The invention belongs to the technical field of
CT imaging, and adopts the specific scheme that a multi-scale feature
generative adversarial network for suppressing artifact
noise in a low-
dose CT image is adopted, an LDCT
image noise reduction model is selected, an LDCT image and NDCT image
data set is constructed, the LDCT image is input into an
error feedback pyramid generator network, he
pyramid generator network extracts cross-scale features of the LDCT image from different angles, the LDCT image is processed by the
error feedback pyramid generator network and then a preliminary
noise reduction result graph is output, the NDCT image and the preliminary
noise reduction result graph are jointly input into an interleaved
convolution discriminator sub-network for iterative training, and afinal
noise reduction result graph is output; the
error feedback pyramid generator can extract shallow features and deep features in the same scale of the image, increase the richness of
feature extraction, improve the discrimination capability of the
discriminator, and the multi-scale feature
generative adversarial network solves the problem of under-
noise reduction or over-
noise reduction caused by high similarity between noise artifacts and tissue structure distribution.