Deep Learning Algorithm for the Diagnosis and Prediction of Hydroxychloroquine Retinopathy: An International,...

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Hydroxychloroquine (HCQ) is a medication used to treat autoimmune diseases.
• Long-term HCQ use can lead to retinopathy, a serious eye condition.
• Early detection of HCQ retinopathy is crucial for preventing vision loss.

Purpose of the Study:

• To develop a deep-learning algorithm (HCQuery) for detecting hydroxychloroquine retinopathy.
• To predict the future occurrence of hydroxychloroquine retinopathy from SD-OCT images.

Main Methods:

• A deep-learning algorithm (EfficientNet-b4) was trained and validated on retrospective SD-OCT images from 409 patients.
• The algorithm processed macular volumes from two SD-OCT devices.
• Performance was assessed using sensitivity, specificity, accuracy, PPV, NPV, AUROC, and AUPRC.

Main Results:

• The HCQuery algorithm demonstrated high accuracy (0.987) in detecting HCQ retinopathy at the time of clinical diagnosis.
• It successfully predicted retinopathy an average of 220.8 days before clinical diagnosis.
• For eyes that developed retinopathy, HCQuery identified it an average of 2.74 years in advance.

Conclusions:

• A deep-learning algorithm, HCQuery, can effectively detect hydroxychloroquine retinopathy.
• HCQuery can predict retinopathy years in advance of clinical diagnosis.
• This algorithm holds potential for early intervention and improved patient outcomes.