Combined diagnostic accuracy of two artificial intelligence systems for glaucoma diagnosis using color fundus photography

Abstract

Purpose To evaluate the diagnostic accuracy of two commercially available artificial intelligence (AI) systems based on color fundus photography (CFP), the Laguna ONhE and VUNO Med-Fundus AI, for glaucoma detection, both independently and in combination. Methods This retrospective cross-sectional study included 370 eyes from 193 patients (248 eyes with primary open-angle glaucoma and 122 healthy eyes). All eyes underwent structural evaluation with swept-source optical coherence tomography (Triton, Topcon) and visual field testing (Octopus 900, Haag-Streit AG). Fundus photographs were analyzed using Laguna ONhE and VUNO Med-Fundus AI systems. Diagnostic accuracy was evaluated. Results Both AI systems demonstrated high diagnostic accuracy. Laguna achieved an AUC of 0.879 using Glaucoma Discriminant Function (GDF), and VUNO showed an AUC of 0.857. When combined, GDF + VUNO achieved an AUC of 0.903. The Global Mean Deviation (GMD) reached the highest diagnostic accuracy (AUC = 0.916), which was not significantly different from GDF + VUNO (p = 0.146). Conclusions Laguna ONhE and VUNO Med-Fundus AI had high diagnostic accuracy in detecting glaucoma using only CFP. Their combined use improved further, achieving accuracy comparable to the GMD. This represents a practical approach for glaucoma screening, particularly in settings without access to OCT or automated perimetry.