A Multi-scale Linear-time Encoder for Whole-Slide Image Analysis
This work addresses efficient and scalable analysis of whole-slide images for medical applications, representing a novel method for a known bottleneck rather than an incremental improvement.
The paper tackled the challenge of whole-slide image analysis, which involves gigapixel resolutions and hierarchical magnifications, by introducing MARBLE, a multi-scale linear-time encoder that improves AUC by up to 6.9%, accuracy by 20.3%, and C-index by 2.3% on five public datasets.
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.