Abstract 7
Category: Basic Science
At the end of the session, participants will be able to:
- Describe the current representation of large language model (LLM) applications across clinical neuroscience subspecialties, with emphasis on neuropathology.
- Compare the scope, study design, and validation of LLM studies in neuropathology versus other pathology subspecialties and radiology.
- Discuss future directions for developing and validating LLMs tailored to the morphologic and molecular complexity of CNS lesions.
COI Disclosure:
None to disclose.
Presenter
Vladimir is entering grade 12 and has an interest in Computer Science and Astronomy. This abstract relates to a branch of computing that is of current interest, artificial intelligence in neuropathology. He has reviewed the existing literature and this will provide a springboard for future work on the reliability of AI in diagnosis.
Vladimir (Vova) R. Auer 1, Pramath Kakodkar1, Daniel Markewich1, Nooshin Shekari2, Bryan Johnston1, Viktor Zherebitskiy1, Roland N. Auer1, and Jay Kalra1
1Department Pathology and Lab Medicine, University of Saskatchewan, Saskatoon, Canada
2Department of Anatomy, Physiology, Pharmacology, University of Saskatchewan, Saskatoon, Canada
Target Audience:
Pathologists, Residents, Medical Students
CanMEDS:
Medical Expert (the integrating role), Communicator, Collaborator, Scholar
Mapping the digital divide: underrepresentation of neuropathology in large language model-driven diagnostic literature
Abstract
An LLM (Large Language Model) is an artificial intelligence (AI) trained on massive amounts of text to predict the next word, letting it generate and work with human‑like language. The objective of this study is to assess the representation and diagnostic utility of LLMs in Clinical Neuroscience with a focus on neuropathology compared to other subspecialties. We performed a PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) compliant review of 539 articles (ChatGPT (n=508), Google Gemini (n=14), Perplexity (n=10), DeepSeek (n=7)) spanning the years 2022-2025. Studies were categorized by subspecialty, input data type (admission note, radiological imaging, radiology report, pathology images, pathology report), and assessed LLM outcomes (diagnosis, grading, prognosis). Only 11.9% (n=64) of included studies focused on clinical neuroscience, and just 3 were dedicated to neuropathology. These studies targeted gliomas and neurodegenerative disorders. The other specialty that had the highest LLM utilization was Radiology (11.3%, n=61). Neuropathology-dedicated studies lagged in sample size, external validation, and model robustness, especially compared with other pathology (n= 37) subspecialties such as head and neck (n=9), general surgical pathology (n=7), gastrointestinal (n=5) and gynecological pathology (n=4). Neuropathology remains significantly underrepresented in the era of diagnostic AI. Given the distinctive morphologic and molecular complexity of CNS (central nervous system) lesions, there is an urgent need for dedicated model development and rigorous validation tailored to neuropathology.