Corneal endothelium segmentation using contextual information under low-contrast conditionition

Abstract

Corneal endothelial cells are non-regenerative, tightly arranged single-cell
layers located in the innermost layer of the cornea. The evaluation of their
health status is of great significance for the diagnosis of corneal endothelial
diseases, preoperative planning, and postoperative rehabilitation analysis.
Segmentation is a crucial step in quantifying corneal endothelial cell density and morphological parameters. However, the low contrast of corneal
endothelial microscope images makes it difficult to distinguish cell boundaries from surrounding tissues, and the elongated structure of cell boundaries easily leads to segmentation breakpoints. In addition, factors such as
uneven illumination, disease, and variation in cell morphology and size further exacerbate the difficulty of robust segmentation. Artificial intelligence
(AI)-based segmentation and quantification methods currently also fail to
provide decision confidence, which prevents clinicians from evaluating the
reliability of the AI model and hinders its widespread implementation in
clinical practice. To address the issues above, we have conducted a series
of studies aimed at enhancing the robustness of corneal endothelial cell
segmentation under the low-contrast condition.

Date of Award	15 Jan 2026
Original language	English
Awarding Institution	University of Nottingham
Supervisor	Jiang Liu (Supervisor), Ruibin Bai (Supervisor) & Dave Towey (Supervisor)