Search papers, labs, and topics across Lattice.
The paper introduces Multi-scale Adaptive Recurrent Biomedical Linear-time Encoder (MARBLE), a novel multiple instance learning (MIL) framework for whole-slide image (WSI) analysis that leverages a purely Mamba-based architecture. MARBLE processes multiple magnification levels of WSIs in parallel and uses a linear-time state-space model to capture cross-scale dependencies, addressing the limitations of single-scale MIL methods and the quadratic complexity of transformer-based approaches. Experiments on five public datasets demonstrate that MARBLE achieves significant improvements in AUC, accuracy, and C-index compared to existing methods, establishing its efficacy for multi-scale WSI analysis.
Ditch the quadratic attention bottleneck: MARBLE uses Mamba to achieve state-of-the-art results on whole-slide image analysis with linear complexity.
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.
