AI-Based Ocula360 Tool Shows High Accuracy in Diabetic Retinopathy Severity Detection

The artificial intelligence platform Ocula360 demonstrated performance similar to human reading centers in identifying diabetic retinopathy (DR) severity levels across 120 ultra-widefield retinal images, according to research presented on June 13, 2026, at the Clinical Trials at the Summit in Las Vegas. Developed by Ophthalytics, the tool uses a web-based algorithm to classify disease severity based on the seven standard Early Treatment Diabetic Retinopathy Study (ETDRS) fields.

Did You Know? The Ocula360 algorithm was trained on a substantial dataset consisting of more than 487,000 images sourced from four distinct collections.

Evaluating AI Performance in Retinal Imaging

To assess the efficacy of the tool, researchers compared its automated classifications against those provided by human reading centers using images from the ONYX-1 study. The analysis showed that Ocula360 assigned 51 images to class 1 (severity level 43 or below), 33 to class 2 (levels 47 or 53), and 31 to class 3 (level 60 or above). In comparison, human graders categorized 50, 32, and 33 images into those same respective classes.

SriniVas R. Sadda, MD, of the Doheny Eye Institute, noted that these results indicate an “excellent” level of agreement between the software and human experts, yielding a kappa value of 0.89. This finding suggests the platform could effectively replicate manual grading processes for monitoring disease progression.

Clinical Implications for Diabetic Retinopathy

While various automated DR assessment tools have already received FDA approval, Sadda pointed out that existing options often face limitations regarding comprehensive disease severity screening. The implementation of Ocula360 may offer a more quantitative and continuous scoring system for clinicians. By automatically detecting lesions on ultra-widefield images, the technology aims to streamline the assessment of retinal health.

Expert Insight: The transition toward AI-driven diagnostics represents a shift in how ophthalmologists monitor chronic conditions like diabetic retinopathy. While human expertise remains the gold standard, the high level of agreement between Ocula360 and reading centers suggests that AI may soon serve as a reliable, scalable support tool, particularly for identifying candidates for complex clinical trials.

Future Applications of Automated Grading

Beyond standard monitoring, Ocula360 could play a role in identifying patients eligible for clinical research trials. Sadda indicated that the platform’s current classification system, which focuses on standard ETDRS fields, could potentially be adapted to incorporate peripheral retinal lesions as well. Future clinical strategies may evolve to include these peripheral findings, potentially offering a more nuanced view of DR severity.

Frequently Asked Questions

How does Ocula360 classify diabetic retinopathy?
The platform uses a three-step system to assign images to specific severity classes: class 1 for a DR severity scale level of 43 or below, class 2 for levels 47 or 53, and class 3 for levels 60 or above.

How accurate is the platform compared to humans?
In the ONYX-1 study, Ocula360 showed “very similar” performance to human reading centers, achieving a kappa value of 0.89, which indicates excellent agreement.

What is the primary benefit of this AI tool?
According to SriniVas R. Sadda, MD, the tool may address limitations in current automated screening tools by facilitating a quantitative and continuous scoring system, while also helping to identify patients suitable for clinical trials.

How might the integration of peripheral lesion detection change the way you approach your own preventative eye health screenings?