CoChat
Sign Up
Cited by 52Artificial Intelligence in Healthcare and EducationEpilepsy research and treatmentMachine Learning in Healthcare

Are AI language models such as ChatGPT ready to improve the care of individuals with epilepsy?

Open access
Boßelmann, Christian M., Leu, Costin, Lal, Dennis·2023·Epilepsia
DOI 10.1111/epi.17570
Open this paper in CoChat — free accountOpen PDF

Abstract

Epilepsy is a neurological disorder characterized by recurrent seizures, which can significantly impact the quality of life of affected individuals. Fortunately, advances in artificial intelligence (AI) are providing new opportunities to improve the diagnosis and treatment of epilepsy. Briefly, examples of ongoing epilepsy-related AI research include (1) algorithms that can analyze large amounts of electroencephalography (EEG) time-series data to label interictal epileptiform discharges both independently and with human supervision,1, 2 (2) diagnostic biomedical imaging with automated magnetic resonance imaging (MRI)–based lesion detection, surgical decision-making support, and outcome prediction,3, 4 and (3) Clinical Decision Support Systems (CDSS) that use patient data to provide physicians with recommendations based on up-to-date evidence and guidelines for an, overall, improved diagnostic and therapeutic accuracy.5, 6 Language models are often used in chatbots and other conversational systems to generate context-aware human-like text in response to an input prompt from a user. Such models are trained on large data sets of human conversations using machine learning (ML) techniques to learn the patterns and structure of natural language. Various artificial intelligence (AI) language models have been developed since the 1950s, but significant advances have only been made in recent years due to improved ML models paired with an increased availability of large amounts of data and computational resources. Some of the earliest examples of such models include ELIZA, developed in the 1960s (one of the first programs to simulate a patient-doctor relationship), and SHRDLU from the 1970s (a program able to emulate dialogue around a simplified world with a limited number of objects, the "blocks world").7, 8 However, these early language models were inherently limited in their capabilities and could perform only a narrow range of tasks. In recent years, more complex, large language models have led to significant progress in natural language processing. Several of these AI language models can be used for dialogue, for example, (1) GPT-3 (Generative Pre-trained Transformer 3), a state-of-the-art language model developed by OpenAI that can generate contextual human-like text for a wide range of applications, including dialogues9; (2) DialoGPT, a language model developed by Microsoft that is trained on a large data set of social media comment chains and can generate responses in single-turn conversations10; (3) Meena, a sensible and specific language model developed by Google that is trained on human–human conversations from public-domain social media and can generate responses that are coherent and contextually appropriate11; and (4) XLNet, a language model developed by Google and Carnegie Mellon University that is capable of several language modeling tasks including question answering, natural language inference, sentiment analysis, and document ranking; and many others.12 Mainly such algorithms enable the analysis of free-text electronic medical records and other written materials (e.g., test results and treatment plans) that are otherwise inaccessible without preprocessing and standardization. By analyzing large amounts of free-text medical records, language models can learn to identify and summarize relevant patterns. Possible outcomes are information on identified hierarchical patient subgroups based on seizure patterns, documented treatment options, and outcome parameters.13-15 This structured information could be queried to provide personalized treatment recommendations based on medical history and other relevant factors. For example, by identifying early candidates for epilepsy surgery, language models can help minimize treatment delays and improve patient outcomes.16, 17 Another example of how language models can improve health care are Clinical Decision Support Systems (CDSS) trained to understand and offer natural responses to queries from health care providers. CDSS can provide medical or surgical treatment recommendations, suggest relevant clinical guidelines or protocols, and alert health care providers to potential errors or risks. Similar methods may be used to create virtual assistants for individuals with epilepsy to answer questions and provide easy access to information about their condition, treatment options, and other related topics, including driving, causes of premature death (including sudden unexpected death in epilepsy [SUDEP]), and status epilepticus.18, 19 Overall, AI language models have the future potential to significantly improve the care and management of individuals with epilepsy by providing natural conversational interfaces to both patients and physicians, allowing for easy access to structured information. We tested ChatGPT (ChatGPT Dec 15 Version, available at chat.openai.com, last accessed 01/07/2023 at 9:30 p.m.) for some of the use cases outlined above and provided the prompts

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1111/epi.17570

Cites

24 of 24 references verified0 citation errors · 0 not checkable

Statistics

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Information incorporated into ChatGPT may have delays and incompleteness due to the nature of its training data.

    “Second, due to the nature of its training data [78], the information incorporated into ChatGPT may have delays and incompleteness [8,79].”

    From Ethical Considerations of Using ChatGPT in Health Care (2023) · cites this paper as [78]

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Cites

Flagged first, then by citations

Research works printed but not linked: 4

  1. Zöllner JP, Wolking S, Weber Y, Rosenow F. Decision support systems, assistance systems and telemedicine in epileptology. Nervenarzt. 2021;92(2):95-106.
  2. Zhang Y, Sun S, Galley M, Chen Y-C, Brockett C, Gao X, et al. DialoGPT: large-scale generative pre-training for conversa- tional response generation [internet]. arXiv. 2020; [cited 2023]. Available from: http://arxiv.org/abs/1911.00536 · arXiv
  3. Adiwardana D, Luong M-T, So DR, Hall J, Fiedel N, Thoppilan R, et al. Towards a human-like open-domain Chatbot [inter- net]. arXiv. 2020; [cited 2023]. Available from: http://arxiv.org/ abs/2001.09977 · arXiv
  4. Venkit PN, Srinath M, Wilson S. A study of implicit bias in Pretrained language models against people with disabilities. In:. Proceedings of the 29th international conference on com- putational linguistics [internet]. Gyeongju, Republic of Korea: International Committee on Computational Linguistics; 2022. p. 1324-32 [cited 2023]; Available from: https://aclan tholo gy.org/2022.colin g-1.113

Cited by

Sorted by citations

Related

Authors
Boßelmann, C. M., Leu, C., & Lal, D.
Affiliations
  • Cleveland Clinic (US)
  • Broad Institute (US)
Corresponding author
Lal, Dennis
ORCID
Venue
Epilepsia
Publisher
Wiley
Published
2023-03-13
Volume
64
Issue
5
Pages
1195-1199
ISSN
  • 0013-9580 (ISSN-L)
  • 1528-1167
DOI
10.1111/epi.17570
OpenAlex
W4323066985
Cited by
52
References
24
Citations by year
2026: 6 · 2025: 12 · 2024: 16 · 2023: 18
FWCI
2.07
Type
journal article
Language
en
Open access
Licence
Locations
  • Epilepsia (publishedVersion, cc-by-nc)
  • PubMed (publishedVersion)
Topics
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Health Sciences › Medicine › Psychiatry and Mental health › Epilepsy research and treatment
  • Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning in Healthcare
Keywords
epilepsy, large language models, ChatGPT, clinical decision support systems, chatbots, epilepsy care, natural language processing, patient-physician communication, epilepsy diagnosis, epilepsy treatment, seizure prediction, EEG signal analysis, MRI lesion detection
MeSH terms
Artificial Intelligence, Epilepsy, Humans

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review