Adaptive Feature Interaction and Enhancement Network for Medical Image Segmentation

In recent years, numerous CNN-ViT hybrid models have been extensively studied for medical image segmentation. However, most of these models face significant challenges such as large feature discrepancies between the encoder and decoder, poor inter-level feature interaction, and high computational complexity. To address these issues in medical image segmentation, this paper proposes a novel Adaptive Feature Interaction and Enhancement Network (AFIENet) including the Adaptive Space-Channel Attention (ASCA) block, the Graph Convolution-based Feature Enhancement (GCFE) block, and the Lightweight Feature Extraction (LFE) block. ASCA can activate relevant features and filter out redundant information to effectively mitigate feature differences from the encoder and the decoder. GCFE can enhance the semantic and structural information of the original features to effectively alleviate blur issue of segmentation edge. LFE can provide additional information from the original image to improve segmentation performance with low complexity. To validate the performance of AFIENet, extensive experiments have been conducted and the results demonstrate that AFIENet offers excellent medical image segmentation performance and can be widely applied to related tasks.