From AI to injury patterns: Insights from emerging ophthalmic research

From surgical innovations to artificial intelligence (AI), ophthalmology continues to push the boundaries of medicine. During the 2025 American Academy of Ophthalmology (AAO) and American Society of Ophthalmic Plastic and Reconstructive Surgery (ASOPRS) meetings, results from 2 distinct yet complementary studies were presented that highlighted the breadth of modern vision science. One explored the diagnostic accuracy of ChatGPT in ophthalmic case interpretation, and the other examined the epidemiology of pediatric oculofacial dog-bite injuries.

Both projects, one grounded in computational modeling, the other in clinical trauma data, reflect a shared mission: using technology and evidence-based insights to improve patient outcomes and guide the next generation of ophthalmologists.

AI meets ophthalmology

Iden Amiri, MD, and Cheng Jiao, MD, with senior author Alice Yang Zhang, MD, presented, “Comprehensive Evaluation of ChatGPT’s Diagnostic Accuracy on Image-Based Ophthalmic Case Interpretations”¹ at AAO 2025. The study represented one of the most detailed investigations to date of a large language model’s clinical reasoning abilities in ophthalmology.

Using 261 real-world cases from the University of Iowa’s EyeRounds repository,² which spans pediatrics, retina, glaucoma, oculoplastics, cornea/cataract, and neuro-ophthalmology, the investigators tested ChatGPT-4o under 2 distinct conditions. In the full-context condition, the model received complete patient histories, examination findings, and structured image descriptions. In the image condition, the model interpreted raw clinical photographs without any textual guidance.

The contrast was striking. When provided with full clinical context, ChatGPT achieved 80.1% diagnostic accuracy, compared with 54.7% when limited to images alone (χ² = 48.00; P < .001). Across subspecialties, accuracy was highest in pediatrics and oculoplastics and lowest in glaucoma and neuro-ophthalmology. Notably, the researchers found that treatment recognition most strongly predicted diagnostic success, underscoring the model’s strength in structured decision-making but its vulnerability when deprived of contextual cues.

“ChatGPT performed like a well-trained resident when it had the full story,” Jiao noted, “but struggled when forced to reason from visuals alone, something that highlights both the promise and the limitations of multimodal AI.”

The analysis also illuminated deeper questions about how AI models think. Logistic regression analysis revealed that, in the image-only condition, success depended heavily on accurate identification of clinical signs and treatment patterns, suggesting that ChatGPT relied on pattern recognition rather than deeper pathophysiologic reasoning.In contrast, when ChatGPT was given complete context, unmeasured factors played a larger role.

The implications reach far beyond diagnostic trivia. Multimodal AI could one day assist in triage, augment resident education, and improve access to subspecialty expertise in under-resourced settings.^3-7 Still,the researchers also warned of pitfalls, including privacy risks when uploading patient data, potential hallucinations, and the danger of overreliance on algorithmic output without clinician oversight.^8-10

As coauthor Zhang emphasized, “AI is a tool, not a replacement for clinical judgment. When integrated thoughtfully, it can streamline care and expand access, but only if we ensure transparency, data security, and clinical accountability.”

ChatGPT’s diagnostic performance across ophthalmic subspecialties revealed a striking contrast between full-text and image-only conditions, with substantially higher accuracy when complete clinical information was provided. This visualization underscores that context remains king, for both human clinicians and artificial intelligence (Figure 1).

Pediatric oculofacial dog-bite injuries: Lessons from 15 years of cases

At ASOPRS 2025, Dhruv Shah, MD, presented a retrospective analysis of pediatric oculofacial injuries from dog bites.¹¹ Although canine bites represent a well-recognized pediatric hazard, few studies have focused specifically on the periocular region.^12,13 This single-site, retrospective chart review examined 818 encounters between 2009 and 2024, identifying 81 eyes with confirmed periocular involvement.

The analysis revealed distinct age-related injury patterns (Figure 2).
 These data suggest that most pediatric dog-bite injuries are lacerating or avulsive, rather than penetrating, and that younger children are uniquely prone to canalicular involvement, likely due to facial height alignment with canine snouts and the smaller overall surface area of children’s faces, which increase the likelihood that a bite targets the medial canthus and canaliculus.^14,15

Management strategies varied from simple closure to multilayer oculoplastic repair, with a minority requiring stenting or secondary intervention.^16,17 Complications included infection, eyelid malposition, and scarring, though long-term outcomes were generally favorable with prompt repair.^12,13,15

“Recognizing high-risk patterns helps emergency physicians and ophthalmologists prioritize early oculoplastic consultation,” Shah explained. “Our findings demonstrate that canalicular assessment, especially in toddlers, is essential to preserving tear-drainage function and cosmesis after dog bites.”

Beyond surgical implications, the study advocates for preventive counseling and collaboration with pediatricians to educate families about dog safety, particularly around very young children. Given that dog-bite incidents often occur in familiar settings and involve household pets, anticipatory guidance may be the most effective intervention of all.^12,16,17

Bridging innovation and compassion

Though methodologically different, both projects converge on a common theme: the fusion of innovation and empathy in ophthalmology. The AI study reminds us that even the most advanced algorithms depend on the clinical narrative, a story only a patient can tell. Results of the pediatric trauma study underscore the human stories behind every statistic: frightened children, worried parents, and surgeons striving to restore both vision and confidence.

Together, they highlight the dual responsibility of the modern ophthalmologist:to embrace technology while never losing sight of the human experience. Whether designing better diagnostic tools or refining reconstructive care, the goal remains unchanged: seeing patients clearly.

Iden Amiri, MD

Amiri is a research fellow at the University of North Carolina at Chapel Hill Department of Ophthalmology. His research interests include glaucoma, oculoplastics, and the integration of AI into ophthalmic care.

Alice Yang Zhang, MD

Zhang is an associate professor and vitreoretinal surgeon at the University of North Carolina at Chapel Hill Department of Ophthalmology. She is also director of the ophthalmology residency program. Her work focuses on medical education, AI in ophthalmology, and clinical and surgical retina.

Financial Disclosures: none

References:

1.Amiri I, Jiao C, Zhang AY. Comprehensive Evaluation of ChatGPT’s Diagnostic Accuracy on Image-Based Ophthalmic Case Interpretations. Poster presented at: American Academy of Ophthalmology Annual Meeting; October 2025; Orlando, Florida.

2. EyeRounds. University of Iowa Ophthalmology and Visual Sciences. Accessed June 12, 2025. https://www.eyerounds.org/#gsc.tab=0

3. Kreso A, Boban Z, Kabic S, et al. Using large language models as decision support tools in emergency ophthalmology. Int J Med Inform. 2025;199:105886. doi:10.1016/j.ijmedinf.2025.105886

4. Kung TH, Cheatham M, Medenilla A, et al. Performance of ChatGPT on USMLE: potential for AI-assisted medical education using large language models. PLOS Digit Health. 2023;2(2):e0000198. doi:10.1371/journal.pdig.0000198

5. Ting DSW, Pasquale LR, Peng L, et al. Artificial intelligence and deep learning in ophthalmology. Br J Ophthalmol. 2019;103(2):167-175. doi:10.1136/bjophthalmol-2018-313173

6. De Fauw J, Ledsam JR, Romera-Paredes B, et al. Clinically applicable deep learning for diagnosis and referral in retinal disease. Nat Med. 2018;24(9):1342-1350. doi:10.1038/s41591-018-0107-6

7. Yim J, Chopra R, Spitz T, et al. Predicting conversion to wet age-related macular degeneration using deep learning. Nat Med. 2020;26(6):892-899. doi:10.1038/s41591-020-0867-7

8. Kleinig O, Gao C, Kovoor JG, Gupta AK, Bacchi S, Chan WO. How to use large language models in ophthalmology: from prompt engineering to protecting confidentiality. Eye (Lond). 2024;38(4):649-653. doi:10.1038/s41433-023-02772-w

9. Salvagno M, Taccone FS, Gerli AG. Artificial intelligence hallucinations. Crit Care. 2023;27(1):180. doi:10.1186/s13054-023-04473-y

10. Ahmed A, Fatani D, Vargas JM, et al. Physicians’ perspectives on ChatGPT in ophthalmology: insights on artificial iIntelligence (AI) integration in clinical practice. Cureus. Published online January 27, 2025. doi:10.7759/cureus.78069

11. Shah D, Kaufmann M, Amiri I, Craft J, Rubinstein D. Pediatric Oculofacial Injuries from Dog Bites: A Retrospective Analysis of Injury Patterns, Management, and Outcomes at a Level 1 Trauma Center. Poster presented at: American Society of Ophthalmic Plastic and Reconstructive Surgery Annual Meeting; October 2025; Orlando, Florida.

12. Patterson KN, Horvath KZ, Minneci PC, et al. Pediatric dog bite injuries in the USA: a systematic review. World J Pediatr Surg. 2022;5(2):e000281. doi:10.1136/wjps-2021-000281

13. Bratton EM, Golas L, Wei LA, Davies BW, Durairaj VD. Ophthalmic manifestations of facial dog bites in children. Ophthalmic Plast Reconstr Surg. 2018;34(2):106-109. doi:10.1097/IOP.0000000000000875

14. Uppuluri S, Nguyen J, Uppuluri A, Langer PD, Bhagat N. Clinical report on the epidemiology of pediatric dog bite-associated ocular injuries. J Pediatr Ophthalmol Strabismus. 2024;61(4):296-297. doi:10.3928/01913913-20240620-04

15. Maurer M, Schlipköter C, Gottsauner M, et al. Animal bite injuries to the face: a retrospective evaluation of 111 cases. J Clin Med. 2023;12(21):6942. doi:10.3390/jcm12216942

16. Chen HH, Neumeier AT, Davies BW, Durairaj VD. Analysis of pediatric facial dog bites. Craniomaxillofac Trauma Reconstr. 2013;6(4):225-232. doi:10.1055/s-0033-1349211

17. Ramgopal S, Macy ML. Pediatric patients with dog bites presenting to US children’s hospitals. Injury Epidemiology. 2021;8(1):55. doi:10.1186/s40621-021-00349-3