DaX: Learning General Pathology Representations Across Scales

2026-06-08 · Source: cs.CV updates on arXiv.org · Field: Science & Research — Artificial Intelligence & Machine Learning, Health & Medical Research, Data Science & Analytics · Depth: Expert, quick

Summary

DaX is a pathology vision foundation model designed to learn general representations that transfer across diverse clinical endpoints and remain robust to variations in magnification, staining, scanner type, and input resolution. It adapts DINOv3-style self-supervised learning for whole-slide histopathology, initialized from natural-image DINOv3 weights. Key designs include continuous magnification training, cross-scale tissue views, orientation-agnostic augmentation, multi-input-size training, and Gram-anchored dense consistency, aiming to link local cellular morphology with global tissue architecture. The model was evaluated on 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. DaX achieved the highest mean performance and consistently strong task-level ranking scores, demonstrating gains across diagnostic pathology, biomarker profiling, tissue context, and prognosis. This establishes DaX as a transferable visual encoder and the benchmark as a standardized evaluation framework.

Key takeaway

For machine learning engineers developing computational pathology solutions, DaX offers a validated foundation model to improve transferability and robustness. You should consider initializing your models with DaX's weights or adopting its multi-scale training and robust augmentation strategies. This approach can significantly enhance performance across diverse clinical endpoints, from diagnostic pathology to biomarker profiling, reducing sensitivity to variations in slide acquisition and preparation.

Key insights

DaX is a pathology foundation model using DINOv3-style self-supervised learning to create robust, transferable visual representations across diverse scales and clinical tasks.

Principles

• Self-supervised learning transfers from natural images to pathology.
• Multi-scale training links local morphology with global architecture.
• Robust augmentation enhances transferability across acquisition types.

Method

DaX adapts DINOv3-style self-supervised learning for histopathology, integrating continuous magnification, cross-scale views, robust augmentation, multi-input-size training, and Gram-anchored dense consistency to stabilize token-level representations across scales.

In practice

• Apply to diagnostic pathology tasks.
• Use for biomarker and molecular profiling.
• Predict patient risk, response, and prognosis.

Topics

• Computational Pathology
• Foundation Models
• Self-supervised Learning
• Histopathology
• DINOv3
• WSI Benchmark

Best for: Computer Vision Engineer, AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• DaX: Learning General Pathology Representations Across Scales — DaX is a pathology vision foundation model using DINOv3-style self-supervised learning for robust, transferable representations across diverse histopathology tasks.
• Clinically-Informed Modeling for Pediatric Brain Tumor Classification from Whole-Slide Histopathology Images — Contrastive learning, especially with expert-guided hard negatives, improves fine-grained pediatric brain tumor classification from WSIs.
• Learning Emergent Modular Representations in Multi-modality Medical Vision Foundation Models — DEX uses modular networks with experts and a director to manage modality-specific and shared representations in multi-modality medical FMs.
• Systematically decoding pathological morphologies and molecular profiles with unified multimodal embedding — Multi-Embed unifies morphological and molecular data for superior disease understanding and prediction.
• Generative Deep Learning for Computational Destaining and Restaining of Unregistered Digital Pathology Images — Preprocessing-based domain adaptation enables cGAN models to generalize for computational pathology across institutions without retraining.
• A Chain-of-thought Reasoning Breast Ultrasound Dataset Covering All Histopathology Categories — BUS-CoT is a large, richly annotated breast ultrasound dataset enabling AI development for chain-of-thought reasoning across all histopathology types.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.