L. John Greenfield Jr, MD, PhD
ARTIFICIAL INTELLIGENCE (AI) is all the rage. Applications range from writing articles or term papers to recognizing faces, generating paintings in the style of a specific painter, and writing computer code. In neurology, AI imaging software known as RapidAI helped turn the tide for interventional stroke care by improving patient selection.1,2
But is there a dark side to AI? Will software replace trained neurologists in making diagnostic or therapeutic decisions? What happens when the software is wrong? What are the implications of AI for academic neurology? To get a feeling for what the future might bring, I had a chat on Microsoft’s Bing with the latest version of OpenAI’s large language model (LLM) software, ChatGPT-4.3 The following is a lightly edited version of this conversation.
Note: ChatGPT-4 provided its own references, and thus will have a separate reference list.
A Chat With ChatGPT-4
LJG: Tell me a little about yourself.
ChatGPT-4: I’m an AI language model created by OpenAI. I’m designed to help you with your queries and provide you with information. I can help you with a wide range of topics such as science, technology, entertainment, and more.
LJG: What can AI do to help us take care of patients with neurological diseases?
ChatGPT-4: AI can help analyze medical data in disease prevention, diagnosis, patient monitoring, and development of new protocols, but can also assist clinicians in dealing with voluminous data in a more accurate and efficient manner.1-5 AI has been helpful in neurodegenerative disorders including Parkinson’s, Alzheimer’s, Amyotrophic Lateral Sclerosis (ALS)1 as well as traumatic brain injury (TBI).3 AI-enabled CT scanning for hemorrhagic patients allows automation of lesion segmentation and detection of hemorrhagic expansion.1 Incorporating advanced AI technologies into the delivery system of neurologic care has tremendous potential to improve health care, but also carries with it unique challenges.4,5
LJG: What databases do you have access to?
ChatGPT-4: As an AI language model, I have access to a wide range of databases such as PubMed, Google Scholar, and more.
LJG: Some physicians are concerned that AI might replace them in making decisions about patient care. Do you think this is possible?
ChatGPT-4: AI is designed to assist physicians in making decisions about patient care, not replace them. AI can help physicians analyze medical data in disease prevention, diagnosis, patient monitoring, and development of new protocols, but can also assist clinicians in dealing with voluminous data in a more accurate and efficient manner. AI can also help to reduce the workload of physicians, allowing them to focus on more complex cases and to spend more time with their patients. AI can help physicians make more informed decisions about patient care, but it cannot replace the human touch and empathy that is required in patient care.
LJG: What role could AI play in teaching neurology to medical students and residents?
ChatGPT-4: AI can play a significant role in teaching neurology.6-9 AI can provide trainees with interactive simulations, virtual reality, and augmented reality experiences.4 Interactive simulations of real-life medical scenarios10 can allow them to practice medical procedures and techniques,11 which help them learn in a more efficient and effective manner.6,12,13 AI can also provide personalized feedback and assessments.4 AI can also provide access to a vast amount of medical data and research.6
LJG: Give an example of how ChatGPT-4 could provide positive feedback to a medical student who had disorganized patient presentations and an average knowledge base, but good neurological exam skills.
ChatGPT-4: “Congratulations on your excellent neuro exam skills! Your ability to accurately assess patients’ neurological function is a valuable asset. I also appreciate your commitment to improving your patient presentations despite their initial disorganization. I have no doubt that with practice, you will continue to improve your presentation skills and become an even more effective communicator with patients and colleagues. Keep up the great work!”
LJG: AI is used in the interpretation of medical images such as brain MRI and CT scans. Could it be used to integrate information and help with diagnoses?
ChatGPT-4: Yes, AI is already being used to interpret medical images.14-17 AI can support the diagnosis and treatment decision-making process by facilitating the integration of the imaging results within the patient electronic medical record.14 Once incorporated, the image can then be used alongside patient clinical data and medical history in computer-aided diagnosis. AI can also help medical professionals identify imaging abnormalities and enhance tissue-based detection and characterization.
LJG: Could AI be used as a medical scribe to help physicians generate notes for the medical record?
ChatGPT-4: Yes. AI medical transcription uses AI and natural language processing to listen into the natural patient-clinician conversation, parse out the medically relevant information, and summarize that information into compliant notes that map back to the appropriate fields of an EHR.18-22 AI-powered medical scribes bring the quality and experience of a human medical scribe, but at the cost and scalability of a dictation service.18 Nuance has created an AI-powered ambient clinical intelligence solution called the Dragon Ambient eXperience (DAX) that can replace medical scribes.19
LJG: What are the limitations of using AI? Have there been examples of bad outcomes when physicians have trusted AI decision support?
ChatGPT-4: Although AI has come a long way in the medical world, human surveillance is still essential.23-26 AI may overlook social variables such as class, gender, ethnicity, age grouping, and group identity.24 Inaccuracies are still possible.23 AI is susceptible to security risks.23 AI may lead to unemployment.24 However, AI can help medical professionals identify imaging abnormalities and enhance tissue-based detection and characterization.23 In 2016, researchers at Beth Israel Deaconess Medical Center reported that an AI-powered diagnostic program correctly identified cancer in pathology slides 92% of the time, just shy of trained pathologists’ 96%. Combining the 2 methods led to 99.5% accuracy.26
[Note: This positive spin on AI’s potential didn’t provide any examples of errors, so I tried again.]
LJG: Provide examples of bad outcomes when physicians have trusted AI decision support systems.
ChatGPT-4: While AI decision support systems have the potential to greatly improve medical care, there have been instances where their use has led to negative outcomes. Here are a few examples:
- Misdiagnosis and delayed treatment: In some cases, AI decision support systems have incorrectly identified a medical condition, leading to a misdiagnosis and delayed treatment. For example, an AI system might fail to recognize a rare disease or misinterpret a symptom, causing the physician to overlook important diagnostic information.
- Overreliance on technology: In some cases, physicians have become overly reliant on AI decision support systems, leading to complacency and a decrease in critical thinking. This can result in errors, particularly if the AI system provides incorrect recommendations or fails to account for all relevant factors.
- Legal and ethical concerns: There are also legal and ethical concerns related to the use of AI decision support systems in health care. For example, if an AI system provides a recommendation that leads to harm, who is responsible for the consequences? Additionally, there are concerns about data privacy and bias in AI algorithms.
- System failure: Like any technology, AI decision support systems can experience technical failures, leading to incorrect or incomplete recommendations. This can result in serious harm to patients, particularly if the physician relies on the AI system without independently verifying its recommendations. It’s important to note that while these negative outcomes are possible, they are not necessarily representative of the typical use of AI decision support systems in health care. With proper design, implementation, and oversight, AI systems can provide valuable support to physicians and improve patient outcomes.
LJG: What are the potential uses of AI for writing academic papers or grant proposals in neurological research?
ChatGPT-4: Here are a few examples:
- Literature review: AI algorithms can quickly search and analyze vast amounts of literature to identify relevant studies and synthesize information. This can help researchers save time and improve the accuracy of their literature reviews.
- Writing assistance: AI systems can provide suggestions for improving the clarity, organization, and coherence of a manuscript. For example, they can suggest changes to sentence structure, word choice, or paragraph organization.
- Data analysis: AI algorithms can analyze complex datasets and identify patterns or relationships that may not be immediately apparent to human researchers. This can help researchers gain new insights into neurological diseases and potential treatments.
- Grant proposal writing: AI systems can assist in drafting grant proposals by providing suggestions for improving the structure and content of the proposal. They can also provide guidance on how to effectively communicate the importance of the proposed research and its potential impact.
- Translation assistance: AI can assist in translating papers or grant proposals from one language to another, which can be particularly useful for international collaborations.
Overall, the use of AI in academic writing can potentially save researchers time and improve the quality of their work. However, it is important to note that AI systems should not replace human expertise and judgment, and that their use should be carefully monitored to ensure that the results are accurate and unbiased.
LJG: How could AI be used to generate new medical knowledge? Has AI been used to look for new diseases or conditions based on large-scale information or data correlations that have not been previously detected?
ChatGPT-4: Yes. Artificial intelligence in medicine is the use of machine learning models to search medical data and uncover insights to help improve health outcomes and patient experiences.27,28 In a proof-of-concept study published in Frontiers in Immunology, a new artificial intelligence (AI) algorithm was developed which uses novel weight engineering methods to extract meaningful biological knowledge learned by AI models.29 Machine learning (ML), large and often unstructured datasets, advanced sensors, natural language processing (NLP) and robotics are all being used in a growing number of health care sectors.28 AI has been used to predict the structures of almost every protein made by the human body.30 Attention maps may be used in the clinic to aid diagnosis, discover new medical knowledge, and identify novel features and characteristics of medical tests.27
LJG: How do patients feel about talking with AI? Could it provide psychotherapy or counseling?
ChatGPT-4: Patients’ feelings about talking with AI vary depending on the clinical application.31-34 For example, a survey from close to 1000 respondents showed that most patients believed that AI would make health care much better or somewhat better (11% and 45%, respectively) with relatively few that feel AI would render health care worse or much worse (4% and 1%, respectively).31 However, participants also voiced concerns that AI tools might increase health care costs and that those costs might be passed on to patients.34 AI can provide psychotherapy or counseling, but it is not a substitute for human interaction. AI can be used to provide support and guidance to patients, but it cannot replace the human touch that is so important in psychotherapy and counseling.
LJG: What will be the future of AI in neurology?
ChatGPT-4: The future of AI in neurology is promising.35-38 AI can be used for predicting outcomes of seizure disorders, grading of brain tumors, upskilling neurosurgical procedures, and rehabilitation of stroke patients with smartphone apps monitoring patient symptoms and progress.35 AI can now diagnose stroke from CT/MRI scans and detect papilledema and diabetic retinopathy from retinal scans.35 AI will cause a paradigm shift in the delivery of health care.36
The Physician’s Take
The ability of ChatGPT-4 to harvest information from the internet and package it into coherent, well-organized, and (mostly) grammatically correct paragraphs is both exciting and a bit frightening. Although the chatty quality of the interface inspires confidence, the answers are often superficial and do not reflect “thoughtful” consideration—they are search results presented in a better organized and linguistically sophisticated way. Moreover, the information provided by AI (or any search engine) is based on unseen algorithms that may be influenced by volume of citations (ie, popularity) or, worse, by undeclared commercial interests. However, ChatGPT-4 and other LLMs are just 1 aspect of AI, and the potential of AI approaches to assist in neurological research, education, diagnosis, and care is enormous.4 Even so, AI lacks judgment based on real-world experience and the intangible information derived directly from patients and the clinical situation. AI has a lot of potential but still requires oversight. For now, our jobs appear safe.
REFERENCES
1.Albers GW, Marks MP, Kemp S, et al. Thrombectomy for stroke at 6 to 16 hours with selection by perfusion imaging. N Engl J Med. 2018;378(8):708-718. doi:10.1056/NEJMoa1713973
2. Nogueira RG, Jadhav AP, Haussen DC, et al. Thrombectomy 6 to 24 hours after stroke with a mismatch between deficit and infarct. N Engl J Med. 2018;378(1):11-21.doi:10.1056/NEJMoa1706442
3. Source: Conversation with Bing. Microsoft Bing.March 16, 2023. https://www.bing.com/search?form=MY0291&OCID=MY0291&q=Bing+AI&showconv=1
4.Hillis JM, Bizzo BC. Use of Artificial Intelligence in Clinical Neurology. Semin Neurol. 2022;42(1):39-47. doi:10.1055/s-0041-1742180
ChatGPT4 REFERENCES
1.Patel UK, Anwar A, Saleem S, Malik P, Rasul B, Patel K, Yao R, Seshadri A, Yousufuddin M, Arumaithurai K. Artificial intelligence as an emerging technology in the current care of neurological disorders. J Neurol. 2021 May;268(5):1623-1642. doi: 10.1007/s00415-019-09518-3. Epub 2019 Aug 26. PMID: 31451912.
2. Six, O. Artificial intelligence in neurology: promising research and proven applications. May 11, 2020. Quantib. http://www.quantib.com/blog/artificial-intelligence-neurology-promising-research-and-proven-application Accessed 3/16/2023.
3. Javaid, S. Top 4 AI Use Cases in Neurology in 2023 – AIMultiple. Updated 1/20/23. https://research.aimultiple.com/neurology-ai/ Accessed 3/16/2023. Old 6
4. Jones DT, Kerber KA. Artificial Intelligence and the Practice of Neurology in 2035: The Neurology Future Forecasting Series. Neurology. 2022 Feb 8;98(6):238-245. doi: 10.1212/WNL.0000000000013200. PMID: 35131918.
5. Savage N. How AI and neuroscience drive each other forwards. Nature. 2019 Jul;571(7766):S15-S17. doi: 10.1038/d41586-019-02212-4. PMID: 31341311.
6. Sandrone S, Carlson CE. Future of Neurology & Technology: Virtual and Augmented Reality in Neurology and Neuroscience Education: Applications and Curricular Strategies. Neurology. 2021 Jun 29:10.1212/WNL.0000000000012413. doi: 10.1212/WNL.0000000000012413. Epub ahead of print. PMID: 34187858.
7. Masters K. Artificial intelligence in medical education. Med Teach. 2019 Sep;41(9):976-980. doi: 10.1080/0142159X.2019.1595557. Epub 2019 Apr 21. PMID: 31007106.Old 12
8. Vishnu VY, Vinny PW. The Neurologist and Artificial Intelligence: Titans at Crossroads. Ann Indian Acad Neurol. 2019 Jul-Sep;22(3):264-266. doi: 10.4103/aian.AIAN_493_18. PMID: 31359934; PMCID: PMC6613413.
9. Keser Z, Rodriguez YA, Tremont J, Hsieh PH, McCullough LD, Sandrone S, Stimming EF. The role of residents in medical students’ neurology education: current status and future perspectives. BMC Med Educ. 2020 Apr 16;20(1):115. doi: 10.1186/s12909-020-02036-1. PMID: 32299428; PMCID: PMC7164350. Old 15
10. Wu Q, Wang Y, Lu L, Chen Y, Long H, Wang J. Virtual Simulation in Undergraduate Medical Education: A Scoping Review of Recent Practice. Front Med (Lausanne). 2022 Mar 30;9:855403. doi: 10.3389/fmed.2022.855403. PMID: 35433717; PMCID: PMC9006810. Old 17
11. How AI Might One Day Revolutionize Clinical Simulation Training. https://www.healthysimulation.com/30417/ai-simulation-technology/ Accessed 3/16/2023.
12. Artificial Intelligence in Medical Education and Simulation – Region H. https://www.regionh.dk/CAMES-english/Research/Research-Areas/Pages/Artificial-Intelligence-in-Medical-Education-and-Simulation.aspx Accessed 3/16/2023.
13. Medscape Patient Simulations. https://www.medscape.org/simulation Accessed 3/16/2023.
14. Oren O, Gersh BJ, Bhatt DL. Artificial intelligence in medical imaging: switching from radiographic pathological data to clinically meaningful endpoints. Lancet Digit Health. 2020 Sep;2(9):e486-e488. doi: 10.1016/S2589-7500(20)30160-6. PMID: 33328116.
15. Artificial Intelligence and Diagnostic Errors | PSNet. https://psnet.ahrq.gov/perspective/artificial-intelligence-and-diagnostic-errors Accessed 3/16/2023.
16. Medical Imaging with AI: Revolutionizing the Healthcare Industry. https://theblue.ai/blog/trends-en/medical-imaging-blog/ Accessed 3/16/2023.
17. Lynch, AH. Realizing the promise of AI for medical imaging and diagnosis. https://www.medtechdive.com/news/realizing-the-promise-of-ai-for-medical-imaging-and-diagnosis/640740/ Accessed 3/16/2023.
18. The Rise of Artificial Intelligence in Medical Transcription – DeepScribe. https://www.deepscribe.ai/resources/the-rise-of-artificial-intelligence-in-medical-transcription Accessed 3/16/2023.
19. How Ambient Clinical Intelligence Replaces Scribes | What’s next. https://whatsnext.nuance.com/healthcare-ai/how-ambient-clinical-intelligence-replaces-scribes/ Accessed 3/16/2023. Old 30
20. ScribeAmerica – Medical Scribe Program for Doctors, Hospitals & Eds. https://www.scribeamerica.com/ Accessed 3/16/2023. Old 31
21. DeepScribe – AI-Powered Medical Scribe. https://www.deepscribe.ai/ Accessed 3/16/2023. Old 32
22. Quantum Health – A.I. Powered Medical Scribe. https://www.quantumai.health/ Accessed 3/16/2023. Old 33
23. Pros & Cons of Artificial Intelligence in Medicine | Drexel CCI. https://drexel.edu/cci/stories/artificial-intelligence-in-medicine-pros-and-cons/ Accessed 3/16/2023.
24. Enedah, V. AI in Medicine: Applications & Limitations | Medium. https://valentineenedah.medium.com/ai-in-medicine-applications-limitations-d9d341ace276 Accessed 3/16/2023.
25. Powell, A. Risks and benefits of an AI revolution in medicine. https://news.harvard.edu/gazette/story/2020/11/risks-and-benefits-of-an-ai-revolution-in-medicine/ Accessed 3/16/2023.
26. Kaul V, Enslin S, Gross SA. History of artificial intelligence in medicine. GastrointestEndosc. 2020 Oct;92(4):807-812. doi: 10.1016/j.gie.2020.06.040.
27. Greenfield D, Wilson S. Artificial Intelligence in Medicine: Applications, implications, and limitations. https://sitn.hms.harvard.edu/flash/2019/artificial-intelligence-in-medicine-applications-implications-and-limitations/ Accessed 3/16/2023.
28. Artificial Intelligence in Medicine | IBM. https://www.ibm.com/topics/artificial-intelligence-medicine Accessed 3/16/2023.
29. Gallagher C.New AI algorithm opens avenues for knowledge discoveries. https://individualizedmedicineblog.mayoclinic.org/2022/08/01/new-ai-algorithm-opens-avenues-for-knowledge-discoveries/ Accessed 3/16/2023.
30. Calloway, E. DeepMind’s AI predicts structures for a vast trove of proteins. https://www.nature.com/articles/d41586-021-02025-4
31. Chang A. How do patients feel about AI in healthcare? – AIMed. https://ai-med.io/literature/how-do-patients-feel-about-ai-in-healthcare/ Accessed 3/16/2023.
32. Richardson JP, Smith C, Curtis S, Watson S, Zhu X, Barry B, Sharp RR. Patient apprehensions about the use of artificial intelligence in healthcare. NPJ Digit Med. 2021 Sep 21;4(1):140. doi: 10.1038/s41746-021-00509-1. PMID: 34548621; PMCID: PMC8455556.
33. Gaudette R. How Do Patients Feel About Using AI in Health Care?. https://medicine.yale.edu/news-article/how-do-patients-feel-about-using-ai-in-healthcare/ Accessed 3/16/2023
34. McNamara D. How Do Patients Feel About AI in Health Care? It Depends – WebMD. https://www.webmd.com/a-to-z-guides/news/20220512/how-do-patients-feel-about-ai-in-health-care-it-depends Accessed 3/16/2023
35. 10 top artificial intelligence (AI) applications in healthcare. https://venturebeat.com/ai/10-top-artificial-intelligence-ai-applications-in-healthcare/ Accessed 3/16/2023
36. Rincon P. AI breakthrough could spark medical revolution – BBC News. https://www.bbc.com/news/science-environment-57929095 Accessed 3/16/2023. Old 47
37. Vinny PW, Vishnu VY, Padma Srivastava MV. Artificial Intelligence shaping the future of neurology practice. Med J Armed Forces India. 2021 Jul;77(3):276-282. doi: 10.1016/j.mjafi.2021.06.003. Epub 2021 Jul 1. PMID: 34305279; PMCID: PMC8282510.
38. Ganapathy K, Abdul SS, Nursetyo AA. Artificial intelligence in neurosciences: A clinician’s perspective. Neurol India. 2018 Jul-Aug;66(4):934-939. doi: 10.4103/0028-3886.236971.
link