DaX: Learning General Pathology Representations Across Scales

2026-06-05 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision & Pattern Recognition, Health & Medical Research · Depth: Expert, quick

Summary

DaX is a pathology vision foundation model designed to create visual representations robust to variations in magnification, staining, scanner type, slide preparation, and input resolution, enabling transfer across diverse clinical endpoints. It adapts DINOv3-style self-supervised learning to whole-slide histopathology, initialized from natural-image DINOv3 weights. DaX incorporates continuous magnification training, cross-scale tissue views, orientation-agnostic augmentation, multi-input-size training, and Gram-anchored dense consistency. These features connect local cellular morphology with global tissue architecture. The model was evaluated on a WSI-level benchmark of 161 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 strong task-level ranking scores, showing gains in diagnostic pathology, biomarker profiling, tissue context, and risk/prognosis prediction.

Key takeaway

For Machine Learning Engineers developing computational pathology solutions, integrate DaX's design principles to enhance model robustness and transferability. Specifically, consider continuous magnification training and cross-scale tissue views. Evaluate your models using a standardized WSI-level benchmark with patient-level cross-validation. This ensures reproducible comparisons and reliable performance across diverse clinical tasks, improving the generalizability of your visual encoders.

Key insights

DaX is a pathology vision foundation model using DINOv3-style self-supervised learning for robust, transferable representations across diverse histopathology tasks.

Principles

• Pathology representations must transfer across diverse clinical endpoints.
• Robustness to magnification, staining, and scanner variations is crucial.
• Connecting local cellular morphology with global tissue architecture improves models.

Method

DaX adapts DINOv3 self-supervised learning with continuous magnification training, cross-scale views, robust augmentation, multi-input-size training, and Gram-anchored dense consistency for whole-slide histopathology.

In practice

• Use DINOv3-style self-supervised learning for pathology.
• Incorporate continuous magnification training for scale robustness.
• Evaluate models on WSI-level benchmarks with patient-level cross-validation.

Topics

• Computational Pathology
• Foundation Models
• Self-supervised Learning
• Histopathology
• Image Representation Learning
• WSI-level Benchmark

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

Related on AIssential

• DaX: Learning General Pathology Representations Across Scales — DaX is a pathology foundation model using DINOv3-style self-supervised learning to create robust, transferable visual representations across diverse scales and clinical tasks.
• v316: Proceedings of COMPAYL 2025 — The workshop highlights advanced AI methods for diverse computational pathology challenges.
• I Gave a Model Eyes and a Genome. Here’s What It Learned About Brain Cancer. — Cross-attention fusion of pathology and genomics significantly improves glioblastoma survival prediction over unimodal approaches.
• 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.
• 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.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.