This algorithm is design to segment nuclei, with a good precision on the edge even in case of nucleus-to-nucleus intensity variation on a given image. It is not designed to split nuclei in close contact.
This algorithm is very robust
Pre-Filtering
This algorithm is sensitive to noise, so a median filter can be applied in order to reduce noise
The Algorithm
It processes in two step:
- Segmentation by global Thresholding. If there are nucleus-to-nucleus intensity variations, nuclei with a weak signal will be under-estimated and nuclei with a strong signal will be over-estimated
- To correct this bias: Each segmented nucleus is locally corrected using a Watershed 3D procedure, that allows to precisely detect edges. This procedure is achieved locally by cropping around the nucleus. In case of under-estimation of the nucleus, the size of the nucleus should be increased, to do so, the image can be with a small border of a user-defined number of pixels)
Post-Filters
If a nucleus contains regions of low intensity (like nucleoli) they have to be filled.
This can be done using morphological filters like 2D/3D fill holes and closing with a large radius.