Artificial intelligence in pediatric ophthalmology: a comparative study of ChatGPT-4.0 and DeepSeek-R1 performance.

Objective

This study aims to evaluate and compare the accuracy and performance of two large language models (LLMs), ChatGPT-4.0 and DeepSeek-R1, in answering pediatric ophthalmology-related questions.

Methods

A total of 44 multiple-choice questions were selected, covering various subspecialties of pediatric ophthalmology. Both LLMs were tasked with answering these questions, and their responses were compared in terms of accuracy.

Results

ChatGPT-4.0 correctly answered 82% of the questions, while DeepSeek-R1 achieved a higher accuracy rate of 93% (p: 0.06). In strabismus, ChatGPT-4.0 answered 70% of questions correctly, while DeepSeek-R1 achieved 82% (p: 0.50). In other subspecialties, ChatGPT-4.0 answered 89% correctly, and DeepSeek-R1 achieved 100% accuracy (p: 0.25).

Conclusion

DeepSeek-R1 outperformed ChatGPT-4.0 in overall accuracy, particularly in pediatric ophthalmology. These findings suggest the need for further optimization of LLM models to enhance their performance and reliability in clinical settings, especially in pediatric ophthalmology.