Machine Learning and Image Processing Methods for the Segmentation and Quantification of the Corneal Endothelium

Research output: ThesisDissertation (TU Delft)

36 Downloads (Pure)


The corneal endothelium, a non-regenerative layer of cells controlling the state of corneal hydration, is the most critical tissue of the cornea. Quantifying its health status is not only important to diagnose and treat certain corneal diseases but also relevant to the execution and evaluation of many eye surgeries. By means of specular microscopy, the endothelium can be visualized and evaluated in vivo, in a non-invasive manner. The estimation of the corneal endothelium parameters, particularly cell density, provides a valuable input for disease diagnosis, prognosis, and monitoring. Manual estimation, which requires cell segmentation, is time consuming and tedious. In this thesis, we have presented and evaluated several automatic techniques for the segmentation and quantification of the corneal endothelium. The final method was evaluated in two clinical studies: one regarding the transplantation of the cornea (41 patients, 1 year follow-up, 383 images), and another regarding the implantation of the glaucoma drainage device Baerveldt (204 patients, 2 year follow-up, 7975 images). Our method detected more than twice the number of cells than the microscope built-in software (Topcon SP-1P), and our estimation error of the corneal parameters was less than one third of Topcon’s error. Overall, our proposed fully-automatic method provided such a high accuracy that the distinctive patterns in the corneal parameters throughout the months were clearly observable and equal to the manual annotations.
Original languageEnglish
Awarding Institution
  • Delft University of Technology
  • van Vliet, L.J., Supervisor
  • Vermeer, K.A., Advisor
Award date13 Jan 2022
Print ISBNs978-94-6421-564-9
Publication statusPublished - 2022


Dive into the research topics of 'Machine Learning and Image Processing Methods for the Segmentation and Quantification of the Corneal Endothelium'. Together they form a unique fingerprint.

Cite this