Benchmarking Pathology Foundation Models for Spatial Domain Understanding

2026-05-25

AI Summary

Current foundation models in computational pathology lack systematic evaluation of their ability to effectively capture tissue spatial architecture. To address this gap, this work proposes SpaPath-Bench—the first representation-level benchmark specifically designed for assessing spatial structure understanding in pathology foundation models. By integrating 42 publicly available paired whole-slide images and spatial transcriptomics datasets, the benchmark establishes a spatial domain identification task. It encompasses 19 encoder architectures and 7 spatial domain identification methods, and systematically evaluates model performance across 83,000 experiments using three criteria: unsupervised spatial consistency, transcriptomic reference consistency, and expert reference consistency. The results reveal distinct differences in how various pretraining paradigms encode tissue spatial organization, offering critical insights for developing next-generation pathology models with enhanced spatial awareness.

Abstract

Pathology foundation models (PFMs) have emerged as a core approach for learning transferable representations from whole slide images (WSIs), and they are typically benchmarked through downstream clinical endpoints. While such task level evaluations are indispensable, they offer limited insight into what the representations themselves encode, particularly whether PFM embeddings can distinguish meaningful tissue regions and capture their spatial relationships. We present SpaPath-Bench, a representation level benchmark designed to diagnose spatial representation capability in PFMs. SpaPath-Bench formulates spatial domain identification (SDI) on paired whole slide image and spatial transcriptomics (ST) data as a diagnostic task. It curates 42 public paired WSI and ST slides, enables large scale evaluation across 19 encoders and seven SDI methods, and measures partition quality using three complementary criteria: unsupervised spatial coherence, transcriptomics referenced agreement, and expert referenced agreement. Across 83K runs, SpaPath-Bench reveals that different pretraining paradigms capture distinct aspects of tissue spatial architecture, and it provides practical guidance for building the next generation of spatially aware computational pathology models. Code and data pipelines are publicly available at https://bokai-zhao.github.io/SpaPath-benchboard/.

Problem

• Pathology foundation models
• spatial domain understanding
• representation benchmarking
• spatial relationships
• whole slide images

Innovation

• spatial domain understanding
• pathology foundation models
• spatial transcriptomics
• representation benchmarking
• whole slide image