Best-Fit Segmentation Created Using Flood-based Iterative Thinning

Demo program: BestFit (exe) (.NET 2.0, Win/Lin/Mac), BestFit4.exe (.NET 4.0, Win/Lin/Mac).

BestFit application is now also a free tool for semiautomatic corneal endothelium image segmentation using KH algorithm and providing manual corrections.

ImageJ plugin (very fast): BestFit_IterativeThinning.class

ImageJ script (using Morphology) - slow, but step by step: BestFit.ijm

The article:

Abstract: Classical methods of segmentation that use binarization in the preprocessing stage often do not provide the precise delineation of the range of objects. For example, this might be useful for images of the corneal endothelium obtained with specular or confocal microscopy. This article presents a solution that makes it possible to adjust the course of the segmentation in the valleys between cells. The algorithm is a combination of iterative thinning and a watershed algorithm that works by the gradual removal of points with increasingly lower brightness levels. The article also contains examples of output images and quality tests.

Keywords: segmentation, iterative thinning, corneal endothelium

modified thinning (Flood-based Iterative Thinning)

cycles of singe dilatation and modified thinning

moving the segmentation lines, final segmentation presented only

Multiple dilatations and next modified thinning - using source image (left) and adjusted (right, bkgrnd removing, normalisation)

---

how to use a demo?

youtube (PL)

Corneal endothelium image	segmentation

Drag and drop example files	use 'full' button and wait	you can see the history of lines moving

Why use bestfit?

input image
	Source image (Y. Gavet)	KH algorithm output
initial segmentation
	manual segmentation (Y. Gavet), CV=30,0	KH output thinned and manually corrected, CV=28,8	difference
segmentation after bestfit
	YG after bestfit, CV=29,6	KH after bestfit, CV=29,5	difference

References:
Images:
- Gavet, Y., Pinoli, J. C.: Comparison and supervised learning of segmentation methods dedicated to specular microscope images of corneal endothelium. Journal of Biomedical Imaging, 2014, 5.
- Piorkowski, A., Nurzynska, K., Gronkowska-Serafin, J., Selig, B., Boldak, C., Reska, D.: Influence of applied corneal endothelium image segmentation techniques on the clinical parameters. Computerized Medical Imaging and Graphics, 2017, 55, 13-27. pdf
- Ruggeri, A., Scarpa, F., De Luca, M., Meltendorf, C., Schroeter, J.: A system for the automatic estimation of morphometric parameters of corneal endothelium in alizarine red stained images., Br J Ophthalmol, 94:643-7, 2010.

Algorithms:
- BestFit algorithm:
Piorkowski A.: Best-fit Segmentation Created Using Flood-based Iterative Thinning. Springer, AISC Vol. 525, pp 61-68, 2017, pdf, bibtex.
- KH algorithm:
Habrat, K., Habrat, M., Gronkowska-Serafin, J., Piorkowski, A.: Cell detection in corneal endothelial images using directional filters. Springer 2016, AISC vol 389, pp. 113-123 pdf, bibtex

Corneal endothelium image normalization: SDA.



Other links:
http://an-fab.kis.p.lodz.pl/cornea/

Datasets:

Gavet, Y., Pinoli, J. C.: Comparison and supervised learning of segmentation methods dedicated to specular microscope images of corneal endothelium. Journal of Biomedical Imaging, 2014, 5. PDF contains images - please contact with Author.
http://bioimlab.dei.unipd.it/Endo%20Aliza%20Data%20Set.htm
http://www.cb.uu.se/~cris/endothelium.html https://user.it.uu.se/~crilu684/endothelium.html
http://www.rodrep.com/confocal-corneal-endothelial-microscopy---description.html
https://doi.org/10.1038/s41598-019-41034-2 contains a link to another dataset