MORPHY

a CNN based tool for the automated morphometry analysis of the human corneal endothelium

This research is in cooperation with the
Physiopatologic Optics Chair - University of Rome Tor Vergata
Dir. Prof. Luciano Cerulli
Dott. Federico Ricci

Physiopathology Background

The cornea is a normally transparent and avascular multilayered tissue forming the anterior part of the eyeball and is the main refractive element of the eye. The structure of the cornea is illustrated in Fig. 1. The great bulk (90% of thickness) is made up of the stroma that is bounded externally by Bowman's membrane and epithelium and internally by Descemet membrane and endothelium. In order to keep maintaining the normal transparency of the cornea, various physiological processes must be active. The corneal stroma is normally in a state of relative deturgescence and if it is allowed to imbibe water it swells and loses its transparency. Both epithelium and endothelium layers represent semipermeable membranes that inhibit the leakage of fluid and ions in the stroma. The epithelium consists of 5-6 layers of cells and therefore it is an effective barrier to the diffusion of tear film fluid into the stroma. Conversely, the endothelium is a single layer of closely packed hexagonal cells whose permeability to small ions is 100 times higher than epithelium. The barrier function of the endothelium is weak. Nevertheless, due to the presence of an active pump mechanism that removes fluid from the stroma, it assumes a critical importance in regulating the optimum level of stromal hydration to maintain corneal transparency. Because of the lack of proliferative capacity of the corneal endothelium, cell damage may be healed by migration and enlargement of remaining cells. This results in a decrease of the cell density, an increase of variation between the cell sizes and a disruption of the normal regular hexagonal pattern. The development of improved systems for the clinical specular microscopy in vivo has strongly contributed to the study and to the comprehension of the physiopathology of the corneal endothelium [1-2]. The commonly employed parameter in the evaluation of the endothelial pattern is the mean cell density expressed as cells/mm2.


Human eyeball

The more recent systems allow both the direct acquisition of the images and automated morphometric analysis. The automated reconstruction of the endothelium cell borders is not always precise, and although it allows to evaluate in a reliable way the Endothelial Cell Density (ECD) and the Coefficient of Variation (CV), it does not allow the correct recognition of the cell shapes, if only in terms of a coefficient of roundness coefficient (approximation to the circle), whose correlation with the cell shape is questionable. The semiautomated devices for morphometric analysis are time consuming and provide results that could be influenced from the operator criterion used in the cell apex definition. The main advantage of such devices is the capability to evaluate the cell shape.
 

[1] RA.Laing, MM.Sandstrom, HM. Liebowitz, "In vivo photomicrography of the corneal endothelium", Arch. Ophthalmol. 93: 143, 1975
[2] WM Bourne, HE Kaufman, "Specular microscopy of human corneal endothelium in vivo", Am. J Ophthalmol. 81: 319, 1976

The MORPHY System

MORPHY has been developed to give a real-time easy-to-use tool to the doctor in order to deal with an useful support in clinical diagnostic. The program consists of two main section relative to the digital acquisition of the microscopic images and their processing respectively.
The acquisition section of the program makes use of a MATROX "Frame Grabber" board and its relative library. It has to acquire the corneal endothelium microscopic images coming form a specular microscopy (KONAN - 5500 SP), with on board a 40x lens. Moreover, the microscopy has been interfaced with an analogue camera (Sony Hi Resolution SSC M370 CE).
A simple hardware interface has been designed and manufactured to enable the user to control the camera focus and the grabbing procedure directly from the console.
The second section of the program has to process the acquired images following the above described algorithm. The image processing steps can be controlled on-line by the doctor who can modify some of the algorithm’s parameters (e.g. the threshold value in the "Threshold" operation). The last processing step will give to the doctor a final report containing both the items of the detected cells and their area. In particular, this step, making use of both the information coming from the skeletonized and branchpoint images by way of a tracking procedure, will be able to detect the cell's shape and calculate the respective areas.
The MORPHY’s output data are:

Moreover, the doctor can introduce, before the tracking operation will be performed,  new corners or border cells. Furthermore, he can delete some of the corners detected by the MORPHY in the case of misleading. The proposed algorithm has already been experimentally proved in medical diagnosis with good results. Moreover, further improvements in algorithm structure for the processing and the analysis of the final images are under development. This is in the aim to grant to the doctor an effective procedure able to supply the required real time support in clinical diagnostic.

Return to the top page

The algorithm


The proposed algorithm is able to the detect the corneal endothelium cell’s apexes allowing the shape recognition as well as the cell’s area measurement. It is composed by two main parts: the first part detects the apexes making wide use of CNN operations, while the second one, mainly using classical techniques, processes the values of the apexes coordinates in order to allow the cell’s shape analysis and the respective area measurement. The first part of the algorithm processes a grey scale image obtained by in vivo specular microscopy techniques and produces as final result a B&W image where the cell’s apexes are detected. In particular, the implemented procedure isolates first the cell borders from the whole image. Therefore, the cell’s borders image is processed in order to obtain a single pixel line representing the actual edge between the cells. Finally, the resulting image is analysed and the cell apexes are extracted. The second part deals with the reconstruction of the cell shapes and calculation of the respective areas. It makes use of a mix of classic and CNN procedures.
Return to the top page

Measures



This picture is an "animated GIF" obtained grouping together the images resulting from MORPHY processing. In this example MORPHY processed a 100 x 200 pixels microscopic image of human corneal endothelium. The whole processing consists of a set of seven automated steps. A further step (not performed by CNN) will process the last image containing the vertexes (apexes) informations and will give to the doctor the final report.


Return to the top page

References

  1. M. Salerno, F. Sargeni, V. Bonaiuto, P. Amerini, L. Cerulli, F. Ricci: "A CNN algorithm for the morphometry analysis of the corneal endothelium" - European Conference of Circuit Theory and Design (ECCTD-97), Budapest. Hungary, August 1997, pp. 148-153.

  2. M. Salerno, F. Sargeni, V. Bonaiuto, P. Amerini, L. Cerulli, F. Ricci: "A new CNN based Tool for an automated morphometry analysis of the corneal endothelium" - 4th International Conference on Neural Networks and their Applications (NEURAP 98), Marseilles, France, March 1998,  pp.243, 249.

  3. M. Salerno, F. Sargeni, V. Bonaiuto, P. Amerini, L. Cerulli, F. Ricci : "MORPHY: a CNN Based Tool for an Automated Morphometry Analysis of the Corneal Endothelium" - 5th IEEE International Workshop on CNN and their Application (CNNA98), London, United Kingdom, 1998.

Return to the top page


Last updated on

For information and comments about this page feel free to contatct the Webmaster