Progressive generative adversarial network for low-dose CT image noise reduction and artifact removal

Abstract

The invention belongs to the technical field of CT imaging, and particularly discloses a progressive generative adversarial network for low-dose CT image noise reduction and artifact removal, a double-generator nested sub-network is designed, each generator comprises a global feature denoiser and a local texture feature intensifier, the global feature denoiser performs feature extraction on a full-resolution input image; global features of the image are obtained; the local texture feature intensifier is used for carrying out feature extraction on an image with relatively low resolution after the input image is subjected to down-sampling, and local detail features of the image are captured; and an LDCT image noise reduction task is completed jointly. According to the invention, a shuffle discriminator network for multi-scale feature extraction is designed, the discrimination capability of a discriminator is improved and the stability and robustness of GAN adversarial training are enhanced while the complexity and operation time of a network structure are not increased; the problem of insufficient noise reduction or excessive noise reduction caused by the fact that noise artifacts are highly similar to organization structure distribution is solved.

Description

technical field [0001] The invention belongs to the technical field of CT imaging, and discloses a deep learning method for efficiently suppressing noise and artifacts in low-dose CT images and facilitating accurate medical diagnosis and analysis in the later stage. Background technique [0002] Since the advent of Computed Tomography (CT) technology in the 1970s, it has been widely used in industrial and agricultural production, safety inspection, biomedical imaging, and industrial nondestructive imaging due to its advantages of simple operation, fast imaging speed, and high sensitivity. Detection, geology and other fields. In the field of medical diagnosis and treatment, CT images have the advantages of clear imaging, high density resolution, and the ability to clearly display three-dimensional information of images, so they are widely used in various congenital developmental abnormalities, inflammatory diseases, metabolic lesions, traumatic changes, Benign and malignant ...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the technical problems of high network complexity, large amount of parameters, and unstable training process existing in the prior art, the present invention provides a deep learning method that can realize fast and efficient LDCT image noise reduction. Under the premise of complexity and calculation time, better artifact noise suppression and detail preservation effect can be achieved

Images