A Multi-scale Linear-time Encoder for Whole-Slide Image Analysis.

2026-02-02

arXiv.org

AI Summary

This work addresses the computational and modeling challenges in whole-slide image (WSI) analysis posed by ultra-high resolution and multi-scale structures, where existing methods are often limited to single-scale processing and suffer from the high complexity of Transformers. To overcome these limitations, we propose MARBLE, a novel framework that introduces a pure Mamba architecture into WSI multiple instance learning for the first time. MARBLE leverages parallel multi-scale processing and state space models to capture coarse-to-fine cross-scale dependencies with linear time complexity, achieving both high efficiency and low parameter overhead. Evaluated on five public datasets, MARBLE demonstrates substantial performance gains, with improvements of up to 6.9% in AUC, 20.3% in accuracy, and 2.3% in C-index, confirming its superior generalization capability and computational efficiency.

Technology Category

• Machine Learning: Multi-instance/Multi-view Learning
• Computer Vision: Multi-modal Vision
• Intelligent Robots: Multi-Robot Systems

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

We introduce Multi-scale Adaptive Recurrent Biomedical Linear-time Encoder (MARBLE), the first \textit{purely Mamba-based} multi-state multiple instance learning (MIL) framework for whole-slide image (WSI) analysis. MARBLE processes multiple magnification levels in parallel and integrates coarse-to-fine reasoning within a linear-time state-space model, efficiently capturing cross-scale dependencies with minimal parameter overhead. WSI analysis remains challenging due to gigapixel resolutions and hierarchical magnifications, while existing MIL methods typically operate at a single scale and transformer-based approaches suffer from quadratic attention costs. By coupling parallel multi-scale processing with linear-time sequence modeling, MARBLE provides a scalable and modular alternative to attention-based architectures. Experiments on five public datasets show improvements of up to \textbf{6.9\%} in AUC, \textbf{20.3\%} in accuracy, and \textbf{2.3\%} in C-index, establishing MARBLE as an efficient and generalizable framework for multi-scale WSI analysis.

Problem

• whole-slide image analysis
• multi-scale processing
• gigapixel resolution
• multiple instance learning
• computational complexity

Innovation

• Mamba
• multi-scale
• linear-time
• whole-slide image
• multiple instance learning