Purchase this article with an account.
Masaki Tanito, Simone Pajaro, Andrea De Giusti; Improved clustering and quantification of color information in images obtained by the 360-degree automatic gonioscopy. Invest. Ophthalmol. Vis. Sci. 2017;58(8):2073.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Gonioscopy is essential to make a diagnosis of glaucoma, however, it requires an examiner’s skill and only provides subjective information. Assessment of irido-corneal angle (ICA) by using currently available modalities (e.g. ultrasound biomicroscopy or anterior segment optical coherence tomography) doesn’t provide any chromatic information, while color analysis is relevant for many pathologies. Recently developed gonioscopic device (NGS-1 Gonioscope, NIDEK Technologies Srl, Italy), was able to automatically acquire true color images of the ICA structures and to combine them in a 360-degree picture of the angle. In this study, image analyses were performed in the pictures obtained by using NGS-1 to quantify the color information of the angle.
The NGS-1 system detects the ICA using a prism with a soft contact to the corneal surface. The prism has 16 mirrored facets, each of them projecting white light to a single portion of the irido-corneal angle (about 4x4 mm2). A rotating 1.3 megapixel camera element scans all the facets to capture 5 images, at different focus depth, for each of the 16 sectors in less than 2 seconds. On each selected image from 16 sectors, a region of interest (ROI) was manually defined in the trabecular meshwork area by means of a dedicated PC software. From each ROI, pixels’ color information was extracted in the Lab color space and statistical clustering was applied.
A set of 450 images collected from 100 patients was considered for training and validation of the proposed method. By image analyses, indexes computed from each cluster were used to perform multi-dimensional data-mining in order to obtain a robust 3-level color classification, similar to the Scheie’s one (Figure 1).
By this further image analysis, the Lab color space and clustering can be used to improve the previously proposed classification (Tanito M., et al., Abstract Number 5118, ARVO 2016). The established method can be applied to understand color/pigment variations among specific directions in normal and glaucomatous eyes.
This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.
Figure 1 - A, B, C: Examples of 0_I, II, and III_IV grade images. D, E, F: Same images with clustering highlighted. G: Distribution of the L index without clustering showing poor separation between the grades. H: Distribution of the L index for the cluster matched to the trabecular meshwork showing a good separation between the grades.
This PDF is available to Subscribers Only