Artificial Intelligence-Based Uveitis Diagnosis Through Retinal Vasculature Analysis: A Paradigm Shift in Ocular Tuberculosis

Introduction: 

Diagnosis of uveitis is complex and often requires a series of investigations. Here, we utilize artificial intelligence (AI) for the quantitative analysis of retinal vasculature parameters from fundus photographs to differentiate confirmed ocular tuberculosis (TB) from QuantiFERON (QFT)-positive uveitis without another identifiable cause and ocular toxoplasmosis. 

Methods: 

Medical records and stored fundus images of patients with uveitis from a cohort at the Department of Ophthalmology, University of Indonesia, were analyzed. Three groups of patients were included: confirmed ocular TB (group A), QFT-positive uveitis (group B), and ocular toxoplasmosis (group C). Fundus images were processed using the Retinalysis models package for segmentation and quantification of retinal vasculature parameters. 

Results: 

The study included nine patients (13 eyes) in group A, 38 patients (48 eyes) in group B, and 26 patients (39 eyes) in group C. Significant differences were found in vein tortuosity parameter, in the eyes within group A showing lower tortuosity score compared to eyes within group B (p = 0.030) and group C (p = 0.013). The area under the curve (AUC) of vein tortuosity for group A compared to group B was 0.749 (95% confidence interval (CI): 0.606–0.892), with a sensitivity of 67.3% and specificity of 76.9%. The AUC of vein tortuosity for group A against group C was 0.803 (95% CI: 0.658–0.948), with a sensitivity of 74.4% and a specificity of 84.6%. In group A, uveitis resolution and vein tortuosity tended to be normalized upon complete antitubercular treatment. 

Conclusions: 

AI-based quantification of retinal vasculature parameters, particularly vein tortuosity, can differentiate confirmed ocular TB from QFT-positive uveitis and ocular toxoplasmosis. This approach shows promise for more precise diagnostic and therapeutic accuracy in ocular TB.