Beyon Human Eye, How Artificial Intelligence Helping In Diagnosing Microbial Keratitis

Microbial keratitis remains a sight-threatening condition in which early and accurate diagnosis is essential for optimal visual outcomes. Clinical differentiation between bacterial, fungal, Acanthamoeba, and viral keratitis is often challenging due to overlapping features, leading to diagnostic delay and inappropriate treatment. The purpose of this paper is to evaluate the role of artificial intelligence as a clinical decision-support tool in the diagnosis of microbial keratitis. It explores how AI-based analysis of ophthalmic imaging and clinical data may enhance diagnostic accuracy, reduce inter-observer variability, and support earlier, more targeted management while complementing specialist clinical judgment. 

This study was conducted at Liaquat National Hospital and Medical College, Karachi, Pakistan, a tertiary-care referral centre with a dedicated cornea service. The hospital manages a high volume of patients with bacterial, fungal, Acanthamoeba, and viral keratitis, providing a representative spectrum for evaluation. Clinical data and ophthalmic images were collected according to institutional protocols and standard care. The study adhered to ethical standards and institutional guidelines.

This study was conducted at Liaquat National Hospital and Medical College, Karachi, Pakistan. Clinical data and ophthalmic images of patients with microbial keratitis were collected from the cornea service. Imaging included slit-lamp photography, anterior segment OCT, and in vivo confocal microscopy when available. Cases were classified by culture-confirmed or clinically presumed etiology, including bacterial, fungal, Acanthamoeba, and viral keratitis. Deep learning AI models were trained and validated on annotated datasets. Performance was assessed by accuracy, sensitivity, specificity, and agreement with corneal specialists to evaluate AI as a diagnostic support tool.

The artificial intelligence–based diagnostic model demonstrated an overall diagnostic accuracy of 80%- 85%  in identifying and classifying microbial keratitis. Agreement between AI predictions and expert clinical diagnosis was high, particularly in distinguishing infectious from non-infectious keratitis and in differentiating common microbial etiologies. Diagnostic performance was strongest for bacterial and fungal keratitis, with comparatively lower accuracy in less prevalent subtypes. These findings support the potential role of AI as a reliable adjunct to clinical assessment in the diagnosis of microbial keratitis.

Artificial intelligence shows significant potential as a clinical decision-support tool in diagnosing microbial keratitis, achieving 80% overall accuracy. AI can aid early etiological identification and improve diagnostic consistency, especially in high-volume tertiary centres. It is particularly valuable where microbiological culture, advanced imaging, or specialist expertise is unavailable, offering guidance when conventional diagnostics are limited. While not a substitute for clinical judgment, AI integration may reduce diagnostic delays, support initial management, and provide practical diagnostic assistance in resource-constrained settings.