MRFAE-UNet: Multi-Receptive Field Attention-Enhanced UNet for Medical Image Segmentation

Automatic and accurate segmentation of pathological tissues or different organs from medical images of different modalities is crucial for various medical diagnoses. Convolutional neural networks (CNNs), especially fully convolutional networks (FCNs) like U-Net, have achieved remarkable success in medical image segmentation tasks. However, existing segmentation networks lack the ability to fully extract and automatically aggregate salient features for objects with large changes in shape, scale, and texture, and the segmentation accuracy needs to be improved. To this end, this paper proposes an effective convolutional neural network based on 3DUnet, Multi-Receptive Field Attention-Enhanced UNet (MRFAE-UNet). Specifically, we propose two modules: (1) Multi-Receptive Field Residual Encoder Module, which is used to replace traditional convolution and enhance the local extraction and global perception capabilities of medical image features. (2) Spatially Attention Gate is integrated into the skip connection to highlight its salient features during the upsampling process. A large number of experiments show that our method has achieved state-of-the-art performance in different modalities and different segmentation tasks on the public BraTS2021 dataset and BTCV dataset, proving the efficiency and robustness of our method.