Abstract:Few-shot Semantic Segmentation(FSS)aim to adapt a pre-trained model to new classes with as few as a single labeled training sample per class. The existing prototypical work used in natural image scenarios biasedly focus on capturing foreground's discrimination while employing a simplistic representation for background, grounded on the inherent observation separation between foreground and background. However, this paradigm is not applicable to medical images where the foreground and background share numerous visual features, necessitating a more detailed description for background. In this paper, we present a new pluggable Background-fused prototype(Bro)approach for FSS in medical images. Instead of finding a commonality of background subjects in support image, Bro incorporates this background with two pivot designs. Specifically, Feature Similarity Calibration(FeaC)initially reduces noise in the support image by employing feature cross-attention with the query image. Subsequently, Hierarchical Channel Adversarial Attention(HiCA)merges the background into comprehensive prototypes. We achieve this by a channel groups-based attention mechanism, where an adversarial Mean-Offset structure encourages a coarse-to-fine fusion. Extensive experiments show that previous state-of-the-art methods, when paired with Bro, experience significant performance improvements. This demonstrates a more integrated way to represent backgrounds specifically for medical image.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to solve the problem of background representation in **Few - Shot Medical Image Segmentation (FSS)**. Specifically, existing FSS methods, when dealing with natural images, usually focus on capturing the discriminative features of the foreground, while the representation of the background is relatively simple. However, in medical images, the foreground and background often share many visual features, resulting in simple background representation methods being unable to effectively distinguish between the foreground and the background. #### Main problems: 1. **High similarity between foreground and background**: In medical images, the visual features between the foreground (such as organs or tissues) and the background (such as other adjacent organs or tissues) are very similar, which makes it difficult for traditional foreground - centered methods to distinguish between the foreground and the background. 2. **Deficiencies of existing methods**: Most of the existing FSS methods are designed based on natural images, ignoring the importance of the background in medical images, resulting in poor performance on medical images. #### Solutions proposed in the paper: To solve the above problems, the author proposes a new plug - gable **Background - fused Prototype (Bro)** method. Bro improves the background representation through the following two modules: 1. **Feature Similarity Calibration (FeaC)**: - Utilize the cross - attention mechanism between the query image and the support image to reduce the noise in the support image and enhance the similarity of the foreground area. - Formula representation: \[ \hat{F_s}=\delta\left(\frac{U_s^T\times U_q}{\|U_s\|\|U_q\|}\right)\times U_s + U_s \] where \( U_s \) and \( U_q \) are the feature matrices of the support image and the query image respectively, and \( \delta \) represents the softmax operation. 2. **Hierarchical Channel - adversarial Attention (HiCA)**: - Generate detailed background representations through self - cross - similarity and adversarial regularization between channel groups. - Formula representation: \[ eF_s = g(G)\times G,\quad G = CG(\hat{F_s}) \] where \( G \) is the channel group generated by \( \hat{F_s} \), and \( g(G) \) is the similarity matrix generated by self - cross - similarity and adversarial regularization. #### Experimental results: The author conducted experiments on three medical image benchmark datasets (ABD - CT, ABD - MRI, and CMR). The results show that the Bro method significantly improves the performance of existing FSS methods, especially in more challenging categories (such as the left and right kidneys). ### Summary: This paper solves the problem of insufficient background representation in few - shot segmentation in medical images by introducing the Bro method, providing a more effective background fusion strategy, thereby improving the accuracy and robustness of the segmentation model.

Is Foreground Prototype Sufficient? Few-Shot Medical Image Segmentation with Background-Fused Prototype

Few-Shot Medical Image Segmentation with High-Fidelity Prototypes

Beyond singular prototype: A prototype splitting strategy for few-shot medical image segmentation

Prototype-Guided Graph Reasoning Network for Few-Shot Medical Image Segmentation

Query Semantic Reconstruction for Background in Few-Shot Segmentation

Support-Query Prototype Fusion Network for Few-shot Medical Image Segmentation

Addressing Background Context Bias in Few-Shot Segmentation Through Iterative Modulation

Self-Supervision with Superpixels: Training Few-shot Medical Image Segmentation without Annotation

Few-Shot Segmentation via Complementary Prototype Learning and Cascaded Refinement.

Few-Shot Medical Image Segmentation via Generating Multiple Representative Descriptors

Learning what and where to segment: A new perspective on medical image few-shot segmentation

Partition-A-Medical-Image: Extracting Multiple Representative Subregions for Few-Shot Medical Image Segmentation

Prototype Correlation Matching and Class-Relation Reasoning for Few-Shot Medical Image Segmentation

SCNet: Enhancing Few-Shot Semantic Segmentation by Self-Contrastive Background Prototypes

PFMNet: Prototype-based feature mapping network for few-shot domain adaptation in medical image segmentation

Few-shot Medical Image Segmentation Regularized with Self-reference and Contrastive Learning

Prototype Comparison Convolutional Networks for One-Shot Segmentation

ABPNet: Adaptive Background Modeling for Generalized Few Shot Segmentation.

Recurrent Mask Refinement for Few-Shot Medical Image Segmentation

Partition-A-Medical-Image: Extracting Multiple Representative Sub-regions for Few-shot Medical Image Segmentation

Dual-Guided Prototype Alignment Network for Few-Shot Medical Image Segmentation