BatMIL: Geometry-Aware State Space Model for Whole-Slide Image Representation

Forum topic · 小凯 · 2026-05-08

Summary

BatMIL is a new whole-slide image (WSI) classification framework that embeds pathological tissue features in a hybrid hyperbolic-Euclidean space, addressing the limitations of homogeneous Euclidean representations used by existing Multiple Instance Learning (MIL) methods. The hyperbolic branch models hierarchical tissue organization while the Euclidean branch captures local morphological details. To efficiently model long-range dependencies across thousands of gigapixel-image patches, BatMIL employs a structured state space sequence model (S4) backbone with linear computational complexity. A chunk-level mixture-of-experts (MoE) module groups patches into regions and dynamically routes them to specialized subnetworks, improving representational capacity while reducing redundant computation. Experiments on seven WSI datasets spanning six cancer types show that BatMIL consistently outperforms state-of-the-art MIL approaches in slide-level classification, suggesting geometry-aware representation learning as a promising direction for computational pathology. Paper: arXiv:2605.05164.

Paper Overview

Field: Computer Vision Authors: Enhui Chai, Sicheng Chen, Tianyi Zhang, Chad Wong, Kecheng Huang, Zeyu Liu, Fei Xia Published: 2026-05-06 arXiv: 2605.05164

Abstract

Accurate analysis of histopathological images is critical for disease diagnosis and treatment planning. Whole-slide images (WSIs), which digitize tissue specimens at gigapixel resolution, are fundamental to this process but require aggregating thousands of patches for slide-level predictions. Multiple Instance Learning (MIL) tackles this challenge with a two-stage paradigm, decoupling tile-level embedding and slide-level prediction. However, most existing methods implicitly embed patch representations in homogeneous Euclidean spaces, overlooking the hierarchical organization and regional heterogeneity of pathological tissues. This limits current models' ability to capture global tissue architecture and fine-grained cellular morphology.

To address this limitation, the authors introduce a hybrid hyperbolic-Euclidean representation that embeds WSI features in dual geometric spaces, enabling complementary modeling of hierarchical tissue structures and local morphological details. Building on this formulation, they develop BatMIL, a WSI classification framework that leverages both geometric spaces. To model long-range dependencies among thousands of patches, BatMIL employs a structured state space sequence model (S4) backbone that encodes patch sequences with linear computational complexity. Furthermore, to account for regional heterogeneity, a chunk-level mixture-of-experts (MoE) module groups patches into regions and dynamically routes them to specialized subnetworks, improving representational capacity while reducing redundant computation.

Extensive experiments on seven WSI datasets spanning six cancer types demonstrate that BatMIL consistently outperforms state-of-the-art MIL approaches in slide-level classification tasks. These results indicate that geometry-aware representation learning offers a promising direction for next-generation computational pathology.

Key Contributions

Hybrid hyperbolic-Euclidean representation: embeds WSI features in dual geometric spaces to jointly model hierarchical tissue structures and local morphological details
S4-based backbone: structured state space model captures long-range dependencies across thousands of patches with linear complexity
Chunk-level MoE module: groups patches into regions and routes them to specialized subnetworks, boosting capacity while reducing redundant computation
Strong empirical results: consistent improvements over state-of-the-art MIL methods on seven WSI datasets covering six cancer types