CN115861470A

CN115861470A - Image artifact correction method, apparatus, device, storage medium and program product

Info

Publication number: CN115861470A
Application number: CN202211703656.0A
Authority: CN
Inventors: 张阳; 廖术
Original assignee: Shanghai United Imaging Intelligent Healthcare Co Ltd
Current assignee: Shanghai United Imaging Intelligent Healthcare Co Ltd
Priority date: 2022-12-29
Filing date: 2022-12-29
Publication date: 2023-03-28

Abstract

The present application relates to an image artifact correction method, apparatus, device, storage medium and program product. The method comprises the following steps: determining at least one candidate facies and low-dose candidate images corresponding to the candidate facies according to low-dose original images of a plurality of facies; the low-dose original image comprises a region of interest; reconstructing the low-dose candidate image, and determining a standard-dose candidate image corresponding to the low-dose candidate image; and performing artifact correction processing according to the standard dose candidate image and a preset first neural network, and determining an artifact correction image of the region of interest. By adopting the method, the radiation damage to the object to be detected can be reduced in the artifact correction process.

Description

Image artifact correction method, apparatus, device, storage medium and program product

Technical Field

The present application relates to the field of computer technologies, and in particular, to a method, an apparatus, a device, a storage medium, and a program product for correcting an image artifact.

Background

In the process of scanning a human body by using a CT (Computed Tomography) device, since the heart of the human body continuously beats, a motion artifact exists in a finally obtained heart coronary image, which affects the subsequent accurate analysis of the heart coronary image, and thus, the artifact correction is performed on the heart coronary image before the image analysis is performed on the heart coronary image.

In the related art, when artifact correction is performed on a coronary artery image of a heart, a coronary artery image of an object to be detected under a standard dose is generally obtained first, then the artifact on the coronary artery image with the standard dose is analyzed to obtain a specific type of the artifact, and standard-dose imaging is performed on the coronary artery again based on the specific type of the artifact to obtain the coronary artery image with the standard dose after artifact correction.

However, although the above-mentioned technique can obtain a coronary image with a standard dose after artifact correction, the radiation damage to the object to be measured is large.

Disclosure of Invention

In view of the foregoing, it is desirable to provide an image artifact correction method, an apparatus, a device, a storage medium, and a program product, which can reduce radiation damage to a subject during artifact correction.

In a first aspect, the present application provides a method for correcting an image artifact, the method comprising:

determining at least one candidate phase and a low-dose candidate image corresponding to the candidate phase according to low-dose original images of a plurality of phases; the low-dose original image comprises a region of interest;

reconstructing the low-dose candidate image to determine a standard-dose candidate image corresponding to the low-dose candidate image;

and performing artifact correction processing according to the standard dose candidate image and a preset first neural network, and determining an artifact correction image of the region of interest.

In one embodiment, the above-mentioned performing an artifact correction process according to the standard dose candidate image and a preset first neural network to determine an artifact corrected image of the region of interest includes:

performing region-of-interest segmentation processing on the standard dose candidate image to obtain a first target segmentation image of the region of interest;

and after establishing a corresponding relation between the standard dose candidate image and the first target segmentation image, inputting the standard dose candidate image and the first target segmentation image into a first neural network for artifact correction processing to obtain an artifact correction image of the region of interest.

In one embodiment, the reconstructing the low-dose candidate image to determine a standard-dose candidate image corresponding to the low-dose candidate image includes:

inputting the low-dose candidate image into a preset second neural network for reconstruction processing, and determining a standard-dose candidate image;

the second neural network is obtained by training according to a plurality of low-dose sample images and standard-dose sample images corresponding to the low-dose sample images.

In one embodiment, the determining at least one candidate facies and low-dose candidate images corresponding to the candidate facies according to low-dose original images of a plurality of facies includes:

reconstructing the low-dose original images of a plurality of phase phases, and determining a standard dose reconstructed image of each phase;

performing region-of-interest segmentation processing on each standard dose reconstruction image to obtain a second target segmentation image corresponding to each standard dose reconstruction image;

at least one candidate phase is determined from the plurality of phase phases based on each second target segmentation image and low dose candidate images corresponding to the candidate phase are determined.

In one embodiment, the determining at least one candidate phase from the plurality of phase phases and determining the low-dose candidate image corresponding to the candidate phase according to each second target segmentation image includes:

performing quality quantization processing on each second target segmentation image according to a preset quality quantization mode to obtain a quality quantization value of each second target segmentation image;

determining at least one candidate phase from the plurality of phases based on the respective mass quantization values;

acquiring raw data corresponding to the candidate period, and performing image reconstruction on the raw data to obtain a low-dose candidate image; the raw data is low dose raw data.

In one embodiment, the image resolution of the low-dose candidate image is greater than the image resolution of the low-dose original image of the corresponding phase.

In one embodiment, the first target segmentation image is a mask image of a region of interest and/or a segmentation probability map of the region of interest.

In one embodiment, the determining an artifact-corrected image of the region of interest by performing artifact correction processing according to the standard dose candidate image and a preset first neural network includes:

performing artifact correction processing according to the standard dose candidate image and a preset first neural network, and determining an initial artifact correction image of the region of interest;

carrying out inverse normalization processing on the initial artifact correction image according to preset normalization parameters to determine an artifact correction image; the normalization parameter is a parameter determined according to a gold standard in the first neural network.

In a second aspect, the present application further provides an image artifact correction apparatus, including:

the candidate image determining module is used for determining at least one candidate phase and a low-dose candidate image corresponding to the candidate phase according to the low-dose original images of the multiple phases; the low-dose original image comprises a region of interest;

the reconstruction module is used for reconstructing the low-dose candidate image and determining a standard-dose candidate image corresponding to the low-dose candidate image;

and the artifact correction module is used for performing artifact correction processing according to the standard dose candidate image and a preset first neural network and determining an artifact correction image of the region of interest.

In a third aspect, the present application further provides a computer device, where the computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

In a fourth aspect, the present application further provides a computer readable storage medium, on which a computer program is stored, the computer program when executed by a processor implementing the steps of:

In a fifth aspect, the present application also provides a computer program product, a computer program product comprising a computer program which, when executed by a processor, performs the steps of:

According to the image artifact correction method, the image artifact correction device, the image artifact correction equipment, the storage medium and the program product, the candidate facies and the corresponding low-dose candidate images are determined through the low-dose original images including the region of interest of the plurality of facies, the low-dose candidate images are subjected to reconstruction processing to obtain the corresponding standard-dose candidate images, artifact correction processing is carried out according to the standard-dose candidate images and the preset first neural network, and the artifact correction images of the region of interest are determined. In the method, because the candidate phase and the low-dose candidate image are selected through the low-dose original image, and the low-dose candidate image is reconstructed to the standard-dose candidate image, the artifact correction of the region of interest is realized through the standard-dose candidate image and the neural network, the high-dose original image is not needed while the artifact correction function is ensured, so that the object to be detected does not need to be repeatedly scanned for a long time to obtain the high-dose original image, the requirement can be met only by scanning the object to be detected for a short time, and the radiation damage to the object to be detected can be reduced in the artifact reconstruction process; in addition, because the artifact correction can be performed by adopting the neural network after the candidate images of the candidate phase are reconstructed to the standard dose, the speed of the process is higher than that of the artifact correction process in the prior art, and therefore the artifact correction efficiency can also be improved.

Drawings

FIG. 1 is a diagram illustrating an internal structure of a computer device according to an embodiment;

FIG. 2 is a flowchart illustrating an exemplary method for image artifact correction;

FIG. 3 is a flowchart illustrating a method for image artifact correction according to another embodiment;

FIG. 4 is a flowchart illustrating a method for image artifact correction according to another embodiment;

FIG. 5 is a flowchart illustrating an image artifact correction method according to another embodiment;

FIG. 6 is a flowchart illustrating an image artifact correction method according to another embodiment;

fig. 7 is a block diagram of an image artifact correction apparatus according to an embodiment.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

The image artifact correction method provided in this embodiment of the present application may be applied to a computer device, where the computer device may be a terminal, and an internal structure diagram of the computer device may be as shown in fig. 1. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The communication interface of the computer device is used for carrying out wired or wireless communication with an external terminal, and the wireless communication can be realized through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. The computer program is executed by a processor to implement an image artifact correction method. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, a key, a track ball or a touch pad arranged on the shell of the computer equipment, an external keyboard, a touch pad or a mouse and the like.

Those skilled in the art will appreciate that the architecture shown in fig. 1 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.

In one embodiment, as shown in fig. 2, an image artifact correction method is provided, which is exemplified by the method applied to the computer device in fig. 1, and the method may include the following steps:

s202, determining at least one candidate phase and a low-dose candidate image corresponding to the candidate phase according to low-dose original images of a plurality of phases; the low dose raw image includes a region of interest.

The low dose can be a dose lower than the standard dose, and an image obtained after the region of interest of the object to be detected is scanned by adopting a low-dose ray can be recorded as a low-dose original image; the region of interest here may be any tissue of any part of the object to be measured, for example, the coronary arteries of the heart of the chest, the kidneys of the abdomen, etc.

Here, the low-dose original image may be an original image with low dose and low resolution, and the low resolution may also be a resolution lower than a conventional resolution, for example, when the object to be measured is scanned with the standard dose of radiation, the obtained image is generally an image with the conventional resolution at the standard dose, for example, the conventional resolution is 512 × 512, where the object to be measured is scanned with the low dose of radiation, and the resolution of the obtained image is generally lower than the conventional resolution, for example, the low resolution is 128 × 128.

Specifically, when the object to be measured is scanned by using the low-dose rays, the same region of interest of the object to be measured may be scanned by using the same low-dose rays in different phases (or referred to as different phases or different moments) to obtain raw data of the region of interest of the object to be measured; the raw data can then be reconstructed to obtain low dose raw images of the region of interest at each phase. It should be noted that, in general, the resolution of the low-dose original images obtained here in each phase is low resolution and equal.

After the low-dose original images of the plurality of phase phases are obtained, one or more superior phase phases can be selected from the low-dose original images of each phase to be used as candidate phase phases, and low-dose candidate images corresponding to the candidate phases are obtained at the same time. As an alternative embodiment, the image resolution of the low-dose candidate image is greater than the image resolution of the low-dose original image of the corresponding phase; for example, the resolution of the low-dose candidate image is the normal resolution and the resolution of the low-dose original image is the low resolution. Further, when obtaining the low-dose candidate image corresponding to the candidate period, the low-dose original image corresponding to the candidate period may be subjected to post-processing such as reconstruction, so as to improve the low resolution of the low-dose original image and obtain the low-dose candidate image at the conventional resolution; or directly carrying out post-processing such as reconstruction on the raw data of the candidate phase to obtain a low-dose candidate image under the conventional resolution. In addition, since the artifact correction is generally performed based on the image with the conventional resolution, a low-dose original image with a lower resolution is initially used, so that the image size can be effectively reduced, and the speed of artifact correction processing is increased.

Further, as can be seen from the above description, by scanning with a low dose of radiation, the less radiation damage to the object to be measured is when scanning the region of interest of the object to be measured.

And S204, reconstructing the low-dose candidate image, and determining a standard-dose candidate image corresponding to the low-dose candidate image.

In this step, after obtaining the low-dose candidate image, since the low-dose candidate image is generally a low-dose image and the artifact correction is generally a standard-dose image, the low-dose candidate image can be reconstructed into an image at a standard dose by reconstructing the low-dose candidate image, and the reconstructed image is referred to as a standard-dose candidate image.

When the low-dose candidate image is reconstructed, the low-dose candidate image may be reconstructed by using a neural network, a reconstruction algorithm, or a mathematical processing method, and the like, and in short, the low-dose candidate image may be reconstructed into a standard-dose candidate image.

It should be noted that, when reconstructing the low-dose candidate image to obtain the standard-dose candidate image, the resolution of the low-dose candidate image is not changed, that is, the obtained standard-dose candidate image and the corresponding low-dose candidate image have the same respective resolutions, which are both the conventional resolutions.

The low-dose candidate image is reconstructed into the standard-dose candidate image in the step for subsequent artifact correction, so that artifact correction can be smoothly executed, and the object to be detected does not need to be scanned by a higher-dose ray, thereby further reducing radiation damage to the object to be detected.

And S206, performing artifact correction processing according to the standard dose candidate image and a preset first neural network, and determining an artifact correction image of the region of interest.

The first neural network may be, for example, a full convolution network (e.g., V-NET, U-NET, etc.) or may be a network generating a countermeasure network (e.g., pix2pix, WGAN, etc.). In addition, in order to reconstruct image details as much as possible for better artifact correction, the first neural network may combine features of each level (representing Hierarchical features of the neural network) using skip-connection, residual block, dense block, and the like.

The first neural network is mainly used for artifact correction processing, the first neural network can be trained in advance before use, during training, a sample image with standard dosage or a sample image with standard dosage and a region-of-interest segmentation image corresponding to the sample image can be used as input of the initial first neural network, a standard artifact correction image of a sample region-of-interest is used as reference output of the initial first neural network, and the initial first neural network is trained to obtain the trained first neural network.

After the standard dose candidate image and the trained first neural network are obtained, the standard dose candidate image is a conventional resolution image under the standard dose, and includes an interested region, and then the trained first neural network is adopted to perform artifact correction processing on the interested region in the standard dose candidate image, so as to obtain an image after artifact correction, and the image is marked as an artifact correction image of the interested region.

It should be noted that the artifact-corrected image obtained here is an artifact-corrected image of a conventional resolution at a standard dose, and reaches a dose and a resolution equivalent to that of the artifact removal of the conventional image.

In the image artifact correction method, candidate phases and corresponding low-dose candidate images are determined through low-dose original images of a plurality of phases and including an interested region, the low-dose candidate images are subjected to reconstruction processing to obtain corresponding standard-dose candidate images, artifact correction processing is carried out according to the standard-dose candidate images and a preset first neural network, and an artifact correction image of the interested region is determined. In the method, because the candidate phase and the low-dose candidate image are selected through the low-dose original image, and the low-dose candidate image is reconstructed to the standard-dose candidate image, the artifact correction of the region of interest is realized through the standard-dose candidate image and the neural network, the high-dose original image is not needed while the artifact correction function is ensured, so that the object to be detected does not need to be repeatedly scanned for a long time to obtain the high-dose original image, the requirement can be met only by scanning the object to be detected for a short time, and the radiation damage to the object to be detected can be reduced in the artifact reconstruction process; in addition, because the artifact correction can be performed by adopting the neural network after the candidate images of the candidate phase are reconstructed to the standard dose, the speed of the process is higher than that of the artifact correction process in the prior art, and therefore the artifact correction efficiency can also be improved.

While the artifact correction processing can be performed by the standard dose candidate image and the neural network in the above embodiment, the following embodiment describes how to perform the artifact correction process.

In another embodiment, another image artifact correction method is provided, and on the basis of the foregoing embodiment, as shown in fig. 3, the foregoing S206 may include the following steps:

s302, performing region-of-interest segmentation processing on the standard dose candidate image to obtain a first target segmentation image of the region of interest.

In this step, a segmentation network may be used to segment the region of interest; the segmentation network may be trained in advance, for example, the segmentation network may be obtained by first obtaining sample images under a plurality of standard doses and a gold standard mask image or a gold standard segmentation probability map of a region of interest in each sample image, and then training the initial segmentation network through each sample image and the gold standard mask image or the gold standard segmentation probability map corresponding to each sample image. The split network may be a full convolution network full volumetric network or the like. In addition, when training the segmentation net, a random image block patch may be taken from the input sample image with the point on the gold standard mask image as the center, as the input of the initial segmentation net, for example, the size of the patch may be [64,128 ], while the input of the segmentation net may be two channels, and then the input may be a sequence connecting two images along the channel dimension, that is, the input of the segmentation net may be represented as [2,64,128 ]. Meanwhile, in order to improve the training speed of the segmentation network and the generalization capability of the segmentation network, an Adam optimizer can be selected as the optimizer of the segmentation network, and the Adam optimizer can quickly converge and has good generalization capability.

The Loss of the segmented network in the training process is composed of two parts, namely BCE (Binary Cross Entropy) Loss and Dice Loss, wherein the BCE Loss (BCE Loss) can be calculated by referring to the following formula (1):

wherein y represents a label of the specimen image (1 or 0,1 represents a region of interest, 0 represents a region of non-interest); y represents the predicted probability map.

Wherein the Dice Loss (Dice Loss) can be calculated by referring to the following equation (2):

wherein,

the probability map of the prediction is shown, and Y represents the gold standard mask image.

In addition, the sample images at the plurality of standard doses may be standard dose low-resolution images, and the images at the standard dose low-resolution may be obtained by obtaining the images at the low dose low-resolution first and reconstructing the images at the low dose low-resolution, and may be used as the sample images at the standard dose here.

After the trained segmentation network is obtained, the trained segmentation network can be used to perform region-of-interest segmentation processing on the standard dose candidate image, so as to obtain a segmentation image corresponding to the region-of-interest, and the segmentation image is recorded as a first target segmentation image. As an alternative embodiment, the first target segmentation image is a mask image of a region of interest and/or a segmentation probability map of the region of interest; the mask image is a binary image, the segmented interesting region is a foreground, and the other parts are backgrounds; the segmentation probability map corresponds to the mask image and includes probabilities that the respective points are regions of interest.

S304, after the standard dose candidate image and the first target segmentation image are in corresponding relation, inputting the standard dose candidate image and the first target segmentation image into a first neural network for artifact correction processing, and obtaining an artifact correction image of the region of interest.

In this step, after the first target segmented image corresponding to each standard dose candidate image is obtained, a corresponding relationship may be established between each standard dose candidate image and the corresponding first target segmented image thereof, and the standard dose candidate images and the corresponding first target segmented images are used as the same group of images and input to the first neural network together to correct the artifact of the region of interest in the standard dose candidate images, so that more features may be provided for the artifact correction process through the segmented images of the region of interest, and meanwhile, the artifact correction process may be concentrated on the region of interest portion, thereby reducing the useless artifact correction process, increasing the speed of the artifact correction process, and finally obtaining the artifact corrected image of the region of interest.

In the embodiment, after the region of interest of the standard dose candidate image is segmented, the segmented image and the standard dose candidate image are input into the first neural network for artifact correction processing after the corresponding relation is established, so that more characteristics can be provided for an artifact correction process, and the accuracy of artifact correction is improved; meanwhile, the artifact correction process can be concentrated in the region of interest, so that useless artifact correction process is reduced, and the speed of artifact correction processing is increased.

In the above embodiments, the low-dose candidate image may be reconstructed, and the following embodiments describe a process of reconstructing the low-dose candidate image by using a neural network.

In another embodiment, another image artifact correction method is provided, and on the basis of the foregoing embodiment, the foregoing S204 may include the following steps:

and step A, inputting the low-dose candidate image into a preset second neural network for reconstruction processing, and determining a standard-dose candidate image.

The second neural network is obtained by training according to a plurality of low-dose sample images and standard-dose sample images corresponding to the low-dose sample images, wherein each standard-dose sample image is used as a gold standard of the corresponding low-dose sample image. The low dose sample image may be a normal resolution sample image at low dose, and the standard dose sample image may be a normal resolution sample image at standard dose.

The second neural network may be similar to the first neural network, and may be, for example, a full capacitive network, or a general adaptive network.

Specifically, after the candidate image with the low dose and the normal resolution of the candidate phase is obtained, the candidate image with the low dose and the normal resolution may be input to a second neural network for reconstruction processing, so as to increase the dose of the input image, and obtain a candidate image with the normal resolution of the standard dose, which is marked as a standard dose candidate image.

In the embodiment, the low-dose candidate image is input into the second neural network for reconstruction processing to obtain the standard-dose candidate image, so that the dose of the image is reconstructed through the neural network, the speed of dose reconstruction can be increased while the low-dose image is converted into the standard-dose image, and the efficiency of whole artifact correction is improved.

While the above embodiments mention the content of candidate facies and low-dose candidate images that can be determined from low-dose original images of multiple facies, the following embodiments describe a process of determining candidate facies and low-dose candidate images from low-dose original images of multiple facies.

In another embodiment, another image artifact correction method is provided, and on the basis of the foregoing embodiment, as shown in fig. 4, the foregoing S204 may include the following steps:

s402, carrying out reconstruction processing on the low-dose original images of the multiple phase phases, and determining a standard dose reconstruction image of each phase.

In this step, low-dose original images of multiple phases can be obtained in the manner of the above step S202, where the low-dose original image is an original image with low resolution at low dose, and the low-resolution original image at low dose can be reconstructed by a neural network to increase the dose thereof, so as to obtain an image with low resolution at standard dose, which is recorded as a standard dose reconstructed image; by operating each phase in this way, a standard dose reconstruction image of each phase can be obtained.

The neural network may be referred to as a third neural network, which is similar to the first neural network or the second neural network, and may be, for example, full capacitive Networks, or generic adaptive Networks. The third neural network is similar to the second neural network during training, and may also be obtained by training a plurality of sample images with low resolution at low dose and sample images with low resolution at standard dose corresponding to the sample images with low resolution at low dose, where each sample image with low resolution at standard dose is used as a gold standard of the corresponding sample image with low resolution at low dose.

In order to more quickly and effectively process the image, the low-dose raw image of each phase may be normalized after the low-dose raw image of each phase is obtained, and then the low-dose raw image after the normalization may be used for the next processing. Assume that the low dose raw image obtained here is denoted as I _ll The normalization process can be seen in the following equation (3):

wherein, I' is a low-dose original image after standardization processing; μ and σ are the mean and standard deviation of the low-resolution sample image at each low dose in the third neural network training process, respectively, and the mean and standard deviation corresponding to the low-resolution sample image at each low dose can be counted as μ and σ here.

Further, after the low-dose original image of each phase is normalized by the formula (3), the normalized low-dose original image of each phase can be obtained, and each low-dose original image is input into the third neural network to be reconstructed, so as to obtain each standard dose image, and the standard dose reconstructed image of each phase is obtained by performing a process opposite to the normalization process on each standard dose image. In addition, the process of performing the inverse of the normalization process here can be performed using the following formula (4):

I＝I _DL *σ _gt +μ _gt (4)

wherein I is a standard dose reconstruction image; i is _DL For each standard dose image output in the third neural network, i.e. the image before the process opposite to the normalization process; mu.s _gt And σ _gt The mean and the standard deviation of the low-resolution sample images under each standard dose in the third neural network training process can be respectively used as the μ _gt And σ _gt 。

S404, performing region-of-interest segmentation processing on each standard dose reconstruction image to obtain a second target segmentation image corresponding to each standard dose reconstruction image.

In this step, similar to the region of interest segmentation process in S302, a similar segmentation network may be used to segment the region of interest, and the training process of the segmentation network may refer to the training process in S302, which is not described herein again.

After the trained segmentation network is obtained, the trained segmentation network can be used to perform region-of-interest segmentation processing on the low-resolution reconstructed image under the standard dose, so as to obtain a segmented image corresponding to the region-of-interest, and the segmented image is recorded as a second target segmented image. As an alternative embodiment, the second target segmentation image is a mask image of the region of interest and/or a segmentation probability map of the region of interest.

By operating all the phase phases in this way, the second target segmentation image corresponding to the standard dose reconstruction image in each phase can be obtained. Each of the second target segmented images is a standard-dose low-resolution segmented image.

And S406, determining at least one candidate phase from the multiple phase phases according to each second target segmentation image and determining a low-dose candidate image corresponding to the candidate phase.

After obtaining the second target segmented image of each phase, a candidate phase can be selected from each phase and a low-dose candidate image of the candidate phase can be obtained accordingly, as an alternative embodiment, as shown in fig. 5, this step can be performed by the following steps:

and S502, performing quality quantization processing on each second target segmentation image according to a preset quality quantization mode to obtain a quality quantization value of each second target segmentation image.

The preset quality quantization mode may be, for example, a quantization mode set for different quality assessment indexes, and different magnitudes of the quality assessment indexes correspond to different quality quantization values. The quality evaluation index here may be determined according to the actual morphological structure of the region of interest, for example, the region of interest is a coronary artery of the heart, and the quality evaluation index here may be a parameter such as circularity or sharpness of the coronary artery in the target segmented image.

The quality quantization processing is carried out on each second target segmentation image through the same or a plurality of quality evaluation indexes, so that the size of the quality evaluation index corresponding to each second target segmentation image can be obtained, and further the corresponding quality quantization value can be obtained.

S504, at least one candidate phase is determined from the multiple phases according to the quality quantization values.

After the quality quantized values corresponding to the periods are obtained, the quality quantized values can be ranked, one or more candidate quality quantized values with larger quality quantized values in the ranking result are obtained, and further, the period phase corresponding to each candidate quality quantized value is obtained and marked as a candidate period phase.

The quality quantization value is larger, the quality of the corresponding segmentation image is better or optimal; in this way, candidate facies with better segmentation image quality can be obtained. The candidate phase selected here may be two adjacent phases before and after the phase with the best quality, for a total of 5 phases.

S506, acquiring raw data corresponding to the candidate period, and carrying out image reconstruction on the raw data to obtain a low-dose candidate image; the raw data is low dose raw data.

After the candidate phase is obtained, image reconstruction processing may be performed on the previously obtained raw data of the region of interest of the object to be measured, so as to obtain a candidate image with a conventional resolution under a low dose, and the candidate image is marked as a low dose candidate image.

In the embodiment, after the low-dose original image of each phase is reconstructed to obtain the standard-dose reconstructed image, the interested segmentation processing is performed to obtain each segmented image, the candidate phase is selected from the multiple phases through each segmented image, and the low-dose candidate image of the candidate phase is determined, so that the phase can be selected according to the characteristics of the region of interest by segmenting the image, the obtained candidate phase and the candidate image can be more accurate, and the subsequent artifact correction is more accurate. Furthermore, the candidate phase is selected through the quantized value of each segmented image after quality quantization processing, and the method is simple and intuitive, so that the accuracy of the selected candidate phase can be further improved.

In the above embodiments, it is mentioned that the artifact correction processing may be performed through the standard dose candidate image and the neural network, and the image may be preprocessed in the actual processing process, and the following embodiments describe the process of preprocessing simultaneously in the artifact correction processing.

In another embodiment, another image artifact correction method is provided, and on the basis of the foregoing embodiment, as shown in fig. 6, the foregoing S206 may include the following steps:

s602, artifact correction processing is carried out according to the standard dose candidate image and a preset first neural network, and an initial artifact correction image of the region of interest is determined.

The standard dose candidate image is an image with a conventional resolution under a standard dose, and before artifact correction processing, the standard dose candidate image may be normalized, and artifact correction processing may be performed through the normalized image and the first neural network.

Taking the above obtained candidate phase as 5 as an example, the standard dose candidate images obtained here are 5 images, which are respectively denoted as Y1 to Y5, the maximum value and the minimum value (for example, the pixel maximum value or the pixel minimum value) of each of Y1 to Y5 can be respectively calculated, and then each standard dose candidate image is normalized by using the following formula (5), as follows:

wherein Y' is a standard dose candidate image after normalization processing; y is a standard dose candidate image; max and min are respectively the maximum value and the minimum value obtained by calculating the respective sequences of the candidate image training set; and (4) normalizing each standard dose candidate image through a formula (5) to obtain the corresponding normalized standard dose candidate images.

After obtaining each normalized standard dose candidate image, the normalized standard dose candidate images may be input to the first neural network to perform artifact correction processing on the region of interest, either individually or after being connected with the respective corresponding first target segmentation images through the channel dimension, so as to obtain an initial artifact correction image of the region of interest.

S604, performing inverse normalization processing on the initial artifact correction image according to preset normalization parameters to determine an artifact correction image; the normalization parameter is a parameter determined according to a gold standard in the first neural network.

In this step, the normalized parameter is a parameter obtained by a gold standard of the first neural network, that is, a maximum value and a minimum value on the standard artifact correction image of the region of interest during the training process of the first neural network, and after the normalized parameter is obtained, the initial artifact correction image may be subjected to inverse normalization processing, which may specifically be performed by using the following formula (6):

P＝P ₀ *(max'-min')+min' (6)

wherein, P is an artifact correction image obtained after inverse normalization processing; p ₀ Correcting the image for the initial artifact; max 'and min' are respectively the maximum value and the minimum value obtained by calculation in the training set of the standard artifact correction image of the region of interest; the initial artifact correction image can be subjected to inverse normalization processing through the formula (6), and a final artifact correction image of the region of interest is obtained.

In order to explain the present solution more clearly, the training process of the first neural network is described in this embodiment. In the first neural network training process, a plurality of sample images with standard dosage and the corresponding region-of-interest segmentation images thereof can be obtained, and a gold standard, namely a standard artifact correction image of the region-of-interest can be obtained; then, normalizing the corresponding images by respectively adopting the maximum value and the minimum value on each image, wherein the normalization can be specifically carried out by adopting the formula (5) to obtain normalized images; and establishing a corresponding relation between each normalized sample image with the standard dosage and the corresponding region-of-interest segmentation image, connecting the normalized sample images through channel dimensions, using the normalized gold standard as the input of the initial first neural network, and training the initial first neural network by using the normalized gold standard as the reference output of the initial first neural network.

In addition, in order to ensure the accuracy of the generated artifact-corrected image of the region of interest, in the present embodiment, 2.5D processing is performed on the input data, so that network learning is always performed in the image block (not the entire voxel volume). Each image after normalization is separately cropped randomly, the size of the crop can be any fixed size that meets the network requirements and image dimensions, for example: [64, 5], it is necessary to ensure a one-to-one correspondence in the positions of the image crop. Data were processed in 2.5D: illustratively, the 2.5D processing refers to 5 slices of a 2D slice image of an image corresponding to 1 slice of a 2D slice image of the golden standard, so the input to the network can be represented as [64, 25] and the corresponding golden standard can be represented as [64, 1].

For the optimizer of the initial first neural network, an Adam optimizer is adopted, mainly because the Adam optimizer can rapidly converge and has good generalization capability. Regarding the selection of the learning rate of the optimizer, first, the best learning rate is selected using the technique of LR Range Test (learning rate Range Test). This technique can be described as: firstly, setting the learning rate to 1 very small value, then simply iterating the network and the data for several times, increasing the learning rate after each iteration is finished, and recording the loss of each training; an LR Range Test (LR Range Test) graph is then plotted, with a generally ideal LR Range Test graph containing three regions: the loss is basically unchanged when the learning rate of the first region is too small, the loss of the second region is reduced and the convergence is fast, and the learning rate of the last region is too large so that the loss begins to diverge; the optimal learning rate can be determined as the learning rate corresponding to the lowest point in the LR Range Test graph, and the optimal learning rate is used as the initial learning rate of the Adam optimizer.

In order to ensure the quality and quantitative accuracy of the finally generated artifact-corrected image of the region of interest, for the loss of the initial first neural network during the training process, the loss function used in this embodiment is composed of two parts, L1 loss and Regional loss, where the expression of L1 loss is shown in the following formula (7):

loss(x _i ,y _i )＝x _i -y _i (7)

wherein x is an output image of the initial first neural network; y is an image corresponding to the gold standard; i is the position index of the pixel; the L1 loss calculates the intensity difference value of each pixel between the network predicted image and the target image, so that the quantitative error between the final output image of the network and the gold standard image is ensured to be as small as possible, and the quantitative accuracy of artifact correction is ensured.

The Regional loss is mainly to calculate the loss between the interested region on the output image of the initial first neural network and the interested region on the gold standard, ensure that the error of the quality of the final output image of the network and the gold standard image at the interested region is as small as possible, and ensure the quality of artifact correction.

In this embodiment, after the initial artifact corrected image of the region of interest is determined by the standard dose candidate image and the first neural network, the final artifact corrected image of the region of interest can be obtained by performing inverse normalization processing on the initial artifact corrected image by using the normalization parameter determined by the gold standard in the first neural network, so that the accuracy of obtaining the artifact corrected image can be higher by the inverse normalization processing.

In the following, a detailed embodiment is given to illustrate the scheme of the embodiment of the present application, and on the basis of the embodiment, the method of the present application may include the following steps:

s1, inputting low-dose low-resolution images of a plurality of phases into a first network for reconstruction processing, and determining standard-dose low-resolution reconstructed images of the phases;

s2, inputting each standard dose low-resolution reconstructed image into a second network for region-of-interest segmentation processing to obtain a target segmentation image corresponding to each standard dose reconstructed image;

s3, performing quality quantization processing on each target segmentation image to obtain a quality quantization value of each target segmentation image;

s4, determining an optimal phase and a plurality of phases around the optimal phase from the plurality of phases according to each quality quantization value to obtain a plurality of candidate phases;

s5, acquiring low-dose raw data corresponding to each candidate period, and performing image reconstruction on the raw data to obtain each low-dose conventional resolution candidate image;

s6, inputting each low-dose conventional resolution candidate image into a third network for reconstruction processing, and determining each standard-dose conventional resolution candidate image;

s7, inputting the standard dose conventional resolution candidate images into a fourth network for region-of-interest segmentation processing to obtain segmentation images of the regions of interest; the segmentation image is a mask image of the region of interest and/or a segmentation probability map of the region of interest;

s8, after establishing a corresponding relation between each standard dose conventional resolution candidate image and each segmented image, inputting the standard dose conventional resolution candidate image into a fifth network for artifact correction processing to obtain an initial artifact correction image of the region of interest;

and S9, carrying out inverse normalization processing on the initial artifact correction, and determining an artifact correction image.

It should be noted that, the first network-the fifth network herein do not correspond to the first neural network and the second neural network mentioned in fig. 2 to 6, and may be determined by respective functions.

It should be understood that, although the steps in the flowcharts related to the embodiments as described above are sequentially displayed as indicated by arrows, the steps are not necessarily performed sequentially as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least a part of the steps in the flowcharts related to the embodiments described above may include multiple steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, and the execution order of the steps or stages is not necessarily sequential, but may be rotated or alternated with other steps or at least a part of the steps or stages in other steps.

Based on the same inventive concept, the embodiment of the present application further provides an image artifact correction apparatus for implementing the image artifact correction method. The implementation scheme for solving the problem provided by the apparatus is similar to the implementation scheme described in the above method, so specific limitations in one or more embodiments of the image artifact correction apparatus provided below may refer to the limitations on the image artifact correction method in the foregoing, and details are not repeated here.

In one embodiment, as shown in fig. 7, there is provided an image artifact correction apparatus including: a candidate image determination module, a reconstruction module, and an artifact correction module, wherein:

and the artifact correction module is used for performing artifact correction processing according to the standard dose candidate image and a preset first neural network to determine an artifact correction image of the region of interest.

Optionally, the image resolution of the low-dose candidate image is greater than the image resolution of the low-dose original image of the corresponding phase.

In another embodiment, another image artifact correction apparatus is provided, and on the basis of the above embodiment, the artifact correction module may include:

the first segmentation unit is used for carrying out region-of-interest segmentation processing on the standard dose candidate image to obtain a first target segmentation image of the region of interest;

and the corresponding and correcting unit is used for establishing a corresponding relation between the standard dose candidate image and the first target segmentation image, inputting the standard dose candidate image and the first target segmentation image into the first neural network for artifact correction processing, and obtaining an artifact correction image of the region of interest.

Optionally, the first target segmentation image is a mask image of the region of interest and/or a segmentation probability map of the region of interest.

In another embodiment, another image artifact correction apparatus is provided, and on the basis of the above embodiment, the reconstruction module may include:

the first reconstruction unit is used for inputting the low-dose candidate image into a preset second neural network for reconstruction processing, and determining a standard-dose candidate image; the second neural network is obtained by training according to a plurality of low-dose sample images and standard-dose sample images corresponding to the low-dose sample images.

In another embodiment, another image artifact correction apparatus is provided, and on the basis of the foregoing embodiment, the candidate image determination module may include:

the second reconstruction unit is used for reconstructing low-dose original images of a plurality of phase phases and determining standard dose reconstructed images of each phase;

the second segmentation unit is used for carrying out region-of-interest segmentation processing on each standard dose reconstruction image to obtain a second target segmentation image corresponding to each standard dose reconstruction image;

and the candidate image determining unit is used for determining at least one candidate phase from the plurality of phase phases according to each second target segmentation image and determining a low-dose candidate image corresponding to the candidate phase.

Optionally, the candidate image determining unit may include:

the quantization subunit is configured to perform quality quantization processing on each second target segmented image according to a preset quality quantization mode to obtain a quality quantization value of each second target segmented image;

a candidate phase determining subunit operable to determine at least one candidate phase from the plurality of phases based on the respective mass quantization values;

the candidate image determining subunit is used for acquiring raw data corresponding to the candidate period and carrying out image reconstruction on the raw data to obtain a low-dose candidate image; the raw data is low dose raw data.

In another embodiment, another image artifact correction apparatus is provided, and on the basis of the above embodiment, the artifact correction module may further include:

the initial correction unit is used for performing artifact correction processing according to the standard dose candidate image and a preset first neural network and determining an initial artifact correction image of the region of interest;

the normalization processing unit is used for carrying out inverse normalization processing on the initial artifact correction image according to preset normalization parameters to determine an artifact correction image; the normalization parameter is a parameter determined according to a gold standard in the first neural network.

The modules in the image artifact correction device can be wholly or partially implemented by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.

In one embodiment, a computer device is provided, comprising a memory and a processor, the memory having a computer program stored therein, the processor implementing the following steps when executing the computer program:

determining at least one candidate phase and a low-dose candidate image corresponding to the candidate phase according to low-dose original images of a plurality of phases; the low-dose original image comprises a region of interest; reconstructing the low-dose candidate image, and determining a standard-dose candidate image corresponding to the low-dose candidate image; and performing artifact correction processing according to the standard dose candidate image and a preset first neural network, and determining an artifact correction image of the region of interest.

In one embodiment, the processor, when executing the computer program, further performs the steps of:

performing region-of-interest segmentation processing on the standard dose candidate image to obtain a first target segmentation image of the region of interest; and after establishing a corresponding relation between the standard dose candidate image and the first target segmentation image, inputting the standard dose candidate image and the first target segmentation image into a first neural network for artifact correction processing to obtain an artifact correction image of the region of interest.

inputting the low-dose candidate image into a preset second neural network for reconstruction processing, and determining a standard-dose candidate image; the second neural network is obtained by training according to a plurality of low-dose sample images and standard-dose sample images corresponding to the low-dose sample images.

reconstructing the low-dose original images of a plurality of phase phases, and determining a standard dose reconstructed image of each phase; performing region-of-interest segmentation processing on each standard dose reconstruction image to obtain a second target segmentation image corresponding to each standard dose reconstruction image; at least one candidate phase is determined from the plurality of phase phases based on each second target segmentation image and low dose candidate images corresponding to the candidate phase are determined.

performing quality quantization processing on each second target segmentation image according to a preset quality quantization mode to obtain a quality quantization value of each second target segmentation image; determining at least one candidate phase from the plurality of phases based on the respective mass quantization values; acquiring raw data corresponding to the candidate period, and carrying out image reconstruction on the raw data to obtain a low-dose candidate image; the raw data is low dose raw data.

In one embodiment, the first target segmentation image is a mask image of the region of interest and/or a segmentation probability map of the region of interest.

In one embodiment, the processor when executing the computer program further performs the steps of:

performing artifact correction processing according to the standard dose candidate image and a preset first neural network, and determining an initial artifact correction image of the region of interest; carrying out inverse normalization processing on the initial artifact correction image according to preset normalization parameters to determine an artifact correction image; the normalization parameter is a parameter determined according to a gold standard in the first neural network.

In one embodiment, a computer-readable storage medium is provided, having a computer program stored thereon, which when executed by a processor, performs the steps of:

determining at least one candidate phase and a low-dose candidate image corresponding to the candidate phase according to low-dose original images of a plurality of phases; the low-dose original image comprises a region of interest; reconstructing the low-dose candidate image to determine a standard-dose candidate image corresponding to the low-dose candidate image; and performing artifact correction processing according to the standard dose candidate image and a preset first neural network, and determining an artifact correction image of the region of interest.

In one embodiment, the computer program when executed by the processor further performs the steps of:

reconstructing the low-dose original images of the multiple phase phases, and determining a standard dose reconstructed image of each phase; performing region-of-interest segmentation processing on each standard dose reconstruction image to obtain a second target segmentation image corresponding to each standard dose reconstruction image; at least one candidate phase is determined from the plurality of phase phases based on each second target segmentation image and low dose candidate images corresponding to the candidate phase are determined.

In one embodiment, a computer program product is provided, comprising a computer program which, when executed by a processor, performs the steps of:

determining at least one candidate phase and a low-dose candidate image corresponding to the candidate phase according to low-dose original images of a plurality of phases; the low-dose original image comprises a region of interest; reconstructing the low-dose candidate image to determine a standard-dose candidate image corresponding to the low-dose candidate image; and carrying out artifact correction processing according to the standard dose candidate image and a preset first neural network, and determining an artifact correction image of the region of interest.

It should be noted that the data referred to in the present application (including but not limited to data for analysis, stored data, presented data, etc.) are data that are fully authorized by the respective parties.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, database, or other medium used in the embodiments provided herein may include at least one of non-volatile and volatile memory. The nonvolatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical Memory, high-density embedded nonvolatile Memory, resistive Random Access Memory (ReRAM), magnetic Random Access Memory (MRAM), ferroelectric Random Access Memory (FRAM), phase Change Memory (PCM), graphene Memory, and the like. Volatile Memory can include Random Access Memory (RAM), external cache Memory, and the like. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM), among others. The databases referred to in various embodiments provided herein may include at least one of relational and non-relational databases. The non-relational database may include, but is not limited to, a block chain based distributed database, and the like. The processors referred to in the embodiments provided herein may be general purpose processors, central processing units, graphics processors, digital signal processors, programmable logic devices, quantum computing based data processing logic devices, etc., without limitation.

The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.

The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the present application. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A method of image artifact correction, the method comprising:

determining at least one candidate facies and low-dose candidate images corresponding to the candidate facies according to low-dose original images of a plurality of facies; the low-dose original image comprises a region of interest;

reconstructing the low-dose candidate image, and determining a standard dose candidate image corresponding to the low-dose candidate image;

2. The method of claim 1, wherein the determining an artifact-corrected image of the region of interest by performing an artifact correction process according to the standard dose candidate image and a preset first neural network comprises:

and after establishing a corresponding relation between the standard dose candidate image and the first target segmentation image, inputting the standard dose candidate image and the first target segmentation image into the first neural network for artifact correction processing to obtain an artifact correction image of the region of interest.

3. The method according to claim 1 or 2, wherein the reconstructing the low-dose candidate image and determining a standard-dose candidate image corresponding to the low-dose candidate image comprises:

inputting the low-dose candidate image into a preset second neural network for reconstruction processing, and determining the standard-dose candidate image;

4. The method according to claim 1 or 2, wherein determining at least one candidate facies and low-dose candidate images corresponding to the candidate facies from low-dose raw images of a plurality of facies comprises:

reconstructing the low-dose original images of the multiple phase phases, and determining a standard dose reconstructed image of each phase;

determining at least one candidate phase from the plurality of phase phases according to each of the second target segmentation images and determining a low dose candidate image corresponding to the candidate phase.

5. The method of claim 4, wherein determining at least one candidate phase from the plurality of phases and determining a low-dose candidate image corresponding to the candidate phase from each of the second target segmented images comprises:

determining at least one candidate phase from the plurality of phases based on each of the quality quantization values;

acquiring raw data corresponding to the candidate period, and performing image reconstruction on the raw data to acquire the low-dose candidate image; the raw data is low dose raw data.

6. The method according to claim 1 or 2, wherein the determining an artifact-corrected image of the region of interest by performing an artifact correction process according to the standard dose candidate image and a preset first neural network comprises:

carrying out inverse normalization processing on the initial artifact correction image according to preset normalization parameters to determine the artifact correction image; the normalization parameter is a parameter determined according to a gold standard in the first neural network.

7. An image artifact correction apparatus, characterized in that the apparatus comprises:

the candidate image determining module is used for determining at least one candidate facies and low-dose candidate images corresponding to the candidate facies according to the low-dose original images of the plurality of facies; the low-dose original image comprises a region of interest;

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor realizes the steps of the method of any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 6.