RUFNet: Query-Guided Support Mask Refinement and Uncertainty Fusion based on Hybrid Mamba for Few-shot Brain Tumor Segmentation

Jul 16, 2026

Upload by Maeve

Few-shot brain tumor segmentation faces three key issues: noisy support masks, inter-patient appearance discrepancies between support and query images, and missing pixel-wise confidence estimation. RUFNet introduces a Hybrid Mamba-based few-shot framework that integrates attention-guided mask refinement with uncertainty-aware posterior fusion. A Hybrid Mamba interaction backbone preserves support-query dependencies at linear cost. Experiments on BraTS 2020 yield Dice of 84.3% (1-way 1-shot) and 86.1% (1-way 5-shot), outperforming prior work.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages9

Reading Time23 min

Abstract

Introduction

Background and motivation

Limitations of existing few-shot methods

Design rationale and challenges

arXiv :2607 .05035v 1 [ cs .CV] 6 Jul 2026 RUFNET: QUERY-GUIDED SUPPORT MASK REFINEMENT AND UNCERTAINTY FUSION BASED ON HYBRID MAMBA FOR FEW-SHOT BRAIN TUMOR SEGMENTATION Dongyi He 1 , Xiangkai Wang2 , Binbing Xu2 , Bin Jiang2 , Hongjie Yan3 , Weixiang Liu4 , Wai Ting Siok1 , and Nizhuan Wang 1,* 1Department of Language Science and Technology, The Hong Kong Polytechnic University, Hong Kong SAR, China 2 School of Artificial Intelligence, Chongqing University of Technology, Chongqing, China 3Affiliated Lianyungang Hospital of Xuzhou Medical University, Lianyungang, China 4 College of Mechatronics and Control Engineering, Shenzhen University, Shenzhen, China * [Correspondence: wangnizhuan1120@gmail.com](Correspondence: wangnizhuan1120@gmail.com) ABSTRACT Few-shot brain tumor segmentation remains challenging due to noisy support masks, inter-patient variations between support and query images, and the lack of pixel-wise confidence estimation. This study proposes RUFNet, a Hybrid Mamba-based few-shot framework that combines support mask refinement with uncertainty-aware posterior fusion. To preserve support-query dependencies with manageable cost, RUFNet adopts a Hybrid Mamba interaction backbone with linear complexity. To reduce support-mask noise, an Attention-Guided Mask Refinement module (AGMR) uses query features to recalibrate support masks and improve prototype consistency. To handle ambiguous predictions, an Uncertainty-Aware Posterior Fusion module (UAPF) estimates pixel-wise variance and adaptively balances few-shot predictions with query-aligned priors. On the Brain Tumor Segmentation Challenge (BraTS) 2020 dataset, RUFNet achieves Dice coefficients of 84.3% and 86.1% in the 1-way 1-shot and 1-way 5-shot settings, respectively, outperforming the compared state-of-the-art methods. These results suggest that Hybrid Mamba interaction, mask refinement and uncertainty modelling can improve the robustness of few-shot medical image segmentation. The official implementation code is available at [https://github.com/hdy6438/RUFNet](https://github.com/hdy6438/RUFNet). Keywords Few-shot learning · Brain tumor segmentation · Mask refinement · Uncertainty estimation · Mamba 1 Introduction Brain tumor segmentation in multi-modal magnetic resonance imaging (MRI) is a fundamental task in neuro-oncology image analysis. Accurate delineation of tumor regions and subregions supports diagnosis, treatment planning and longitudinal assessment, but manual annotation is time-consuming and depends on specialist expertise. Deep learning has therefore become an important approach for automated brain tumor segmentation [1] . Despite some progress, high-performing segmentation networks remain annotation-intensive. nnU-Net adapts preprocessing and architecture to each dataset, but still assumes enough labelled cases for reliable configuration and training [2] . Swin-UNet introduces Transformer-style long-range modelling, but its attention-based design is data hungry and computationally heavier when dense image interactions are required [3] . These requirements are difficult to satisfy in clinical settings, where rare tumor subtypes, small patient cohorts and expensive voxel-level annotation limit labelled data. Few-shot learning can reduce this burden, but brain tumor models remain vulnerable because tumor appearance, size and location vary substantially across patients [4] . Existing few-shot segmentation methods address parts of this problem, but leave important gaps. PANet uses prototype alignment, yet its prototypes can be corrupted by inaccurate support masks [5] . SENet improves channel recalibration, RUFNet Figure 1: Overall architecture of RUFNet. Support and query images are encoded by a shared backbone, the support mask is refined by AGMR, and Hybrid Mamba blocks model support-query interaction. UAPF then fuses the metaprediction with the refined prior to produce the final segmentation. but does not expl

FAQs

What problems does RUFNet target in few-shot brain tumor segmentation?

It addresses noisy support masks, appearance variations between support and query images across patients, and the absence of pixel-wise confidence estimation for uncertain boundaries.

How does RUFNet refine support masks?

RUFNet uses an Attention-Guided Mask Refinement (AGMR) module that recalibrates support masks using query features to improve prototype consistency.

How does RUFNet handle uncertainty during prediction fusion?

The Uncertainty-Aware Posterior Fusion (UAPF) module estimates pixel-wise variance and adaptively balances few-shot predictions with query-aligned priors.

Jul 16, 2026

Upload by Maeve