WILSON - Multi-magnification composites as a foundation for patient-level pathology analysis and diagnostic text generation
Abstract
Pathologists examine tissue sections across multiple magnifications and slides, integrating morphological information to establish a diagnosis. In contrast, most pathology foundation models encode thousands of image patches from a single representative ("best") slide and aggregate their features, limiting multiscale and patient-level contextual integration. Here we introduce WILSON, a vision-language foundation model that represents whole-slide images and multi-slide cases through multi-magnification composites. Trained on approximately 189,000 slides spanning 42 organs and 838 diagnostic entities, WILSON enables patient-level analysis, multimodal retrieval and diagnostic text generation. Without task-specific training, WILSON outperformed a dedicated case-level model (macro-F1, 0.52 versus 0.38) and matched substantially larger slide-level models while requiring 400-1,350 times fewer computational operations. End-to-end fine-tuning improved breast cancer subtyping and stromal lymphocyte grading. WILSON also surpassed PRISM in image-text retrieval (recall@1, 75.6% versus 58.1%) and generated diagnostic descriptions that received higher ratings from five board-certified pathologists than competing models. These findings demonstrate that compact, multiscale representations can support computationally efficient, clinically grounded pathology foundation models.
Related
- Towards Trustworthy Biological Alignment in TabPFN-Probed Pathology Foundation Models
- Diagnostic challenge in veterinary pathology: Late term abortion in a cow
- Decomposing scanner batch effects and their impact on pathology foundation models
- OmniTCR: a foundation model unifying T cell receptor recognition prediction and conditional sequence generation
- DAPE-PD: Diffusion adaptation of genomic foundation model enables tailored-generation of cis-regulatory DNA
BibTeX
@misc{alfasly2026wilson,
title = {{WILSON - Multi-magnification composites as a foundation for patient-level pathology analysis and diagnostic text generation}},
author = {Saghir Alfasly and Wataru Uegami and Sobhan Hemati and Wenchao Han and Xiaojia Tang and Kevin Thompson and Daniel Stone and Ghazal Alabtah and Chris Hartley and Chady Meroueh and John C. Cheville and Thomas J. Flotte and Saba Yasir and Michael R. Lucas and Eric W. Klee and Cheryl L. Willman and Judy C. Boughey and Matthew P. Goetz and Krishna R. Kalari and H. R. Tizhoosh},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2609.25123},
eprint = {2609.25123},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.25123},
}
RIS
TY - UNPB
TI - WILSON - Multi-magnification composites as a foundation for patient-level pathology analysis and diagnostic text generation
AU - Saghir Alfasly
AU - Wataru Uegami
AU - Sobhan Hemati
AU - Wenchao Han
AU - Xiaojia Tang
AU - Kevin Thompson
AU - Daniel Stone
AU - Ghazal Alabtah
AU - Chris Hartley
AU - Chady Meroueh
AU - John C. Cheville
AU - Thomas J. Flotte
AU - Saba Yasir
AU - Michael R. Lucas
AU - Eric W. Klee
AU - Cheryl L. Willman
AU - Judy C. Boughey
AU - Matthew P. Goetz
AU - Krishna R. Kalari
AU - H. R. Tizhoosh
PY - 2026
DA - 2026/10/18
PB - arXiv
DO - 10.48550/arxiv.2609.25123
UR - https://arxiv.org/abs/2609.25123
AB - Pathologists examine tissue sections across multiple magnifications and slides, integrating morphological information to establish a diagnosis. In contrast, most pathology foundation models encode thousands of image patches from a single representative ("best") slide and aggregate their features, limiting multiscale and patient-level contextual integration. Here we introduce WILSON, a vision-language foundation model that represents whole-slide images and multi-slide cases through multi-magnification composites. Trained on approximately 189,000 slides spanning 42 organs and 838 diagnostic entities, WILSON enables patient-level analysis, multimodal retrieval and diagnostic text generation. Without task-specific training, WILSON outperformed a dedicated case-level model (macro-F1, 0.52 versus 0.38) and matched substantially larger slide-level models while requiring 400-1,350 times fewer computational operations. End-to-end fine-tuning improved breast cancer subtyping and stromal lymphocyte grading. WILSON also surpassed PRISM in image-text retrieval (recall@1, 75.6% versus 58.1%) and generated diagnostic descriptions that received higher ratings from five board-certified pathologists than competing models. These findings demonstrate that compact, multiscale representations can support computationally efficient, clinically grounded pathology foundation models.
ER -