SCA-Net: A Lightweight Medical Image Segmentation Model Based on Spatial-Channel Feature Extraction

Medical image segmentation faces challenges such as inconsistent image quality, complex tissue structures, and multi-modal variability. While deep learning methods like CNNs have improved performance, they struggle with global information capture. This paper proposes SCA-Net, a lightweight network integrating a Spatial-Channel feature extraction module (SC block), Channel Prior Convolutional Attention(CPCA) attention mechanism, and Irregular Convolution Fusion Block (IFCB). SCA-Net achieves Dice coefficients of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$87.9 \%, 88.0 \%$</tex>, and 92.0% on Kvasir-SEG, ISIC2018, and GLaS datasets, respectively, with only 0.167 M parameters and 0.360 GFLOPs.