Comprehensive Summary
This paper, presented by Kuang et al., investigates the development of ChatGPT in ophthalmology and its application in the clinical field. The most recent breakthroughs in AI stem from the fact that the large language models (LLM) no longer utilize unimodal text but use multimodal synergy, enabling more complex and accurate responses as well as the ability to create images, data, and textual data. This allows a more seamless and efficient use of AI in the clinical setting of ophthalmology. Data was collected through MEDLINE and Scopus, with emphasis on studies between 2018 and 2025. 2018 was chosen as the starting point because of the release date for the very first model of ChatGPT. The paper lists main benefits of how large language models can be applied into the field of ophthalmology. The major application that LLM’s can have is to assist with clinical diagnosis. Due to the nature of ophthalmology, diagnosis requires a comprehensive evaluation of the patient’s condition beyond the eyes. If LLM’s are able to access patient’s data, it would be able to analyze the overall condition of the patient and make a diagnosis. In the study, it was illustrated that ChatGPT Plus v4.0 had similar accuracy results to neuro-ophthalmologists, indicating a positive outlook for the future. Apart from diagnosis, LLM’s can be utilized for education, research, and medical history collection. The ability for an LLM to automate writing would be able to save enormous amounts of time for physicians. However, the paper noted challenges and improvements needed for advances in applying LLM’s in ophthalmology. Firstly, LLM’s utilize a black box, making their decisions more analytical rather than intuitive. This will result in the inability to effectively explain the relevant information and can cross ethical boundaries through informed consent. Additionally, the information needed for ChatGPT to effectively be applied may be concerning to patients for the amount of their information that needs to be fed to a dataset. Lastly, ChatGPT and LLM’s are still prone to producing inaccurate mistakes, and if presented to a patient could lead to negative results.
Outcomes and Implications
Kuang et al., in this study evaluates the applications of ChatGPT in clinical practice and the benefits and drawbacks. This paper has highlighted how the utilization of ChatGPT can assist with more thorough and accurate diagnosis and treatment plans due to its ability to sift through patient data and compare with datasets. Additionally, being able to automate writing and busy tasks, it would allow physicians to spend less time facing a computer and writing up information and instead see more patients. The accuracy of ChatGPT and other LLM’s in diagnosis continues to increase as we see continued development and advancements in building LLMs. However, there are still many issues that need to be addressed. The two main issues that arise when applying LLM’s to clinical practice are ethical concerns and validity of data. ChatGPT is still prone to making mistakes, which can be detrimental to the patient and impractical to the physician. For future directions to this issue, researching more about how LLM models function and the collection of more data into a dataset that LLMs can rely on to produce a result. From an ethical standpoint, the usage of LLMs can create issues surrounding patient privacy as well as full disclosure of any result that is produced from the LLMs. To combat this, data that LLMs utilize should be encrypted and many security measures should be in place to keep data secured. Additionally, further understanding of how the LLM approaches an answer needs to occur to provide necessary information to the patient. Overall, Kuang et al. implies that the utilization of LLMs in clinical practice has a positive outlook, but is not quite ready to be applied until further development and safety measures are applied to these large language models.