Deep learning post-processing method for solving low-dose CT image over-smoothing

Abstract

The invention discloses a deep learning post-processing method for solving low-dose CT image over-smoothing, and belongs to the technical field of image processing. The method comprises the steps: carrying out preprocessing of a plurality of normal-dose CT images and low-dose CT image pairs; inputting the processed low-dose CT images into a generator of a generative adversarial network to obtain generated images; inputting the generated images and the processed normal-dose CT images into a discriminator of the generative adversarial network, and optimizing the discriminator according to a discriminator loss function; based on the optimized discriminator, optimizing the generator by taking a joint loss function formed by minimizing the generator loss, perception loss and structural similarity loss as a target; generating new generated images by using the optimized generator, and inputting the new generated images into the optimized discriminator; and continuing to alternately optimize the discriminator and the generator. Noise can be effectively suppressed, more image structures and high-frequency information can be reserved, and the images are prevented from being over-smooth.

Description

technical field [0001] The invention belongs to the technical field of image processing, and more specifically relates to a deep learning post-processing method for solving over-smoothing of low-dose CT images. Background technique [0002] X-ray computed tomography (X-Ray computedtomography, CT) is currently widely used in clinical disease inspection, treatment and other aspects. However, excessive X-ray radiation may cause some diseases such as cancer and gene mutation, so low-dose CT imaging technology has become a research hotspot in recent years. [0003] At present, the way to reduce the dose in low-dose CT reconstruction is to reduce the current of the ray tube or reduce the number of projections. Reducing the current of the ray tube will reduce the number of photons received by the detector, resulting in noise in the CT image; reducing the number of projections will make CT reconstruction an underdetermined problem, resulting in serious bar artifacts in the image. ...

AI Technical Summary

Problems solved by technology

[0007] In view of the above defects or improvement needs of the prior art, the present invention provides a deep learning post-processing method to solve the over-smoothing of low-dose CT images, aiming to solve the problem of over-smoothing of the output image caused by the deep learning network in the post-processing part

Images