KR102254971B1 - Method and apparatus for converting contrast enhanced image and non-enhanced image using artificial intelligence - Google Patents

Abstract

The present invention relates to a method and apparatus for converting contrast-enhanced images and non-contrast-enhanced images using artificial intelligence, and the method for converting contrast-enhanced and non-contrast-enhanced images using artificial intelligence And generating a non-enhanced CT or MRI image and a contrast enhanced CT or MRI image for a specific region of a patient using a CT or MRI device; Generating a contrast-enhanced CT or MRI image through learning of an input image using an artificial neural network to obtain an output image; Comparing the output image generated through learning and the target image to obtain similarity or dissimilarity, and generating a cost function; And feedback the cost function to the artificial neural network to learn the input image to minimize the cost function.
According to the present invention, it is possible to reduce the radiation dose by re-taking the contrast-enhanced CT image to see lesions that are not visible in the non-contrast-enhanced CT image, and it is difficult to use a contrast agent due to kidney disease or allergy. In this case, it is possible to solve the problem of insufficient diagnosis with only non-contrast-enhanced CT or MRI images by generating contrast-enhanced CT or MRI images without using a contrast agent.

Description

TECHNICAL FIELD [Method and apparatus for converting contrast enhanced image and non-enhanced image using artificial intelligence}

The present invention uses artificial intelligence to convert non-enhanced CT (Computed Tomography) or MRI (Magnetic Resonance Imaging) images into contrast enhanced CT or MRI images, and vice versa. enhanced) It relates to a method and apparatus for converting a CT or MRI image into a non-enhanced CT or MRI image.

In general, lesions that can be seen with non-enhanced CT images and contrast enhanced CT images are different. Therefore, there are cases in which both non-enhanced CT images and enhanced CT images are taken at the same time, or if the lesion is not visible in one image, the image is taken again. That is, if the lesion is not visible on the non-enhanced CT image, the enhanced CT image is retaken or, conversely, if the lesion is not seen on the enhanced CT image, the non-enhanced CT image is retaken. In this case, there is a problem that the radiation dose increases.

Also, it may be difficult to use a contrast agent due to kidney disease or allergy. In this case, there is a problem in that the non-enhanced image is insufficiently diagnosed, or the contrast-enhanced image is photographed with the risk of allergy and the like.

Accordingly, there is a need for a method and apparatus for generating a non-enhanced CT or MRI image as a contrast enhanced CT or MRI image, and vice versa, that is, converting a contrast enhanced CT or MRI image to a non-enhanced CT or MRI image.

Korean Patent Application Publication No. 10-2018-0117009 Republic of Korea Patent Application Publication No. 10-2018-0123810

The problem to be solved by the present invention is created to solve the problems and inconveniences of the prior art, and contrast enhanced CT or MRI images using artificial intelligence Contrast-enhanced image and non-contrast-enhanced image conversion method and apparatus using artificial intelligence that converts to an image and, conversely, a contrast-enhanced CT or MRI image to a non-enhanced CT or MRI image Is to provide.

Contrast-enhanced image and non-contrast-enhanced image conversion method using artificial intelligence according to the present invention for achieving the above technical problem is Non-enhanced CT or MRI image and Contrast enhanced CT for a specific part of a patient using a CT or MRI device. Or generating an MRI image; When the generated non-enhanced CT or MRI image is used as an input image and a contrast enhanced CT or MRI image is referred to as a target image, the input image is learned using an artificial neural network to generate and output a contrast enhanced CT or MRI image. Obtaining an image; Comparing the output image generated through the learning and the target image to obtain a degree of similarity or dissimilarity, and generating a cost function; And learning the input image by feeding back the cost function to the artificial neural network to minimize the cost function.

Contrast-enhanced image and non-contrast-enhanced image conversion method using artificial intelligence according to the present invention for achieving the above technical problem is Non-enhanced CT or MRI image and Contrast enhanced CT for a specific part of a patient using a CT or MRI device. Or generating an MRI image; When the generated contrast enhanced CT or MRI image is used as an input image and a non-enhanced CT or MRI image is a target image, the input image is learned using an artificial neural network to generate a non-enhanced CT or MRI image. Obtaining an output image; Comparing the output image generated through the artificial neural network learning with the target image to obtain a degree of similarity or dissimilarity, and generating a cost function; And learning the input image by feeding back the cost function to the artificial neural network to minimize the cost function.

The method of converting a contrast-enhanced image and a non-contrast-enhanced image using artificial intelligence according to the present invention includes a rotational displacement amount (tilting) and a left-right vertical displacement amount (offset value) according to the amount of change in the patient's posture movement of the target image with respect to the input image. It may further include a pre-treatment step of adjusting. The pre-processing step may further include selecting an image of the target image closest to the input image in order to compensate for a difference according to the section thickness of the target image with respect to the input image. In addition, the pre-processing step may further include selecting an entire target image selected as a position closest to the entire input image in order to compensate for a difference according to a section thickness of the target image with respect to the input image.

A contrast-enhanced image and non-contrast-enhanced image conversion apparatus using artificial intelligence according to the present invention to achieve the above technical problem is generated using a CT (Computed Tomography) or MRI (Magnetic Resonance Imaging) device, When a non-enhanced CT or MRI image is referred to as an input image, and a contrast enhanced CT or MRI image generated using the CT or MRI device is referred to as a target image, the non-enhanced CT or MRI image is referred to as a target image. An artificial intelligence learning unit that learns an input image using artificial intelligence, generates a contrast-enhanced CT or MRI image, and provides an output image; And comparing the output image provided through the learning of the artificial intelligence learning unit with the target image to obtain a similarity or dissimilarity, generating a cost function, and feeding back the cost function to the artificial intelligence learning unit to learn the cost. It includes a cost function that minimizes the value of the function.

A contrast-enhanced image and non-contrast-enhanced image conversion apparatus using artificial intelligence according to the present invention to achieve the above technical problem is generated using a CT (Computed Tomography) or MRI (Magnetic Resonance Imaging) device, When a contrast enhanced CT or MRI image is referred to as an input image, and a non-enhanced CT or MRI image generated using the CT or MRI device is referred to as a target image, the contrast enhancement input An artificial intelligence learning unit that learns an image using artificial intelligence, generates a non-contrast enhanced CT or MRI image, and provides an output image; And comparing the output image provided through the learning of the artificial intelligence learning unit with the target image to obtain a similarity or dissimilarity, generating a cost function, and feeding back the cost function to the artificial intelligence learning unit to learn the cost. It includes a cost function that minimizes the value of the function.

The apparatus for converting a contrast-enhanced image and a non-contrast-enhanced image using artificial intelligence according to the present invention calculates a rotational displacement amount (tilting) and a left-right vertical displacement amount (offset value) according to the amount of change in the patient's posture movement with respect to the input image. It may further include a pre-processing unit to adjust. The preprocessor may further include selecting an image of the target image closest to the input image in order to compensate for a difference according to the section thickness of the target image with respect to the input image. In addition, the preprocessor may further include selecting an entire target image selected as a position closest to the entire input image in order to compensate for a difference according to a section thickness of the target image with respect to the input image.

According to the method and apparatus for converting contrast-enhanced and non-contrast-enhanced images using artificial intelligence according to the present invention, it is possible to reduce the radiation dose by re-taking the contrast-enhanced CT image to view the lesions that are not visible in the non-contrast-enhanced CT image. .

And according to the present invention, in the case of a patient having difficulty using a contrast agent due to kidney disease or allergy, by generating a contrast-enhanced CT or MRI image without using a contrast agent, a non-contrast-enhanced CT or MRI image is insufficiently diagnosed. You can solve the problem. In addition, it is possible to solve the difficulty of taking a contrast enhanced image even at the expense of difficulties such as allergies.

1 is a block diagram showing an embodiment of an apparatus for converting a contrast-enhanced image and a non-contrast-enhanced image using artificial intelligence according to the present invention.
FIG. 2 shows the artificial intelligence learning unit 120 in more detail in the apparatus for converting contrast-enhanced images and non-contrast-enhanced images using artificial intelligence according to the present invention.
3 shows an embodiment of an artificial intelligence-learned contrast-enhanced image and non-contrast-enhanced image conversion apparatus according to the present invention.
Figure 4 shows the bone window and bone HU (Hounsfield Unit) values in the CT image.
5 is a flowchart illustrating an embodiment of converting a non-contrast-enhanced image into a contrast-enhanced image in the method for converting a contrast-enhanced image and a non-contrast-enhanced image using artificial intelligence according to the present invention.
6 is a flowchart illustrating an embodiment of converting a contrast-enhanced image into a non-contrast-enhanced image in the method for converting a contrast-enhanced image and a non-contrast-enhanced image using artificial intelligence according to the present invention.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. Since the embodiments described in the present specification and the configurations shown in the drawings are only preferred embodiments of the present invention, and do not represent all the technical spirit of the present invention, various equivalents that can replace them at the time of the present application It should be understood that there may be variations and variations.

1 is a block diagram showing an embodiment of an apparatus for converting a contrast-enhanced image and a non-contrast-enhanced image using artificial intelligence according to the present invention. Referring to FIG. 1, an embodiment of a contrast-enhanced image and non-contrast-enhanced image conversion apparatus using artificial intelligence according to the present invention includes an artificial intelligence learning unit 120 and a cost function unit 130, and preprocessing It may further include a unit 110.

Non-enhanced CT or MRI images and contrast-enhanced CT or MRI images of specific areas of the patient using a CT (Computed Tomography) or MRI (Magnetic Resonance Imaging) device (not shown). Take a picture. The specific region of the patient is often a specific region of the same patient, but if the region to be photographed is the same, non-enhanced and contrast enhanced CT or MRI images of a specific region of a non-identical patient may be photographed. The photographed non-enhanced CT or MRI image may be an input image or a target image. Contrast enhanced CT or MRI image taken using the CT or MRI device becomes a target image when the non-contrast enhanced CT or MRI image is an input image, and the non-contrast enhanced CT or MRI image is a target In the case of an image, it becomes an input image. Here, non-enhanced CT or MRI images and contrast enhanced CT or MRI images and contrast enhanced CT or MRI images are not only a few minutes or hours of parallax, but also a significant parallax, e.g. For example, it can be taken at intervals of days, months, or even years.

The artificial intelligence learning unit 120 learns the non-contrast-enhanced input image using artificial intelligence, converts it into a contrast-enhanced CT or MRI image, and provides it as an output image. In addition, the artificial intelligence learning unit 120 learns the contrast-enhanced input image using artificial intelligence, converts it into a non-contrast-enhanced CT or MRI image, and provides it as an output image. The artificial intelligence learning unit 120 may include an encoder 122, a deep neural network 124, and a decoder 126.

FIG. 2 shows the artificial intelligence learning unit 120 in more detail in the apparatus for converting contrast-enhanced images and non-contrast-enhanced images using artificial intelligence according to the present invention. Referring to FIG. 2, the encoder 122 may be implemented as a convolution neural network, and may include a plurality of convolution layers 210 and a plurality of pooling layers 220. In addition, the deep neural network 124 may be configured as a fully connected layer, and the decoder 126 may be implemented as a convolution neural network, and a plurality of convolution layers 230 and a plurality of unpooling layers 240 may be used. Can be configured. The artificial intelligence learning unit 120 in the apparatus for converting contrast-enhanced and non-contrast-enhanced images using artificial intelligence according to the present invention may use a simple neural network and a recurrent neural network as well as a CNN. It is also possible to use an existing prediction model. For example, a regression model or surrogate management framework can be used.

The cost function unit 130 compares the output image provided through the learning of the artificial intelligence learning unit 120 and the target image to obtain the similarity or dissimilarity between the output image and the target image, and generates a cost function, The cost function is fed back to the artificial intelligence learning unit 120 to learn, and the feedback is repeated to minimize the cost function value. The similarity or dissimilarity between the output image and the target image includes a difference between the output image and the target image, and may be generated as a cost function by obtaining the difference between the output function and the target image.

In addition, a device or program that distinguishes whether the output image and the target image are the same or different may be used as a cost function. That is, the degree of similarity or dissimilarity between the output image and the target image may be calculated using a device or a program that distinguishes whether the output image and the target image are the same or different. For example, in the case of CT, two images are extracted by allowing overlapping of non-enhanced CT and contrast enhanced CT images of patients 　 artificial intelligence (e.g., Convoluted neural network) 　 or input to other predictive models 　If the image is actually the same image, it is output as'same image' (in the program, it can also be output as a number of '0'), and all other images are output as'different image' (in the program, the number '1' or '0~) It is also possible to use a device or program that has been trained to be (may be output with a probability of different images between 1').

In the case of MRI, it is also possible to configure a device that distinguishes whether two images are the same or different from each other. It is also possible to use a device that distinguishes whether two images trained using both CT and MRI images are the same or different. The device used in the present invention can use a device that distinguishes whether two images configured as described above are the same or different from each other as a cost function. The device of the present invention is trained until the device that distinguishes between the contrast-enhanced image and the actual contrast-enhanced image using the patient's non-enhanced CT recognizes the same image as each other.

In the opposite case, in the present invention, until the device that distinguishes whether the two images are the same or different from the image converted to non-enhanced using the contrast enhanced CT image of the patient and the actual non-enhanced CT image are recognized as the same image. Train a device to be used as a cost function. In this case, a cost function can be generated without using the difference between the two images at all.

In the case of MRI, the image converted to Contrast enhanced using the patient's non-enhanced MRI image and the actual Contrast enhanced MRI image using a device that distinguishes whether two images are the same or different are the same or different from each other. The device of the present invention is trained until the devices that distinguish them recognize each other as the same image. Even in the opposite case, the device of the present invention uses the patient's Contrast enhanced MRI image to convert the image into non-enhanced and the actual non-enhanced MRI image until the device that distinguishes whether the two images are the same or different from each other recognizes the same image. To train.

The preprocessor 110 adjusts a rotational displacement amount (tilting) and a left-right vertical displacement amount (offset value) according to the amount of change in the patient's posture movement of the target image with respect to the input image. That is, the pre-processing unit 110 may have a difference between a patient's posture when taking a non-contrast-enhanced CT or MRI image and a patient's posture when taking a non-contrast-enhanced CT or MRI image due to a minute movement of the patient. The difference can be expressed as a rotational displacement and an offset value. The amount of rotational displacement (tilting) is the degree to which the patient's photographing area is twisted, and the left-right vertical displacement (offset value) is the value in which the patient's photographing area is moved up, down, left and right.

The pre-processing unit 110 adjusts the alignment of the patient's posture change to match the enhancement-enhanced image and the non-contrast-enhanced image to increase accuracy. The alignment adjustment may be performed manually by a person, but manual adjustment may take a lot of time and cost.

In the present invention, in the case of a CT image, the region with the least change among the human body parts is bone, so if a cutoff value is set near the bone HU (about 1000) like a bone window, the difference between the non-contrast enhanced image and the contrast enhanced image is the most. You can adjust the alignment to be small. Figure 4 shows the bone window and bone HU (Hounsfield Unit) values in the CT image. W is the width and L is the level. HU (Hounsfield Unit) is a relative expression of the degree of absorption by the density of each part when X-rays pass through the body. Referring to FIG. 4, the HU of bone is 1000, and the HU of water (H ₂ O) is 0. , HU of air is -1000.

In CT scans, the image value of the bone has a value of about 1000, and there is a separate bone window when viewing the CT. If you look through the bone window, the data is still there, but only about 1000 is white, and the rest is black, so that only the bone can be seen. Adjusting the alignment while rotating the enhancement image and the non-contrast enhancement image, where the visible bone is highlighted, can be obtained by adjusting the alignment. By reflecting the obtained tilting and offset values, the image is rotated and the position is changed to match the enhancement-enhanced image and the non-contrast-enhanced image.

In the case of an MRI image, a cutoff value is set near the bone signal value for each imaging mode, and the alignment is adjusted so that the difference between the non-contrast-enhanced image and the contrast-enhanced image is smallest.

In addition, the preprocessor 110 may select a position of an image closest to the input image to compensate for a difference according to the section thickness of the target image with respect to the input image and set it as a target image. After adjusting the alignment, select a video at a nearby location as the target video. In addition, the preprocessor 110 may select the entire target image selected as the closest position to the entire input image in order to compensate for a difference according to the section thickness of the target image with respect to the input image. Accuracy can be improved by learning using the entire input image and the selected target image.

That is, put the whole image together and train it all at once. For example, after taking 100 input images, if there are 10 values per image, there are 1000 data for 100 images. With 1000 data as one data set, one data set per person becomes one, and one data set is trained at a time. In this way, the coefficient determination of the CNN can converge quickly. Without doing so, if one image is trained one by one as one data set, if the number of images is large, the CNN coefficients may be different, which may lower the learning accuracy.

3 shows an embodiment of an artificial intelligence-learned contrast-enhanced image and non-contrast-enhanced image conversion apparatus according to the present invention. 3, when inputting a non-contrast-enhanced CT or MRI image as an input image 310 to the artificial intelligence-learned contrast-enhanced image and non-contrast-enhanced image conversion apparatus 320 according to the present invention, contrast-enhanced CT or MRI The image is output as an output image 330. Conversely, when a contrast-enhanced CT or MRI image is input to the artificial intelligence-learned contrast-enhanced image and non-contrast-enhanced image conversion device 320 according to the present invention as the input image 310, the non-contrast-enhanced CT or MRI image is output image It is output as 330.

5 is a flowchart illustrating an embodiment of converting a non-contrast-enhanced image into a contrast-enhanced image in the method for converting a contrast-enhanced image and a non-contrast-enhanced image using artificial intelligence according to the present invention. Referring to FIG. 5, a non-enhanced CT or MRI image and a contrast enhanced CT or MRI image are generated by photographing the human body of a patient using a CT or MRI device. (S510) The human body of the patient is the same as the human body of the patient. In many cases, if the body part to be photographed is the same, a non-enhanced and contrast enhanced CT or MRI image may be generated by photographing the human body of a non-identical patient.

When the generated non-enhanced CT or MRI image is used as an input image, and a contrast enhanced CT or MRI image is referred to as a target image, a tilting amount according to the amount of change in the patient's posture movement with respect to the input image and Pre-processing of adjusting the left and right displacement amount (offset value) is performed. (S520) The pre-processing (S520) is the closest to the input image in order to compensate for the difference according to the section thickness of the target image with respect to the input image. It may further include selecting an image of the target image of the location. In addition, in the pre-processing (S520), in order to compensate for a difference according to the section thickness of the target image with respect to the input image, the entire target image image selected as the closest position to the entire input image may be selected.

When the pre-processing process is performed, the input image is acquired as an output image by generating a contrast enhanced CT or MRI image through learning using an artificial neural network (step S530) by comparing the output image generated through the learning with the target image. The difference is obtained and generated as a cost function (step S540). The cost function is fed back to the artificial neural network to learn the input image to minimize the cost function (step S550).

6 is a flowchart illustrating an embodiment of converting a contrast-enhanced image into a non-contrast-enhanced image in the method for converting a contrast-enhanced image and a non-contrast-enhanced image using artificial intelligence according to the present invention. Referring to FIG. 6, a non-enhanced CT or MRI image and a contrast enhanced CT or MRI image are generated by photographing the human body of a patient using a CT or MRI device. (S610) The human body of the patient is the same as the human body of the patient. In many cases, if the body parts to be photographed are the same, a non-enhanced and contrast enhanced CT or MRI image may be generated by photographing the human body of a non-identical patient. The generated contrast enhanced CT or MRI image is used as an input image. And, when a non-enhanced CT or MRI image is a target image, pre-processing is performed to adjust the amount of rotational displacement (tilting) and the amount of left and right vertical displacement (offset value) according to the amount of change in the patient's posture movement with respect to the input image. (Step S620) The preprocessing (Step S620) further includes selecting an image of the target image closest to the input image to compensate for a difference according to the section thickness of the target image with respect to the input image. I can. In addition, in the pre-processing (S620), in order to compensate for a difference according to the section thickness of the target image with respect to the input image, the entire target image image selected as the closest position to the entire input image may be selected.

When the pre-processing process is performed, a non-enhanced CT or MRI image is generated through learning using an artificial neural network to obtain an input image as an output image (step S630). The output image generated through the learning and the target image are compared. The difference is calculated and generated as a cost function (step S640). The cost function is fed back to the artificial neural network to learn the input image to minimize the cost function (step S650).

The present invention can be implemented as a computer-readable code (including all devices having an information processing function) on a computer-readable recording medium. The computer-readable recording medium includes all types of recording devices that store data that can be read by a computer system. Examples of computer-readable recording devices include ROM, RAM, CD-ROM, magnetic tapes, floppy disks, and optical data storage devices. In addition, in the present specification, the “unit” may be a hardware component such as a processor or a circuit, and/or a software component executed by a hardware configuration such as a processor.

The present invention has been described with reference to the embodiments shown in the drawings, but these are merely exemplary, and those of ordinary skill in the art will understand that various modifications and equivalent other embodiments are possible therefrom. Therefore, the true technical protection scope of the present invention should be determined by the technical idea of the attached registration claims.

110: pre-processing unit 120: artificial intelligence learning unit
122: Encoder (CNN) 124: Deep Neural Network
126: decoder (CNN) 130: cost function unit
210: Convolution Layer 220: Pooling Layer
230: Unpooling Layer 240: Convolution Layer
310: input image 320: image conversion device
330: output image

Claims (10)

Generating, by an artificial intelligence learning unit, a non-enhanced CT or MRI image and a contrast enhanced CT or MRI image for a specific region of the patient using a CT or MRI device;
When the artificial intelligence learning unit uses the generated non-enhanced CT or MRI image as an input image and the contrast enhanced CT or MRI image is a target image, the input image is learned using an artificial neural network. Generating an image and obtaining an output image;
Calculating, by the cost function unit, a similarity or dissimilarity by comparing the output image generated through the learning with the target image, and generating a cost function; And
Including, by a cost function unit, learning the input image by feeding back the cost function to the artificial neural network to minimize the cost function,
The pre-processing unit further comprises a pre-processing step of adjusting a rotational displacement amount (tilting) and a left-right vertical displacement amount (offset value) according to the amount of change in the patient's posture movement of the target image with respect to the input image; Augmented image and non-contrast enhanced image conversion method. Generating, by an artificial intelligence learning unit, a non-enhanced CT or MRI image and a contrast enhanced CT or MRI image for a specific region of the patient using a CT or MRI device;
When the artificial intelligence learning unit uses the generated Contrast enhanced CT or MRI image as an input image and a non-enhanced CT or MRI image as a target image, the input image is non-enhanced CT or through learning using an artificial neural network. Generating an MRI image and obtaining an output image;
Comparing the output image generated through the artificial neural network learning and the target image to obtain a degree of similarity or dissimilarity, and generating a cost function as a cost function; And
Including, by a cost function unit, learning the input image by feeding back the cost function to the artificial neural network to minimize the cost function,
The pre-processing unit further comprises a pre-processing step of adjusting a rotational displacement amount (tilting) and a left-right vertical displacement amount (offset value) according to the amount of change in the patient's posture movement of the target image with respect to the input image; Augmented image and non-contrast enhanced image conversion method. delete The method of claim 1 or 2, wherein The pretreatment step is
A contrast-enhanced image using artificial intelligence, further comprising the step of selecting an image of the target image closest to the input image in order to compensate for a difference according to the section thickness of the target image with respect to the input image. Non-contrast enhanced image conversion method. The method of claim 1 or 2, wherein The pretreatment step is
Artificial intelligence, characterized in that it further comprises the step of selecting an entire target image image selected as a position closest to the entire input image in order to compensate for a difference according to the section thickness of the target image with respect to the input image. Contrast-enhanced image and non-contrast-enhanced image conversion method using. A non-enhanced CT or MRI image of a specific part of the patient, created using a CT (Computed Tomography) or MRI (Magnetic Resonance Imaging) device, is called an input image, and the CT or MRI device is used. When the resulting contrast enhanced CT or MRI image is referred to as a target image, artificial intelligence that learns the non-contrast enhanced input image using artificial intelligence and generates a contrast-enhanced CT or MRI image to provide an output image Learning department; And
The artificial intelligence learning unit compares the output image provided through learning with the target image to obtain a similarity or dissimilarity, generates a cost function, and feeds the cost function back to the artificial intelligence learning unit to learn the cost function. Including; a cost function unit for minimizing the value,
A pre-processing unit for adjusting the amount of rotational displacement (tilting) and the amount of left and right vertical displacement (offset value) according to the amount of change in the patient's posture movement of the target image with respect to the input image; Augmented image and non-contrast enhanced image conversion device. Contrast enhanced CT or MRI image for a specific part of the patient, created using a Computed Tomography (CT) or Magnetic Resonance Imaging (MRI) device, is called an input image, and is created using the CT or MRI device. When a non-enhanced CT or MRI image is referred to as a target image, artificial intelligence that learns the contrast-enhanced input image using artificial intelligence and generates a non-enhanced CT or MRI image to provide an output image Learning department; And
The artificial intelligence learning unit compares the output image provided through learning with the target image to obtain a similarity or dissimilarity, generates a cost function, and feeds the cost function back to the artificial intelligence learning unit to learn the cost function. Including; a cost function unit for minimizing the value,
A pre-processing unit for adjusting the amount of rotational displacement (tilting) and the amount of left and right vertical displacement (offset value) according to the amount of change in the patient's posture movement of the target image with respect to the input image; Augmented image and non-contrast enhanced image conversion device. delete The method of claim 6 or 7, wherein the pre-processing unit
Contrast-enhanced image and ratio using artificial intelligence, characterized in that it further comprises selecting an image of the target image closest to the input image in order to compensate for the difference according to the section thickness of the target image with respect to the input image. Contrast-enhanced image conversion device. The method of claim 6 or 7, wherein the pre-processing unit
A contrast-enhanced image using artificial intelligence, further comprising selecting the entire target image selected as a position closest to the entire input image to compensate for a difference according to the section thickness of the target image with respect to the input image. And a non-contrast-enhanced image conversion device.

Images