Fetal brain MR image reconstruction method based on multi-scale fusion attention residual dense network

The invention discloses a fetal brain MR image reconstruction method based on a multi-scale fusion attention residual dense network, and the method comprises the steps: inputting a low-resolution fetal brain MR image into a superficial layer feature extraction network, and extracting the superficial layer features of the contour, texture and edge of the fetal brain MR image; inputting the obtained feature map of the superficial layer features into a deep feature extraction network of a non-local attention residual dense block to extract deep features to obtain an approximate fetus brain MR image; the extracted deep features of the fetal brain MR image are input into a cavity space pyramid pooling module ASPP, so that the texture of the reconstructed fetal brain MR image is clearer; adding the deep features output by the ASPP and the output of a first convolutional layer of the shallow extraction network, and then performing up-sampling to a high-resolution space to obtain a preliminary high-resolution fetal b