Editorial. AtlasGPT: dawn of a new era in neurosurgery for intelligent care augmentation, operative planning, and performance

Benjamin S. Hopkins Department of Neurological Surgery, Keck Medicine of University of Southern California, Los Angeles, California;

Search for other papers by Benjamin S. Hopkins in
Current site
Google Scholar
PubMed
Close
 MD, MBA
,
Bob Carter Harvard University and Massachusetts General Hospital, Boston, Massachusetts;

Search for other papers by Bob Carter in
Current site
Google Scholar
PubMed
Close
 MD, PhD
,
Jesse Lord The Neurosurgical Atlasand Atlas Meditech, Carmel, Indiana;

Search for other papers by Jesse Lord in
Current site
Google Scholar
PubMed
Close
,
James T. Rutka Editor-in-Chief, Journal of Neurosurgery Publishing Group; and

Search for other papers by James T. Rutka in
Current site
Google Scholar
PubMed
Close
 MD, PhD
, and
Aaron A. Cohen-Gadol The Neurosurgical Atlasand Atlas Meditech, Carmel, Indiana;
Department of Neurological Surgery, Indiana University, Indianapolis, Indiana

Search for other papers by Aaron A. Cohen-Gadol in
Current site
Google Scholar
PubMed
Close
 MD, MSc, MBA
Free access

ABBREVIATIONS

AI = artificial intelligence; GPT = generative pretraining transformer; LLM = large language model; RAG = retrieval-augmented generation.

The primary objective of this editorial is to introduce what we believe to be the next wave of technology poised to revolutionize neurosurgical care and knowledge (all subspecialties of neurosurgery) as we currently understand it—the development of extended specialty-focused large language models (LLMs) in medicine. We describe AtlasGPT, an LLM based on reliable peer-reviewed neurosurgery-specific data sources.

Over the past 5 years, artificial intelligence (AI), accelerated dramatically by the advent of ChatGPT and GPT-4, has emerged as a mainstream force at the forefront of innovation, with applications relevant across industries.112 Although the prowess of LLMs such as ChatGPT and GPT-4 is undeniably impressive, these algorithms are optimized to plausibly complete language patterns, not serve as databases of trustworthy information.

Scientists and specialists alike are actively pushing the boundaries of these technologies to solve complex problems that challenge the world’s most foremost experts in their respective fields,1,9,10 and the field of neurosurgery stands as no exception to this transformative trend. Whether it is striving to successfully pass neurosurgical boards or streamlining modern workflows, AI has already begun to revolutionize modern-day medicine.1,9,10

As evident from the numerous recent publications focused on LLMs within the realms of medicine and neurosurgery, generative pretraining transformer (GPT) models have showcased remarkable capabilities.1,5,6,912 Nevertheless, considering the intricacy and technical nature of neurosurgical knowledge and decision-making, the credibility of sources used by GPT models and the imperative for them to return precise answers devoid of any error take on paramount significance.1,5,6,912

In light of this need, we have extended the application of ChatGPT and other GPT models by introducing an innovative modification specifically tailored to cater to the needs of neurosurgeons—AtlasGPT. This newly proposed cutting-edge adaptation, created with a specialized training regimen, provides neurosurgeons with incredibly accurate answers to their prompts and questions.

AtlasGPT uses cutting-edge retrieval-augmented generation (RAG) techniques to mitigate risks and ensure the highest standards of output.13,14 These models, notable for their dual-process architecture, combine the generative strength of LLMs with an external data retrieval mechanism. This approach is rapidly gaining popularity as a result of its ability to integrate up-to-date and relevant information from external sources into its responses. By doing so, RAG models enhance the accuracy and relevance of generated content, which addresses a key limitation of traditional LLMs. RAG architecture introduces a pivotal filtering and ranking step in the initial query—the retrieval step (Fig. 1).13,14

FIG. 1.
FIG. 1.

Schematic of selective RAG of AtlasGPT for improving the accuracy and validity of LLM responses.

Upon receiving any prompt, the model initiates a process to source and retrieve pertinent data and sources, predetermined during the training phase of model development. After retrieving relevant articles, passages, or partitions from a large, predefined language corpus, the model integrates the information with the user’s initial prompt.13,14 This collaborative approach provides the GPT model with sufficient context to deliver a truthful and accurate answer. In addition, the complexity and tone of the response can be adjusted according to the level of medical knowledge of the user or target audience; options include patient, medical student, resident, and practicing neurosurgeon styles.

The implementation architecture can be scaled and integrated easily. Two bespoke integrations have already been completed using the representational state transfer architectural style application programming interface (REST API) of the platform. Automated ingestion pipelines are also in place to ensure that AtlasGPT remains up to date and relevant over time.

AtlasGPT places significant emphasis on the retrieval step and extensively leverages pretrained neurosurgical literature, encompassing sources such as the Journal of Neurosurgery, The Neurosurgical Atlas, and various peer-reviewed neurosurgical publications. As of December 2023, AtlasGPT had undergone training using 250,000 pages of meticulously selected peer-reviewed neurosurgical articles. This careful curation is aimed at streamlining and facilitating the production of reliable outputs that align with the exacting standards set by the world’s leading neurosurgical journals. The potential uses for such a highly specialized and expansive language model are virtually limitless; select applications of AtlasGPT are listed in Table 1, and we present an example query and response from AtlasGPT in Fig. 2 to show its potential applications. AtlasGPT will continue to evolve with time to include image recognition and other applications of computer vision. In addition, its knowledge base will continue to increase, further expanding its capabilities for transforming neurosurgery.

FIG. 2.
FIG. 2.

Response from AtlasGPT to a question regarding the management of a Chance fracture. This response is in "surgeon style," but it could easily be adjusted to the knowledge level of a resident, medical student, or patient. MEP = motor evoked potential; SSEP = somatosensory evoked potential.

As Sir Francis Bacon noted, "ipsa scientia potestas est" ("knowledge itself is power"). However, with the advent of AtlasGPT we could rephrase this quote as "scientia accessibilis est potentia accessibilis" ("accessible knowledge is accessible power"). AI LLMs such as AtlasGPT are opening up a future in which vast amounts of knowledge can be shared by creative minds for better neurosurgical care.

We hope that you share our enthusiasm for AtlasGPT and its potential to facilitate the acquisition of new knowledge to enable you to provide the best possible care for your patients.

References

  • 1

    Hopkins BS, Nguyen VN, Dallas J, et al. ChatGPT versus the neurosurgical written boards: a comparative analysis of artificial intelligence/machine learning performance on neurosurgical board–style questions. J Neurosurg. 2023;139(3):904911.

    • PubMed
    • Search Google Scholar
    • Export Citation
  • 2

    Hopkins BS, Murthy NK, Texakalidis P, et al. Mass deployment of deep neural network: real-time proof of concept with screening of intracranial hemorrhage using an open data set. Neurosurgery. 2022;90(4):383389.

    • PubMed
    • Search Google Scholar
    • Export Citation
  • 3

    Hopkins BS, Yamaguchi JT, Garcia R, et al. Using machine learning to predict 30-day readmissions after posterior lumbar fusion: an NSQIP study involving 23,264 patients. J Neurosurg Spine. 2019;32(3):399406.

    • PubMed
    • Search Google Scholar
    • Export Citation
  • 4

    Hopkins BS, Weber KA II, Kesavabhotla K, Paliwal M, Cantrell DR, Smith ZA. Machine learning for the prediction of cervical spondylotic myelopathy: a post hoc pilot study of 28 participants. World Neurosurg. 2019;127:e436e442.

    • PubMed
    • Search Google Scholar
    • Export Citation
  • 5

    Ali R, Tang OY, Connolly ID, et al. Performance of ChatGPT, GPT-4, and Google Bard on a neurosurgery oral boards preparation question bank. Neurosurgery. Published online June 12, 2023. doi:10.1227/neu.0000000000002551

    • PubMed
    • Search Google Scholar
    • Export Citation
  • 6

    Ali R, Tang OY, Connolly ID, et al. Performance of ChatGPT and GPT-4 on neurosurgery written board examinations. Neurosurgery. 2023;93(6):13531365.

  • 7

    Danilov G, Kotik K, Shevchenko E, et al. Predicting the length of stay in neurosurgery with RuGPT-3 language model. Stud Health Technol Inform. 2022;295:555558.

  • 8

    Danilov G, Kotik K, Shevchenko E, et al. Length of stay prediction in neurosurgery with Russian GPT-3 language model compared to human expectations. Stud Health Technol Inform. 2022;289:156159.

    • PubMed
    • Search Google Scholar
    • Export Citation
  • 9

    Guerra GA, Hofmann H, Sobhani S, et al. GPT-4 artificial intelligence model outperforms ChatGPT, medical students, and neurosurgery residents on neurosurgery written board-like questions. World Neurosurg. 2023;179:e160e165.

    • PubMed
    • Search Google Scholar
    • Export Citation
  • 10

    Kuang YR, Zou MX, Niu HQ, Zheng BY, Zhang TL, Zheng BW. ChatGPT encounters multiple opportunities and challenges in neurosurgery. Int J Surg. 2023;109(10):28862891.

    • PubMed
    • Search Google Scholar
    • Export Citation
  • 11

    Li W, Fu M, Liu S, Yu H. Revolutionizing neurosurgery with GPT-4: a leap forward or ethical conundrum? Ann Biomed Eng. 2023;51(10):21052112.

  • 12

    Resnick DK. Commentary: Performance of ChatGPT, GPT-4, and Google Bard on a neurosurgery oral boards preparation question bank. Neurosurgery. Published online July 19, 2023. doi:10.1227/neu.0000000000002618

    • PubMed
    • Search Google Scholar
    • Export Citation
  • 13

    Wang C, Ong J, Wang C, Ong H, Cheng R, Ong D. Potential for GPT technology to optimize future clinical decision-making using retrieval-augmented generation. Ann Biomed Eng. Published online August 2, 2023. doi:10.1007/s10439-023-03327-6

    • PubMed
    • Search Google Scholar
    • Export Citation
  • 14

    Lewis P, Perez E, Piktus A, et al. Retrieval-augmented generation for knowledge-intensive NLP tasks. arXiv. Preprint posted online May 22, 2020. doi:10.48550/arXiv.2005.11401

    • PubMed
    • Search Google Scholar
    • Export Citation
  • Collapse
  • Expand
Illustration from Srinivasan et al. (pp 1335–1343). © Sandeep Kandregula, published with permission.
  • FIG. 1.

    Schematic of selective RAG of AtlasGPT for improving the accuracy and validity of LLM responses.

  • FIG. 2.

    Response from AtlasGPT to a question regarding the management of a Chance fracture. This response is in "surgeon style," but it could easily be adjusted to the knowledge level of a resident, medical student, or patient. MEP = motor evoked potential; SSEP = somatosensory evoked potential.

  • 1

    Hopkins BS, Nguyen VN, Dallas J, et al. ChatGPT versus the neurosurgical written boards: a comparative analysis of artificial intelligence/machine learning performance on neurosurgical board–style questions. J Neurosurg. 2023;139(3):904911.

    • PubMed
    • Search Google Scholar
    • Export Citation
  • 2

    Hopkins BS, Murthy NK, Texakalidis P, et al. Mass deployment of deep neural network: real-time proof of concept with screening of intracranial hemorrhage using an open data set. Neurosurgery. 2022;90(4):383389.

    • PubMed
    • Search Google Scholar
    • Export Citation
  • 3

    Hopkins BS, Yamaguchi JT, Garcia R, et al. Using machine learning to predict 30-day readmissions after posterior lumbar fusion: an NSQIP study involving 23,264 patients. J Neurosurg Spine. 2019;32(3):399406.

    • PubMed
    • Search Google Scholar
    • Export Citation
  • 4

    Hopkins BS, Weber KA II, Kesavabhotla K, Paliwal M, Cantrell DR, Smith ZA. Machine learning for the prediction of cervical spondylotic myelopathy: a post hoc pilot study of 28 participants. World Neurosurg. 2019;127:e436e442.

    • PubMed
    • Search Google Scholar
    • Export Citation
  • 5

    Ali R, Tang OY, Connolly ID, et al. Performance of ChatGPT, GPT-4, and Google Bard on a neurosurgery oral boards preparation question bank. Neurosurgery. Published online June 12, 2023. doi:10.1227/neu.0000000000002551

    • PubMed
    • Search Google Scholar
    • Export Citation
  • 6

    Ali R, Tang OY, Connolly ID, et al. Performance of ChatGPT and GPT-4 on neurosurgery written board examinations. Neurosurgery. 2023;93(6):13531365.

  • 7

    Danilov G, Kotik K, Shevchenko E, et al. Predicting the length of stay in neurosurgery with RuGPT-3 language model. Stud Health Technol Inform. 2022;295:555558.

  • 8

    Danilov G, Kotik K, Shevchenko E, et al. Length of stay prediction in neurosurgery with Russian GPT-3 language model compared to human expectations. Stud Health Technol Inform. 2022;289:156159.

    • PubMed
    • Search Google Scholar
    • Export Citation
  • 9

    Guerra GA, Hofmann H, Sobhani S, et al. GPT-4 artificial intelligence model outperforms ChatGPT, medical students, and neurosurgery residents on neurosurgery written board-like questions. World Neurosurg. 2023;179:e160e165.

    • PubMed
    • Search Google Scholar
    • Export Citation
  • 10

    Kuang YR, Zou MX, Niu HQ, Zheng BY, Zhang TL, Zheng BW. ChatGPT encounters multiple opportunities and challenges in neurosurgery. Int J Surg. 2023;109(10):28862891.

    • PubMed
    • Search Google Scholar
    • Export Citation
  • 11

    Li W, Fu M, Liu S, Yu H. Revolutionizing neurosurgery with GPT-4: a leap forward or ethical conundrum? Ann Biomed Eng. 2023;51(10):21052112.

  • 12

    Resnick DK. Commentary: Performance of ChatGPT, GPT-4, and Google Bard on a neurosurgery oral boards preparation question bank. Neurosurgery. Published online July 19, 2023. doi:10.1227/neu.0000000000002618

    • PubMed
    • Search Google Scholar
    • Export Citation
  • 13

    Wang C, Ong J, Wang C, Ong H, Cheng R, Ong D. Potential for GPT technology to optimize future clinical decision-making using retrieval-augmented generation. Ann Biomed Eng. Published online August 2, 2023. doi:10.1007/s10439-023-03327-6

    • PubMed
    • Search Google Scholar
    • Export Citation
  • 14

    Lewis P, Perez E, Piktus A, et al. Retrieval-augmented generation for knowledge-intensive NLP tasks. arXiv. Preprint posted online May 22, 2020. doi:10.48550/arXiv.2005.11401

    • PubMed
    • Search Google Scholar
    • Export Citation

Metrics

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 2487 2487 450
PDF Downloads 2332 2332 369
EPUB Downloads 0 0 0