Summary
Histopathology is considered the gold standard for determining the presence and nature of disease, particularly cancer. However, the process of analyzing tissue samples and producing a final pathology report is time-consuming, labor-intensive, and non-standardized. Therefore, new technological solutions are being sought to reduce the workload of pathologists. In this work, we present HistoGPT, a vision language model that takes digitized slides as input and generates reports that match the quality of human-written reports, as confirmed by natural language processing metrics and domain expert evaluations. We show that HistoGPT generalizes to five international cohorts and can predict tumor subtypes and tumor thickness in a zero-shot fashion. Our work represents an important step toward integrating AI into the medical workflow. We publish both model code and weights so that the scientific community can apply and improve HistoGPT to advance the field of computational pathology.
Highlights
A large vision language model is trained to generate dermatopathology reports
It takes as input multiple whole slide images and outputs tissue descriptions
Generated reports match human-written reports as confirmed by pathologists
It predicts tumor subtypes and thickness zero-shot better than current methods
Competing Interest Statement
M.T. is employed by Roche Diagnostics GmbH but conducted his research independently of his work at Roche Diagnostics GmbH as a guest scientist at Helmholtz Munich (Helmholtz Zentrum Muenchen - Deutsches Forschungszentrum fuer Gesundheit und Umwelt GmbH).
Funding Statement
M.T., S.W.J., and V.K. are supported by the Helmholtz Association under the joint research school "Munich School for Data Science - MUDS". C.M. acknowledges funding from the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation program (grant agreement no. 866411) and support from the Hightech Agenda Bayern.
Author Declarations
I confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained.
Yes
The details of the IRB/oversight body that provided approval or exemption for the research described are given below:
All research procedures were conducted in accordance with the Declaration of Helsinki. Ethics approval was granted by the Ethics Commission of the Technical University Munich (reference number 2024-98-S-CB) and the Ethics Commission of Westfalen-Lippe (reference number 2024-157-b-S).
I confirm that all necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived, and that any patient/participant/sample identifiers included were not known to anyone (e.g., hospital staff, patients or participants themselves) outside the research group so cannot be used to identify individuals.
Yes
I understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance).
Yes
I have followed all appropriate research reporting guidelines, such as any relevant EQUATOR Network research reporting checklist(s) and other pertinent material, if applicable.
Yes
Footnotes
↵# Lead contact carsten.marr{at}helmholtz-munich.de
Data Availability
The public data sets used are available from the respective providers. The 100 patient cases from the Munich cohort used in the blinded study will be made available upon publication. The model code and weights are available at https://github.com/marrlab/HistoGPT.
https://portal.gdc.cancer.gov/projects/TCGA-SKCM
https://www.cancerimagingarchive.net/collection/cptac-cm/
