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

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Dwarampudi, Jagan Mohan Reddy, Wong, Joshua, Van Nguyen, Hien, Banerjee, Tania
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866908807911702528
author Dwarampudi, Jagan Mohan Reddy
Wong, Joshua
Van Nguyen, Hien
Banerjee, Tania
author_facet Dwarampudi, Jagan Mohan Reddy
Wong, Joshua
Van Nguyen, Hien
Banerjee, Tania
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2602_02918
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle A Multi-scale Linear-time Encoder for Whole-Slide Image Analysis
Dwarampudi, Jagan Mohan Reddy
Wong, Joshua
Van Nguyen, Hien
Banerjee, Tania
Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
Tissues and Organs
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.
title A Multi-scale Linear-time Encoder for Whole-Slide Image Analysis
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
Tissues and Organs
url https://arxiv.org/abs/2602.02918