CT image black band artifact elimination method and system based on deep learning

Abstract

The invention relates to a CT image black band artifact elimination method based on deep learning. The method comprises the steps that a CT image with black band artifacts and a CT image without blackband artifacts which is not matched with the CT image with black band artifacts are collected; artifact elimination is carried out on the CT image with the black band artifacts; noise is added to theCT image without black band artifacts; normalization processing is respectively performed on the four CT images; a Cycle-GAN network model is constructed according to the model; network training is performed on the network model by utilizing the processed CT image with the black band artifacts, the CT image without the black band artifacts, the CT image with the CT value offset and the CT image with the noise until network parameters of the network model reach target conditions, so as to obtain a trained network model; and artifact elimination is performed on the CT image to be processed, andthe CT images before and after artifact elimination are input into the network model to obtain a CT image. According to the method, under the condition that the overall CT value of the CT image is basically unchanged, the black band artifacts in the CT image are effectively relieved.

Description

technical field [0001] The invention belongs to the technical field of CT image processing, and in particular relates to a method and system for eliminating black band artifacts in CT images based on deep learning. Background technique [0002] Computed Tomography (CT) uses precisely collimated X-ray beams and highly sensitive detectors to perform cross-sectional scans one after another around a certain part of the human body, so that the tissue structure of the human body can be shown to the audience in the form of pictures. Doctor, for doctor's diagnosis. The traditional CT data acquisition time period is long. In order to prevent motion artifacts in the CT image, the patient needs to remain still for a long time, which is somewhat impractical for the patient. Therefore, the spiral CT came into being. However, it is easy to produce cone beam artifacts when reconstructing CT images. [0003] In CT images, high-density substances (such as bones) and low-density substances ...

AI Technical Summary

Problems solved by technology

[0006] 1. The collected CT images with black-band artifacts and the corresponding CT images without black-band artifacts are directly used to train the convolutional depth network or generate adversarial networks. The trained model can alleviate the black-band artifacts, However, there will be a phenomenon that the overall CT value of the CT image is too large;

[0007] 2. In order to prevent the phenomenon that the overall CT value of the CT image is too large, try to segment the black belt artifact first, and then use the segmented image to train the network. This method seems feasible, but due to the position of the black belt artifact in the CT image It is not fixed and the shape is irregular, and it is impossible to accurately segment the black band artifacts of the CT image, and the image segmentation is also more complicated

[0010] 1. The information loss at the black band artifact in the CT image is very serious. If only CT images with black band artifacts are input to the black band artifact removal network (G_AtoB), the network cannot compensate for the lost information;

[0011] 2. The two loss functions that play a major role in Cycle-GAN are adversarial loss, which pulls the source domain image distribution to the target domain image distribution and cycle consistency loss (cycle consistency loss), which can ensure that the generated image retains the input image If you only rely on the above two loss constraints to generate an image, compared with the original image, some image details may change or even be lost.

Images