FusionU-Net: U-Net with Enhanced Skip Connection for Pathology Image Segmentation

TL;DR

FusionU-Net introduces a novel two-round fusion module to reduce semantic gaps in U-Net, significantly improving pathology image segmentation performance by enhancing skip connection effectiveness.

Contribution

The paper proposes a new FusionU-Net architecture with a two-round fusion module that better exchanges information between encoder layers, reducing semantic gaps and improving segmentation accuracy.

Findings

FusionU-Net outperforms existing methods on pathology datasets.

The fusion module effectively reduces semantic gaps in skip connections.

FusionU-Net can be integrated into other networks for performance enhancement.

Abstract

In recent years, U-Net and its variants have been widely used in pathology image segmentation tasks. One of the key designs of U-Net is the use of skip connections between the encoder and decoder, which helps to recover detailed information after upsampling. While most variations of U-Net adopt the original skip connection design, there is semantic gap between the encoder and decoder that can negatively impact model performance. Therefore, it is important to reduce this semantic gap before conducting skip connection. To address this issue, we propose a new segmentation network called FusionU-Net, which is based on U-Net structure and incorporates a fusion module to exchange information between different skip connections to reduce semantic gaps. Unlike the other fusion modules in existing networks, ours is based on a two-round fusion design that fully considers the local relevance between adjacent encoder layer outputs and the need for bi-directional information exchange across multiple layers. We conducted extensive experiments on multiple pathology image datasets to evaluate our model and found that FusionU-Net achieves better performance compared to other competing methods. We argue our fusion module is more effective than the designs of existing networks, and it could be easily embedded into other networks to further enhance the model performance.