agree, it would be nice to have additional shape descriptors. Other descriptors already exist (notably compactness which could fit your purpose maybe), available via the ‘compute feature’ Task.
One possibility would be to launch a dialog to ask the user which additional descriptors should be computed when launching the editor…
Regarding roundness, can you be more precise? Do you have a screenshot?
I think because the sample I am segmenting is quite complex the training of the classifier becomes a trade off between over or under segmentation of areas, and ultimately in this case I was left with a lot of segmented gaps.
Attached is an example of what I am talking about. The ‘edgy’ gap (E) between very round cells (*) is segmented. If these could be colour-coded as ‘edgy’ then they would be very quick to find and delete.
mmh, I think that if the classifier does not make the distinction, then you need a quite “clever” descriptor. I know what you mean by ‘edgy’ but there are multiple ways to translate this into computer vision.
Regarding ‘Compactness’, you could check how it could discriminate between correct and ‘edgy’ cells by creating nuclei objects from this image, then load the nuclei in the image view, right click on a few ‘edgy’ shape and see if the value of compactness differs between the 2 groups. If this is the case then it will be worth adding compactness as a default descriptor in the editor.
In the future, Fourier descriptors could be useful for this (also as a feature to feed the classifier in Nessys)!