DCAU-Net: Medical Image Segmentation based on Deformable Convolution and Attention Gate

Automatic segmentation of cardiac magnetic resonance imaging (MRI) is critical for clinical analysis and diagnosis of cardiac function. With the advancement of deep learning in medical image segmentation, an increasing number of cardiac segmentation methods based on deep learning have been proposed. This paper introduces a new neural network model, DCAU-Net, which uses a u-shaped encoder-decoder architecture. The deformable convolution is introduced to overcome limitations of traditional convolution in processing features of varying shapes, enabling the model to better adapt to irregular structures and deformations in cardiac MRI images. The improved attention gate module helps the model concentrate on important regions within images, distinguishing between target and background areas, and capturing long-range contextual information more accurately. Comprehensive experiments were conducted on the Automatic Cardiac Diagnosis Challenge (ACDC) dataset. The results show our method achieves accurate segmentation of target areas and outperforms other networks in accuracy.