JP7407618B2 - information processing system - Google Patents

Description

The present invention relates to an information processing system.

Conventionally, screening using MRI (Magnetic Resonance Imaging) has been performed (for example, see Patent Document 1).
For example, women with BRCA1/2 gene mutations are said to have a high risk of developing breast cancer. For this reason, breast MRI screening is recommended for breast cancer screening for high-risk groups of breast cancer. In this breast MRI, a contrast agent is used for a subject (subject) for breast cancer screening.

JP2006-247113A

However, for those undergoing breast cancer screening, using a contrast agent every year when breast MRI is performed has disadvantages such as allergies, brain deposition of gadolinium, financial burden, and time burden.
Therefore, if similar findings can be obtained by simple MRI without using a contrast agent, it is estimated that the burden on not only breast cancer screening candidates but also medical institutions will be reduced.
In other words, there is a demand for the establishment of a technology that can obtain findings similar to those obtained using a contrast medium even with simple MRI without using a contrast medium. This is a situation that cannot be met with technology.

The present invention has been made in view of this situation, and aims to establish a technique that allows simple MRI without using a contrast agent to obtain findings similar to those obtained using a contrast agent.

To achieve the above object, an information processing system according to one embodiment of the present invention includes:
An information processing system including a first information processing device that generates or updates a model by machine learning, and a second information processing device that executes information processing using the model,
The first information processing device includes:
Type 1 image data obtained as a result of actually performing MRI on an object without using a contrast agent, and type 1 image data obtained as a result of actually performing MRI on the same object using a contrast agent. By performing predetermined machine learning using a set with the second type image data as learning data, when the first type image data is input, a third type image equivalent to the second type image data is generated. Model generation/update means for generating or updating a model that outputs data;
The second information processing device includes:
acquisition means for acquiring the first type image data to be processed;
generation means for generating the third type image data based on the first type image data acquired by the acquisition means and the model generated by the first information processing device;
Equipped with.

According to the present invention, it is possible to obtain findings similar to those obtained using a contrast agent even by simple MRI without using a contrast agent.

1 is a diagram showing an example of the configuration of an information processing system according to a first embodiment of the present invention. 2 is a block diagram showing an example of the hardware configuration of a learning device according to a first embodiment of the information processing device of the present invention in the information processing system of FIG. 1. FIG. 2 is a functional block diagram showing an example of the functional configuration of the information processing system of FIG. 1 of the first embodiment, that is, the learning device and contrast MRI device of FIG. 2. FIG. 4 is a diagram showing an example of a "contrast MRI image generation model" generated or updated by the learning device of FIG. 3. FIG. FIG. 4 is a diagram showing an overview of the model of FIG. 3 that is generated or updated using training data of the first embodiment. FIG. 2 is a functional block diagram showing an example of the functional configuration of the information processing system of FIG. 1 of the second embodiment, that is, the learning device and contrast MRI device of FIG. 2; FIG. 4 is a diagram showing an overview of the model in FIG. 3 that is generated or updated using training data according to the second embodiment. FIG. 7 is a diagram showing evaluation results of contrast-enhanced MRI image equivalent data generated by the contrast-enhanced MRI generating apparatus of FIG. 6 of the second embodiment. 4 is a diagram showing an example of contrast-enhanced MRI image data generated by the contrast-enhanced MRI generating apparatus of FIG. 3 of the first embodiment. FIG.

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a diagram showing an example of the configuration of an information processing system according to a first embodiment of the present invention.
The information processing system shown in FIG. 1 is configured to include a learning device 1, a contrast MRI generating device 2, and a model DB 3.

The learning device 1 performs machine learning on data of breast MRI images performed during breast cancer screening.

Specifically, for example, a combination of MRI image data actually taken without using a contrast agent and MRI image data actually taken with a contrast agent for a breast cancer patient. is input to the learning device 1 as learning data.
Hereinafter, among the learning data, data of MRI images actually captured without using a contrast agent will be referred to as "learning MRI image data." Further, among the learning data, data of an MRI image actually captured using a contrast agent is referred to as "contrast MRI image data for learning."

The learning device 1 can generate the following model by performing machine learning using a large amount of learning data.
Here, data of an MRI image taken of a person who underwent breast MRI screening without using a contrast medium for breast cancer screening is referred to as "MRI image data." When this MRI image data is input, a model that outputs image data equivalent to an MRI image captured using a contrast agent (hereinafter referred to as "contrast MRI image equivalent data") is created in the learning device. 1 is generated or updated. Note that such a model will hereinafter be referred to as a "contrast MRI image generation model."
The "contrast MRI image generation model" generated by the learning device 1 is stored in the model DB3.
Details of the functional configuration and processing of the learning device 1 will be described later with reference to the drawings from FIG. 3 onwards.

The contrast-enhanced MRI generation device 2 acquires the "contrast-enhanced MRI image generation model" from the model DB3. The contrast-enhanced MRI generation device 2 receives MRI image data and generates contrast-enhanced MRI image equivalent data using a "contrast MRI image generation model." The contrast-enhanced MRI generation device 2 outputs this contrast-enhanced MRI image equivalent data as output data.
The functional configuration and processing details of the contrast MRI generating device 2 will be described later with reference to the drawings from FIG. 3 onwards.

The contrast-enhanced MRI image equivalent data output as output data from the contrast-enhanced MRI generator 2 is used as feedback data (hereinafter referred to as "FB data") in the learning device along with the MRI image data input to the contrast-enhanced MRI generator 2. 1 is input.
The learning device 1 updates the "contrast MRI image generation model" by performing relearning using the FB data, and stores it in the model DB3.

FIG. 2 is a block diagram showing an example of the hardware configuration of a learning device according to the first embodiment of the information processing device of the present invention in the information processing system of FIG.

The learning device 1 includes a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, a bus 14, an input/output interface 15, an output section 16, and an input section 17. , a storage section 18 , a communication section 19 , and a drive 20 .

The CPU 11 executes various processes according to programs recorded in the ROM 12 or programs loaded into the RAM 13 from the storage section 18 .
The RAM 13 also appropriately stores data necessary for the CPU 11 to execute various processes.

The CPU 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input/output interface 15 is also connected to this bus 14 . An output section 16 , an input section 17 , a storage section 18 , a communication section 19 , and a drive 20 are connected to the input/output interface 15 .
The output unit 16 is composed of a display, a speaker, etc., and outputs images and sounds.
The input unit 17 is composed of a keyboard, a mouse, etc., and inputs various information according to user's instruction operations.
The storage unit 18 is composed of a hard disk or the like, and stores data of various information.

The communication unit 19 controls communication with other objects (for example, the model DB 3 in FIG. 1) via the network.
A removable medium 30, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is appropriately installed in the drive 20. Programs and image data read from the removable medium 30 by the drive 20 are installed in the storage unit 18 as necessary. Further, the removable medium 30 can also store various data (for example, image data, etc.) stored in the storage section 18 in the same way as the storage section 18 .

Although not shown, the contrast MRI generating apparatus 2 of the information processing system in FIG. 1 has basically the same hardware configuration as the one shown in FIG. 2. Therefore, a description of the hardware configuration of the contrast MRI generating apparatus 2 will be omitted.
Further, for convenience of explanation, the learning device 1 is provided separately from the contrast-enhanced MRI generating device 2; however, the present invention is not particularly limited to this, and each function of the learning device 1 and the contrast-enhanced MRI generating device 2 is integrated into one information It may be integrated into a processing device.

FIG. 3 is a functional block diagram showing an example of the functional configuration of the information processing system of FIG. 1 of the first embodiment, that is, the learning device and contrast MRI device of FIG. 2.

First, the functional configuration of the learning device 1 of the first embodiment will be explained.
In the learning device 1 of the first embodiment, an MRI image acquisition section 51, a contrast MRI image acquisition section 52, a color image generation section 53, a learning data generation section 54, a model generation update section 55, and an FB data acquisition section The section 56 functions.

The MRI image acquisition unit 51 acquires learning MRI image data about breast cancer patients.
The contrast-enhanced MRI image acquisition unit 52 acquires learning contrast-enhanced MRI image data for the same breast cancer patient.

Here, as an example, out of 26,819 contrast-enhanced MRI images of a case diagnosed with primary breast cancer, data on fat-suppressed T1-weighted images (hereinafter referred to as "T1 data") and data on fat-suppressed T2 images (hereinafter referred to as " 3040 images (hereinafter referred to as "T2 data"), diffusion weighted image DWI data (hereinafter referred to as "DWI data"), and contrast-enhanced early phase image (early phase) data in which the cut planes all coincided. 760 slices) is input to the learning device 1 as learning data. Here, matching of the cut planes means that the deviation width is 1 mm or less.
That is, it is assumed that T1 data, T2 data, DWI data, and early phase data, all of which have matching cut planes, are used as a set of learning data. In this case, among the set of learning data, T1 data, T2 data, and DWI data are acquired by the MRI image acquisition unit 51 as learning MRI image data. On the other hand, among the set of learning data, early phase data is acquired by the contrast MRI image acquisition unit 52 as learning contrast MRI image data.
Note that the above 3040 images (760 slices) are hereinafter referred to as "learning data of this example" as learning data.

In this embodiment, data of contrast-enhanced early phase images (early phase) was used since it is applied to the detection of breast cancer. However, for MRI images after anticancer drugs are administered before surgery, delayed phase data is used. used.
That is, in this example, the early phase is used for convenience of explanation, but in reality, it is not limited to the early phase, and it is sufficient to use contrast image data.

The color image generation unit 53 will be described later, but the learning data generation unit 54 generates training data (in the case of learning with a teacher image, a set with the teacher data) as learning data based on the learning data of this example. generate.
In this embodiment, as will be described later, a convolutional neural network is employed as the "contrast MRI image generation model." Therefore, the data generated by the color image generation unit 53, which will be described later, based on the learning MRI image data (T1 data, T2 data, and DWI data) is used as the training data, which is the learning contrast-enhanced MRI image data (early phase data). are respectively generated as teacher data by the learning data generation unit 54.

The model generation/update unit 55 generates or updates a “contrast MRI image generation model” by performing machine learning using the learning data output from the learning data generation unit 54 .

Here, it is sufficient that the "contrast MRI image generation model" can be generated or updated by any method of machine learning using some learning data (supervised or unsupervised); however, in this embodiment, It is assumed that model M shown in FIG. 4 is adopted.

FIG. 4 shows an example of a "contrast MRI image generation model" generated or updated by the learning device of FIG. 3.
Model M shown in Fig. 4 uses the artificial intelligence Convolutional Autoencoder method to learn to estimate contrast-enhanced MRI image equivalent data (early phase data) from T1 data, T2 data, and DWI data (learning data). This is an example of a "contrast-enhanced MRI image generation model" that is reproduced or updated as a result.
In this model M, after image data is compressed using a normal convolutional neural network, the size of the image data is returned to the original size.
The model generation/updating unit 55 extracts information that captures the features within the image so that the image data obtained by correcting the original image data (learning data) is output as contrast-enhanced MRI image equivalent data (early phase data). and execute machine learning.
Here, the algorithm used in machine learning by the model generation/updating unit 55 is not particularly limited, but in this embodiment, one based on U-Net, which is a typical algorithm, is adopted.
Furthermore, the model generation/updating unit 55 uses the least squares method (MSE) to evaluate the similarity of images to perform machine learning so that the training data approaches early phase data.
Furthermore, as a method for evaluating the degree of image deterioration, this embodiment employs a method of measuring the peak signal-to-noise ratio (PSNR) (see FIG. 8, which will be described later).

As described above, training data used for machine learning to generate or update the model M shown in FIG. 4 as a "contrast MRI image generation model" is T1 data, T2 data, and DWI data.
Here, each of the T1 data, T2 data, and DWI data can be used as separate image data, and the three image data can be used as training data.
Here, general color image data has a configuration in which each pixel has pixel values of three elements, for example, a configuration in which each pixel has RGB (red, green, blue) pixel values.
On the other hand, each of the MRI image data, that is, T1 data, T2 data, and DWI data, is configured as grayscale image data, that is, each pixel has one element of pixel value (luminance value). There is.
Therefore, the learning device 1 of the first embodiment is provided with the color image generation section 53 shown in FIG. 3. That is, the color image generation unit 53 converts each pixel value of T1 data, T2 data, and DWI data that constitute the learning MRI image data into each pixel value of three elements of color image data (for example, R, By assigning the data to each of the G and B pixel values, data of one temporary color image is generated as training data.
That is, in the first embodiment, the model M in FIG. 3 is generated or updated using temporary color image data in which T1, T2, and DWI are assigned to each of R, G, and B as training data. Machine learning is being performed.

FIG. 5 shows an overview of the model of FIG. 3 that is generated or updated using the training data of the first embodiment.
Although the model M in FIG. 3 is employed as a "contrast MRI image generation model" in the second embodiment described later, the training data is different between the first embodiment and the second embodiment.
Therefore, the training data of the first embodiment, that is, the temporary color image data in which T1, T2, and DWI are assigned to each of R, G, and B, is used to generate or update the image shown in FIG. As shown in FIG. 5, the model M is called "model M1" (to distinguish it from that of the second embodiment described later).
That is, as shown in FIG. 5, each pixel value of input MRI image data, that is, T1 data IT1, T2 data IT2, and DWI data IDWI, is changed to each pixel value of three elements of color image data ( For example, each pixel value of R, G, and B is assigned to each pixel value, thereby generating data for one temporary color image. In the first embodiment, a model M1 that outputs contrast-enhanced MRI image equivalent data Out when data of this temporary color image is input is employed as a "contrast-enhanced MRI image generation model."

Returning to FIG. 3, the FB data acquisition unit 56 acquires the above-mentioned FB data.
The model generation/update unit 55 can update the "contrast MRI image generation model" by performing machine learning (relearning) using this FB data.
Note that during this relearning, each pixel value of the MRI image data constituting the FB data, that is, T1 data IT1, T2 data IT2, and DWI data IDWI, is changed to each pixel value of the three elements of color image data. By assigning each value (for example, each pixel value of R, G, and B), data of one temporary color image is generated.
Learning using four image data of this provisional color image data and contrast-enhanced MRI image data (early phase data in this example), or provisional color image data and four contrast-enhanced MRI image equivalent data. Re-learning is performed using the image data of 2 images.

The functional configuration of the learning device 1 of the first embodiment has been described above.
Next, the functional configuration of the contrast MRI generating apparatus 2 of the first embodiment will be described.

In the contrast-enhanced MRI generation device 2 of the first embodiment, an MRI image acquisition section 61, a color image generation section 62, and a contrast-enhanced MRI generation section 63 function.

In this embodiment, MRI image data (T1 data, T2 data, and DWI data) of a person who underwent breast MRI screening without using a contrast agent for breast cancer screening is input to the contrast MRI generation device 2. Ru.
The MRI image acquisition unit 61 acquires the MRI image data.

The color image generation unit 62 converts each pixel value of T1 data, T2 data, and DWI data constituting the MRI image data into pixel values of three elements of color image data (for example, R, G, and B). By assigning the color image data to each pixel value (each pixel value), one temporary color image data (hereinafter referred to as "color MRI image data") is generated.
That is, in the first embodiment, temporary color image data in which T1, T2, and DWI are assigned to each of R, G, and B is color MRI image data. This color MRI image data is used as the source data for generating contrast-enhanced MRI image data, that is, as input data for a "contrast-enhanced MRI image generation model."

The contrast-enhanced MRI generation unit 63 acquires the model M1 in FIG. 5 as a "contrast-enhanced MRI image generation model." Further, the contrast MRI generation unit 63 acquires the color MRI image data generated by the color image generation unit 62 as input data to the “contrast MRI image generation model”. Then, the contrast-enhanced MRI generation unit 63 inputs this color MRI image data into a "contrast-enhanced MRI image generation model" to generate contrast-enhanced MRI image equivalent data and outputs it.

The information processing system according to the first embodiment of the present invention has been described above with reference to FIGS. 1 to 5.
Next, an information processing system according to a second embodiment of the present invention will be described with reference to the drawings from FIG. 6 onwards.

The hardware configuration of the information processing system according to the second embodiment is the same as that according to the first embodiment.
That is, the information processing system according to the second embodiment can have a configuration similar to that in FIG. 1.
Further, each of the learning device 1 and the contrast MRI generating device 2 that constitute the information processing system of the second embodiment can have the same configuration as in FIG. 2 .
Here, in the second embodiment, as in the first embodiment, for convenience of explanation, the learning device 1 is provided separately from the contrast-enhanced MRI generating device 2, but the learning device 1 and the contrast-enhanced MRI generator 2 are not particularly limited to this. Each function of the MRI generation device 2 may be integrated into one information processing device.
Based on the above, description of the hardware configuration of the information processing system according to the second embodiment will be omitted.

FIG. 6 is a functional block diagram showing an example of the functional configuration of the information processing system of FIG. 1 of the second embodiment, that is, the learning device and contrast MRI device of FIG. 2.

First, the functional configuration of the learning device 1 of the second embodiment will be explained.
In the learning device 1 of the second embodiment, an MRI image acquisition section 151, a color image generation section 153, a learning data generation section 154, a model generation/update section 155, and an FB data acquisition section 156 function.

The MRI image acquisition unit 151, color image generation unit 153, and learning data generation unit 154 to FB data acquisition unit 156 in the second embodiment are respectively the MRI image acquisition unit 51, color image generation unit, and color image generation unit in the first embodiment. 53 and the learning data generation section 54 to the FB data acquisition section 56, respectively, and have basically the same functions and configurations.
That is, the difference between the first embodiment and the second embodiment regarding the functional configuration of the learning device 1 is that the contrast MRI image acquisition unit 52 is provided in the first embodiment, whereas the contrast MRI image acquisition unit 52 is provided in the second embodiment. However, such a functional block is not provided.
In other words, the difference between the first embodiment and the second embodiment is that the format of learning data used for machine learning for generating or updating the "contrast MRI image generation model" is different.
That is, during machine learning in the first embodiment, MRI image data and contrast-enhanced MRI image data were used as learning data.
On the other hand, during machine learning in the second embodiment, MRI image data and contrast-enhanced MRI equivalent image data are used as learning data.

Therefore, in the following explanation of the functional configuration of the learning device 1 of the second embodiment, only the differences from the first embodiment, that is, the points related to the format of training data, will be explained, and the rest will be omitted. .

FIG. 7 shows an overview of the model of FIG. 5 that is generated or updated using the training data of the second embodiment.
Note that the model M in FIG. 5 that is generated or updated using the training data of the second embodiment is referred to as a "model M2" as shown in FIG. 7 (to distinguish it from that of the first embodiment described above).
In the second embodiment, the color image generation unit 153 converts each pixel value of the MRI image data (T1 data IT1, T2 data IT2, and DWI data IDWI) acquired by the MRI image acquisition unit 151 into a color image. One sheet of color MRI image data is generated by assigning each pixel value of three elements of data (for example, each pixel value of R, G, and B).
The model generation/updating unit 155 executes machine learning using the color MRI image data and the contrast-enhanced MRI image equivalent data Out obtained from the color MRI image data. As a result, when MRI image data of a person who was screened by breast MRI without using a contrast agent for breast cancer screening is given, when the MRI image data is input, T1 data IT1, T2 data IT2 and DWI data IDWI are used, and contrast MRI image equivalent data Out is output. As a result, model M2 is generated or updated as a "contrast MRI image generation model."

Hereinafter, the model M2 of the second embodiment as a "contrast MRI image generation model" will be further explained from the perspective of comparison with the model M1 of the first embodiment (FIG. 5).
The contrast-enhanced MRI image equivalent data as the output data of the model M2 of the second embodiment is higher resolution data when compared with the contrast-enhanced MRI image equivalent data as the output data of the model M1 of the first embodiment.
This is due to the following reason. That is, the early phase data is image data with a contrast effect added to the T1 data IT1. Taking this into consideration, the "contrast MRI image generation model" obtained by creating a shortcut to connect the T1 data IT1 from the input layer to the output layer is the model M2 of the second embodiment. Such shortcuts will result in higher resolution.
That is, the model generation/update unit 155 generates or updates the model M2 of the second embodiment by executing machine learning that learns only the contrast effect, which is the difference, instead of the image data itself.

The functional configuration of the learning device 1 of the second embodiment has been described above.
Next, the functional configuration of the contrast MRI generating apparatus 2 of the second embodiment will be described.

In the contrast-enhanced MRI generation device 2 of the second embodiment, as shown in FIG. 6, an MRI image acquisition section 161, a color image generation section 162, and a contrast-enhanced MRI generation section 163 function.

In the second embodiment, similarly to the first embodiment, MRI image data (T1 data, T2 data, and DWI data) of breast cancer screening performed by breast MRI without using a contrast agent is , is input to the contrast MRI generating device 2.
The MRI image acquisition unit 161 acquires the MRI image data.

The color image generation unit 162 converts each pixel value of the T1 data, T2 data, and DWI data that constitute the MRI image data acquired by the MRI image acquisition unit 161 into each pixel of three elements of the color image data. By assigning each value (for example, each pixel value of R, G, and B), one sheet of color MRI image data is generated.

The contrast-enhanced MRI generation unit 163 acquires the model M2 in FIG. 7 as a "contrast-enhanced MRI image generation model." Further, the contrast MRI generation unit 163 obtains the color MRI image data generated by the color image generation unit 162 as input data to the “contrast MRI image generation model”. Then, the contrast-enhanced MRI generation unit 163 outputs contrast-enhanced MRI image equivalent data by inputting this color MRI image data into the "contrast-enhanced MRI image generation model."

FIG. 8 shows evaluation results of contrast-enhanced MRI image equivalent data generated by the contrast-enhanced MRI generating apparatus of the second embodiment.
FIG. 8(A) shows the result of MSE (Mean Squared Error) for the contrast-enhanced MRI image equivalent data generated by the contrast-enhanced MRI generating apparatus 2 of the second embodiment. Ideally, the value of MSE is 0, and the smaller the value, the better. It can be seen that the MSE value obtained as the evaluation result has decreased to 264, which is a suitable value.
FIG. 8(B) shows the PSNR (Peak Signal-to-Noise Ratio) of the contrast-enhanced MRI image equivalent data generated by the contrast-enhanced MRI generating apparatus 2 of the second embodiment. The standard PSNR value is 30 to 50 dB, and 50 to 65 dB is said to be good. It can be seen that the PSNR value obtained as the evaluation result is 55.1 dB, which is good.

FIG. 9 shows an example of contrast-enhanced MRI image data generated by the contrast-enhanced MRI generating apparatus of the first embodiment.
FIG. 9(A) shows T1 data IT1 (called "T1-weighted image" in FIG. 9(A)) obtained from MRI image data (image data obtained as a result of MRI performed without using a contrast agent). display).
FIG. 9B shows contrast-enhanced MRI image equivalent data (FIG. 9B ) is displayed as "generated image").
FIG. 9(C) is data of an image actually captured when MRI was performed using a contrast agent (indicated as "contrast image (early phase)" in FIG. 9(C)).
The contrast-enhanced MRI image equivalent data (displayed as "generated image" in FIG. 9(B)) obtained by model M2 in FIG. 9(B) includes the actual image data in FIG. When compared with the contrast-enhanced image (early phase) in ), it can be seen that the contrast-enhanced MRI generating apparatus 2 of the first embodiment has obtained an image that emphasizes only the breast cancer part like the actual early stage layer.

By applying the information processing system of the first embodiment or the second embodiment as described above, the following effects can be achieved.
First, the effects on the subject (subject) of breast cancer screening include, for example, the following first to fourth effects. That is, the first effect is that allergy caused by the contrast medium can be avoided. The second effect is that side effects caused by gadolinium contrast agent deposition in the brain can be avoided. The third effect is that the economic burden and time burden can be reduced. The fourth effect is that inspection opportunities are increased due to the simplification of inspection.
Next, the effects on the medical institution side include, for example, the following fifth to ninth effects. The fifth effect is that the risk of emergency response to allergies can be avoided. The sixth effect is that equipment resources can be reallocated to other subjects by shortening the testing time, and revenue can be increased by increasing the number of tests. The seventh effect is that the weight of the image data stored in the PACS server can be reduced and the cost can be reduced.
Examples of effects in terms of impact on the field include the following eighth and ninth effects. The eighth effect is that new equipment is not required and introduction is easy. The ninth effect is that it can be applied not only to breast cancer screening but also to all MRI examinations.

Although one embodiment of the present invention has been described above, the present invention is not limited to the above-described embodiment, and modifications, improvements, etc. are included in the present invention as long as the purpose of the present invention can be achieved. It is.

For example, in the embodiment described above, the model obtained by machine learning was to generate contrast-enhanced MRI image equivalent data from MRI image data (obtained without using a contrast agent) regarding breast cancer; Not limited.
For example, when inspecting cells, we sometimes use image data with colored cells, but we use machine learning to create a model that generates image data equivalent to colored cells from image data without colored cells. Or you may update it.
For example, there are objects that can be inspected both destructively and non-destructively, and a model that generates image data equivalent to that obtained by destructive inspection from image data obtained by non-destructive inspection. It may be generated or updated by machine learning.

Further, for example, the model that can obtain high resolution by shortcuts is the model M2 of the second embodiment in the above-described embodiment, but it can also be the model M1.

Further, for example, the series of processes described above can be executed by hardware or by software.
In other words, the functional configurations in FIGS. 3 and 6 are merely examples and are not particularly limited.
In other words, it is sufficient that the information processing system is equipped with a function that can execute the above-mentioned series of processes as a whole, and what kind of functional blocks are used to realize this function is particularly explained in the examples of FIGS. 3 and 6. Not limited. Furthermore, the locations of the functional blocks are not particularly limited to those shown in FIGS. 3 and 6, and may be arbitrary.

Further, for example, when a series of processes is executed by software, a program constituting the software is installed on a computer or the like from a network or a recording medium.
The computer may be a computer built into dedicated hardware.
Further, the computer may be a computer that can execute various functions by installing various programs, such as a server, a general-purpose smartphone, or a personal computer.

For example, a recording medium containing such a program may not only be a removable medium (not shown) distributed separately from the device itself to provide the program to the user, but also a recording medium that is pre-installed in the device body. It consists of recording media etc. provided to users.

Note that in this specification, the term "system" refers to an overall device composed of a plurality of devices, a plurality of means, and the like.

In other words, the information processing system to which the present invention is applied can take various embodiments having the following configurations.
That is, the information processing system to which the present invention is applied,
A first information processing device that generates or updates a model by machine learning (for example, the learning device 1 in FIG. 1), and a second information processing device that executes information processing using the model (for example, the contrast MRI generation device 2 in FIG. 1). ), the information processing system comprising:
The first information processing device includes:
Type 1 image data obtained as a result of actually performing MRI on an object that does not use a contrast agent (for example, MRI image data for learning), and type 1 image data obtained as a result of actually performing MRI on an object that does not use a contrast agent, and type 1 image data that is actually obtained on the same object that uses a contrast agent. By performing predetermined machine learning using a set with type 2 image data (for example, contrast-enhanced MRI image data for learning) obtained as a result of MRI as learning data, the type 1 image data is input. Model generation/update means (for example, model generation/update unit 55 in FIG. 3 or Model generation update unit 55 in FIG. 6)
Equipped with
The second information processing device includes:
an acquisition unit (for example, the MRI image acquisition unit 61 in FIG. 3 or the MRI image acquisition unit 161 in FIG. 6) that acquires the first type image data to be processed;
generating the third type image data (for example, contrast-enhanced MRI image equivalent data) based on the first type image data acquired by the acquisition means and the model generated by the first information processing device; means and
Equipped with.
As a result, it becomes possible to easily establish a technique that allows simple MRI without using a contrast agent to obtain findings similar to those obtained using a contrast agent.

DESCRIPTION OF SYMBOLS 1... Learning device, 2... Contrast MRI generation device, 3... Model DB, 11... CPU, 51... MRI image acquisition part, 52... Contrast MRI image acquisition part, 53. ...Color image generation unit, 54...Learning data generation unit, 55...Model generation update unit, 56...FB data acquisition unit, 61...MRI image acquisition unit, 62...Color image generation 63... Contrast MRI generation unit, 151... MRI image acquisition unit, 153... Color image generation unit, 154... Learning data generation unit, 155... Model generation update unit, 156...・FB data acquisition unit, 161... MRI image acquisition unit, 162... Color image generation unit, 163... Contrast MRI generation unit

Claims (1)

An information processing system including a first information processing device that generates or updates a model by machine learning, and a second information processing device that executes information processing using the model,
The first information processing device includes:
A first image containing T1 data of a fat-suppressed T1-weighted image, T2 data of a fat-suppressed T2 image, and DWI data of a diffusion-weighted image DWI obtained as a result of actually performing MRI on an object that does not use a contrast agent. By executing a predetermined machine learning according to the Convolutional Autoencoder method using the seed image data as learning data, when the first type image data is input , the same object using a contrast agent is detected. The third type image data is equivalent to the second type image data obtained as a result of actually performing MRI on an object, and is image data with a contrast effect added to the T1 data. a model that outputs the T1 data, the model generating and updating means generating or updating a model obtained as a result of the machine learning of only the contrast effect by creating a shortcut to connect the T1 data from the input layer to the output layer. ,
The second information processing device includes:
acquisition means for acquiring the first type image data to be processed;
generation means for generating the third type image data based on the first type image data acquired by the acquisition means and the model generated by the first information processing device;
An information processing system equipped with.

Images