FMaMIL: Frequency-Driven Mamba Multi-Instance Learning for Weakly Supervised Lesion Segmentation in Medical Images

2025-06-09

AI Summary

Weakly supervised lesion segmentation in medical imaging is hindered by the scarcity of pixel-level annotations. To address this, we propose a two-stage frequency-domain enhanced Mamba framework. In Stage I, a learnable frequency-domain encoder collaborates with a Mamba-based encoder to generate high-quality class activation maps (CAMs), pioneering the integration of spectral modeling into multi-instance learning (MIL). In Stage II, soft pseudo-labels are generated under CAM guidance, augmented by a self-correction mechanism to suppress label noise. By embedding frequency-domain priors, our method enhances spatial-structural awareness without requiring explicit pixel-wise supervision. Evaluated on multiple public and private histopathological datasets, it significantly outperforms existing weakly supervised state-of-the-art methods, achieving substantial gains in segmentation accuracy and robustness. This work validates the efficacy of frequency-driven sequential modeling for weakly supervised medical image analysis.

Technology Category

• Machine Learning: Multi-instance/Multi-view Learning
• Computer Vision: Segmentation
• Search and Optimization: Metareasoning and Metaheuristics

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Accurate lesion segmentation in histopathology images is essential for diagnostic interpretation and quantitative analysis, yet it remains challenging due to the limited availability of costly pixel-level annotations. To address this, we propose FMaMIL, a novel two-stage framework for weakly supervised lesion segmentation based solely on image-level labels. In the first stage, a lightweight Mamba-based encoder is introduced to capture long-range dependencies across image patches under the MIL paradigm. To enhance spatial sensitivity and structural awareness, we design a learnable frequency-domain encoding module that supplements spatial-domain features with spectrum-based information. CAMs generated in this stage are used to guide segmentation training. In the second stage, we refine the initial pseudo labels via a CAM-guided soft-label supervision and a self-correction mechanism, enabling robust training even under label noise. Extensive experiments on both public and private histopathology datasets demonstrate that FMaMIL outperforms state-of-the-art weakly supervised methods without relying on pixel-level annotations, validating its effectiveness and potential for digital pathology applications.

Problem

• Weakly supervised lesion segmentation using image-level labels
• Long-range dependency capture in medical image patches
• CAM-guided pseudo label refinement for robust training

Innovation

• Mamba-based encoder for long-range dependencies
• Frequency-domain encoding enhances spatial sensitivity
• CAM-guided self-correction for robust training