GigaPath-Flash and GigaTIME-Flash: Efficient Pathology Foundation Models for Whole-Slide and Tumor Microenvironment Analysis
Naoto Usuyama, Jeya Maria Jose Valanarasu, Sicong Yao, Hanwen Xu, Jaspreet Bagga, Guanghui Qin, Robert E. Kramer, Cliff Wong, Soohee Lee, Hao Qiu, Theodore Zhengde Zhao, Racheli Ben Shimol, Angela Crabtree, Kevin Matlock, Eduardo Alejandro Lozano Garcia, Naiteek Sangani, Alberto Santamaria-Pang, Jason Entenmann, Alexandra Q. Bartlett, Bill J. Wright, Bernard A. Fox, Brian Piening, Sheng Zhang, Sheng Wang, Tristan Naumann, Carlo Bifulco, Hoifung Poon
Read on arXiv →
Key claim
GigaPath-Flash retains high performance with 50x less compute.
In plain English
In computational pathology, existing models often struggle with high computational costs and limited accessibility, hindering their use in clinical settings. Current pretrained models typically operate at the image-tile level, which is inefficient for whole-slide analysis. GigaPath-Flash and GigaTIME-Flash address these issues by offering efficient, pretrained models that maintain high performance while significantly reducing compute requirements. Builders in the field can leverage these open-weight models to enhance cancer diagnosis and treatment selection.
Introduces efficient models for whole-slide pathology AI with significant performance retention.
Demonstrates improvements over existing models with clear performance metrics.
Deep reliability assessment
The methodology supports efficient whole-slide pathology AI with reduced computational costs, but the claim of democratizing pathology AI may be overclaimed without broader adoption evidence.
Reproducibility
yes, the models and weights are released under an Apache-2.0 license, allowing for reproducibility.
Key figure
Figure 1 provides an overview of the GigaPath/GigaTIME model family, highlighting the efficient tile and slide encoders used in GigaPath-Flash and GigaTIME-Flash.