Low-dose CT reconstruction method based on iteration data refinement and self-attention

The invention relates to a low-dose CT reconstruction method based on iteration data refinement and self-attention, and belongs to the field of CT image reconstruction. According to the method, a guide image filtering technology is introduced to filter a low-dose CT image to generate a new image, the filtered image is used as input data of network training, and an original low-dose CT image is used as label data, so that the problem of lack of labels is solved. When the convolutional neural network is designed, a self-supervised learning network in unsupervised learning is selected to be applied to low-dose CT image reconstruction. A self-attention mechanism is introduced into a classical self-encoding-decoding residual convolutional neural network, and a new self-supervised learning network is constructed and serves as a network trained by the method. And finally, training the self-supervised learning network by adopting an iterative data refinement technology so as to alleviate the problem of data offset, t