Deep learning-based quantification of anterior segment optical coherence tomography parameters

ObjectiveTo develop and validate a deep learning (DL) algorithm that could automate the annotation of scleral spur (SS) segmentation anterior chamber (AC) structures for measurements AC, iris, angle width parameters in segment optical coherence tomography (ASOCT) scans.DesignCross-sectional studySubjectsData from two population-based studies (i.e., Singapore Chinese Eye Study Malay Study) one clinical study on closure disease were included development. A separate was used external validation.MethodImage contrast ASOCT scans first enhanced with CycleGAN. We utilized heat map regression approach coarse-to-fine framework SS annotation. Then, an ensemble network U-Net, full resolution residual (FRRnet), U-Net (FRRUnet) structure segmentation. Measurements obtained predicted measured compared to manual ground truth).Main outcome measuresWe Euclidean distance intra-class correlation coefficients (ICC) evaluate annotation, Dice similarity coefficient (DSC) The ICC, Bland-Altman plot, repeatability (RC) agreement precision measurements.ResultsFor our achieved 124.7μm, ICC ≥0.95, 3.3% error rate. For segmentation, we DSC ≥0.91 cornea, AC measurements, ≥95% data points within 95% limits-of-agreement (LOA) plot insignificant systematic bias (all P>0.12). ranged 0.71 0.87 0.54 IT750, 0.83 0.85 other iris 0.89 0.99 measurements. Using same coordinates human expert, generally less variable than semi-automated assessment program.ConclusionWe DL like experts, both open eyes. This reduces time needed subjectivity obtaining