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

Dimensions overall score 8.0

• MARBLE is the first purely Mamba-based multi-state multiple instance learning (MIL) framework for whole-slide image (WSI) analysis.method 70%
  Evidencepartial

  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.

  Implicationpartial

  The abstract explicitly states this as a novel contribution.

  Verificationpartial

  partial

• MARBLE processes multiple magnification levels in parallel and integrates coarse-to-fine reasoning within a linear-time state-space model.method 70%
  Evidencepartial

  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.

  Implicationpartial

  This describes the core architectural innovation of MARBLE, as stated in the abstract.

  Verificationpartial

  partial

• MARBLE efficiently captures cross-scale dependencies with minimal parameter overhead.technical 70%
  Evidencepartial

  efficiently capturing cross-scale dependencies with minimal parameter overhead.

  Implicationpartial

  The abstract directly states this benefit of the MARBLE architecture.

  Verificationpartial

  partial

• Existing MIL methods typically operate at a single scale, and transformer-based approaches suffer from quadratic attention costs.limitation 70%
  Evidencepartial

  while existing MIL methods typically operate at a single scale and transformer-based approaches suffer from quadratic attention costs.

  Implicationpartial

  This highlights the limitations of prior art that MARBLE aims to overcome, as stated in the abstract.

  Verificationpartial

  partial

• MARBLE provides a scalable and modular alternative to attention-based architectures.technical 70%
  Evidencepartial

  By coupling parallel multi-scale processing with linear-time sequence modeling, MARBLE provides a scalable and modular alternative to attention-based architectures.

  Implicationpartial

  This positions MARBLE as a superior alternative to existing methods, as stated in the abstract.

  Verificationpartial

  partial

• MARBLE achieved improvements of up to 6.9% in AUC on five public datasets.result 70%
  Evidencepartial

  Experiments on five public datasets show improvements of up to \textbf{6.9\%} in AUC

  Implicationpartial

  Specific quantitative results are provided in the abstract.

  Verificationpartial

  partial

• MARBLE achieved improvements of up to 20.3% in accuracy on five public datasets.result 70%
  Evidencepartial

  \textbf{20.3\%} in accuracy

  Implicationpartial

  Specific quantitative results are provided in the abstract.

  Verificationpartial

  partial

• MARBLE achieved improvements of up to 2.3% in C-index on five public datasets.result 70%
  Evidencepartial

  and \textbf{2.3\%} in C-index

  Implicationpartial

  Specific quantitative results are provided in the abstract.

  Verificationpartial

  partial

Competitive landscape

MARBLE offers a scalable, efficient tool for multi-scale whole-slide image analysis with significant accuracy improvements.