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