Abstract
Purpose :
NIDEK GS-1 (NIDEK CO., LTD Japan) is a novel ophthalmic device capable of automatically providing a complete view of the irido-corneal angle, accessible only through manual inspection in traditional gonioscopy. The optical trade-off between image resolution and depth of field makes it necessary to take several pictures of an area, each in a different focus plane, in order to capture the whole angle depth. Let us call this image sequence “focus stack”. The aim of this study is to develop a software tool capable of merging the information contained in a focus stack into a single augmented-focus image.
Methods :
The considered strategy has been tested on 1040 image sequences, acquired during the GS-1 clinical trial. They are made up by 17 RGB images with a resolution of 1280x960 pixels. Pictures are first pre-processed to enhance true detail components while attenuating noise, then they are rigidly registered. The 2-D derivatives of the images are used to locate the detail distribution over the sequence and save it in a matrix, which provides a depth map of the angle. This map is filtered through a non-linear regularization process which also takes into account a colour-based metric of the anatomical region and the spatial layout of its structures. The final depth map is used to extract non-overlapping sets of pixels from the original sequence and to merge them in a single image which simulates an augmented depth of field.
Results :
A visual evaluation of the results has been independently carried out by four experts considering the actual presence of all the information held by each focus stack in the corresponding output picture. From this assessment, the 81% of the sequences were considered as correctly processed. About the 80% of failures are caused by registration errors, but also artefacts due to filtering may be present. An example of correct outcome may be assessed by comparing Figure 1 and Figure 2.
Conclusions :
Augmented-focus visualization of gonioscopic images acquired by NIDEK GS-1 may be a valid tool to support and speed up the diagnosis of diseases related to irido-corneal angle appearance (e.g. glaucoma). Further evaluation has to be performed in order to define a quantitative measure of the algorithm outcome.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.