↓Skip to main content
Fernando Quintero

Deep Regression of Signed Distance Maps for Corneal Endothelium Image Segmentation

· loading · loading · ·
Fernando Quintero👨‍💻
, 
Juan Sierra
, 
Kevin Mendoza
, 
Lenny A. Romero
, 
Andres G. Marrugo
BEST PAPER AWARD Publication Presentation
Fernando Quintero
Author
Fernando Quintero
Researcher / Computer Vision / ML

IEEE Colombian Caribbean Conference (C3), 2023

🔗Paper

Abstract:
#

This paper describes a corneal endothelial image segmentation strategy based on a deep regression of a signed distance map (UNet-dm) compared to a classical pixel-wise classification (UNet-Mask). The proposed approach generates cell masks closer to reference masks, improving the mapping of well-defined cell and guttae boundaries. The results reveal enhanced morphometric parameters that align closer to reference values. The study emphasizes a new technique for continuous segmentation, employing a UNet model, demonstrating its promise for accurate segmentation of corneal endothelial cells and presenting it as a valuable alternative to other methods.

(a) Specular microscopy image. (b) Profile of row 300 from the distance map. (c) Final segmentation after applying watersheed transformation.
(a) Specular microscopy image. (b) Profile of row 300 from the distance map. (c) Final segmentation after applying watersheed transformation.

Citation:
#

F. Quintero, J. Sierra, K. D. Mendoza, A. G. Marrugo and L. Romero, “Deep Regression of Signed Distance Maps for Corneal Endothelium Image Segmentation,” 2023 IEEE Colombian Caribbean Conference (C3), Barranquilla, Colombia, 2023, pp. 1-6, doi: 10.1109/C358072.2023.10436286.