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

TL;DR

MARBLE is a novel multi-scale linear-time encoder for whole-slide image analysis that efficiently captures cross-scale dependencies, outperforming transformer-based methods in accuracy and scalability.

Contribution

It introduces the first purely Mamba-based multi-scale MIL framework with linear-time sequence modeling for WSI analysis, addressing scalability and efficiency issues.

Findings

Up to 6.9% AUC improvement on public datasets

20.3% accuracy increase over baselines

2.3% C-index enhancement in experiments

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.