KR102492950B1 - Apparatus and method for processing artifacts in medical images - Google Patents

Abstract

According to the present invention, an apparatus for processing artifacts in medical images, which removes a beam hardening artifact from the medical image, comprises: a first artifact removal module for receiving the medical image to convert the medical image into a sinogram form and correct artifacts from the sinogram; and a second artifact removal module for correcting residual artifacts from an image-dimensional medical image from which the artifacts are removed to generate a high-quality readout image, wherein the first and second artifact removal modules may have different correction coefficients. According to the present invention, the apparatus can quickly obtain a high-quality readout image.

Description

Artifact processing device and processing method of medical images {APPARATUS AND METHOD FOR PROCESSING ARTIFACTS IN MEDICAL IMAGES}

The present invention relates to a medical image artifact processing apparatus and processing method, and more particularly, to a medical image artifact processing apparatus and processing method for removing artifacts from a medical image.

In general, medical equipment such as X-ray, CT, and MRI is used to acquire medical images. Medical images acquired from these medical devices are used as very important grounds for decision-making by discriminating the presence or absence of lesions and their characteristics in the course of diagnosis and treatment of patients in modern medicine.

Conventional medical image processing technology has already been disclosed by "Korean Patent Publication No. 2014-0134903 (Medical image quality improvement method and apparatus, 2014.11.25.)". The disclosed invention is a technique for improving noise in medical images.

However, various causes of noise generation in medical images are known. For example, a beam hardening artifact occurs between bones when X-rays pass through the bones, and some energy is absorbed or transmitted through the bones. Accordingly, efforts are being made to remove the beam hardening artifact included in the medical image. However, due to the problem that distortion occurs in medical images in the removal of beam hardening artifacts, research on effectively removing artifacts continues, but no specific technology has been presented yet.

Republic of Korea Patent Publication No. 2014-0134903 (Medical image quality improvement method and apparatus, 2014.11.25.)

An object of the present invention is to provide an apparatus and method for processing artifacts of medical images to remove beam hardening artifacts included in medical images.

An artifact processing apparatus for a medical image according to the present invention is a medical image artifact processing apparatus for removing a beam hardening artifact from a medical image, wherein the medical image is received and the medical image is converted into a sinogram (Sinogram). ) dimension and a first artifact removal module for correcting artifacts from the sinogram and a second artifact removal module for generating a high-quality reading image by correcting residual artifacts from the medical image of the image dimension from which the artifacts are removed. and the correction coefficients of the first and second artifact removal modules may be different from each other.

The first artifact removal module removes the artifact from the sinogram dimension with a higher correction factor compared to the second artifact removal module, and the second artifact removal module removes the artifact from the sinogram dimension with a lower correction compared to the first artifact removal module. A coefficient can remove the residual artifact from the image dimension.

The second artifact removal module includes a deep learning model for removing residual artifacts of the medical image in the image dimension, and the deep learning model is to be learned based on the medical image for which artifact removal has been completed in the first artifact removal module. can

The deep learning model is learned by pairing a medical image from which artifact removal has been completed in the first artifact removal module and a medical image that does not include the residual artifact, and in the learning of the deep learning model, the residual artifact is not included. Based on the medical images, repeated training may be performed to minimize a difference between a medical image for which artifact removal has been completed in the first artifact removal module and a medical image without residual artifacts.

The first artifact removal module may convert the medical image into a sinogram dimension based on medical information of the medical image, and remove the artifact from the sinogram dimension according to a pre-set rule.

The first artifact removal module may remove the artifact by filtering the synthesized sinogram of the medical image with a filter kernel according to a pre-set rule.

The first artifact removal module may remove the artifact by applying a 2D Fourier Transform to the synthesized sinogram of the medical image.

The first artifact removal module may remove the artifact by applying a 2D wavelet transform to the synthesized sinogram of the medical image.

The first artifact removal module may remove the artifact by applying the synthesized sinogram of the medical image to eigencomponent decomposition of a Hessian matrix.

Meanwhile, in the artifact processing method of a medical image according to the present invention, in the artifact processing method of a medical image for removing a beam hardening artifact from a medical image, the medical image is received and the medical image is converted into a sinogram. An artifact removal step of converting to (Sinogram) dimension and correcting artifacts from the sinogram, a conversion step of converting the artifact-removed sinogram into an image dimension, and correcting residual artifacts from the image dimension medical image A residual artifact removal step of generating a high-quality readout image may be included, and correction coefficients in the artifact removal step and the residual artifact removal step may be different from each other.

In the artifact removal step, the artifact is removed from the sinogram dimension with a correction factor higher than in the residual artifact removal step, and in the residual artifact removal step, the artifact is removed from the image dimension with a correction coefficient lower than in the artifact removal step. The remaining artifacts may be removed.

In the residual artifact removal step, a deep learning model for removing residual artifacts of the medical image in the image dimension is applied, and the deep learning model may be learned based on the medical image from which artifact removal is completed in the artifact removal step.

The deep learning model is trained by pairing a medical image from which artifact removal has been completed in the artifact removal step and a medical image that does not contain the residual artifact, and in learning the deep learning model, the medical image that does not include the residual artifact Based on , training may be repeated to minimize a difference between a medical image in which artifact removal is completed in the artifact removal step and a medical image in which the residual artifact is not included.

In the artifact removal step, the medical image may be converted into a synthesized sinogram dimension based on medical information of the medical image, and the artifact may be removed from the sinogram dimension according to a pre-set rule.

In the artifact removal step, the artifact may be removed by filtering the synthesized sinogram of the medical image with a filter kernel according to a preset rule.

In the artifact removal step, the artifact may be removed by applying a 2D Fourier Transform to the synthesized sinogram of the medical image.

In the artifact removal step, the artifact may be removed by applying a 2D wavelet transform to the synthesized sinogram of the medical image.

The artifact removal step may remove the artifact by applying the synthetic sinogram of the medical image to eigencomponent decomposition of a Hessian matrix.

The apparatus for processing artifacts of medical images according to the present invention has an effect of generating high-quality readout images without image distortion by removing artifacts in the sinogram dimension and the image dimension based on mutually different correction coefficients.

In addition, the apparatus for processing artifacts of medical images according to the present invention reliably removes residual artifacts by learning a deep learning model based on images from which artifact removal has been performed in the sinogram dimension, thereby producing high-quality read images in a short time. has the effect of obtaining it.

The technical effects of the present invention as described above are not limited to the effects mentioned above, and other technical effects not mentioned will be clearly understood by those skilled in the art from the description below.

1 is a conceptual diagram schematically illustrating an apparatus for processing artifacts of a medical image according to an exemplary embodiment;
2 is a flowchart showing a method for removing artifacts of a first artifact removal module according to this embodiment;
3 is a conceptual diagram showing a method for removing artifacts of a first artifact removal module according to this embodiment;
4 is a flowchart showing a training method of a second artifact removal module according to this embodiment;
5 is a conceptual diagram showing a training method of a second artifact removal module according to this embodiment;
6 is a flowchart illustrating a method for processing artifacts of a medical image according to an exemplary embodiment.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, this embodiment is not limited to the embodiments disclosed below and may be implemented in various forms, but this embodiment only makes the disclosure of the present invention complete, and the scope of the invention to those skilled in the art. It is provided for complete information. The shapes of elements in the drawings may be exaggeratedly expressed for more clear description, and elements indicated by the same reference numerals in the drawings mean the same elements.

1 is a conceptual diagram schematically illustrating an apparatus for processing artifacts of a medical image according to an exemplary embodiment.

As shown in FIG. 1 , the medical image artifact processing apparatus 1000 (hereinafter, referred to as a processing apparatus) according to the present embodiment may remove beam hardening artifacts. That is, the processing device 1000 removes artifacts included in the medical image 30 obtained from the medical equipment 10 to generate a high-quality readout image 50 .

Here, the medical equipment 10 may be a computed tomography (CT), a magnetic resonance imaging (MRI), a positron emission tomography (PET), or the like, and a medical image (30 ) may be a CT image. However, the types of medical equipment 10 and medical images 30 are not limited.

Meanwhile, the processing device 1000 may include a communication module 100 , a first artifact removal module 200 and a second artifact removal module 300 .

The communication module 100 receives the medical image 30 provided from the medical equipment 10 or a database (not shown).

Accordingly, the first artifact removal module 200 acquires the medical image 30 provided through the communication module 100 . The first artifact removal module 200 converts the medical image 30 into a synthesized sinogram dimension to perform artifact removal. Here, the first artifact removal module 200 may perform artifact removal using a pre-trained deep learning model, but the artifact removal method is not limited. Also, the first artifact removal module 200 may convert the artifact-removed sinogram dimension medical image 30 into an image dimension.

Accordingly, the second artifact removal module 300 receives the image-dimensional medical image 30 provided from the first artifact removal module 200 . And the second artifact removal module 300 performs residual artifacts on the medical image 30 in the image dimension. Here, the second artifact removal module 300 may perform residual artifact removal using a pre-trained deep learning model. The deep learning model may be trained based on medical images output from the first artifact removal module 200 .

Meanwhile, the first artifact removal module 200 and the second artifact removal module 300 may perform artifact removal and residual artifact removal with mutually different correction coefficients.

For example, the first artifact removal module 200 removes artifacts by applying a high correction coefficient in artifact removal performed in the synthesized sinogram dimension. The second artifact removal module 300 removes artifacts by applying a low correction coefficient in residual artifact removal performed at the image level.

For example, artifact removal can be performed using a weak correction factor in the artifact removal process performed at the synthetic sinogram level, and residual artifact removal can be performed using a high correction factor in the residual artifact removal process performed at the image level. there is. At this time, there is a problem in that artifact reduction efficiency is lowered at the synthesized sinogram level and distortion may occur in the image at the image level.

Accordingly, the first artifact removal module 200 according to the present embodiment may apply a high correction coefficient, and the second artifact removal module 300 may apply a low correction coefficient. That is, the first artifact removal module 200 strongly removes artifacts in the synthesized sinogram dimension, and the second artifact removal module 300 relatively weakly removes residual artifacts in the image dimension, effectively removing artifacts and residual artifacts while There is an advantage of being able to solve the distortion of the image.

Hereinafter, the first artifact removal module 200 and the second artifact removal module 300 will be described in detail with reference to the accompanying drawings.

2 is a flowchart illustrating a method of removing artifacts of a first artifact removal module according to an embodiment, and FIG. 3 is a conceptual diagram illustrating a method of removing artifacts of a first artifact removal module according to an embodiment.

As shown in FIGS. 2 and 3 , the first artifact removal module 200 according to the present embodiment receives a medical image 30 from the communication module 100 (S210).

The first artifact removal module 200 converts the medical image 30 into a synthetic sinogram dimension (S210). At this time, the first artifact removal module 200 is based on the medical information of the medical image 30, the attenuation coefficient for each pixel of the image, distance information between the focus of the X-ray source and the detector, and distance information between the focus of the X-ray source and the patient. decide

For example, the first artifact removal module 200 obtains information on the contact pressure corresponding to the medical image 30 and determines the attenuation coefficient for each pixel. Also, the first artifact removal module 200 may determine distance information between the focal point of the X-ray source and the detector and distance information between the focal point of the X-ray source and the patient based on image information of the medical image 30 .

The first artifact removal module 200 generates a synthesized sinogram based on attenuation coefficients for each pixel, distance information between the focal point of the X-ray source and the detector, and distance information between the focal point of the X-ray source and the patient. In this case, the synthesized sinogram may be generated by performing a projection image for each rotation angle based on the determined attenuation coefficient for each pixel, distance information between the focus of the X-ray source and the detector, and distance information between the focus of the X-ray source and the patient.

Thereafter, the first artifact removal module 200 removes artifacts from the synthesized sinogram of the medical image 30 (S230).

As an example, the first artifact removal module 200 determines a filter kernel according to rules specified in advance for the synthesized sinogram of the medical image 30 . Also, the first artifact removal module 200 may remove artifacts based on the determined filter kernel. At this time, the first artifact removal module 200 determines a filter kernel according to a pre-established rule to facilitate separation of artifact components based on the difference in local variation between the artifact component and the structural component. Also, the first artifact removal module 200 may remove artifact components by filtering the synthesized sinogram of the medical image 30 with a filter kernel.

For example, the first artifact removal module 200 may remove artifact components by applying a 2D Fourier Transform to the synthesized sinogram of the medical image 30 . In this case, the first artifact removal module 200 may remove artifact components from the synthesized sinogram of the medical image 30 by performing 2D Fourier transform, multiplying the high frequency band by a predetermined weight, and inversely transforming the result. there is.

For example, the first artifact removal module 200 may remove artifact components by applying a 2D wavelet transform to the synthesized sinogram of the medical image 30 . In this case, the first artifact removal module 200 may remove the artifact component by using a characteristic that the artifact component is located in a high frequency region different from the structural component in the 2D wavelet domain. The first artifact removal module 200 performs two-dimensional wavelet transform on the synthesized sinogram of the medical image 30, multiplies it by a predetermined weight, and inversely transforms it to obtain artifacts from the synthesized sinogram of the medical image 30. components can be removed.

For example, the first artifact removal module 200 may remove artifact components from the synthesized sinogram of the medical image 30 based on eigencomponent decomposition of a Hessian matrix. A Hessian matrix is a matrix of second order partial derivatives in vertical and horizontal directions at each pixel. When the eigencomponents are decomposed in the Hessian matrix, the first eigencomponent that can be obtained has the property of representing the structural stone component and the second eigencomponent representing the artifact component. Accordingly, the first artifact removal module 200 includes the second eigencomponent of the Hessian matrix at each pixel of the synthesized sinogram of the medical image 30 to generate artifacts from the synthesized sinogram of the medical image 30. components can be removed.

Meanwhile, the first artifact removal module 200 converts the medical image 30 into an image dimension by applying a filtered back projection operation to the synthesized sinogram of the medical image 30 from which artifact components have been removed. Convert (S240).

However, this is for explanation of the present embodiment, and a separate module may be used to convert the synthesized sinogram-dimensional medical image 30 into an image dimension.

Meanwhile, the second artifact removal module 300 may remove residual artifact components of the medical image 30 from which the artifact components are removed in the first artifact removal module 200 . In this case, the second artifact removal module 300 may remove residual artifact components based on the deep learning model.

To this end, the deep learning model may be trained based on medical images from which artifact components are removed in the first artifact removal module 200 . Hereinafter, a method for training a deep learning model will be described in detail with reference to the accompanying drawings.

4 is a flowchart showing a training method of the second artifact removal module according to this embodiment, and FIG. 5 is a conceptual diagram showing a training method of the second artifact removal module according to this embodiment.

As shown in FIGS. 4 and 5 , the second artifact removal module 300 according to the present embodiment may remove residual artifacts based on a deep learning model. To this end, the deep learning model may be trained based on medical images from which artifact components are removed in the first artifact removal module 200 .

As an example, the deep learning model receives the medical image 30 output from the first artifact removal module 200 (S410). The deep learning model receives a clear medical image from the outside, for example, a medical image without residual artifact components (S420). Further, the deep learning model may be trained by pairing an image with residual artifact components with a medical image without residual artifact components (S430). Accordingly, the deep learning model may have a function of removing residual artifact components from the medical image input from the first artifact removal module 200 .

That is, the deep learning model performs repetitive training so that the difference between the medical image output from the first artifact removal module 200 and the clear medical image can be minimized based on the clear medical image, so that from the first artifact removal module 200 It may have a function of removing residual artifacts from the input medical image 30 .

At this time, in the training of the deep learning model, the deep learning model is trained to remove residual artifacts with a correction coefficient lower than the correction coefficient of the first artifact removal module 200, thereby preventing distortion of the image at the image level. .

On the other hand, when training of the deep learning model is completed, the second artifact removal module 300 removes residual artifact components of the medical image 30 provided from the first artifact removal module 200 to obtain a high-quality read image ( 50) can be created.

Hereinafter, a method for processing artifacts of medical images according to the present embodiment will be described in detail with reference to the accompanying drawings. However, detailed descriptions of the above-described components will be omitted and the same reference numerals will be given for description.

6 is a flowchart illustrating a method for processing artifacts of a medical image according to an exemplary embodiment.

As shown in FIG. 6 , the processing apparatus 1000 according to the present embodiment may be applied to removing artifacts of a beam hardening phenomenon after training of a deep learning model is completed.

First, the communication module 100 receives the medical image 30 to be artifact removed (S610). In this case, the medical image 30 may be provided from the medical equipment 10 or a database (not shown).

Accordingly, the first artifact removal module 200 converts the medical image 30 into a sinogram dimension to generate a synthesized sinogram for the medical image 30 (S620). The synthesized sinogram may be generated by performing a projection image for each rotation angle based on predetermined attenuation coefficients for each pixel, distance information between the focal point of the X-ray source and the detector, and distance information between the focal point of the X-ray source and the patient.

Thereafter, the first artifact removal module 200 removes artifacts from the synthesized sinogram of the medical image 30 (S630). In this case, the first artifact removal module 200 may perform artifact removal on the synthesized sinogram with a higher correction coefficient than the second artifact removal module 300 to achieve high removal performance.

In this case, the first artifact removal module 200 may perform artifact removal on the synthesized sinogram based on a filter kernel, 2D Fourier transform, 2D wavelet transform, and Hessian matrix.

However, this is for explanation of the present embodiment and does not limit the artifact removal method, and it may be possible to apply a deep learning model to which the above algorithm is applied in artifact removal.

Meanwhile, when artifact removal for the synthesized sinogram is completed, the sinogram-dimensional medical image 30 may be converted into an image dimension (S640).

At this time, the processing device 1000 converts the medical image 30 into an image dimension by applying a filtered back projection operation to the synthesized sinogram. The conversion from the synthesized sinogram dimension to the image dimension may be performed in the first artifact removal module 200, but is not limited thereto.

Then, the second artifact removal module 300 removes residual artifacts from the image-dimensional medical image 30 (S650).

At this time, the second artifact removal module 300 performs residual artifact removal on the image-dimensional medical image 30 with a correction coefficient lower than that of the first artifact removal module 200 . Accordingly, the second artifact removal module 300 may generate the readout image 50 without distortion of the medical image 30 . Here, the second artifact removal module 300 may process the medical image 30 without distortion based on a deep learning model in which a correction coefficient is set in advance.

Thereafter, the processing device 1000 may transmit the high-quality reading image 50 to the reading doctor so that an image diagnosis may be performed (S660).

As described above, the apparatus for processing artifacts of medical images according to the present invention has an effect of generating a high-quality readout image without image distortion by removing artifacts in the sinogram dimension and the image dimension based on mutually different correction coefficients.

In addition, the apparatus for processing artifacts of medical images according to the present invention reliably removes residual artifacts by learning a deep learning model based on images from which artifact removal has been performed in the sinogram dimension, thereby producing high-quality read images in a short time. has the effect of obtaining it.

One embodiment of the present invention described above and shown in the drawings should not be construed as limiting the technical idea of the present invention. The protection scope of the present invention is limited only by the matters described in the claims, and those skilled in the art can improve and change the technical spirit of the present invention in various forms. Therefore, such improvements and changes will fall within the protection scope of the present invention as long as they are obvious to those skilled in the art.

Claims (18)

A medical image artifact processing apparatus for removing beam hardening artifacts from medical images,
a first artifact removal module for receiving the medical image, converting the medical image into a sinogram dimension, and correcting an artifact from the sinogram; and
A second artifact removal module for generating a high-quality readout image by correcting residual artifacts from the image-dimensional medical image from which the artifacts are removed;
The correction coefficients of the first and second artifact removal modules are different from each other,
The first artifact removal module
removing the artifact from the sinogram dimension with a higher correction factor compared to the second artifact removal module;
The second artifact removal module
The artifact processing device for medical images, characterized in that for removing the residual artifact from the image dimension with a correction coefficient lower than that of the first artifact removal module.
delete According to claim 1,
The second artifact removal module
A deep learning model for removing residual artifacts of the medical image in the image dimension;
The deep learning model is
Artifact processing device for medical images, characterized in that learning is performed based on the medical images for which artifact removal has been completed in the first artifact removal module.
According to claim 3,
The deep learning model is
A medical image from which artifact removal has been completed in the first artifact removal module and a medical image that does not include the residual artifact are paired and learned;
In the learning of the deep learning model,
Based on the medical image that does not include the residual artifact, the medical image characterized by repeatedly training to minimize the difference between the medical image for which the artifact removal is completed in the first artifact removal module and the medical image that does not include the residual artifact. Artifact processing unit of .
According to claim 1,
The first artifact removal module
An artifact processing device for a medical image, characterized in that the medical image is converted into a sinogram dimension based on medical information of the medical image, and the artifact is removed from the sinogram dimension according to a pre-set rule.
According to claim 5,
The first artifact removal module
The artefact processing device of the medical image, characterized in that for removing the artifact by filtering the synthesized sinogram of the medical image with a filter kernel according to a pre-set rule.
According to claim 5,
The first artifact removal module
The artifact processing apparatus of the medical image, characterized in that for removing the artifact by applying a two-dimensional Fourier transform to the synthesized sinogram of the medical image.
According to claim 5,
The first artifact removal module
The artifact processing apparatus of the medical image, characterized in that for removing the artifact by applying a two-dimensional wavelet transform to the synthesized sinogram of the medical image.
According to claim 5,
The first artifact removal module
The artifact processing apparatus of the medical image, characterized in that for removing the artifact by applying eigencomponent decomposition of a Hessian matrix to the synthesized sinogram of the medical image.
A method for processing artifacts of medical images to remove beam hardening artifacts from medical images,
an artifact removal step of receiving the medical image, converting the medical image into a sinogram dimension, and correcting an artifact from the sinogram;
a conversion step of converting the artifact-removed sinogram into an image dimension; and
A residual artifact removal step of generating a high-quality readout image by correcting residual artifacts from the image-dimensional medical image;
The correction coefficients in the artifact removal step and the residual artifact removal step are different from each other,
In the artifact removal step,
Eliminate the artifact from the sinogram dimension with a higher correction factor compared to the residual artifact removal step;
In the residual artifact removal step,
A method for processing artifacts in medical images, characterized in that for removing the residual artifacts from the image dimension with a lower correction factor compared to the artifact removal step.
delete According to claim 10,
In the residual artifact removal step,
A deep learning model for removing residual artifacts of the medical image in the image dimension is applied,
The deep learning model is
Artifact processing method of a medical image, characterized in that learning is based on the medical image for which the artifact removal is completed in the artifact removal step.
According to claim 12,
The deep learning model is
In the artifact removal step, a medical image from which artifact removal has been completed and a medical image that does not include the residual artifact are paired and learned;
In the learning of the deep learning model,
Artifacts of medical images, characterized in that for repetitive training to minimize the difference between the medical images from which artifact removal has been completed in the artifact removal step and the medical images without the residual artifacts based on the medical images that do not contain the residual artifacts. processing method.
According to claim 10,
The artifact removal step is
A method for processing artifacts of a medical image, characterized in that the medical image is converted into a synthesized sinogram dimension based on medical information of the medical image, and the artifact is removed from the sinogram dimension according to a pre-set rule.
According to claim 14,
The artifact removal step is
The artifact processing method of the medical image, characterized in that for removing the artifact by filtering the synthesized sinogram of the medical image with a filter kernel according to a pre-set rule.
According to claim 14,
The artifact removal step is
The artifact processing method of the medical image, characterized in that for removing the artifact by applying a synthesized sinogram of the medical image to a 2-dimensional Fourier transform.
According to claim 14,
The artifact removal step is
The artifact processing method of the medical image, characterized in that for removing the artifact by applying a synthesized sinogram of the medical image to a two-dimensional wavelet transform.
According to claim 14,
The artifact removal step is
The artifact processing method of the medical image, characterized in that for removing the artifact by applying the synthesized sinogram of the medical image to the eigencomponent decomposition of the Hessian matrix.

Images