Long-Term Outcomes Of Therapeutic Corneal Grafts Using Artificial Intelligence Analysis At A Tertiary Eye Care Centre

To evaluate the demographics, clinical characteristics, and long-term outcomes of therapeutic penetrating keratoplasty (TPK) performed for severe infectious keratitis, and to analyze risk factors influencing outcomes using an artificial intelligence (AI) model at a tertiary eye care hospital.

A retrospective review of 1,046 therapeutic keratoplasties performed between 2013 and 2022 was conducted, with outcomes assessed up to 3 years  at a tertiary eye care centre.

Demographic data, clinical presentation, indications for keratoplasty, microbiological profile, and need for secondary surgical interventions were analyzed. Final outcomes were categorized as therapeutic (eradication of infection), anatomical (globe integrity), and functional (visual outcome). An AI-based predictive model was developed to identify risk factors for adverse outcomes. Model performance was evaluated using area under the curve (AUC), sensitivity (Se), specificity (Sp.), and overall accuracy (Ac).

The median recipient age was 58 years, with 72% males. Advanced microbial keratitis, including perforated infectious ulcers, non-healing ulcers, and graft infections, constituted the major indications. The AI model identified mixed infections (fungal + bacterial), followed by fungal infections (Aspergillus flavus), as significant predictors of adverse outcomes. Greater distance from the eye care centre, associated with higher likelihood of missed follow-up, also emerged as an important risk factor. Therapeutic success was achieved in 72% at 6 months. At 3 years, 15% maintained a clear graft following primary TPK. The predictive model demonstrated 82% accuracy, 73% sensitivity, and 86% specificity.

Therapeutic keratoplasty is effective in eradicating infection and restoring anatomical integrity in severe keratitis; however, long-term graft clarity and functional success remain limited. Early intervention and timely referral are critical to improving outcomes. Identification of key risk factors, including infection type and socio-economic barriers affecting follow-up, enables better prognostication and individualized management strategies. Artificial intelligence–based models may serve as valuable tools in guiding clinical decision-making and optimizing long-term results.