Edge-Enhanced Dense Network Based on Attention for Low-Dose CT Denoising

In recent years, due to the influence of the new crown pneumonia, the problem of low-dose CT image denoising has become a hot research direction. With the rapid development of deep learning technology, many algorithms that apply convolutional neural networks to this have also obtained good results. However, the current denoising algorithms still have problems such as excessive smoothing of images and obvious noise. Influenced by EDCNN network, in this algorithm, an edge-enhanced dense network based on attention mechanism (EDACNN) is proposed. In the network model, it is proposed to extract image features using attention mechanism and learnable Sobel convolutional kernel. The learnable Sobel convolution kernel enables good feature extraction where the edges of the image are uneven. The attention mechanisms introduced include channel attention mechanism and spatial attention mechanism. Not only can the channel attention mechanism feedback each pixel of the image in the process of feature extraction, but also it can pay attention to the largest feature point in the image. The spatial attention mechanism can focus on areas of the image where feature information is rich. Compared with the existing low-dose CT image denoising algorithm, the proposed model has significant improvement in all aspects of the denoising image.