Abstract
Introduction:
Assessing the capabilities of ChatGPT-4.0 and ChatGPT-3.5 for diagnosing corneal eye diseases based on case reports and compare with human experts.
Methods:
We randomly selected 20 cases of corneal diseases including corneal infections, dystrophies, degenerations, and injuries from a publicly accessible online database from the University of Iowa. We then input the text of each case description into ChatGPT-4.0 and ChatGPT3.5 and asked for a provisional diagnosis. We finally evaluated the responses based on the correct diagnoses then compared with the diagnoses of three cornea specialists (Human experts) and evaluated interobserver agreements.
Results:
The provisional diagnosis accuracy based on ChatGPT-4.0 was 85% (17 correct out of 20 cases) while the accuracy of ChatGPT-3.5 was 60% (12 correct cases out of 20). The accuracy of three cornea specialists were 100% (20 cases), 90% (18 cases), and 90% (18 cases), respectively. The interobserver agreement between ChatGPT-4.0 and ChatGPT-3.5 was 65% (13 cases) while the interobserver agreement between ChatGPT-4.0 and three cornea specialists were 85% (17 cases), 80% (16 cases), and 75% (15 cases), respectively. However, the interobserver agreement between ChatGPT-3.5 and each of three cornea specialists was 60% (12 cases).
Conclusions:
The accuracy of ChatGPT-4.0 in diagnosing patients with various corneal conditions was markedly improved than ChatGPT-3.5 and promising for potential clinical integration.
Keywords: ChatGPT, Large Language Models (LLM), Generative Pre-trained Transformer (GPT), Artificial Intelligence (AI), Corneal eye diseases, Provisional Diagnosis
INTRODUCTION
The cornea is a clear, non-vascularized tissue that serves as a structural barrier, offering defense against infections to the eye1. Corneal eye diseases encompass a diverse range of conditions, including but not limited to corneal infections, dystrophies, degenerations, and injuries2. Identifying corneal diseases can be challenging and time-consuming particularly when access to specialized eye care provider is limited 3,4. Accurate and timely diagnosis of these conditions is paramount to preserving visual acuity and ensuring optimal patient outcomes.
In recent years, the integration of artificial intelligence (AI) into various medical disciplines has paved the way for innovative approaches to diagnosis and patient care5. Ophthalmology, one of the most imaging intensive fields of medicine, has witnessed a significant transformation with the emergence of AI-powered diagnostic tools6–8. However, AI applications in the anterior segment parts of the eye including cornea9–12 have received less attention compared to the AI applications in posterior segment of the eye including retina. 13–16
Among AI tools, ChatGPT, a cutting-edge large language model (LLM) developed by OpenAI (San Francisco, California), has recently received attention, and holds great potential for comprehending clinical expertise and delivering relevant information17,18. ChatGPT employs deep learning techniques to generate coherent and contextually relevant text based on user inputs19. This AI-driven tool has shown remarkable capabilities in diverse domains since its inception20–22, and its use in the field of ophthalmology is highly promising particularly in the landscape of diagnostics20,23.
This article explores the capabilities of ChatGPT-4.0 (commercially available version 4.0, updated on March 13, 2023) and ChatGPT-3.5 (publicly available version 3.5, updated on August 3, 2022) in diagnosing corneal eye diseases based on detailed case descriptions and comparing with human experts. Gaining insight into the capacities and limitations of such tools can shape the creation of enhanced systems for supporting the diagnoses in an automated way. This, in turn, may enhance triaging as well as patient care for those with corneal eye diseases and mitigate the demands for specialized ophthalmic services particularly in underserved regions.
METHODS
Case Collection
We selected a total of 20 cases with various corneal eye diseases from the openly available database offered by the Department of Ophthalmology and Visual Sciences at the University of Iowa (https://webeye.ophth.uiowa.edu/eyeforum/cases.htm). These 20 cases were selected from over 200 cases which were categorized based on ophthalmic subspecialty. The underlying corneal conditions included corneal infections, dystrophies, degenerations, and injuries including Acanthamoeba keratitis, Acute corneal Hydrops, Atopic Keratoconjunctivitis, Calcific Band Keratopathy, Cogan’s syndrome, Corneal Marginal Ulcer, Cystinosis, Cytarabine induced keratoconjunctivitis, Exposure Keratopathy, Fabry disease, Fuchs’ Endothelial Corneal Dystrophy, Herpes Simplex Viral (HSV) Keratitis, Infectious Crystalline Keratopathy (ICK), Lattice corneal dystrophy type II (Meretoja’s syndrome), Megalocornea, Peripheral Ulcerative Keratitis, Posterior Polymorphous Corneal Dystrophy (PPCD), Pseudophakic Bullous Keratopathy (PBK), Salzmann’s Nodular Degeneration (SND), and Amiodarone-Induced Corneal Deposits (Corneal Verticillata). Details of every case encompassed patient’s demographics, chief complaint, present illness, and major examination findings. Case reports that required specialized exam maneuvers to establish the diagnosis (e.g., Fungal Keratitis-Fusarium) or case reports that are overly obvious (e.g., Chemical Eye Injury) were excluded. Institutional review board (IRB) approval was not required per the direction of our local IRB office as we used a publicly accessible dataset with no patient’s information in this analysis. This study was compliant to the tenets of the Helsinki declaration and ethical aspects was approved by our local research ethics office.
ChatGPT
ChatGPT represents a derivation of the GPT (Generative Pre-trained Transformer) language model tailored for producing text within conversational settings. Through extensive refinement on substantial conversational datasets, it possesses the capability to produce pertinent and logically connected responses in correspondence with provided input24. ChatGPT-4.0 emerged as OpenAI’s most recent language model, embodying a substantial enhancement over its forerunners. GPT-3 was initially trained based on about 176 billion parameters while GPT-4.0 is trained based on approximately 1.75 trillion parameters25. GPT-4.0 stands as an advanced multimodal model that leverages diverse data formats to elevate its performance. As such, ChatGPT possesses some level of computer vision-based image interpretation capabilities that are however not yet appropriate for disease diagnosis24.
ChatGPT Diagnosis
We input identical case descriptions into ChatGPT-4.0 and ChatGPT-3.5 and assessed whether the model was able to provide the correct provisional diagnoses. Specifically, we asked: “What is the most likely diagnosis?” (Fig. 1)
Figure 1.
A sample case description input into the ChatGPT-4.0 model and corresponding responses.
We then compared the accuracy of ChatGPT-4.0 and ChatGPT-3.5 with respect to the correct diagnosis. Due to possible reinforcement learning from human feedback (RLHF) capabilities of ChatGPT, we recorded and used the first answer of ChatGPT for the comparisons.
RESULTS
A comprehensive assessment was conducted, utilizing 20 case presentations spanning various corneal eye diseases, to appraise ChatGPT’s proficiency in generating provisional diagnoses. ChatGPT-4 made the correct diagnosis in 17 out of 20 cases (85%), while ChatGPT-3.5 correctly diagnosed 12 out of 20 cases (60%). Moreover, the three cornea specialists made correct diagnosis of 100% (20 cases), 90% (18 cases), and 90% (18 cases), respectively. The Interobserver agreement between ChatGPT-4.0 and ChatGPT-3.5 was 13 out of 20 cases (65%) while the interobserver agreement between ChatGPT-4.0 and three cornea specialists were 85% (17 cases), 80% (16 cases), and 75% (15 cases), respectively. However, the interobserver agreement between ChatGPT-3.5 and each of three cornea specialists was 60%. Table 1 shows the details of the provisional diagnosis provided by ChatGPT-4.0, ChatGPT-3.5, and human experts. It took approximately 20–40 minutes for the graders to diagnose 20 cases while it took around a couple of minutes (each case around a few seconds) for ChatGPTs to complete diagnosis.
Table 1.
Provisional diagnoses provided by ChatGPT-4.0, ChatGPT-3.5 and Human Experts
| No | Corneal Eye diseases | ChatGPT4 Diagnosis | ChatGPT3.5 Diagnosis | Human Expert Diagnosis |
|---|---|---|---|---|
| 1 | Cystinosis | Cystinosis | Cystinosis | H1: Cystinosis H2: Cystinosis H3: Cystinosis |
| 2 | Fuchs’ Endothelial Corneal Dystrophy (FECD) | FECD | FECD | H1: FECD H2: FECD H3: FECD |
| 3 | Pseudophakic Bullous Keratopathy (PBK) | PBK | Fuchs′ Endothelial Corneal Dystrophy | H1: PBK H2: PBK H3: PBK |
| 4 | Amiodarone-Induced Corneal Deposits (Corneal Verticillata) | Amiodarone-Induced Corneal Deposits | Amiodarone-Induced Corneal Deposits | H1: Amiodarone-Induced Corneal Deposits H2: Amiodarone-Induced Corneal Deposits H3: Amiodarone-Induced Corneal Deposits |
| 5 | Acanthamoeba Keratitis | Acanthamoeba Keratitis | Acanthamoeba Keratitis | H1: Acanthamoeba Keratitis H2: Acanthamoeba Keratitis H3: Acanthamoeba Keratitis |
| 6 | Cogan’s Syndrome (Interstitial Keratitis & Vertigo) | Scleritis | Ocular Rosacea | H1: Cogan’s Syndrome H2: Episcleritis H3: Cogan’s Syndrome |
| 7 | Infectious Crystalline Keratopathy (ICK) | Fungal Keratitis | Recurrent Herpes Simplex Virus Keratitis | H1: ICK H2: ICK H3: ICK |
| 8 | Megalocornea | Megalocornea | Positional Pseudophacodonesi s | H1: Megalocornea H2: Megalocornea H3: Megalocornea |
| 9 | Herpes Simplex Viral Keratitis | Herpes Simplex Viral Keratitis | Herpes Simplex Viral Keratitis | H1: Herpes Simplex Viral Keratitis H2: Herpes Simplex Viral Keratitis H3: Herpes Simplex Viral Keratitis |
| 10 | Atopic Keratoconjunctivitis | Atopic Keratoconjunctivitis | Ocular Cicatricial Pemphigoid (OCP) | H1: Atopic Keratoconjunctivitis H2: Atopic Keratoconjunctivitis H3: OCP |
| 11 | Lattice Corneal Dystrophy Type II (Meretoja’s syndrome) | Meesmann Corneal Dystrophy (MCD) | MCD | H1: Lattice Corneal Dystrophy Type II H2: Lattice Corneal Dystrophy Type II H3: Lattice Corneal Dystrophy Type II |
| 12 | Salzmann’s Nodular Degeneration (SND) | SND | SND | H1: SND H2: SND H3: SND |
| 13 | Exposure Keratopathy | Exposure Keratopathy | Exposure Keratopathy | H1: Exposure Keratopathy H2: Exposure Keratopathy H3: Exposure Keratopathy |
| 14 | Peripheral Ulcerative Keratitis | Peripheral Ulcerative Keratitis | Peripheral Ulcerative Keratitis | H1: Peripheral Ulcerative Keratitis H2: Peripheral Ulcerative Keratitis H3: Peripheral Ulcerative Keratitis |
| 15 | Calcific Band Keratopathy | Calcific Band Keratopathy | Superficial Corneal Scar | H1: Calcific Band Keratopathy H2: Calcific Band Keratopathy H3: Calcific Band Keratopathy |
| 16 | Posterior Polymorphous Corneal Dystrophy (PPCD) | PPCD | Congenital Hereditary Endothelial Dystrophy (CHED) | H1: PPCD H2: Granular Corneal Dystrophy H3: PPCD |
| 17 | Acute Corneal Hydrops | Acute Corneal Hydrops | Acute Corneal Hydrops | H1: Acute Corneal Hydrops H2: Acute Corneal Hydrops H3: Acute Corneal Hydrops |
| 18 | Corneal Marginal Ulcer | Corneal Ulceration | Corneal Ulcer | H1: Corneal Marginal Ulcer H2: Corneal Marginal Ulcer H3: Mooren Ulcer |
| 19 | Fabry Disease | Fabry Disease | Fabry Disease | H1: Fabry Disease H2: Fabry Disease H3: Fabry Disease |
| 20 | Cytarabine Induced Keratoconjunctivitis | Cytarabine Induced Keratoconjunctivitis | Cytarabine Induced Keratoconjunctivitis | H1: Cytarabine Induced Keratoconjunctivitis H2: Cytarabine Induced Keratoconjunctivitis H3: Cytarabine Induced Keratoconjunctivitis |
DISCUSSION
We conducted a prospective study to examine the performance of ChatGPT-4.0 and ChatGPT-3.5 based on 20 cases with different types of corneal eye diseases. The accuracy of ChatGPT-4.0 was 85% while the accuracy of ChatGPT-3.5 was 60%. The interobserver agreement between ChatGPT-4 and the ChatGPT-3.5 was reasonable (65%). We observed that compared to the publicly available ChatGPT-3.5, ChatGPT-4.0, the commercial version, generate markedly improved provisional diagnosis for different corneal eye diseases. These models may assist healthcare providers in generating consistent and useful information regarding the underlying corneal condition.
Some of the capabilities and limitations of ChatGPT in ophthalmology have been discussed previously.20 Recently, ChatGPT was investigated in responding to multiple choice questions from the USMLE and it was observed that ChatGPT correctly responded to over 50% of questions and also provided relevant supporting explanations for the selected choices. 21 More relevant to our study, a recent investigation showed that ChatGPT-3.0 correctly diagnosed 9 out of 10 general ophthalmology case (90%).26 Our accuracy was based on ChatGPT-3.5 was significantly lower (60%) but our ChatGPT-v4.0 accuracy was comparable (85%). Nevertheless, that study assessed general ophthalmology cases while we investigated various corneal conditions that inherently are more challenging to diagnose.
Utilizing conversational AI language models such as ChatGPT could significantly assist frontline healthcare professionals in delivering prompt and precise diagnoses to their patients. In near real-time, such language models could assist primary care and emergency doctors in not only assessing and treating patients but also directing patients to specialized care when required.
Successful integration of ChatGPT or similar LLMs into ophthalmology and cornea services may offer multifaceted benefits. Firstly, ChatGPT’s capability to quickly process large amounts of medical data enhances diagnostic speed and efficiency, leading to quicker patient management and consistent identification of underlying conditions. ChatGPT may transform medical education as well. It could enable students and practitioners to generate interactive and case-based learning materials to foster a deeper understanding of ophthalmic diseases22. Another benefit of ChatGPT-based models is versatility in responding to various kinds of questions, rather than just offering diagnosis based on input images or disease-related parameters. Indeed, ChatGPT was not initially planned to respond to diagnostic support, however, its capability in learning from large corpus has provided ChatGPT to even be applicable in narrowed and specialized areas of disease diagnosis.
In addition to diagnostic capabilities and educational purposes, ChatGPT has the potential to be used for patient education as well. For instance, ChatGPT, as a tool with natural language processing (NLP) capabilities, can translate complex medical terms into simple and accessible language leading to enhanced active patient participation. Collectively, these versatile capabilities of ChatGPT makes it a potential tool that may enhance diagnostics, education, and patient engagement.
Although ChatGPT-4.0 was relatively accurate in making a correct diagnosis on most of the cases, we observed that human expert is more accurate on rare cases. For instance, both ChatGPT versions were incorrect on two rare cases including ICK (case # 7, Table 1) and Lattice Corneal Dystrophy Type II (case #11, Table 1), human experts were correct on both cases. As such, the use of ChatGPT in real-world clinical practice should be considered with caution.
While ChatGPT presents remarkable advantages from several aspects, its potential limitations should be acknowledged as well. The accuracy of ChatGPT lies on the quality and diversity of the training data that it has been exposed to until September 202127. Therefore, the model may encounter challenges when faced with rare or emerging corneal conditions that lack representation in its training dataset. Additionally, ChatGPT’s recommendations should always be validated based on clinical evaluations, as its insights may be derived from non-scientific and publicly available knowledge and historical cases. As such, ChatGPT may generate responses that appear fluent and believable however it may contain factual inaccuracies, a phenomenon often termed as hallucination28.
Although our study is one of the first investigations of ChatGPT capabilities in diagnosing corneal conditions, it has several limitations as well. First, we have used an online and publicly available dataset to evaluate ChatGPT thus there is a concern that this database has been exposed to ChatGPT previously. To address this concern, we reviewed the years that the cases were added to this online database and noticed that case # 20 in Table 1 has been added in 2023 to this database, that is after September 2021 that the latest ChatGPT training completed and both ChatGPT versions were correct on this case. Additionally, both ChatGPT versions were incorrect on numerous cases that have been added to this database prior to September 2021. Therefore, the likelihood that ChatGPT has seen this online database is slight. Second, we have evaluated ChatGPT based on 20 cases thus follow up studies are warranted to evaluate ChatGPT based on larger number of cases to verify our findings. However, obtaining larger databases with a greater number of case reports is highly challenging and requires larger multi-center and multi-disciplinary collaborations. One major obstacle however is ethical considerations and data privacy issues. The utilization of patient data for diagnostic purposes raises concerns about data security and patient confidentiality. Therefore, rigorous safeguards and compliance with regulatory standards are imperative to ensure responsible and ethical use of ChatGPT in cornea research and clinical practice.
The integration of ChatGPT into the diagnosis of corneal eye diseases marks a significant milestone in the evolution of ophthalmic practice. As AI continues to reshape healthcare, ChatGPT’s potential to enhance diagnostic accuracy, expedite patient care, empower medical education, and stimulate research is evident. While challenges exist, a balanced approach that combines AI-generated insights with clinical expertise holds the key to unlocking the full potential of ChatGPT for the diagnosis of corneal conditions. As we peer into the future, the collaboration between AI and ophthalmology promises to redefine the standards of care and elevate patient outcomes in the realm of corneal eye diseases.
CONCLUSION
Corneal diseases encompass a diverse variety of conditions that could be challenging to diagnose. We showed that the accuracy of ChatGPT-4.0 in diagnosing patients with various corneal eye diseases is promising and such models may enhance corneal diagnostics. Additionally, ChatGPT may improve patient interaction and experience as well as medical education. A balanced approach that combines AI-generated insights with clinical findings holds the promise to enhance eye care.
Key summary points:
The aim of this work was to evaluate the performance of ChatGPT-4 and ChatGPT-3.5 for providing the provisional diagnosis of different corneal eye diseases based on case descriptions and compared them with three cornea specialists.
The accuracy of ChatGPT-4.0 in diagnosing patients with various corneal conditions was significantly better than ChatGPT-3.5 based on the specific cases.
The interobserver agreement between ChatGPT-4.0 and ChatGPT-3.5 was 65% while the interobserver agreement between ChatGPT-4.0 and three cornea specialists were 85%, 80%, and 75%, respectively.
ACKNOWLEDGEMENT
Funding
This work was supported by NIH Grants R01EY033005 (SY), R21EY031725 (SY), grants from Research to Prevent Blindness (RPB), New York (SY), and supports from the Hamilton Eye Institute (SY). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Abbreviations:
- LLM
Large Language Model
- AI
Artificial Intelligence
- ChatGPT
Chat Generative Pretrained Transformer
- ICK
Infectious Crystalline Keratopathy
- PPCD
Posterior Polymorphous Corneal Dystrophy
- PBK
Pseudophakic Bullous Keratopathy
- SND
Salzmann’s Nodular Degeneration
- IRB
Institutional Review Board
- RLHF
Reinforcement Learning from Human Feedback
- NLP
Natural Language Processing
- FECD
Fuchs’ Endothelial Corneal Dystrophy
- MCD
Meesmann Corneal Dystrophy
- CHED
Congenital Hereditary Endothelial Dystrophy
Footnotes
Conflict of Interest
Mohammad Delsoz: None.
Yeganeh Madadi: None
Wuqaas M Munir: None
Brendan Tamm: None
Shiva Mehravaran: None
Mohammad Soleimani: None
Ali Djalilian: None
Siamak Yousefi: Remidio, M&S Technologies, Visrtucal Fields, InsihgtAEye, Enolink
Medical Writing/Editorial Assistance
Not applicable.
Ethical Approval
Institutional review board (IRB) approval was not required per the direction of our local IRB office as we used a publicly accessible dataset with no patient’s information in this analysis. This study was compliant to the ethical tenets of the Helsinki declaration and was approved by our local ethical team.
Data Availability
Dataset is online and publicly available.
REFERENCES
- 1.Yang AY, Chow J, Liu J. Corneal Innervation and Sensation: The Eye and Beyond. Yale J Biol Med. Mar 2018;91(1):13–21. [PMC free article] [PubMed] [Google Scholar]
- 2.Clinic C. doi:https://my.clevelandclinic.org/health/diseases/8586-corneal-disease [Google Scholar]
- 3.Solomon SD, Shoge RY, Ervin AM, et al. Improving Access to Eye Care: A Systematic Review of the Literature. Ophthalmology. Oct 2022;129(10):e114–e126. doi: 10.1016/j.ophtha.2022.07.012 [DOI] [PubMed] [Google Scholar]
- 4.Gelston CD, Patnaik JL. Ophthalmology training and competency levels in care of patients with ophthalmic complaints in United States internal medicine, emergency medicine and family medicine residents. J Educ Eval Health Prof. 2019;16:25. doi: 10.3352/jeehp.2019.16.25 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Liu PR, Lu L, Zhang JY, Huo TT, Liu SX, Ye ZW. Application of Artificial Intelligence in Medicine: An Overview. Curr Med Sci. Dec 2021;41(6):1105–1115. doi: 10.1007/s11596-021-2474-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Ting DSJ, Foo VH, Yang LWY, et al. Artificial intelligence for anterior segment diseases: Emerging applications in ophthalmology. Br J Ophthalmol. Feb 2021;105(2):158–168. doi: 10.1136/bjophthalmol-2019-315651 [DOI] [PubMed] [Google Scholar]
- 7.Wawer Matos PA, Reimer RP, Rokohl AC, Caldeira L, Heindl LM, Große Hokamp N. Artificial Intelligence in Ophthalmology - Status Quo and Future Perspectives. Semin Ophthalmol. Apr 2023;38(3):226–237. doi: 10.1080/08820538.2022.2139625 [DOI] [PubMed] [Google Scholar]
- 8.Yousefi S. Clinical Applications of Artificial Intelligence in Glaucoma. J Ophthalmic Vis Res. Jan-Mar 2023;18(1):97–112. doi: 10.18502/jovr.v18i1.12730 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Al-Timemy AH, Alzubaidi L, Mosa ZM, et al. A Deep Feature Fusion of Improved Suspected Keratoconus Detection with Deep Learning. Diagnostics (Basel). May 10 2023;13(10)doi: 10.3390/diagnostics13101689 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Abdelmotaal H, Hazarbasanov R, Taneri S, et al. Detecting dry eye from ocular surface videos based on deep learning. The ocular surface. Jan 26 2023;28:90–98. doi: 10.1016/j.jtos.2023.01.005 [DOI] [PubMed] [Google Scholar]
- 11.Yousefi S, Yousefi E, Takahashi H, et al. Keratoconus severity identification using unsupervised machine learning. PLoS One. 2018;13(11):e0205998. doi: 10.1371/journal.pone.0205998 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Kamiya K, Ayatsuka Y, Kato Y, et al. Keratoconus detection using deep learning of colour-coded maps with anterior segment optical coherence tomography: a diagnostic accuracy study. BMJ Open. Sep 27 2019;9(9):e031313. doi: 10.1136/bmjopen-2019-031313 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Gulshan V, Peng L, Coram M, et al. Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs. JAMA. Dec 13 2016;316(22):2402–2410. doi: 10.1001/jama.2016.17216 [DOI] [PubMed] [Google Scholar]
- 14.Abramoff MD, Lou Y, Erginay A, et al. Improved Automated Detection of Diabetic Retinopathy on a Publicly Available Dataset Through Integration of Deep Learning. Invest Ophthalmol Vis Sci. Oct 1 2016;57(13):5200–5206. doi: 10.1167/iovs.16-19964 [DOI] [PubMed] [Google Scholar]
- 15.Yousefi S, Elze T, Pasquale LR, et al. Monitoring Glaucomatous Functional Loss Using an Artificial Intelligence-Enabled Dashboard. Ophthalmology. Sep 2020;127(9):1170–1178. doi: 10.1016/j.ophtha.2020.03.008 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Thakur A, Goldbaum M, Yousefi S. Predicting Glaucoma before Onset Using Deep Learning. Ophthalmol Glaucoma. Jul-Aug 2020;3(4):262–268. doi: 10.1016/j.ogla.2020.04.012 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Nath S, Marie A, Ellershaw S, Korot E, Keane PA. New meaning for NLP: the trials and tribulations of natural language processing with GPT-3 in ophthalmology. Br J Ophthalmol. Jul 2022;106(7):889–892. doi: 10.1136/bjophthalmol-2022-321141 [DOI] [PubMed] [Google Scholar]
- 18.Singhal K, Azizi S, Tu T, et al. Large language models encode clinical knowledge. Nature. Aug 2023;620(7972):172–180. doi: 10.1038/s41586-023-06291-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Ramponi M. How ChatGPT actually works. AssemblyAI. Dec 23, 2022; [Google Scholar]
- 20.Antaki F, Touma S, Milad D, El-Khoury J, Duval R. Evaluating the Performance of ChatGPT in Ophthalmology: An Analysis of Its Successes and Shortcomings. Ophthalmol Sci. Dec 2023;3(4):100324. doi: 10.1016/j.xops.2023.100324 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.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. Feb 2023;2(2):e0000198. doi: 10.1371/journal.pdig.0000198 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Dave T, Athaluri SA, Singh S. ChatGPT in medicine: an overview of its applications, advantages, limitations, future prospects, and ethical considerations. Front Artif Intell. 2023;6:1169595. doi: 10.3389/frai.2023.1169595 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Michael Balas EBI. Conversational AI Models for ophthalmic diagnosis: Comparison of ChatGPT and the Isabel Pro Differential Diagnosis Generator. JFO Open Ophthalmology. 20x23;doi: 10.1016/j.jfop.2023.100005 [DOI] [Google Scholar]
- 24.Ortiz S. What is ChatGPT and why does it matter? ZDNET Tech. Aug 2023; [Google Scholar]
- 25.Lubbad M. GPT-4 Parameters: Unlimited guide NLP’s Game-Changer. March 2023; [Google Scholar]
- 26.Balas M, Ing EB. Conversational AI Models for ophthalmic diagnosis: Comparison of ChatGPT and the Isabel Pro Differential Diagnosis Generator. JFO Open Ophthalmology. 2023/March/01/ 2023;1:100005. doi: 10.1016/j.jfop.2023.100005 [DOI] [Google Scholar]
- 27.Jungwirth D, Haluza D. Artificial Intelligence and Public Health: An Exploratory Study. Int J Environ Res Public Health. Mar 3 2023;20(5)doi: 10.3390/ijerph20054541 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Ji Z, Lee N, Frieske R, et al. Survey of hallucination in natural language generation. ACM Computing Surveys. 2023;55(12):1–38. [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
Dataset is online and publicly available.