FFR-UNet: Feature Filter-Refinement UNet for Medical Image Segmentation


Weixin Xu (Beihang University)
The 35th British Machine Vision Conference
PDFPosterVideo (Right click to download)

Abstract

Medical image segmentation poses a significant challenge in the field of computer vision. Traditional approaches leverage Convolutional Neural Networks (CNNs) and Transformer-based methods to address the intricacies of medical image segmentation. However, inherent limitations persist: CNN-based methods often neglect long-range dependencies, and Transformer-based methods may overlook local context information. Moreover, in contrast to natural images, medical images present a distinct challenge wherein the foreground targets requiring segmentation are typically smaller, accompanied by a greater abundance of background information (considered as irrelevant information). This inherent characteristic often interferes with segmentation networks and leads to segmentation results that may lack the desired refinement. To overcome these deficiencies, we propose a novel Feature Filter Module (FFM) designed to discern between informative and non-informative features. These features seamlessly transition into our proposed Feature Refinement Module (FRM), assigning them distinct roles to establish a robust connection between the two input features. This strategy empowers our module to concurrently focus on both long-range dependencies and local context information by skillfully merging convolution operations with cross-attention mechanisms. Moreover, by integrating our proposed FFM and FRM into the encoder block of the UNet architecture, we introduce a novel framework named Feature Filter-Refinement UNet (FFR-UNet). Extensive experiments demonstrate the superiority of FFR-UNet, consistently achieving state-of-the-art (SOTA) performance compared to existing methods. Codes will be publicly available at https://github.com/xuweixinxxx/FFR-UNet.

Citation

@inproceedings{Xu_2024_BMVC,
author    = {Weixin Xu},
title     = {FFR-UNet: Feature Filter-Refinement UNet for Medical Image Segmentation},
booktitle = {35th British Machine Vision Conference 2024, {BMVC} 2024, Glasgow, UK, November 25-28, 2024},
publisher = {BMVA},
year      = {2024},
url       = {https://papers.bmvc2024.org/0046.pdf}
}


Copyright © 2024 The British Machine Vision Association and Society for Pattern Recognition
The British Machine Vision Conference is organised by The British Machine Vision Association and Society for Pattern Recognition. The Association is a Company limited by guarantee, No.2543446, and a non-profit-making body, registered in England and Wales as Charity No.1002307 (Registered Office: Dept. of Computer Science, Durham University, South Road, Durham, DH1 3LE, UK).

Imprint | Data Protection