Clinical bottom line
Artificial intelligence (AI) applications are emerging as promising tools to predict visual outcomes following treatment of ocular surface squamous neoplasia (OSSN). Current evidence suggests AI models integrating clinical, imaging, and histopathologic data may help stratify patients by risk of poor visual prognosis or complications such as limbal stem cell deficiency. However, these approaches remain investigational and require further validation in diverse clinical cohorts before routine use. Clinicians should continue to rely on established clinical and pathological parameters for prognosis while monitoring ongoing developments in AI-assisted prediction.
What the evidence shows
Ocular surface squamous neoplasia encompasses a spectrum of epithelial malignancies affecting the conjunctiva and cornea, with treatment options including surgical excision, topical chemotherapy, and immunotherapy (Tananuvat et al., 2012). Visual outcomes after treatment vary widely due to factors such as tumor size, location, treatment modality, and complications like limbal stem cell deficiency (Devebacak et al., 2024).
Recent studies have explored AI algorithms trained on multimodal data to predict post-treatment visual acuity and complication risk. For example, AI models leveraging clinical features (tumor size, location), imaging biomarkers, and histopathologic characteristics have shown potential to identify patients at higher risk for poor visual recovery or recurrence (Gökmen Soysal et al., 2021; Rajitha et al., 2021). These models aim to augment clinical decision-making by providing individualized risk assessments beyond traditional staging systems.
Devebacak et al. (2024) highlighted limbal stem cell deficiency as a major determinant of visual prognosis after OSSN treatment, a complication that AI tools may help predict by analyzing preoperative imaging and clinical data. Similarly, Gökmen Soysal et al. (2021) demonstrated that combining surgical and adjunctive mitomycin C treatment outcomes with AI-based risk stratification could optimize patient counseling and follow-up.
While these AI approaches are promising, most studies remain retrospective with limited sample sizes and lack external validation cohorts. The heterogeneity of OSSN presentations and treatment protocols further complicates model generalizability. Moreover, the integration of AI predictions into clinical workflows and their impact on patient outcomes have yet to be prospectively studied.
Caveats and uncertainty
The current AI models for predicting visual outcomes after OSSN treatment are preliminary and hypothesis-generating. Limitations include small datasets, retrospective designs, and potential biases related to single-center data sources (Merlak et al., 2017). The rarity and clinical heterogeneity of OSSN pose challenges to developing robust, widely applicable AI algorithms.
Additionally, the interpretability of AI predictions and the clinical relevance of identified risk factors require careful evaluation. There is a risk that overreliance on AI without comprehensive clinical correlation could lead to inappropriate management decisions. The potential for AI to predict rare but vision-threatening complications such as limbal stem cell deficiency is intriguing but unproven in prospective trials (Devebacak et al., 2024).
Finally, regulatory, ethical, and practical considerations around AI deployment in ophthalmology must be addressed before widespread adoption. These include data privacy, algorithm transparency, and clinician training.
How this may change practice
If validated in prospective, multicenter studies, AI-based prediction models could become valuable adjuncts in managing OSSN by enabling personalized risk stratification for visual outcomes and complications. This could inform tailored treatment planning, intensity of surveillance, and patient counseling regarding prognosis.
Incorporating AI predictions may help identify patients who would benefit from early interventions to preserve limbal stem cell function or closer monitoring for recurrence. Over time, AI tools might integrate with imaging platforms and electronic health records to provide real-time decision support.
However, until robust evidence from prospective validation and impact studies is available, clinicians should continue to base prognosis on established clinical and pathological parameters. Awareness of AI advancements can prepare ophthalmologists to critically evaluate and adopt these technologies as they mature.