[Paper Review] MEGANet: Multi-Scale Edge-Guided Attention Network for Weak Boundary Polyp Segmentation

Nhat-Tan Bui, Dinh-Hieu Hoang+3 more|arXiv (Cornell University)|Sep 6, 2023

Colorectal Cancer Screening and Detection7 citations

TL;DR

MEGANet introduces a multi-scale edge-guided attention framework for polyp segmentation, leveraging high-frequency edge features and decoded boundary/reverse attention to improve boundary delineation in weak-boundary scenarios.

ABSTRACT

Efficient polyp segmentation in healthcare plays a critical role in enabling early diagnosis of colorectal cancer. However, the segmentation of polyps presents numerous challenges, including the intricate distribution of backgrounds, variations in polyp sizes and shapes, and indistinct boundaries. Defining the boundary between the foreground (i.e. polyp itself) and the background (surrounding tissue) is difficult. To mitigate these challenges, we propose Multi-Scale Edge-Guided Attention Network (MEGANet) tailored specifically for polyp segmentation within colonoscopy images. This network draws inspiration from the fusion of a classical edge detection technique with an attention mechanism. By combining these techniques, MEGANet effectively preserves high-frequency information, notably edges and boundaries, which tend to erode as neural networks deepen. MEGANet is designed as an end-to-end framework, encompassing three key modules: an encoder, which is responsible for capturing and abstracting the features from the input image, a decoder, which focuses on salient features, and the Edge-Guided Attention module (EGA) that employs the Laplacian Operator to accentuate polyp boundaries. Extensive experiments, both qualitative and quantitative, on five benchmark datasets, demonstrate that our MEGANet outperforms other existing SOTA methods under six evaluation metrics. Our code is available at https://github.com/UARK-AICV/MEGANet.

Motivation & Objective

• Motivate robust polyp segmentation under weak boundaries and complex mucous membranes.
• Develop an edge-guided attention mechanism to focus on polyp boundaries across scales.
• Leverage high-frequency edge information and decoded attention to improve segmentation accuracy.

Proposed method

• Introduce an Edge-Guided Attention (EGA) module that ingests high-frequency features from the Laplacian pyramid to guide attention to boundary regions.
• Utilize a decoded predicted feature at each layer that is decomposed into boundary attention and reverse attention to refine predictions.
• Combine boundary attention and reverse attention to emphasize object regions and suppress background leakage.
• Operate within a multi-scale encoder-decoder architecture to propagate edge-guided cues across decoder levels.
• Visualize and leverage high-frequency features to preserve texture and edge details critical for polyp delineation.

Experimental results

Research questions

• RQ1 How can high-frequency edge information improve boundary precision in weak-boundary polyp segmentation?
• RQ2 Can an edge-guided attention mechanism at multiple scales consistently enhance polyp boundary delineation across diverse datasets?
• RQ3 What is the effect of decomposing decoded features into boundary and reverse attention on segmentation quality?
• RQ4 How does Laplacian-based boundary attention interact with multi-scale decoding to improve overall segmentation?

Key findings

• The Edge-Guided Attention module highlights boundary regions via high-frequency features from the Laplacian pyramid, guiding decoder layers to focus on polyp edges.
• Decoded features decomposed into boundary attention and reverse attention improve the distinction between polyps and surrounding mucosa.
• Combining boundary and reverse attention helps the model recognize whole object regions and suppress background noise.
• The approach emphasizes boundary accuracy in weak-boundary scenarios through multi-scale guidance and edge information preservation.