DaX: Learning General Pathology Representations Across Scales

2026-06-05

AI Summary

Computational pathology urgently requires general-purpose visual representations that are robust to variations in staining protocols, scanner types, and image resolution, and that generalize across diverse clinical endpoints. This work proposes a multi-scale training strategy for whole-slide images based on the DINOv2 self-supervised framework, incorporating continuous magnification sampling, cross-scale tissue view fusion, and a Gram-anchored dense consistency mechanism to jointly model cellular morphology and tissue architecture. The authors establish a large-scale, standardized evaluation benchmark encompassing 161 clinical tasks and validate their approach on 34,394 whole-slide images. Their method significantly outperforms existing approaches, achieving state-of-the-art average performance across diagnostic tasks, biomarker prediction, tissue context understanding, and prognosis.

Abstract

Computational pathology requires visual representations that transfer across diverse clinical endpoints and remain robust to variation in magnification, staining, scanner type, slide preparation, and input resolution. We present DaX, a pathology vision foundation model that adapts DINOv3-style self-supervised learning to whole-slide histopathology. DaX is initialized from natural-image DINOv3 weights and incorporates continuous magnification training, cross-scale tissue views, orientation-agnostic and acquisition-robust augmentation, multi-input-size training, and Gram-anchored dense consistency. These designs aim to connect local cellular morphology with global tissue architecture while stabilizing dense token-level representations across input scales. We further construct a WSI-level benchmark comprising 161 clinically meaningful tasks from 44 public datasets, covering 28,182 patients and 34,394 slides across four clinical domains and nine task categories. All models are evaluated under a fixed patient-level cross-validation protocol with fold-level statistical ranking, enabling reproducible comparisons that are less sensitive to split-dependent variation. Across this benchmark, DaX achieves the highest mean performance across tasks and consistently strong task-level ranking scores, with gains spanning diagnostic pathology, biomarker and molecular profiling, tissue/specimen context, and risk, response, and prognosis. These results support DaX as a transferable visual encoder for computational pathology and provide a standardized evaluation framework for future pathology foundation models. Project page: https://alibaba-damo-academy.github.io/DaX/benchboard/.

Problem

• computational pathology
• visual representation
• magnification robustness
• stain variation
• whole-slide imaging

Innovation

• computational pathology
• vision foundation model
• self-supervised learning
• multi-scale representation
• whole-slide image