Purchase this article with an account.
Alfredo Ruggeri, Fabio Scarpa; Automatic estimation of optimal Region Of Interest for morphometry assessment in corneal endothelium images. Invest. Ophthalmol. Vis. Sci. 2017;58(8):3552.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
In order to reliably estimate the morphometric parameters of endothelium cells, the selection of the optimal region of interest (ROI) where cells are clearly recognizable is usually manually performed by the user. Since the estimation of these parameters is now often carried out by computer programs, we have developed an automatic procedure able to identify the optimal ROI.
The method we developed quantitatively expresses the information content of each image pixel using a quality measure based on entropy (E) and power spectral density (PSD). Values of E and PSD are then combined with a weighted linear combination to derive the global quality measure Q, where weights are determined by maximizing the difference in Q between focused and unfocused pixels, manually identified in 5 training images. The optimal ROI in any image under exam is then selected by choosing the connected region where all pixels have a Q value higher than a specified threshold.To evaluate the performance of the proposed procedure, three morphometric parameters (cell density, pleomorphism, polymegethism) were estimated in 15 endothelium images acquired with a Confoscan 4 microscope (Nidek Technologies, Italy) in normal and pathological subjects. Parameter values were estimated first in the original manually selected ROI (RES_1) and then in the ROI identified by the proposed automatic procedure (RES_2).
The differences, for each of the three parameters values, between RES_1 and RES_2 are reported in Table 1.
The proposed procedure for automatic ROI selection was shown to allow an estimation of morphometric parameters equivalent to that performed on a manual ROI. With this step, the whole procedure for morphometry assessment of corneal endothelium can now be carried out by computer programs.
This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.
This PDF is available to Subscribers Only