Skip to content

Latest commit

 

History

History
39 lines (38 loc) · 3.84 KB

README.md

File metadata and controls

39 lines (38 loc) · 3.84 KB
Raw

FG-UNet: Rethinking Feature Guidance for medical image segmentation

This is an officially public repository of FG-UNet source code. Our repository will continue to be updated based on subsequent research.

Abstract

Despite the evident advantages of variants of UNet in medical image segmentation, these methods still exhibit limitations in the extraction of foreground, background, and boundary features. This paper introduces an enhanced UNet network, FG-UNet, rooted in feature guidance principles. Specifically, adjacent high-level and low-level features are used to gradually guide the network to perceive organizations and details of lesions, respectively. Subsequently, to accommodate lesion features of different scales, the multi-order gated aggregation (MGA) block is designed based on multi-order feature interactions to extract multi-scale features. Furthermore, a novel feature-guided context-aware (FGCA) block is devised to enhance the capability of FG-UNet to segment lesions by fusing boundary-enhancing features, object-enhancing features, and uncertain areas. Eventually, a bi-dimensional interaction attention (BIA) block is designed to enable the network to highlight crucial features more effectively. To appraise the efficacy of FG-UNet, extensive experiments were conducted on Kvasir-seg, ISIC2018, and COVID-19 datasets. The experimental results illustrate that FG-UNet achieves a DSC score of 92.70% on the Kvasir-seg dataset, which is 1.15% higher than that of the latest SCUNet++, 4.70% higher than that of ACC-UNet, and 5.17% higher than that of UNet. FG-UNet obtains state-of-the-art metrics in different medical image segmentation tasks compared to other networks.

Overview

Overall architecture of FG-UNet

Overall architecture.

Experiments

Comparative Experiments

Experiments.

Ablation Experiments

Ablation experiments

Visualization Results

Grad-CAM visualization

To validate the effectiveness of the proposed blocks, visualizations using Grad-CAM are presented, corresponding to the ablation experiments in Table 2, numbered (a) through (f). The specific details are as follows:

  • (a): w/o high-level and low-level feature guidance;
  • (b): w/o MGA;
  • (c): w/o BFF and object enhancement features;
  • (d): w/o FGCA;
  • (e): w/o BIA;
  • (f): Take only the prediction of the final decoder. Results of visualization of different blocks.

These visualizations provide compelling evidence for the proposed blocks. For example, when the MGA block is removed, the network is slightly challenged in extracting COVID-19 pneumonia of different sizes. Compared to (c), the FG-UNet with the help of the BFF block shows excellent performance in boundary segmentation. Furthermore, the network struggles to capture pneumonia areas when the FGCA block is removed, effectively proving the contextual awareness capability of the FGCA block. Finally, the attention of the network to lesions decreases significantly when the BIA block is removed.

Qualitative comparison

The qualitative analyses are conducted on three datasets. In the third line, FG-UNet accurately segments COVID-19 pneumonia. Other methods may produce false positives (e.g. UNet, Swin-UNet, SCUNet++) or false negatives (e.g. CTO, ACC-UNet). Results of qualitative comparison of different networks.

Environment:

conda create -n FGUNet python=3.10.9

conda install pytorch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 pytorch-cuda=11.8 -c pytorch -c nvidia

Datasets

  1. Kvasir-Seg link.
  2. ISIC2018 link.
  3. COVID-19 CT scan lesion segmentation dataset link.

Acknowledgment