Innovations in ophthalmology were prominently showcased during the 2025 American Academy of Ophthalmology (AAO) and American Society of Ophthalmic Plastic and Reconstructive Surgery (ASOPRS) meetings, focusing on the intersection of artificial intelligence (AI) and clinical practice. Two studies presented at the events illustrated the evolving landscape of vision science: one assessed the diagnostic capabilities of ChatGPT in interpreting ophthalmic cases, while the other investigated the epidemiology of pediatric oculofacial injuries resulting from dog bites.
The first study, titled “Comprehensive Evaluation of ChatGPT’s Diagnostic Accuracy on Image-Based Ophthalmic Case Interpretations,” was led by Iden Amiri, MD, and Cheng Jiao, MD, with Alice Yang Zhang, MD, as senior author. This investigation is among the most thorough to date examining a large language model’s proficiency in clinical reasoning within the ophthalmology field. Utilizing 261 real-world cases from the University of Iowa’s EyeRounds repository, which encompasses various subspecialties, the study tested ChatGPT-4o under two conditions: one where the model had access to complete patient histories and examination findings, and another where it interpreted raw clinical images without accompanying text.
The findings revealed significant disparities in diagnostic accuracy linked to the level of context provided. When given the complete clinical narrative, ChatGPT achieved an 80.1% diagnostic accuracy rate, a stark contrast to the 54.7% accuracy noted when relying solely on images (χ² = 48.00; P < .001). Accuracy varied across subspecialties, being highest in pediatrics and oculoplastics, while glaucoma and neuro-ophthalmology proved more challenging. The study underscored that treatment recognition was a crucial predictor of diagnostic success, suggesting the model excels in structured decision-making yet falters without contextual information.
“ChatGPT performed like a well-trained resident when it had the full story,” Jiao stated, emphasizing the dual nature of AI’s capabilities and limitations. The researchers noted that, in the image-only scenario, the model’s success hinged on its ability to identify clinical signs and treatment patterns, indicating a reliance on pattern recognition rather than deeper pathophysiological understanding. However, when provided with full context, unmeasured factors appeared to play a more significant role in diagnostic accuracy.
The implications of this study extend beyond the realm of diagnostic trivia. With the potential to assist in triage and enhance educational experiences for residents, multimodal AI could significantly improve access to subspecialty care, especially in underserved areas. Nonetheless, the researchers cautioned against potential risks such as privacy concerns when handling patient data, hallucinations by the AI, and the dangers of overreliance on algorithmic outputs without clinician oversight. “AI is a tool, not a replacement for clinical judgment,” Zhang asserted, advocating for the thoughtful integration of technology to enhance care while maintaining accountability and data security.
In another notable presentation at the ASOPRS 2025, Dhruv Shah, MD, unveiled findings from a retrospective analysis of pediatric oculofacial injuries due to dog bites, based on data from 818 encounters between 2009 and 2024. This focused study highlighted the unique challenges posed by canine bites in the periocular region, an area previously overlooked in pediatric research. The analysis identified 81 eyes with confirmed periocular involvement, revealing that most injuries were either lacerating or avulsive rather than penetrating.
Age-related patterns in injury were discerned, indicating that younger children are particularly at risk of canalicular damage, likely due to their facial structure aligning with canine snouts. Management strategies ranged from simple closure to complex oculoplastic repairs, with some cases necessitating stenting or secondary interventions. Complications such as infection and eyelid malposition were noted, but overall, long-term outcomes were favorable with prompt management.
“Recognizing high-risk patterns helps emergency physicians and ophthalmologists prioritize early oculoplastic consultation,” Shah explained. His study not only emphasizes the importance of canalicular assessment, especially in toddlers, but also advocates for preventive counseling and collaboration with pediatricians to educate families about dog safety, particularly around young children. Given that most dog-bite incidents occur in familiar settings involving household pets, proactive guidance may serve as the most effective preventative measure.
Despite the methodological differences, both studies converge on a vital theme: the combination of innovation and compassion within ophthalmology. The AI study reminds practitioners that even sophisticated algorithms depend on clinical narratives, emphasizing the importance of patient context. Meanwhile, the pediatric trauma analysis sheds light on the human experiences underlying clinical cases, highlighting the emotional stakes for children and families, as well as the commitment of surgeons striving to restore vision and confidence. Collectively, these findings underscore the dual obligation of modern ophthalmologists to embrace technological advancements while remaining attuned to the human experiences that define their practice.
See also
ASU Grad Aashritha Machiraju Advances AI for Health Solutions, Wins Impact Award
Axiom Launches AI Frontier Agents at AWS re:Invent 2025, Pioneered by Carina Hong
Human Researchers Outperform AI in Medical Systematic Reviews, Study Finds
