Woodward-Court, Peter;
Hogg, Jeffry;
Lee, Terry;
Taribagil, Priyal;
Zhao, Cindy S;
Otti, Vanessa;
Tucker, William R;
... Dow, Eliot R; + view all
(2025)
Deep-learning algorithm for the diagnosis and prediction of hydroxychloroquine retinopathy: An International, multi-institutional study.
Ophthalmology Retina
10.1016/j.oret.2025.06.003.
(In press).
Preview |
Text
1-s2.0-S246865302500274X-main.pdf Download (5MB) | Preview |
Abstract
Purpose: We sought to develop a deep-learning algorithm - HCQuery - to detect the presence of hydroxychloroquine retinopathy and predict its future occurrence from spectral-domain optical coherence tomography (SD-OCT) images. // Design: We trained and validated a deep-learning algorithm using retrospective SD-OCT images from patients taking hydroxychloroquine. // Participants: The study involved a retrospective, non-consecutive collection of 409 patients (171 positive for hydroxychloroquine retinopathy, 238 negative for retinopathy) and 8251 SD-OCT b-scans (1988 volumes) from five independent international clinical locations. // Methods: Imaging macular volumes from two different SD-OCT devices (Heidelberg Spectralis, Zeiss Cirrus) at two clinical sites were used to train and validate a convolutional neural network (EfficientNet-b4) to produce a Likelihood of Retinopathy Score (LRS) for each SD-OCT b-scan. LRS scores were processed across SD-OCT volumes for an eye- and patient-level binary decision output of the presence or absence of retinopathy. The adjudicated consensus of up to three independent retina specialists using patient clinical data and multimodal testing served as the reference standard for hydroxychloroquine retinopathy. The algorithm was tested on four withheld test sets, one internal (Data Set 1) and three external (Data Sets 3, 4, and 5). The test sets were obtained in two countries (United States, United Kingdom) and represented two SD-OCT devices each with diverse acquisition parameters. // Main Outcome Measures: The algorithm was assessed with sensitivity, specificity, accuracy, negative predictive value (NPV), positive predictive value (PPV), area under the receiver-operator characteristic (AUROC), and area under the precision-recall curve (AUPRC) for the detection of hydroxychloroquine retinopathy either at the time of clinical diagnosis or up to 18 months in advance of clinical diagnosis. // Results: The algorithm demonstrated discriminated hydroxychloroquine retinopathy at the time of clinical diagnosis as well as in advance of clinical diagnosis (Mean 220.8 days prior to clinical diagnosis; Accuracy: 0.987 (95% CI: 0.962-1.00), Sensitivity: 1.00 (95% CI: 0.833-1.00), Specificity: 0.983 (95% CI: 0.952-1.00), PPV: 0.944 (95% CI: 0.836-1.00), NPV: 1.00 (95% CI: 0.937-1.00)). For eyes that developed retinopathy, it was identified as positive by the algorithm on average 2.74 years in advance of the clinical diagnosis. // Conclusions: We report a deep learning algorithm that can detect hydroxychloroquine retinopathy at all stages of disease as well as predict retinopathy years in advance of clinical diagnosis. // Financial Disclosure(s): Authors with financial interests or relationships to disclose are listed.
Archive Staff Only
 |
View Item |
