A Deep Learning Approach for Automated Detection of Shallow Anterior Chamber Depth Based on Hidden Features of Fundus Photographs

Published: 4 June 2021| Version 1 | DOI: 10.17632/zz768t4fs5.1
TaeKeun Yoo


The datasets are not redistributable to researchers other than those engaged in the Institutional Review Board-approved research collaborations with the B&VIIt Eye Center, South Korea. The datasets utilized during this study are not publicly available due to reasonable privacy and security concerns. Instead, a sample anonymized fundus photography data with shallow and deep ACD is available in the publicly accessible source. Note that this was not the exact data used in the research, but is a cleaned-up reproduction of our study's key insight. In the current study, we analyzed the preoperative ocular data of healthy subjects who intended to undergo refractive surgery at the B&VIIT Eye Center from January 2015 to December 2016. All patients underwent preoperative measurements of best-corrected distance visual acuity and manifest refraction, slit-lamp examinations of the anterior segment. We used macula-centered fundus photographs containing posterior pole of eyes which were obtained during preoperative examination. Non-mydriatic fundus photographs were taken using two different retinal cameras including Maestro2 non-mydriatic retinal camera (Topcon Corporation, Tokyo, Japan) and CR-2 non-mydriatic retinal camera (Canon, Tokyo, Japan). Non-contact tonometry and pachymetry device (NT-530P; Nidek Co., Ltd., Aichi, Japan) was used to evaluate the intraocular pressure and central corneal thickness. All these measurements were carried out by trained medical workers and ophthalmologists. The measurement of ACD was performed using a Pentacam HR Scheimpflug device (Oculus Optikgeräte GmbH, Wetzlar, Germany). Currently, there is the lack of universal definition of shallow ACD due to ethnic differences. In this study, we defined shallow ACD as an ACD measurement of 2.80 mm or lower according to the previous study (Malyugin et al., 2012). This criterion was adopted in both classification and feature map generation. All fundus photographs were de-identified to share publicly. Reference Malyugin, B.E., Shpak, A.A., Pokrovskiy, D.F., 2012. Accommodative changes in anterior chamber depth in patients with high myopia. Journal of Cataract & Refractive Surgery 38, 1403–1407. https://doi.org/10.1016/j.jcrs.2012.04.030