Self-supervised Learning for Cancer Pathology Foundation Models (SLC-PFM)

Welcome to the SLC-PFM Competition - an unprecedented platform for advancing the development of the next generation of pathology foundation models. This competition provides exclusive access to the largest pathology dataset ever assembled for building pathology foundation models (PFMs), enabling researchers to focus on technical innovation in self-supervised learning for pre-training PFMs for clinically relevant computational pathology tasks.

Competition Highlights

• Comprehensive Pre-training Data: Access to MSK-SLCPFM dataset with ~300M images from 39 cancer types
• Robust Validation Framework: Multi-institutional evaluation across 23 clinically relevant pathology tasks
• Focus on Technical Innovation: Participants can concentrate on novel architectures and learning approaches without data curation barriers
• Global Collaboration: Evaluation conducted by leading cancer centers including top-ranked institutions in the USA and globally

🎯 Your Challenge: Build the Next Generation of Pathology Foundation Models

This competition is specifically designed for developing novel self-supervised learning algorithms and architectures to create powerful pathology foundation models.

What You’ll Develop:

• Self-supervised learning approaches that can learn meaningful representations from pathology images without requiring labeled data
• Novel neural network architectures optimized for the unique characteristics of histopathology data
• Foundation models that can serve as powerful feature extractors for diverse downstream clinical tasks

The Innovation Opportunity:

SLC-PFM focuses on pre-training methodologies - giving you the freedom to explore cutting-edge self-supervised techniques like:

• Contrastive Learning: SimCLR, MoCo, SwAV, VICReg
• Self-Distillation: DINO, DINOv2, BYOL
• Masked Image Modeling: MAE, iBOT, BEiT
• Multi-Scale Learning: Hierarchical representations across 20× and 40× magnifications
• Pathology-Specific Innovations: Novel augmentations, rotation invariance, cross-tile relationship modeling
• Transformer Architectures: Vision Transformers (ViTs) optimized for gigapixel pathology images

No labels. No supervision. Pure algorithmic innovation.

Design self-supervised learning frameworks that extract clinically meaningful patterns from 300M pathology images to create foundation models that will advance cancer diagnosis worldwide.

Competition Phases

Phase 1: Foundation Model Development (June - October 2025)

• Access to MSK-SLCPFM (Phase 1) dataset with ~300 million pathology image tiles
• Self-supervised learning approach for foundation model pre-training
• 51,578 whole slide images spanning 39 cancer types
• Multi-resolution tile formats for hierarchical learning

Phase 2: Clinical Task Evaluation (October - November 2025)

• Evaluation across 23 clinically relevant downstream tasks
• Multi-institutional assessment by leading cancer centers
• Tasks include biomarker prediction, cancer subtyping, image retrieval, and survival prediction
• Independent evaluation ensuring fair comparison

Timeline

Who Can Participate

This competition is designed for a diverse audience including:

• Machine learning and AI practitioners
• Computer scientists and engineers
• Bioinformaticians
• Medical professionals and specialists
• Academic researchers and industry teams

No prior experience in pathology or medical image processing is required - the competition removes domain knowledge barriers to encourage broad participation.

Impact and Innovation

The SLC-PFM competition addresses critical challenges in computational pathology:

• Data Accessibility: Most current pathology foundation models rely on proprietary datasets, creating entry barriers for researchers
• Clinical Integration: Healthcare institutes vary in infrastructure and patient populations - models must generalize across diverse settings
• Standardized Evaluation: Lack of standardized benchmarking across multiple institutions

By providing access to the largest pathology dataset (~300M images) for pre-training PFMs and establishing a rigorous multi-institutional evaluation framework, this competition will advance self-supervised learning techniques and improve diagnostic capabilities for cancer patients worldwide.

Ready to participate? Visit our Registration page to get started!