# SCEAF-UNet: Medical image segmentation based on spatial-channel feature enhancement and adaptive fusion

## TL;DR

This paper introduces a new medical image segmentation model that improves accuracy by enhancing spatial and channel features.

## Contribution

The novel SCEAF module combines spatial and channel attention mechanisms for better medical image segmentation.

## Key findings

- SCEAF-UNet outperforms existing models on Synapse and ACDC datasets.
- The SCEAF module improves spatial detail recovery and channel feature discrimination.

## Abstract

Achieving a balance between spatial and channel feature representations is critical for improving performance in medical image segmentation. This paper proposes the spatial-channel feature enhancement and adaptive fusion (SCEAF) module. This module is composed of a multi-scale spatial attention gated block (MSAGBlock) and a channel attention modulation block (CAMBlock) operating in parallel. The MSAGBlock enhances spatial detail recovery, while the CAMBlock strengthens channel feature discrimination, and achieves dynamic fusion between the two blocks by means of gated weighting. Building upon the RWKV-UNet backbone network, we integrate the SCEAF module into the decoder to construct the novel SCEAF-UNet architecture. In addition, we introduce the lightweight edge attention fusion (EAF) module at the skip connection, which captures edge information and highlights structural contours, helping the network better delineate organ borders. Experiments conducted on the public Synapse and ACDC datasets indicate that SCEAF-UNet significantly surpasses current models of various architectures. Further ablation experiments verify the effectiveness and scalability of the designed modules, which are suitable for integration into diverse medical image segmentation architectures.