EXAONE Path 2.0: Pathology Foundation Model with End-to-End Supervision

2025-07-09

AI Summary

In digital pathology, representation learning for gigapixel whole-slide images (WSIs) faces two key challenges: (1) patch-level self-supervised learning (SSL) relies on generic natural-image augmentations that poorly capture domain-specific tissue patterns associated with biomarkers; and (2) SSL suffers from low data efficiency and high computational cost. To address these, we propose the first end-to-end slide-level supervised foundation model for pathology, directly leveraging WSI-level labels—such as genomic mutation status—to drive representation learning, thereby eliminating the conventional two-stage paradigm of “patch-level SSL followed by multiple-instance learning (MIL) aggregation.” Trained on only 37,000 WSIs, our model achieves state-of-the-art average performance across ten biomarker prediction tasks. It significantly improves domain-specific feature capture, data efficiency, and generalizability.

Technology Category

• Machine Learning: Representation Learning
• Computer Vision: Representation Learning for Vision
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Scalable techniques for the creation, curation, publication, maintenance, and consumption of large, Web-based, structured, reusable, knowledge graphs and ontologies

Abstract

In digital pathology, whole-slide images (WSIs) are often difficult to handle due to their gigapixel scale, so most approaches train patch encoders via self-supervised learning (SSL) and then aggregate the patch-level embeddings via multiple instance learning (MIL) or slide encoders for downstream tasks. However, patch-level SSL may overlook complex domain-specific features that are essential for biomarker prediction, such as mutation status and molecular characteristics, as SSL methods rely only on basic augmentations selected for natural image domains on small patch-level area. Moreover, SSL methods remain less data efficient than fully supervised approaches, requiring extensive computational resources and datasets to achieve competitive performance. To address these limitations, we present EXAONE Path 2.0, a pathology foundation model that learns patch-level representations under direct slide-level supervision. Using only 37k WSIs for training, EXAONE Path 2.0 achieves state-of-the-art average performance across 10 biomarker prediction tasks, demonstrating remarkable data efficiency.

Problem

• Patch-level SSL overlooks domain-specific biomarker features
• SSL methods are less data-efficient than supervised approaches
• Gigapixel WSIs are challenging for current pathology models

Innovation

• End-to-end slide-level supervised pathology foundation model
• Direct patch-level representation learning for biomarker prediction
• Achieves state-of-the-art performance with 37k WSIs