Dynamic analysis of iris changes and a deep learning system for automated angle-closure classification based on AS-OCT videos

• DOI: 10.1186/s40662-022-00314-1
• Publication Date: 2022-11-05T00:19:32Z
• Authors: Luoying Hao; Yan Hu; Yanwu Xu; Huazhu Fu; Hanpei Miao; Ce Zheng; Jiang Liu
• Keywords: Constriction; IRIS (biosensor)

To study the association between dynamic iris change and primary angle-closure disease (PACD) with anterior segment optical coherence tomography (AS-OCT) videos develop an automated deep learning system for screening as well validate its performance.A total of 369 AS-OCT (19,940 frames)-159 subjects 210 normal controls (two datasets using different capturing devices)-were included. The correlation changes (pupil constriction) PACD was analyzed based on clinical parameters diameter) under guidance a senior ophthalmologist. A temporal network then developed to learn discriminative features from videos. were randomly split into training, test sets fivefold stratified cross-validation used evaluate performance.For parameter evaluation, mean velocity pupil constriction (VPC) significantly lower in eyes (0.470 mm/s) than (0.571 (P < 0.001), acceleration (APC, 3.512 mm/s2 vs. 5.256 mm/s2; P 0.001). For our network, areas curve images, original videos, aligned 0.766 (95% CI: 0.610-0.923) 0.820 0.680-0.961) 0.905 0.802-1.000) (for Casia dataset) 0.767 0.620-0.914) 0.837 0.713-0.961) 0.919 0.831-1.000) Zeiss dataset).The results showed, comparatively, that stretches less response illumination eyes. Furthermore, feature motion could assist classification.