Abstract
Purpose :
To develop a fully automatic, reliable cell detection system for corneal endothelium images from clinical specular and confocal microscopy. The system should eventually allow the measurement of the morphometric parameters that provide an objective clinical assessment of corneal endothelium (density, pleomorphism, polymegethism).
Methods :
The proposed method detects corneal endothelial cells contour in three main steps. 1: the centers of the endothelial cells are automatically detected by convolving the original image with customized two-dimensional kernels (Fig 1 - top); 2: a structure made by connected vertices is derived from the centers using the Euclidean distance (Fig 1 - middle); 3: the structure is fine-tuned by means of a genetic algorithm, which combines information about the typical regularity of endothelial cells shape with the pixels intensity of the actual image (Fig 1 - bottom). The final structure of connected vertices forms a set of polygons that fit the underlying cells contours. From these contours the morphometric parameters of clinical interest can then be easily computed.
Results :
The procedure was applied to 15 images acquired with the SP-3000P (Topcon) specular microscope and 15 images acquired with the Confoscan4 (Nidek Technologies) confocal microscope, from both healthy and pathological subjects. Ground truth values for the three morphometric parameters were obtained from manually carefully drawn cell contours. Results show that an accurate automatic estimation is achieved: the mean absolute percent difference between manual and automated estimation is 0.82 (range 0.00-4.38) for density, 3.13 (0.00-6.24) for pleomorphism and 3.95 (0.00-6.86) for polymegethism in images from specular microscopy, and 1.49 (0.26-2.82) for density, 2.91 (0.00-6.49) for pleomorphism and 5.26 (0.00-8.43) for polymegethism in images from confocal microscopy. The procedure analyzes about 100 cells per image and requires few minutes per image.
Conclusions :
The proposed totally automatic method appears capable of obtaining the cell contours in regions containing also hundreds of cells and thus the clinically reliable computation of all the important morphometric parameters used in clinical practice.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.