Deep Learning-Based Automatic Detection of Ellipsoid Zone Loss in Spectral-Domain OCT for Hydroxychloroquine Retinal Toxicity Screening

ABSTRACT Purpose Retinal toxicity due to hydroxychloroquine (HCQ) use manifests photoreceptor loss and disruption of the ellipsoid zone (EZ) reflectivity band detectable on spectral domain optical coherence tomography (SD-OCT) imaging. This work investigates whether an automatic deep-learning based algorithm can detect and quantitate EZ loss in SD-OCT images with an accuracy comparable to human annotations. Design Retrospective analysis of data acquired in a prospective, single-center, case-control study. Subjects, Participants, Controls 85 patients (168 eyes) who were long term HCQ users (average exposure time = 14 ± 7.2 years). Methods Mask-RCNN network was implemented and trained on individual OCT B-scans. Scan-by-scan detections were aggregated to produce an enface map of EZ loss per 3D SD-OCT volume image. To improve the accuracy and robustness of the EZ loss map, a dual network architecture is proposed that learns to detect EZ loss in parallel using horizontal (M-RCNNH) and vertical (M-RCNNV) B-scans independently. To quantify accuracy ten-fold cross validation was performed. Main Outcome Measures Precision, recall, F1-score metrics and measured total EZ loss area were compared against human grader annotations and to the determination of toxicity based on the recommended screening guidelines. Results The combined model (CPN) demonstrated the best overall performance with precision = 0.90 ± 0.09, recall = 0.88 ± 0.08, and F1 score = 0.89 ± 0.07. The combined model performed superiorly to horizontal only (M-RCNNH) (precision = 0.79 ± 0.17, recall = 0.96 ± 0.04, IOU = 0.78 ± 0.15, and F1 score =0.86 ± 0.12) and vertical only (M-RCNNV) (precision = 0.71 ± 0.21, recall = 0.94 ± 0.06, IOU = 0.69 ± 0.21, and F1 score = 0.79 ± 0.16) models. The accuracy was comparable to the variability of human experts with precision = 0.85 ± 0.09, recall = 0.98 ± 0.01, IOU= 0.82 ± 0.12, and F1 score = 0.91 ± 0.06. Automatically generated enface EZ loss maps provide quantitative SD-OCT metrics for accurate toxicity determination combined with other functional testing. Conclusions The algorithm can provide a fast, objective, automatic method for measuring areas with EZ loss and serve as a quantitative assistance tool to screen patients for the presence and extent of toxicity.