A public generalizable AI tool for automated segmentation of coronal brain tissue slabs for 3D neuropathology

Jonathan Williams Ramirez , Dina Zemlyanker, Lucas Deden-Binder, Rogeny Herisse, Erendira Garcia Pallares, Karthik Gopinath, Harshvardhan Gazula, Christopher Mount, Liana N. Kozanno, Michael S. Marshall, Theresa R. Connors, Matthew P. Frosch, Mark Montine, Derek H. Oakley, Christine L. Mac Donald, C. Dirk Keene, Bradley T. Hyman, Bruce Fischl, Juan E. Iglesias

Abstract
Advances in image registration and machine learning have recently enabled volumetric analysis of postmortem brain tissue from conventional photographs of coronal slabs, which are routinely collected in brain banks and neuropathology laboratories around the world. One caveat of this methodology is the requirement of segmentation of the tissue from the background and out-of-slice tissue in photographs, which currently requires laborious manual intervention. Manual delineation is a bottleneck in this process and poses challenges in scalability, and resources, restricting adoption of these methods.

Introduction
The fundamental mechanisms underlying many neurodegenerative diseases remain poorly understood, largely due to their molecular complexity, clinical comorbidities, overlapping phenotypes, and pathological heterogeneity [1,2]. As the discipline dedicated to studying and diagnosing diseases of the nervous system through the examination of brain tissue, Neuropathology plays a central role in addressing these challenges.

Materials and method
Three photographic datasets were obtained from two distinct sites, each using different imaging setups and tissue types (e.g., single hemisphere vs. whole brain, frozen vs. fixed). All specimens were sectioned in the coronal plane in an anterior-to-posterior orientation and subsequently photographed using an overhead camera. Each image contains a variable number of coronal slabs, imaged from the posterior side, and includes a spatial reference object (e.g., rulers, grids, or fiducial markers) to enable pixel size scaling and, when possible, perspective correction.

Methods
Calibrating pixel size in medical image segmentation is essential to ensure that spatial measurements reflect true anatomical dimensions. Eliminating pixel size variability enables better generalization across photographs from different sites, resolutions, or zoom levels

Experiments and results
To contextualize the performance of the automated method, we first conducted an experiment involving only human labelers, with two main objectives: to assess labeling time efficiency (i.e., the time required for manual annotation), and to evaluate variability both within and between annotators (i.e., intra- and inter-rater variability).

Discussion
This work introduces a fully automated segmentation tool for brain dissection photographs, offering a practical and scalable alternative to resource-intensive annotation workflows. By leveraging the nnU-Net framework, our method achieves a level of accuracy that is comparable to that of expert human annotators. It yields strong performance on in-distribution data, and generalizes effectively to out-of-distribution images from a previously unseen site.

Conclusion
The authors would like to thank the research participants and their families without whom this work would be impossible.

Citation: Williams Ramirez J, Zemlyanker D, Deden-Binder L, Herisse R, Garcia Pallares E, Gopinath K, et al. (2026) A public generalizable AI tool for automated segmentation of coronal brain tissue slabs for 3D neuropathology. PLoS One 21(8): e0355740. https://doi.org/10.1371/journal.pone.0355740

Editor: Federico Giove, Museo Storico della Fisica e Centro Studi e Ricerche Enrico Fermi, ITALY

Received: January 13, 2026; Accepted: July 24, 2026; Published: August 17, 2026

Copyright: © 2026 Williams Ramirez et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: All training images and labels for the segmentation model are publicly available at https://zenodo.org/records/20647553 with DOI: https://doi.org/10.48550/arXiv.2508.09805.

Funding: This research was primarily funded by the National Institute of Aging, https://www.nia.nih.gov/, (1R01AG070988 (JEI)). Additional support was provided by the National Institute of Mental Health, https://www.nimh.nih.gov/, (1RF1MH123195 (JEI, BF), 1UM1MH130981), the National Institute of Aging, https://www.nia.nih.gov/, (1RF1AG080371 (JEI)), the National Institute of Biomedical Imaging and Bioengineering, https://www.nibib.nih.gov/, (1R01EB031114), and the National Institute of Neurological Disorders and Stroke, https://www.ninds.nih.gov/, (1R21NS138995). The UW BioRepository and Integrated Neuropathology (BRaIN) laboratory is supported by the National Institutes of Health (NIH), by the National Institute of Mental Health, https://www.nimh.nih.gov/, through the the BRAIN Initiative Cell Atlas Network (UM1MH134812 (CDK)); also by the National Institute of Aging, https://www.nia.nih.gov/, through the UW Alzheimer’s Disease Research Center (P30 AG066509 (CDK)), the Kaiser Permanente Washington Adult Changes in Thought (ACT) study (U19 AG066567 (CDK)), the Seattle Alzheimer’s Disease Brain Cell Atlas (U19AG060909 (CDK)), also multiple collaborative projects supporting imaging, digital pathology, neuropathology, and related research tools (U24AG072458 (CDK); through the National Institute of Neurological Disorders and Stroke, https://www.ninds.nih.gov/, U24NS133949; U24NS133945; U24NS135651; U01NS137500; and U01NS137484 (CDK)), the US Department of Defense, https://www.war.gov/, (DoD W81XWH-21-S- TBIPH2), the Allen Institute for Brain Science, and the Nancy and Buster Alvord Endowment (to CDK). Support for this research was also provided in part by the BRAIN Initiative Cell Atlas Network (BICAN) grants, through the National Institute of Mental Health, https://www.nimh.nih.gov/, U01MH117023 (BF), UM1MH134812 (CDK) and UM1MH130981, the Brain Initiative Brain Connects consortium, through the National Institute of Neurological Disorders and Stroke, https://www.ninds.nih.gov/, (U01NS132181 (BF), 1UM1NS132358-01), the National Institute for Biomedical Imaging and Bioengineering, https://www.nibib.nih.gov/, (1R01EB023281 (BF), R21EB018907 (BF), R01EB019956 (BF), P41EB030006 (BF)), the National Institute on Aging, https://www.nia.nih.gov/, (R21AG082082 (BF), 1R01AG064027 (BF), R01AG016495, 1R01AG070988 (JEI)), the National Institute of Mental Health, https://www.nimh.nih.gov/, (UM1MH130981, R01 MH123195 (BF, JEI), R01 MH121885, 1RF1MH123195 (BF, JEI)), the National Institute for Neurological Disorders and Stroke, https://www.ninds.nih.gov/, (1U24NS135561-01 (BF), R01NS070963 (BF), 2R01NS083534 (BF), R01NS105820 (BF), R25NS125599 (BF)), and was made possible by the resources provided by Shared Instrumentation Grants, through National Center for Research Resources, http://www.ncrr.nih.gov/, 1S10RR023401 (BF), 1S10RR019307 (BF), and 1S10RR023043. Additional support was provided by the NIH Blueprint for Neuroscience Research through National Institute of Mental Health, https://www.nimh.nih.gov/, (5U01-MH093765), part of the multi institutional Human Connectome Project. Much of the computation resources required for this research was performed on computational hardware generously provided by the Massachusetts Life Sciences Center (https://www.masslifesciences.com). The authors would like to thank the research participants and their families without whom this work would be impossible. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. There was no additional external funding received for this study. The commercial affiliations declared in the competing interests statement provided support in the form of salaries for authors BH and BF, but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: Dr Hyman owns stock in Novartis; he serves on the SAB of Dewpoint and has an option for stock. He serves on a scientific advisory board or is a consultant for AbbVie, Alexion, Ambagon, Aprinoia Therapeutics, Arbor Bio, Arvinas, Avrobio, AstraZenica, Biogen, Bioinsights, BMS, Cell Signaling, Cure Alz Fund, CurieBio, Dewpoint, Etiome, Latus, Merck, Novartis, Paragon, Pfizer, Sanofi, Sofinnova, SV Health, Takeda, TD Cowen, Vigil, Violet, Voyager, WaveBreak. His laboratory is supported by research grants from the National Institutes of Health, Cure Alzheimer’s Fund, Tau Consortium, and the JPB Foundation – and sponsored research agreement from Abbvie and Sanofi. He has a collaborative project with Biogen and Neurimmune. Dr Fischl is an advisor to DeepHealth, a company whose medical pursuits focus on medical imaging and measurement technologies. BF’s interests were reviewed and are managed by Massachusetts General Hospital and Mass. General Brigham in accordance with their conflict of interest policies. These affiliations did not play a role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript and only provided financial support in the form of authors’ salaries (BH, BF). These affiliations do not alter our adherence to PLOS ONE policies on sharing data and materials.