KR102128990B1 - A method for transforming MRI data and device for the same - Google Patents

Abstract

When input layer data, which includes information specifying an MRI sequence and source data which is MRI data obtained from a subject part by using the MRI sequence, is inputted to an input layer of a neural network unit, output layer data having the same format as ρ, T1, and T2 parameters is outputted. Disclosed is a technology for generating synthesized MRI data of the subject part by using the output layer data and an arbitrary MRI sequence together. The present invention provides a technology for converting an image obtained with an arbitrary MRI sequence into an image obtained with another MRI sequence.

Description

A method for transforming MRI data and a device therefor

The present invention relates to a computing technique for a technique for converting MRI data.

When the first MRI scanner and the second MRI scanner are scanned using different MRI sequences, comparing the first image and the second image obtained by the first MRI scanner and the second MRI scanner with respect to the same test portion It is difficult. This is true even when the first MRI scanner and the second MRI scanner are identical to each other.

A first image obtained by scanning by an MRI scanner using a first MRI sequence with respect to an arbitrary test portion is given at a first time point, and an MRI scanner using a second MRI sequence with respect to the arbitrary test portion at a second time point. It may be assumed that the second image obtained by scanning is given. In this case, the first image and the second image may be compared with each other in order to observe a change in the MRI image of the arbitrary test portion over time, but the first image and the second image may have different MRI sequences. Since it is obtained by using, it is difficult to directly compare them.

Alternatively, it may be assumed that a third image obtained by scanning by an MRI scanner using a third MRI sequence with respect to the arbitrary test portion is given at the first time point, and at the second time point, the arbitrary test portion It can be assumed that the fourth image obtained by the MRI scanner is scanned using the third MRI sequence. At this time, the third image and the fourth image may be compared with each other in order to observe a change in the MRI image of the arbitrary test portion over time. In this case, unlike the above, since the third image and the fourth image are obtained using the same third MRI sequence, it is easy to directly compare them.

Accordingly, if a technology capable of converting the first image and the second image into the third image and the fourth image is provided, the first image and the second image are respectively respectively the third image and the first image. After converting to the fourth image, the third image and the fourth image are compared with each other, so that the effect of comparing the first image and the second image with each other may be obtained.

If such a technique is provided, images obtained by any MRI sequence may be converted into a standardized format.

The above is considered by the inventor in the process of derivation of the present invention, and is not necessarily recognized as a prior art.

In the present invention, it is intended to provide a technique for converting an image obtained by an arbitrary MRI sequence into an image obtained by another MRI sequence.

According to an aspect of the present invention, upon receiving a first MRI image and a first MRI sequence used to obtain the first MRI image of a subject, physical parameters including ρ, T1, and T2 parameters of the subject The learned neural network unit outputting the may be used. In addition, when the ρ, T1, and T2 parameters and an arbitrary second MRI sequence are input, a transform unit that outputs a second MRI image related to the subject may be used. The conversion unit may be according to a conventional technique, and the neural network unit may be proposed by the present invention.

A network part 10 according to an aspect of the present invention; And a transforming part 20 may be provided. In the input layer 110 of the network unit, information specifying an MRI sequence (3 ₁₁ ), and MRI data obtained for a measuring part using the MRI sequence (an MRI datum) An input layer datum 3 ₁ including in-source data 3 ₁₂ is input. In addition, the output layer 120 of the network unit is configured to output an output layer datum 40 including a set of physical parameters of the test unit. In addition, the conversion unit is configured to generate synthesized MRI data 50 for the test unit using the physical parameter set output from the network unit.

At this time, the network unit may be machine-learned using a set of training data for the input layer and a set of training data for the output layer. .

In this case, the information is a set of operating parameters specifying the MRI sequence, and may include the MRI sequence and scan parameters used in the MRI sequence.

In this case, the subject may be a specific measurement target slice of the MRI subject.

In this case, the source data may be MRI images or k-space data.

In this case, the physical parameter set may include at least one or all of ρ, T1, and T2.

In this case, the physical parameter set may include all of ρ, T1, and T2.

In this case, the network unit may be a convolutional neural network using an automatic encoder.

In this case, the synthesized MRI data may be an MRI image or k-space data.

In this case, the machine learning is based on supervised learning, and any first learning output layer data among the set of learning output layer data is the set of physical parameters related to an arbitrary first test unit, and the set of learning input layers The first learning input layer data corresponding to the first learning output layer data may include information for specifying an MRI sequence, and source data that is MRI data obtained for the first test unit using the MRI sequence. Can.

At this time, the input layer of the network unit includes information for specifying a first type MRI sequence (3 ₁₁ ), and a first source that is MRI data obtained for the first inspected unit using the first type MRI sequence. Using the first part 110 ₁ into which data 3 ₁₂ is input, information specifying the MRI sequence of the second type (3 ₂₁ ), and the MRI sequence of the second type to the first test unit the acquired MRI data is the second source data ₍₃₂₂₎ may comprise a second portion (110 ₂₎ to be input.

At this time, in the first portion of the input layer of the network unit where learning has been completed, information identifying the first type MRI sequence, and MRI data obtained for the second test unit using the first type MRI sequence The third source data is input, and no information may be input to the second part of the input layer of the learned network part.

According to another aspect of the present invention, in a data conversion apparatus including a network portion including an input layer and an output layer, a method for supervising a neural network for supervising a network portion may be provided. The input layer is configured to receive information specifying an arbitrary MRI sequence and source data that is MRI data obtained for an arbitrary subject using the arbitrary MRI sequence. In addition, the output layer is configured to output output layer data having the same format as ρ, T1, and T2 parameters for an arbitrary test unit. The supervised learning method comprises: providing first input data, which is information for specifying a first MRI sequence, and first source data, which is MRI data obtained for a first subject using the first MRI sequence, to the input layer; And updating the weight assigned in the network unit to reduce an error between the output layer data output from the output layer and ρ, T1, and T2 parameters related to the first test unit.

At this time, the MRI data obtained for the first test unit, MRI for the first test unit synthesized using the ρ, T1, and T2 parameters and the first MRI sequence for the first test unit together It may be image or k-space data.

According to another aspect of the present invention, an MRI data conversion method for converting MRI data using a data conversion device including a network unit and a conversion unit may be provided. The output layer of the network unit is configured to output output layer data having the same format as ρ, T1, and T2 parameters for an arbitrary test unit. In the MRI data conversion method, input layer data including the network unit, information specifying an MRI sequence on the input layer of the network unit, and source data that is MRI data obtained for the test unit using the MRI sequence is input. If it is, outputting the output layer data; And the conversion unit generating the synthesized MRI data for the test unit by using the output layer data output from the output layer of the network unit and an arbitrary MRI sequence together.

According to the present invention, it is possible to provide a technique for converting an image obtained by an arbitrary MRI sequence into an image obtained by another MRI sequence.

1 is a block diagram showing functions included in a data conversion device provided according to an embodiment of the present invention.
FIG. 2 is related to the network unit shown in FIG. 1, and is for explaining a method of learning a network unit having a state in which machine learning has not been completed.
3 is a diagram for explaining a learning sequence for training the network unit shown in FIG. 2 according to an embodiment of the present invention.
FIG. 4A is for explaining a method of obtaining input layer data for learning and output layer data for learning for learning of a network unit shown in FIG. 2 according to an embodiment of the present invention.
4B is for explaining a method of obtaining input layer data for learning and output layer data for learning for learning of a network unit shown in FIG. 2 according to another embodiment of the present invention.
5 is a flowchart illustrating a method of supervising a neural network network according to an embodiment of the present invention.
6 is a flowchart illustrating a method of converting MRI data according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. However, the present invention is not limited to the embodiments described herein and may be implemented in various other forms. The terminology used herein is for the purpose of understanding the embodiments, and is not intended to limit the scope of the present invention. In addition, the singular forms used hereinafter include plural forms unless the phrases express the opposite meaning.

Data converter

1 is a block diagram showing functions included in a data conversion device provided according to an embodiment of the present invention.

The data conversion device 1 may be a computing device including a CPU, a storage unit, a user interface, and a communication unit.

The data conversion device 1 may be, for example, a server. At this time, the data conversion device 1 may communicate with the MRI scanner.

Alternatively, the data conversion device 1 may be a computing device included in the MRI scanner.

The data conversion device 1 may include a network part 10 and a transforming part 20.

The network unit 10 is a machine learning module, and may be a multi-layered machine learning module including an input layer 110 and an output layer 120. The network unit 10 may be provided with dedicated hardware included in the data conversion device 1. Alternatively, the network unit 10 is provided as a program including instructions, and can be loaded into the memory of the data converter 1 whenever necessary and executed by the processing unit (CPU).

The network unit 10 may have completed machine learning. At this time, the input layer 110 may include a plurality of input nodes. The input layer 110 may be divided into a plurality of sub-input sets. Each of the sub-input sets may include one or more of the input nodes. An input layer datum 3 ₁ , 3 ₂ , 3 ₃ , ..., 3 _M in a predetermined format may be input to each of the sub-input sets (M is a natural number). In addition, the output layer 120 may include a plurality of output nodes. An output layer datum 40 in a predetermined format may be output from each output node.

Network Department Learning

FIG. 2 is related to the network unit shown in FIG. 1, and is for explaining a method of learning a network unit having a state in which machine learning has not been completed.

The network unit 10' in which learning is not completed may be learned by a supervised learning method. In the input layer of the network unit 10' where learning is not completed, training data for the input layer having the same format as input layer data having the predetermined format (4_n = {4 ₁ , 4 ₂ , 4 ₃ , ..., 4 _M }) can be input. Training data for the output layer 5_n corresponding to the learning input layer data 4_n may be prepared. Here, the subscript n is a natural number from 1 to N, and N can represent, for example, the number of repetitions of the learning of the network unit 10'.

The index n is not indicated in the reference number set {4 ₁ , 4 ₂ , 4 ₃ , ..., 4 _M } shown in FIG. 2. That is, the reference number set {4 ₁ , 4 ₂ , 4 ₃ , ..., 4 _M } shown in FIG. 2 is a reference number commonly provided for any n.

The output layer data 40 in the predetermined format may be output from each output node of the network unit 10 ′ in which learning is not completed.

The pair of learning input layer data 4_n and the corresponding learning output layer data 5_n can be obtained from a specific measuring part, for example, the nth testing part 50_n shown in FIG. 2. A specific embodiment of obtaining information regarding a pair of input layer data for learning 4_n and corresponding output layer data for learning 5_n will be described later in FIGS. 4A and 4B.

Input learning data layer (4_n) may include specific information (4 _k1), and the data source (4 _k2) of the MRI sequence (an MRI sequence). Where k is a natural number from 1 to M.

The information (4 _k1 ) specifying the MRI sequence is a set of operating parameters specifying the MRI sequence (#k), and scan parameters used for the MRI sequence (#k) and the MRI sequence (#k) ( #k).

The different information specifying the MRI sequence may include different MRI sequences and/or different scan parameters. That is, for example, the MRI sequence and scan parameters included in the first information specifying the MRI sequence #1 may be different from the MRI sequence and scan parameters included in the second information specifying the MRI sequence #2. .

In one embodiment, can be a source of data (4 _k2) is MRI data (an MRI datum) that can be obtained for the first n insertable portion (50_n) using said MRI sequence (#k). Two methods that can be used for the acquisition will be described below with reference to FIGS. 4A and 4B.

In a preferred embodiment, the MRI data may be an MRI image.

In another embodiment, the MRI data may be k-space data.

The nth test unit 50_n may be living or killed tissue that is an object to be scanned using an MRI scanner. The tissue can be human or animal or plant. In particular, the nth test unit 50_n may be a specific measurement target slice of the tissue.

The learning output layer data 5_n may include a set of physical parameters obtainable from the nth test unit 50_n.

In this specification, the physical parameter set may be referred to as a physical information set in other words.

In one preferred embodiment, the physical parameter set may include at least one or all of ρ, T1, and T2 that can be obtained from the nth test unit 50_n.

The ρ, T1, and T2 are well-known parameters or well-known information in the MRI field, and a method of obtaining the ρ, T1, and T2 from the subject is also known. This method may be performed by the parameter acquisition device 400 described later in FIG. 4A.

The output layer data 40 output from the output layer 120 of the network unit 10 ′ may be data having the same format as a set of physical parameters obtainable from the test unit.

The network leadership unit 60 may calculate an error value between the output layer data 5_n for learning and the output layer data 40, and update the weight values inside the network unit 10' to reduce the error value. . The signal for updating the weight values may be transmitted through the path 70.

3 is a diagram for explaining a learning sequence for training the network unit shown in FIG. 2 according to an embodiment of the present invention.

For example, the network unit 10' may be learned by repeating N times. Where N is a natural number. That is, the learning of the network unit 10', as shown in Fig. 3, repeats #1, repeats #2, repeats #3, ..., repeats #n, ..., repeats the learning process of #N. Can. In Figure 3, the horizontal axis is the time axis.

In different repetition intervals, different learning input layer data and learning output layer data may be provided to the network unit 10 ′. For example, in the first repetition section (repeat #1), the first learning input layer data 4_1 and the first learning output layer data 5_1 may be provided to the network unit 10 ′. In the second repetition section (repeat #2), the second learning input layer data 4_2 and the second learning output layer data 5_2 may be provided to the network unit 10 ′. At this time, the first learning input layer data 4_1 and the first learning output layer data 5_1 are obtained from the first test unit 50_1, the second learning input layer data 4_2 and the second learning output layer data 5_2 ) May be obtained from the second subject 50_2.

Hereinafter, embodiments for obtaining specific learning input layer data and specific learning output layer data from a specific test unit will be described with reference to FIGS. 4A and 4B.

FIG. 4A is for explaining a method of obtaining input layer data for learning and output layer data for learning for learning of a network unit shown in FIG. 2 according to an embodiment of the present invention.

Referring to (a) of FIG. 4A, the MRI scanner 300 may output the MRI signal of the nth test unit 50_n using the kth MRI sequence #k. The output MRI signal can be converted to the data source (4 _k2) who described with regard to Figure 2 and using the MRI image pickup device 500. In addition, information (4 _k1 ) for specifying the MRI sequence (an MRI sequence) described with reference to FIG. 2 may be prepared from the kth MRI sequence (#k). Here, if the processes shown in (a) of FIG. 4A are executed for k=1 to M, the input layer data 4_n for learning shown in FIG. 2 can be obtained.

In one embodiment, the source data _4k2 may be an MRI image. Or, in another embodiment, the source data 4 _k2 may be k-space data. In FIG. 4A, the source data _4k2 is an MRI image.

Referring to FIG. 4A (b), the MRI scanner 300 may output an MRI signal of the nth test unit 50_n using an arbitrary qth MRI sequence (#q). The output MRI signal may acquire the learning output layer data (ρ, T1, T2) 5_n described in connection with FIG. 2 using the parameter acquisition device 400.

The parameter retrieval device 400 may also be referred to as a'physical information acquisition unit under test' in other words.

The MRI image acquiring device 500 and the parameter acquiring device 400 may be implemented by a technique known in the art, and may be implemented by a computing device embedded in the MRI scanner 300 or connected to the MRI scanner 300. have.

FIG. 4B is for explaining a method of obtaining input layer data for learning and output layer data for learning for learning of a network unit shown in FIG. 2 according to another embodiment of the present invention.

Referring to FIG. 4B, the MRI scanner 300 may output the MRI signal of the nth test unit 50_n using an arbitrary qth MRI sequence (#q). The output MRI signal may acquire the learning output layer data (ρ, T1, T2) 5_n described in connection with FIG. 2 using the parameter acquisition device 400.

Then, the conversion unit 20', the learning output layer data (ρ, T1, T2) (5_n) output from the parameter acquiring device 400 and the k-th MRI sequence (#k) (40 _k1 ) together By doing so, source data (4 _k2 ) for the nth test unit 50_n may be synthesized. Here, k=1 to M may be a natural number.

For example, given ρ, T1, and T2, and given an arbitrary MRI sequence, a method by which the transform unit 20' generates one MRI image from them has already been disclosed. Of course, k-space data can be obtained from the MRI image.

For example, when the physical parameter set including n-{ρ, T1, and T2} pairs for the nth test unit 50_n is prepared using a conventional MRI technique, the prepared n-{ρ, T1, and T2} pairs may be used to synthesize various MRI images for the nth test unit 50_n.

For example, by driving the MRI scanner 300 using the first MRI sequence 40 ₁₁ , the first MRI image 4 ₁₂ for the nth test unit 50_n may be directly obtained. However, instead of this, the first MRI image 4 ₁₂ using the n-{ρ, T1, and T2} pairs and the first MRI sequence 40 ₁₁ for the nth test unit 50_n together Can also be synthesized.

In addition, by driving the MRI scanner 300 using the second MRI sequence 40 ₂₁ , the second MRI image 4 ₂₂ for the nth test unit 50_n may be directly obtained. However, instead of this, the first MRI image 4 ₂₂ using the n-{ρ, T1, and T2} pairs and the second MRI sequence 40 ₂₁ for the nth test unit 50_n together Can also be synthesized.

As described above, various MRI images for the nth test unit 50_n may be synthesized by using the n-{ρ, T1, and T2} pairs together with an arbitrary MRI sequence.

That is, using the prior art, if {ρ, T1, and T2} pairs are provided for a specific subject, various MRI images for the specific subject can be obtained without directly driving the MRI scanner. This synthesis function can be executed by the conversion unit 20'.

In one embodiment, the synthesized source data _4k2 may be an MRI image. Or In another embodiment, the synthetic data source (4 _k2) may be a k- space data. In FIG. 4B, the synthesized source data _4k2 is a MRI image.

Information (4 _k1 ) specifying the MRI sequence may be obtained from the kth MRI sequence (40 _k1 ).

Now, for the learning output layer data (ρ, T1, T2) (5_n) output from the parameter acquisition device 400, using a different MRI sequence (40 _k1 ) (k = 1 to M natural number) 4b It is understood that if the operation of the conversion unit 20' shown in is repeatedly executed, the input layer data for learning 4_n shown in FIG. 2 can be obtained.

The conversion unit 20' may be provided with dedicated hardware included in the data conversion device 1 or other learning computing device. Alternatively, the conversion unit 20' is provided as a program including instructions, and can be loaded into the memory of the data conversion device 1 or other learning computing device whenever necessary and executed by the processing unit (CPU).

The method shown in FIG. 4A uses more MRI scanners than the method shown in FIG. 4B. A large number of MRI scanners are disadvantageous in terms of cost and time. Therefore, the method according to FIG. 4B may be preferred. However, one embodiment of the present invention does not exclude the method according to FIG. 4A.

As described above, the weight of the link between the nodes in the network unit 10' is updated so as to minimize the error between the data 40 output from the network unit 10' and the output layer data 5_n prepared in advance. Can. In order to train the network unit 10', one set of input layer data for learning may be provided, and one set of output layer data for learning may be provided. Learning of the network unit 10' may be completed by repeating the update using each set of learning data.

Function of data converter

Now, functions included in a data conversion device provided according to an embodiment of the present invention will be described again with reference to FIG. 1.

In one embodiment of the invention, part of the input layer data (3 _k) is MRI sequence (an MRI sequence) for identifying information (3 _k1), and the source input to the input layer 110 of the network unit 10 Data 3 _k2 . At this time, k is a natural number of 1 to M.

The information (3 _k1 ) specifying the MRI sequence is a set of operational parameters specifying the k-th MRI sequence (#k), and is used for the k-th MRI sequence (#k) and the k-th MRI sequence (#k). K-th scan parameters.

Different information specifying different MRI sequences may include different MRI sequences and/or different scan parameters. For example, the first MRI sequence #1 and the first scan parameter included in the first information 3 ₁₁ specifying the first MRI sequence #1 may include: a second MRI sequence #2; The second MRI sequence (#2) and the second scan parameter included in the 2 information 3 ₂₁ may be different.

In one embodiment, the source data 3 _k2 may be MRI data (an MRI datum) acquired for an arbitrary subject using a k-th MRI sequence (#k).

In a preferred embodiment, the MRI data may be an MRI image.

In another embodiment, the MRI data may be k-space data.

The MRI image and the k-space data can be mutually transformed by Fourier transform.

The optional test part may be living or dead tissue that is an object to be scanned using the MRI scanner 300. The tissue can be human or animal or plant. In particular, the test portion may be a specific measurement target slice of the tissue.

The output layer data 40 output from the output layer 120 of the network unit 10 may include data in the same format as a set of physical parameters obtainable from the test unit.

In one preferred embodiment, the set of physical parameters may include at least one or all of ρ, T1, and T2, which can be obtained from the arbitrary subject.

As described in connection with FIG. 2, ρ, T1, and T2 are well-known parameters in the field of MRI, and a method of obtaining the ρ, T1, and T2 from the arbitrary subject is also known.

The converter 20 may be provided with dedicated hardware included in the data converter 1. Alternatively, the conversion unit 20 is provided as a program including instructions, and can be loaded into the memory of the data conversion device 1 whenever necessary and executed by the processing unit (CPU).

The conversion unit 20 may be configured to generate synthesized MRI data 50 for the test unit using the physical parameter set output from the network unit 10.

The conversion unit 20 shown in FIG. 1 may perform the same function as the conversion unit 20' described with reference to FIG. 4B.

In one embodiment, the synthesized MRI data 50 may be an MRI image. Or in other embodiments, the synthesized MRI data 50 may be k-space data.

In one embodiment, the synthesized MRI data 50 may be data having the same format as the MRI data 3 _k2 .

For example, given ρ, T1, and T2, a method for generating one MRI image from the given ρ, T1, and T2 is already known. The details of this have already been described in FIG. 4B.

For example, using the conventional MRI technique, it is possible to obtain the physical parameter set including {ρ, T1, and T2} pairs for an arbitrary subject. Now, the {ρ, T1, and T2} pairs can be used to synthesize various MRI images for the arbitrary subject.

For example, by driving the MRI scanner using the first MRI sequence, the first MRI image for the arbitrary subject may be directly obtained. In this case, the first MRI image may be synthesized by using the {ρ, T1, and T2} pair and the first MRI sequence for the arbitrary test part together.

In addition, for example, by driving the MRI scanner using the second MRI sequence, the second MRI image for the arbitrary subject may be directly obtained. In this case, the second MRI image may be synthesized by using the {ρ, T1, and T2} pair and the second MRI sequence for the arbitrary test part together.

In this way, by using the {ρ, T1, and T2} pairs together with an arbitrary MRI sequence, various MRI images for the arbitrary subject can be synthesized.

That is, if {ρ, T1, and T2} pairs are provided for an arbitrary subject using a conventional technique, various MRI images for the arbitrary subject can be obtained without directly driving the MRI scanner. The synthesis function can be executed by the conversion unit 20.

Example 1-Supervised Learning

5 is a flowchart illustrating a method for supervising a neural network network according to an embodiment of the present invention.

The supervised learning method is a method for supervising and learning the network unit in a data conversion apparatus including a network unit including an input layer and an output layer.

The input layer is configured to receive information specifying an arbitrary MRI sequence, and source data, which is MRI data obtained for an arbitrary test unit using the arbitrary MRI sequence, and the output layer relates to an arbitrary test unit. It may be configured to output the output layer data having the same format as the ρ, T1, and T2 parameters.

In step S10, information specifying a first MRI sequence and first source data, which is MRI data obtained for a first subject using the first MRI sequence, may be provided to the input layer.

In step S20, the weight assigned in the network unit may be updated to reduce an error between the output layer data output from the output layer and the ρ, T1, and T2 parameters for the first test unit.

At this time, the MRI data obtained for the first test unit, the MRI for the first test unit synthesized using the ρ, T1, and T2 parameters and the first MRI sequence for the first test unit together Can be image or k-space data

Example 2-MRI data conversion

6 is a flowchart illustrating a method of converting MRI data according to an embodiment of the present invention.

The MRI data conversion method may convert MRI data using a data conversion device including a network unit and a conversion unit. The output layer of the network unit may be configured to output output layer data having the same format as ρ, T1, and T2 parameters for an arbitrary test unit.

In step S110, when the input layer data including the network unit, the information specifying the MRI sequence in the input layer of the network unit, and the source data that is the MRI data obtained for the test unit using the MRI sequence are input, , The output layer data can be output.

Then, in step S120, the conversion unit may generate the combined MRI data for the test unit by using the output layer data output from the output layer of the network unit and an arbitrary MRI sequence together.

By using the above-described embodiments of the present invention, those who belong to the technical field of the present invention will be able to easily implement various changes and modifications without departing from the essential characteristics of the present invention. The contents of each claim of the claims can be combined with other claims without citation within the scope understood through the present specification.

Claims (19)

Network unit 10; And
Converter 20;
It includes,
Input layer data (3) including information (3 _k1 ) for specifying an MRI sequence, and source data (3 _k2 ), which is MRI data obtained for a subject using the MRI sequence, in the input layer 110 of the network unit _k ) is input,
The output layer 120 of the network unit is configured to output the output layer data 40 including the physical parameter set of the test unit,
The conversion unit is configured to generate the synthesized MRI data 50 for the subject using the physical parameter set,
Data converter. The data conversion apparatus of claim 1, wherein the conversion unit is configured to generate the synthesized MRI data using the physical parameter set and an arbitrary MRI sequence together. The data conversion apparatus according to claim 1, wherein the network unit is machine-learned using a set of input layer data for learning and a set of output layer data for learning. The data conversion apparatus of claim 1, wherein the information specifying the MRI sequence is a set of operational parameters specifying the MRI sequence, and includes the MRI sequence and scan parameters used in the MRI sequence. The data conversion device according to claim 1, wherein the subject is a specific measurement target slice of an MRI subject. The data conversion apparatus of claim 1, wherein the source data is MRI image or k-space data. The data conversion apparatus of claim 1, wherein the physical parameter set includes at least one or all of ρ, T1, and T2. The data conversion apparatus of claim 1, wherein the physical parameter set includes all of ρ, T1, and T2. The data conversion apparatus according to claim 1, wherein the network unit is a convolutional neural network using an automatic encoder. The data conversion apparatus of claim 1, wherein the synthesized MRI data is an MRI image or k-space data. The data conversion apparatus of claim 1, wherein the synthesized MRI data is data in the same format as the source data. According to claim 3,
The machine learning is by supervised learning,
Any first output layer data for learning among the set of output layer data for learning is the set of physical parameters related to an arbitrary first test unit,
The first learning input layer data corresponding to the first learning output layer data among the set of input layer data for learning includes MRI sequence information and MRI acquired for the first test unit using the MRI sequence. Including source data that is data,
Data converter. The method of claim 12, wherein the MRI data obtained for the first test unit is MRI image or k-space data generated by an MRI signal obtained from the first test unit by the MRI scanner using the MRI sequence. Data converter. The MRI of the first test unit synthesized by using the ρ, T1, and T2 parameters and the MRI sequence of the first test unit together with the MRI data obtained for the first test unit. Data converter, which is image or k-space data. The method of claim 12,
The input layer of the network unit,
Information for specifying the first type of MRI sequence (3 ₁₁ ), and first source data (3 ₁₂ ), which is MRI data obtained for the first subject using the first type of MRI sequence, is input. 1 part (110 ₁ ), and
Information for specifying the MRI sequence of the second type (3 ₂₁ ), and the second source data (3 ₂₂ ), which is the MRI data obtained for the first subject using the MRI sequence of the second type, is input. 2 parts (110 ₂ )
Containing,
Data converter. The method of claim 15,
In the first portion of the input layer of the network unit is completed learning,
Information for specifying the MRI sequence of the first type, and third source data that is MRI data obtained for an arbitrary subject using the first type of MRI sequence are input,
Data converter. A data conversion apparatus comprising a network portion including an input layer and an output layer, the method comprising: a supervised learning method of a neural network for supervising and learning the network portion;
The input layer is configured to receive information specifying an arbitrary MRI sequence and source data, which is MRI data obtained for an arbitrary subject using the arbitrary MRI sequence,
The output layer is configured to output output layer data having the same format as ρ, T1, and T2 parameters for an arbitrary test unit,
Providing, to the input layer, information specifying a first MRI sequence and first source data that is MRI data obtained for a first subject using the first MRI sequence; And
Updating a weight assigned in the network unit to reduce an error between the output layer data output from the output layer and ρ, T1, and T2 parameters for the first test unit;
Containing,
Method of supervised learning of neural networks. 18. The method of claim 17, wherein the MRI data obtained for the first test section is synthesized by using the first MRI sequence together with ρ, T1, and T2 parameters related to the first test section. A method of supervised learning of neural networks, which is an MRI image of wealth or k-space data. A MRI data conversion method for converting MRI data using a data conversion device including a network unit and a conversion unit,
The output layer of the network unit is configured to output output layer data having the same format as ρ, T1, and T2 parameters for an arbitrary test unit,
The network unit outputs the output layer data when input layer data including information specifying an MRI sequence on the input layer of the network unit and source data that is MRI data obtained for the test unit using the MRI sequence is input. To do; And
Generating, by the conversion unit, MRI data synthesized for the test unit by using the output layer data output from the output layer of the network unit and an arbitrary MRI sequence together;
Containing,
MRI data conversion method.

Images