MERCI: a machine learning approach to identifying hydroxychloroquine retinopathy using mfERG

Area of Science:

• Ophthalmology
• Medical Technology
• Computational Biology

Background:

• Hydroxychloroquine (HCQ) is crucial for autoimmune diseases but can cause vision loss due to retinal toxicity.
• Early detection of HCQ retinopathy is vital to prevent irreversible vision impairment.
• Multifocal electroretinogram (mfERG) is a key diagnostic tool for assessing retinal function.

Purpose of the Study:

• To evaluate the diagnostic accuracy of a machine learning model for detecting HCQ retinopathy.
• To present updated reference thresholds for mfERG analysis in HCQ toxicity.
• To identify patients at risk of vision loss from long-term HCQ treatment.

Main Methods:

• A retrospective study analyzed mfERG and OCT data from patients undergoing HCQ retinopathy screening.
• A Support Vector Machine (SVM) model was trained using mfERG features.
• Retinopathy diagnosis was confirmed using established clinical reference standards and OCT imaging.

Main Results:

• The SVM model achieved 85.3% accuracy, 90.9% sensitivity, and 84.0% specificity in detecting HCQ retinopathy.
• The model identified 50 out of 55 retinopathy cases, demonstrating strong diagnostic capability.
• 1463 eyes from 748 patients were included in the initial study cohort.

Conclusions:

• Machine learning analysis of mfERG data offers a promising method for early detection of HCQ retinopathy.
• This technology can aid in identifying at-risk patients, potentially preventing vision loss.
• Further application of AI in mfERG analysis can enhance patient care for those on HCQ therapy.