JP7355532B2 - Medical information processing device, method and program - Google Patents

Description

Embodiments of the present invention relate to a medical information processing device, method, and program.

In machine learning using medical data such as medical image data and its raw data, a deep neural network (DNN) that is trained from a large amount of training data is used to restore the original data from partially missing medical data. ). For example, in magnetic resonance imaging (MRI), DNN is applied to k-space data that has been undersampled by Cartesian acquisition to generate k-space data with missing parts restored, and the restored k-space data is There is a method to obtain reconstructed images based on There is also a method of directly obtaining a restored image by applying DNN to undersampled k-space data.

US Patent Application Publication No. 2015/0276953 US Patent No. 9672617 US Patent No. 9668699

The problem to be solved by the present invention is to improve the accuracy of output data using machine learning .

The medical information processing device according to the embodiment includes: a first acquisition unit that acquires medical data regarding a subject; a second acquisition unit that acquires numerical data that digitizes the collection conditions of the medical data; and a generation unit that applies a machine learning model to input data including medical data to generate output data based on the medical data.

FIG. 1 is a diagram showing the configuration of a magnetic resonance imaging apparatus that implements a medical information processing apparatus according to this embodiment. FIG. 2 is a diagram showing a typical flow of MR imaging by the processing circuit of FIG. FIG. 3 is a diagram schematically showing the mask data generation process in steps S2 and S3 of FIG. FIG. 4 is a diagram schematically showing the flow of the first DNN reconfiguration according to this embodiment. FIG. 5 is a diagram schematically showing the flow of the second DNN reconfiguration according to this embodiment. FIG. 6 is a diagram schematically showing the flow of the third DNN reconfiguration according to this embodiment. FIG. 7 is a diagram showing the configuration of the model learning device according to this embodiment. FIG. 8 is a diagram schematically showing processing by the learning sample generation function of FIG. 7. FIG. 9 is a diagram showing another configuration of the medical information processing apparatus according to this embodiment. FIG. 10 is a diagram schematically showing numerical data according to this embodiment. FIG. 11 is a diagram schematically showing input of the numerical data of FIG. 10 to the machine learning model. FIG. 12 is a diagram schematically showing input of other numerical data to the machine learning model. FIG. 13 is a diagram schematically showing input of other numerical data to the machine learning model. FIG. 14 is a diagram schematically showing other numerical data. FIG. 15 is a diagram schematically showing the network structure of the machine learning model according to this embodiment. FIG. 16 is a diagram schematically showing the network structure of another machine learning model according to this embodiment. FIG. 17 is a diagram schematically showing input and output of a machine learning model having multiple CNNs.

Hereinafter, a medical information processing apparatus, method, and program according to the present embodiment will be described with reference to the drawings.

The medical information processing device according to this embodiment is a device that is equipped with a processing circuit that processes medical information. The medical information processing apparatus according to this embodiment is realized by, for example, a computer installed in a medical image diagnostic apparatus. The medical image diagnostic apparatus according to this embodiment includes a magnetic resonance imaging apparatus (MRI apparatus), an X-ray computed tomography apparatus (CT apparatus), an X-ray diagnostic apparatus, a PET (Positron Emission Tomography) apparatus, and a SPECT apparatus (Single Photon It may be a single modality device such as an ultrasonic diagnostic device or an ultrasonic diagnostic device, or a multimodality device such as a PET/CT device, a SPECT/CT device, a PET/MRI device, or a SPECT/MRI device. good. As another example, the medical information processing apparatus according to the present embodiment may be a computer communicably connected to a medical image diagnostic apparatus via a cable, a network, etc., or may be independent of the medical image diagnostic apparatus. It may be a computer. Hereinafter, it is assumed that the medical information processing apparatus according to this embodiment is implemented in a magnetic resonance imaging apparatus.

FIG. 1 is a diagram showing the configuration of a magnetic resonance imaging apparatus 1 that implements a medical information processing apparatus 50 according to the present embodiment. As shown in FIG. 1, the magnetic resonance imaging apparatus 1 includes a pedestal 11, a bed 13, a gradient magnetic field power supply 21, a transmitting circuit 23, a receiving circuit 25, a bed driving device 27, a sequence control circuit 29, and a medical information processing device 50. .

The pedestal 11 has a static magnetic field magnet 41 and a gradient magnetic field coil 43. The static magnetic field magnet 41 and the gradient magnetic field coil 43 are housed in the casing of the pedestal 11. A hollow bore is formed in the casing of the pedestal 11 . A transmitting coil 45 and a receiving coil 47 are arranged within the bore of the pedestal 11.

The static magnetic field magnet 41 has a hollow, substantially cylindrical shape, and generates a static magnetic field substantially inside the cylinder. As the static field magnet 41, for example, a permanent magnet, a superconducting magnet, a normal conducting magnet, or the like is used. Here, the central axis of the static magnetic field magnet 41 is defined as the Z-axis, the axis vertically orthogonal to the Z-axis is defined as the Y-axis, and the axis horizontally orthogonal to the Z-axis is defined as the X-axis. The X-axis, Y-axis, and Z-axis constitute an orthogonal three-dimensional coordinate system.

The gradient magnetic field coil 43 is a coil unit that is attached inside the static magnetic field magnet 41 and formed in a hollow, substantially cylindrical shape. The gradient magnetic field coil 43 receives current from the gradient magnetic field power supply 21 and generates a gradient magnetic field. More specifically, the gradient magnetic field coil 43 has three coils corresponding to the X-axis, Y-axis, and Z-axis that are orthogonal to each other. The three coils form a gradient magnetic field whose magnetic field strength changes along each of the X-axis, Y-axis, and Z-axis. The gradient magnetic fields along the X, Y, and Z axes are combined to form a mutually orthogonal slice selection gradient magnetic field Gs, phase encode gradient magnetic field Gp, and frequency encode gradient magnetic field Gr in a desired direction. The slice selection gradient magnetic field Gs is used to arbitrarily determine an imaging section. The phase encoding gradient magnetic field Gp is used to change the phase of the MR signal depending on the spatial position. The frequency encode gradient magnetic field Gr is used to change the frequency of the MR signal depending on the spatial position. In the following description, it is assumed that the gradient direction of the slice selection gradient magnetic field Gs is the Z axis, the gradient direction of the phase encode gradient magnetic field Gp is the Y axis, and the gradient direction of the frequency encode gradient magnetic field Gr is the X axis.

The gradient magnetic field power supply 21 supplies current to the gradient magnetic field coil 43 according to a sequence control signal from the sequence control circuit 29 . The gradient magnetic field power supply 21 supplies current to the gradient magnetic field coils 43, thereby causing the gradient magnetic field coils 43 to generate gradient magnetic fields along the X, Y, and Z axes. The gradient magnetic field is applied to the subject P while being superimposed on the static magnetic field formed by the static magnetic field magnet 41.

The transmitting coil 45 is disposed, for example, inside the gradient magnetic field coil 43, receives current from the transmitting circuit 23, and generates a high-frequency magnetic field pulse (hereinafter referred to as an RF magnetic field pulse).

The transmitting circuit 23 supplies current to the transmitting coil 45 in order to apply an RF magnetic field pulse to the subject P via the transmitting coil 45 for exciting target protons present in the subject P. The RF magnetic field pulse oscillates at a resonant frequency unique to the target protons and excites the target protons. A magnetic resonance signal (hereinafter referred to as an MR signal) is generated from the excited target protons and detected by the receiving coil 47. The transmitting coil 45 is, for example, a whole body coil (WB coil). Whole body coils may be used as transmit and receive coils.

The receiving coil 47 receives the MR signal emitted from the target protons present in the subject P under the action of the RF magnetic field pulse. The receiving coil 47 has a plurality of receiving coil elements capable of receiving MR signals. The received MR signal is supplied to the receiving circuit 25 via wire or wirelessly. Although not shown in FIG. 1, the receiving coil 47 has a plurality of receiving channels implemented in parallel. The reception channel includes a reception coil element that receives the MR signal, an amplifier that amplifies the MR signal, and the like. The MR signal is output for each receiving channel. The total number of reception channels and the total number of reception coil elements may be the same, or the total number of reception channels may be greater or less than the total number of reception coil elements.

The receiving circuit 25 receives the MR signal generated from the excited target protons via the receiving coil 47. The receiving circuit 25 processes the received MR signal to generate a digital MR signal. Digital MR signals can be expressed in k-space defined by spatial frequencies. Therefore, hereinafter, the digital MR signal will be referred to as k-space data. K-space data is a type of raw data used for image reconstruction. The k-space data is supplied to the medical information processing device 50 via wire or wireless.

Note that the above-mentioned transmitting coil 45 and receiving coil 47 are merely examples. Instead of the transmitting coil 45 and the receiving coil 47, a transmitting/receiving coil having a transmitting function and a receiving function may be used. Moreover, the transmitting coil 45, the receiving coil 47, and the transmitting/receiving coil may be combined.

A bed 13 is installed adjacent to the pedestal 11. The bed 13 has a top plate 131 and a base 133. A subject P is placed on the top plate 131. The base 133 supports the top plate 131 so as to be slidable along each of the X-axis, Y-axis, and Z-axis. The bed driving device 27 is accommodated in the base 133. The bed driving device 27 moves the top plate 131 under control from the sequence control circuit 29. The bed driving device 27 may include any motor such as a servo motor or a stepping motor.

The sequence control circuit 29 has, as hardware resources, a processor such as a CPU (Central Processing Unit) or an MPU (Micro Processing Unit), and a memory such as a ROM (Read Only Memory) or a RAM (Random Access Memory). The sequence control circuit 29 synchronously controls the gradient magnetic field power supply 21, the transmitting circuit 23, and the receiving circuit 25 based on the imaging protocol determined by the imaging protocol setting function 511 of the processing circuit 51, and generates pulses according to the imaging protocol. The sequence is executed to perform MR imaging of the subject P, and k-space data regarding the subject P is collected.

As shown in FIG. 1, the medical information processing device 50 is a computer device having a processing circuit 51, a memory 52, a display 53, an input interface 54, and a communication interface 55.

The processing circuit 51 includes a processor such as a CPU, a GPU (Graphics Processing Unit), or an MPU, and a memory such as a ROM or RAM as hardware resources. The processing circuit 51 functions as the core of the magnetic resonance imaging apparatus 1 . For example, the processing circuit 51 has an imaging protocol setting function 511, a mask data acquisition function 512, a raw data acquisition function 513, an image generation function 514, an image processing function 515, and a display control function 516 by executing various programs.

In the imaging protocol setting function 511, the processing circuit 51 automatically sets the imaging protocol for MR imaging of the target by a user instruction via the input interface 54 or automatically. An imaging protocol is a collection of various imaging parameters regarding one MR imaging. As the imaging parameters according to this embodiment, various imaging parameters that are directly or indirectly set for performing MR imaging, such as imaging time, type of k-space filling method, type of pulse sequence, TR, and TE, can be applied. be.

In the mask data acquisition function 512, the processing circuit 51 acquires mask data regarding MR imaging of the target. The mask data is data in which a numerical value corresponding to the number of times of acquisition and/or the direction of acquisition in object imaging is assigned to a plurality of data acquisition locus candidates having a finite number of elements. The data collection trajectory refers to the collection trajectory of raw data on k-space, and is related to the type of k-space filling method. For example, if the k-space filling method is a radial method, the data collection trajectory is a general line (spoke) passing through the center of k-space. When the k-space filling method is a spiral method or a variable density spiral method, the data collection trajectory is a spiral curve from the center of the k-space to the outer periphery. The acquisition of mask data includes processes such as generation of mask data by the processing circuit 51, selection of any one mask data from a plurality of mask data, and reception, transfer, or transmission of mask data from another device.

The raw data acquisition function 513 acquires k-space data. Acquisition of k-space data includes collection of k-space data by MR imaging performed under the control of the processing circuit 51, selection of any one k-space data from a plurality of k-space data, and acquisition of k-space data from another device. This shall include processing such as reception, transfer, or transmission of spatial data.

In the image generation function 514, the processing circuit 51 performs reconstruction processing using the machine learning model 521 on the mask data acquired by the mask data acquisition function 512 and the k-space data acquired by the raw data acquisition function 513. , generates an MR image regarding the subject P. Machine learning model 521 is stored in memory 52. Machine learning model 521 is a parameterized composite function defined by a combination of multiple adjustable functions and parameters (weighting matrices or biases). The machine learning model 521 is realized by a multilayer network model (DNN: Deep Neural Network) having an input layer, a hidden layer, and an output layer. Hereinafter, the reconstruction process using the machine learning model 521 will be referred to as DNN reconstruction.

In the image processing function 515, the processing circuit 51 performs various image processing on the MR image. For example, the processing circuit 51 performs image processing such as volume rendering, surface rendering, pixel value projection processing, MPR (Multi-Planer Reconstruction) processing, and CPR (Curved MPR) processing.

In the display control function 516, the processing circuit 51 displays various information on the display 53. For example, the processing circuit 51 displays an MR image generated by the image generation function 514, an MR image generated by the image processing function 515, an imaging protocol setting screen, etc. on the display 53.

The memory 52 is a storage device such as an HDD (Hard Disk Drive), an SSD (Solid State Drive), or an integrated circuit storage device that stores various information. Further, the memory 52 may be a drive device or the like that reads and writes various information to/from a portable storage medium such as a CD-ROM drive, a DVD drive, or a flash memory. For example, the memory 52 stores k-space data, control programs, machine learning models 521, and the like.

Display 53 displays various information. For example, the display 53 displays an MR image generated by the image generation function 514, an MR image generated by the image processing function 515, an imaging protocol setting screen, and the like. As the display 53, for example, a CRT display, a liquid crystal display, an organic EL display, an LED display, a plasma display, or any other display known in the art can be used as appropriate.

The input interface 54 includes input devices that accept various commands from the user. As input devices, keyboards, mice, various switches, touch screens, touch pads, etc. can be used. Note that the input device is not limited to those equipped with physical operation parts such as a mouse and a keyboard. For example, an electrical signal processing circuit that receives electrical signals corresponding to input operations from an external input device provided separately from the magnetic resonance imaging apparatus 1 and outputs the received electrical signals to various circuits is also input. Included in the example of interface 54.

The communication interface 55 connects the magnetic resonance imaging apparatus 1 to a workstation, a PACS (Picture Archiving and Communication System), an HIS (Hospital Information System), an RIS (Radiology Information System), etc. via a LAN (Local Area Network) or the like. This is the interface to connect. The network IF sends and receives various information to and from connected workstations, PACS, HIS, and RIS.

Note that the above configuration is an example and is not limited thereto. For example, the sequence control circuit 29 may be incorporated into the medical information processing device 50. Furthermore, the sequence control circuit 29 and the processing circuit 51 may be mounted on the same substrate. The sequence control circuit 29, the gradient magnetic field power supply 21, the transmission circuit 23, and the reception circuit 25 may be implemented in a single control device different from the medical information processing device 50, or may be implemented in a distributed manner in multiple devices. Also good.

Hereinafter, an example of the operation of the magnetic resonance imaging apparatus 1 and the medical information processing apparatus 50 according to this embodiment will be described.

FIG. 2 is a diagram showing a typical flow of MR imaging by the processing circuit 51. The process shown in FIG. 2 starts from setting the imaging protocol for the target MR imaging.

As shown in FIG. 2, the processing circuit 51 executes the imaging protocol setting function 511 (step S1). In step S1, the processing circuit 51 sets an imaging protocol regarding the subject P. Imaging time, type of k-space filling method, type of pulse sequence, TR, TE, etc. are set as imaging parameters included in the imaging protocol. In this embodiment, the k-space filling method may be any method such as a radial method, a spiral method, or a Cartesian method. However, in order to specifically explain the following, it is assumed that the k-space filling method is a radial method.

When step S1 is performed, the processing circuit 51 executes the mask data acquisition function 512 (steps S2 and S3). In steps S2 and S3, the processing circuit 51 generates mask data regarding MR imaging of the target. Mask data is data in which a numerical value indicating that data is to be collected or a numerical value indicating that data is not to be collected is assigned to a plurality of candidates for data collection trajectories (spokes) having a finite number of elements. In other words, the mask data is data indicating a trajectory to be collected among candidate trajectories with a finite number of elements.

FIG. 3 is a diagram schematically showing the generation process of the mask data 91 in steps S2 and S3. As shown in FIG. 3, the data collection trajectory for the radial method radially passes approximately the center of k-space. As shown in the upper part of FIG. 3, in MR imaging according to the present embodiment, data to be collected is collected from a set 70 of candidates for data collection trajectories (hereinafter referred to as candidate trajectories) having a predetermined finite number of elements. A trajectory (hereinafter referred to as a collection target trajectory) is selected. In FIG. 3, candidate trajectories are indicated by dotted lines. The number of elements (number of lines) and angle of the candidate trajectory are set in advance. The number of elements of the candidate trajectory is set to such a number that sufficient image quality can be guaranteed when data collection is performed for all candidate trajectories with the number of elements. It is preferable that the number of elements of the candidate trajectory is finite. That is, the number of elements of the candidate trajectory may be a large number, such as one million, as long as it is not an irrational number. Note that FIG. 3 shows eight candidate trajectories for the sake of simplicity.

The angle of the candidate trajectory according to this embodiment is defined as an angle from a reference angle, for example, 0 degrees. The angular interval between adjacent candidate trajectories is set to a predetermined angle. Specifically, the angular intervals between adjacent candidate trajectories may be set to approximately equal intervals. In this way, by limiting the number of elements and angles, it becomes possible to also limit the number of elements and angles of candidate trajectories of machine learning learning data.

The collection angle of each candidate trajectory is set to a multiple of the base angle. That is, when the collection angle of the first data collection trajectory is 0 degrees, the collection angle of the second candidate trajectory = basic angle, the collection angle of the third candidate trajectory = "basic angle x 2", and the fourth candidate The collecting angle of the locus is set as follows: "basic angle x 3", . . .

If the number of elements is finite, the basic angle may be set to a golden angle, for example, an irrational number of approximately 111.25 degrees. For example, if the number of elements is 1000, the collection angle of the first candidate trajectory is 0 degrees, the collection angle of the second candidate trajectory is a golden angle, and the collection angle of the third candidate trajectory is "golden angle x 2". , the collection angle of the fourth candidate trajectory is set as "golden angle x 3", etc. up to the 1000th candidate trajectory.

The collection angle may be set to 360 degrees divided by the number of elements. For example, if the number of elements is 1000, the basic angle may be set to 360/1000 degrees = 0.36 degrees. Further, the basic angle may be set to a multiple of 360/1000 degrees. For example, if the multiple is 309, the basic angle will be set to 360/1000 degrees x 309 = 111.24 degrees. This allows the fundamental angle to be approximately equal to the golden angle.

A number (hereinafter referred to as a trajectory number) is assigned to each candidate trajectory for identification. The trajectory number is used to generate mask data. Any rule may be used for assigning trajectory numbers. For example, numbers may be assigned in order from candidate trajectories with small or large collection angles, or numbers may be assigned according to the setting order of candidate trajectories.

First, the processing circuit 51 selects a collection target trajectory based on the imaging time and the candidate trajectory (step S2). For example, the processing circuit 51 selects a trajectory to be collected from among the candidate trajectories included in the set 70 according to a predetermined rule (hereinafter referred to as a selection rule) based on the imaging time set in step S1. Specifically, first, the processing circuit 51 determines the number of trajectories to be collected according to the imaging time. Next, the processing circuit 51 selects the determined number of collection target trajectories from the set 70 according to the selection rule. As a selection rule, for example, it is preferable to select such that the angles between the collection target trajectories are approximately the same. For example, as shown in the middle part of FIG. 3, three candidate trajectories with approximately equal angles, that is, the first, fourth, and seventh candidate trajectories, are selected as collection target trajectories from among the eight candidate trajectories. . In FIG. 3, the trajectory to be collected is shown by a solid line.

The trajectory to be collected is set for each slice or volume, or for each frame. That is, the collection target locus is set for each image. Image quality obtained by the radial method is typically guaranteed by about 800 spokes. However, according to the present embodiment, equivalent image quality can be guaranteed with approximately 30 to 50 spokes per image. One reason for this is that the collection locus according to this embodiment has substantially equal angular intervals.

Next, the processing circuit 51 generates mask data 91 based on the selected trajectory to be collected (step S3). As shown in the lower part of FIG. 3, the mask data 91 includes a numerical value indicating that a plurality of candidate trajectories have been selected (i.e., data will be collected) or a numerical value indicating that they have not been selected (i.e., data will not be collected). is the assigned numerical data. For example, "1" is assigned as a numerical value indicating that the item has been selected, and "0" is assigned as a numerical value indicating that it is not selected. The processing circuit 51 determines whether or not each candidate trajectory has been selected, and assigns "1" if selected, and assigns "0" if not selected. The numerical values of "0" or "1" are arranged in the order of the trajectory numbers of the candidate trajectories. For example, in the case of FIG. 3, the mask data 91 indicates that the first, fourth, and seventh candidate trajectories are selected, and the second, third, fifth, sixth, and eighth candidate trajectories are not selected. , (1, 0, 0, 1, 0, 0, 1, 0).

When step S3 is performed, the processing circuit 51 executes the raw data acquisition function 513 (step S4). In step S4, the processing circuit 51 instructs the sequence control circuit 29 to execute MR imaging. The sequence control circuit 29 synchronously controls the gradient magnetic field power source 21, the transmitting circuit 23, and the receiving circuit 25, executes MR imaging for the collection target trajectory selected in step S2, and acquires k-space data regarding the collection target trajectory. collect. The collected k-space data is k-space data corresponding to the collection target trajectory represented by the mask data generated in step S3. Since the trajectory to be collected is selected from among the candidate trajectories corresponding to the number of elements, typically, the k-space data collected in step S4 is sparse.

After step S4 is performed, the processing circuit 51 executes the image generation function 514 (step S5). In step S5, the processing circuit 51 performs DNN reconstruction on the mask data generated in step S3 and the k-space data collected in step S4 to generate an MR image.

4, 5, and 6 are diagrams schematically showing the flow of DNN reconfiguration according to this embodiment. In the first DNN reconstruction shown in FIG. 4, the processing circuit 51 applies the machine learning model 521-1 to the k-space data RD and the mask data MD, and generates the MR image RI corresponding to the k-space data RD. . The machine learning model 521-1 receives k-space data and mask data and has learned parameters so as to output an MR image corresponding to the k-space data. The structure of the machine learning model 521-1 is not particularly limited. The MR image RI generated by the machine learning model 521-1 based on the k-space data RD and mask data MD is compared to the MR image generated by an analytical reconstruction method such as Fourier transform for the k-space data RD. Therefore, artifacts caused by signal loss included in the k-space data RD are reduced. This is because the machine learning model 521-1 uses, in addition to the k-space data RD, mask data MD indicating the collection target locus corresponding to the k-space data RD as input.

Note that the signal loss according to the present embodiment is a concept that includes any difference between actual k-space data and desired k-space data, including sparsity. For example, signal loss includes, in addition to sparsity, signal deterioration due to noise caused by various causes, information loss due to conversion from continuous values to discrete values that occurs in the A/D conversion process, and the like.

In the second DNN reconstruction shown in FIG. 5, the processing circuit 51 applies the machine learning model 521-2 to the k-space data RD and the mask data MD, and the signal missing portion included in the k-space data RD is restored. Generate k-space data RDM. The machine learning model 521-2 inputs k-space data and mask data and has learned parameters so as to output denoised k-space data. The k-space data RDM generated by the machine learning model 521-2 based on the k-space data RD and mask data MD has less artifacts caused by signal loss included in the k-space data RD than the k-space data RD. reduced. This is because the machine learning model 521-2 uses, in addition to the k-space data RD, mask data MD indicating the collection target trajectory corresponding to the k-space data RD as input.

As shown in FIG. 5, when the k-space data RDM is generated, the processing circuit 51 performs Fourier transformation on the k-space data RDM to generate the k-space data RD or the MR image RI corresponding to the k-space data RDM. do. Since the MR image RI is generated by Fourier transformation of the k-space data RDM in which the signal loss portion has been restored, the image quality is lower than that of the MR image generated by Fourier transformation of the k-space data RD including the signal loss. In good condition.

In the third DNN reconstruction shown in FIG. 6, the processing circuit 51 performs Fourier transform on the k-space data RD to generate a temporary MR image RIM corresponding to the k-space data RD. The temporary MR image RIM is a reconstructed image generated by performing Fourier transformation on the k-space data RD including signal defects, and therefore includes many signal defects.

As shown in FIG. 6, when the temporary MR image RIM is generated, the processing circuit 51 applies the machine learning model 521-3 to the temporary MR image RIM and the mask data MD to An MR image RI in which the missing portion is restored is generated. The machine learning model 521-3 inputs the temporary MR image RIM and the mask data MD, and the parameters are learned so as to output a denoised MR image. The MR image RI generated by the machine learning model 521-3 based on the temporary MR image RIM and the mask data MD has fewer artifacts caused by signal loss than the temporary MR image RIM. This is because the machine learning model 521-3 uses as input, in addition to the temporary MR image RIM, the mask data MD indicating the acquisition target trajectory corresponding to the k-space data RD used for the temporary MR image RIM.

When step S5 is performed, the processing circuit 51 executes the display control function 516 (step S6). In step S6, the processing circuit 51 displays the MR image generated in step S5 on the display 53.

With the above steps, MR imaging by the processing circuit 51 is completed.

Note that the process flow shown in FIG. 2 is an example, and the present embodiment is not limited thereto. For example, in step S2, the processing circuit 51 automatically selects the trajectory to be collected based on the imaging time. However, this embodiment is not limited to this. For example, the processing circuit 51 may select a candidate trajectory specified by the user via the input interface 54 as the collection target trajectory. In this case, for example, the processing circuit 51 displays on the display 53 a schematic diagram graphically representing candidate trajectories with a finite number of elements in a selectable manner. A good example of a schematic diagram is an image depicting candidate trajectories arranged in k-space and their trajectory numbers, as shown in the upper part of FIG. 3, for example. The user, via the input interface 54, specifies any trajectory from among the candidate trajectories included in the displayed schematic diagram as the trajectory to be collected. The processing circuit 51 may select the designated locus as the locus to be collected.

The above embodiment is based on the premise that data collection is performed once or zero times for each acquisition target locus for one image of MR imaging. However, this embodiment is not limited to this. For example, data collection may be performed two or more times for a certain collection target trajectory. By collecting data two or more times, it is possible to improve the reliability of the k-space data regarding the trajectory to be collected. For example, if data collection is performed twice for the 1st and 4th trajectories, once for the 2nd, 5th and 7th trajectories, and 0 times for the 3rd, 6th and 8th trajectories, then the mask data becomes (2, 1, 0, 2, 1, 0, 1, 0).

In the mask data according to the above embodiment, it is assumed that a numerical value corresponding to the number of times of collection is assigned to each data collection trajectory. However, this embodiment is not limited to this. For example, the numerical value "1" may be uniformly assigned to data collection trajectories in which data collection is performed two or more times. Further, the numerical value "1" may be assigned to 0 times and 2 or more times, and the numerical value "0" may be assigned only to 1 time. Any numerical value may be assigned to the mask data according to this embodiment according to other rules.

Note that in the description of this embodiment, non-negative values are used as mask data, but the mask data is not limited to this, and may include negative values, for example, -1.

Further, the numerical value of the mask data was assumed to have a value according to the number of times of collection. However, this embodiment is not limited to this. For example, consider a case where the same trajectory is collected in the positive direction and in the negative direction in radial collection. When the data collection trajectory in the positive direction and the data collection trajectory in the negative direction are treated as different trajectories, a numerical value corresponding to the number of times of collection is assigned to each data collection trajectory. No distinction is made between data collection trajectories in the positive direction and data collection trajectories in the negative direction, and a value corresponding to the collection direction or a combination of the number of collections and the collection direction is assigned to one data collection trajectory. It's okay to be beaten. For example, when data is collected 100 times in the positive direction and in the negative direction with respect to a data collection trajectory of 0 degrees, (1, 0), etc. are assigned to the data collection trajectory. Note that "1" in (1,0) is an example of a numerical value indicating a positive direction, and "0" is an example of a numerical value indicating a negative direction.

The structure of the machine learning model 521 can be changed in design as appropriate. For example, in the case of the second DNN reconstruction, the machine learning model 521 is an arbitrary multilayer network (hereinafter referred to as the main network layer) that receives k-space data RD and mask data MD as input and outputs k-space data RDM. Subsequently, an FFT (Fast Fourier Transfer) layer may be incorporated. K-space data is input to the FFT layer, FFT is applied to the input k-space data, and an MR image is output. Thereby, the machine learning model 521 alone can output the MR image RI based on the k-space data RD and the mask data MD.

Further, the machine learning model 521 may have a chain structure in which unit network structures including the main network layer, the FFT layer, and an IFFT (Inverse Fast Fourier Transfer) layer are cascaded. An MR image is input to the IFFT layer, IFFT is applied to the input MR image, and k-space data is output. The chain structure makes it possible to improve the accuracy of restoring signal-deficient portions.

In the chain structure, a matching layer may be provided after the IFFT layer. The matching layer receives the k-space data based on the MR image output from this network layer and the unprocessed k-space data input to this network layer, and performs processing using the unprocessed k-space data. Matching processing is performed on the completed k-space data, and matched k-space data is output. In the matching process, the processed k-space data is weighted and added to the unprocessed k-space data for each pixel according to the degree of signal loss. For example, the lower the degree of signal loss is, the higher the weight is given to the pixel value of the unprocessed k-space data, and the higher the degree of signal loss is, the lower the weight is given to the pixel value of the unprocessed k-space data. This makes it possible to ensure consistency between the processed k-space data and the unprocessed k-space data.

Furthermore, in the machine learning model 521, an element product calculation layer may be provided before the main network layer. The element product calculation layer calculates the element product of the k-space data and the mask data, and outputs the element product data. This network layer receives k-space data and element product data as input, and outputs MR image data in which signal-defective portions of the k-space data are restored. The unit network structures of this network layer and the element product calculation layer may be connected in cascade.

In the above description, it is assumed that the k-space data input to the machine learning model 521 is the original k-space data collected by MR imaging. However, this embodiment is not limited to this. The k-space data according to this embodiment may be calculated k-space data generated by performing forward projection processing on an MR image. Furthermore, the k-space data or MR image according to this embodiment may be raw data that has been subjected to arbitrary signal processing such as signal compression processing, resolution decomposition processing, signal interpolation processing, resolution synthesis processing, etc. Further, the raw data according to this embodiment may be hybrid data generated by performing Fourier transform on k-space data in the lead-out direction.

Next, learning of the machine learning model 521 will be explained. Machine learning model 521 is generated by a model learning device. The model learning device causes a machine learning model to perform machine learning according to a model learning program based on learning data including a plurality of learning samples, and generates a trained machine learning model (hereinafter referred to as a trained model). The model learning device is a computer such as a workstation having a processor such as a CPU and a GPU. The model learning device and the medical information processing device 50 may or may not be communicably connected via a cable or a communication network. Further, the model learning device and the medical information processing device 50 may be configured by an integrated computer.

FIG. 7 is a diagram showing the configuration of the model learning device 6 according to this embodiment. As shown in FIG. 7, the model learning device 6 includes a processing circuit 61, a memory 62, an input interface 63, a communication interface 64, and a display 65.

The processing circuit 61 includes a processor such as a CPU or GPU. By starting the model learning program installed in the memory 62 or the like, the processor executes the learning sample generation function 611, the learning function 612, the display control function 613, and the like. Note that each of the functions 611 to 613 is not limited to being implemented by a single processing circuit. A processing circuit may be configured by combining a plurality of independent processors, and the functions 611 to 613 may be realized by each processor executing a program.

In the learning sample generation function 611, the processing circuit 61 generates a learning sample that is a combination of input data and output data. The processing circuit 61 generates learning samples based on the k-space data collected along all candidate trajectories.

In the learning function 612, the processing circuit 61 causes the machine learning model to learn parameters based on learning data regarding a plurality of learning samples. By learning the parameters by the learning function 612, a trained machine learning model 521 shown in FIG. 1 is generated.

In the display control function 613, the processing circuit 61 displays learning data, learning results, etc. on the display 65.

The memory 62 is a storage device such as a ROM, RAM, HDD, SSD, or integrated circuit storage device that stores various information. The memory 62 stores, for example, a model learning program for learning a multilayer network. In addition to the above storage devices, the memory 62 may also be a drive device that reads and writes various information to/from a portable storage medium such as a CD, DVD, or flash memory, or a semiconductor memory element such as a RAM. Furthermore, the memory 62 may be located in another computer connected to the model learning device 6 via a network.

The input interface 63 accepts various input operations from the user, converts the received input operations into electrical signals, and outputs the electrical signals to the processing circuit 61 . Specifically, the input interface 63 is connected to input devices such as a mouse, keyboard, trackball, switch, button, joystick, touch pad, and touch panel display. The input interface 63 outputs an electrical signal to the processing circuit 61 according to an input operation to the input device. Furthermore, the input device connected to the input interface 63 may be an input device provided in another computer connected via a network or the like.

The communication interface 64 is an interface for data communication with the medical information processing device 50, the medical image diagnostic device, and other computers.

The display 65 displays various information according to the display control function 613 of the processing circuit 61. For example, the display 65 displays learning data, learning results, and the like. Furthermore, the display 65 outputs a GUI and the like for accepting various operations from the user. For example, as the display 65, a liquid crystal display, a CRT display, an organic EL display, a plasma display, or any other display can be used as appropriate.

Next, an example of the operation of the model learning device 6 according to this embodiment will be described.

As described above, processing circuit 61 generates training samples based on k-space data collected along all candidate trajectories. The types of input data and output data differ depending on the type of machine learning model 521 shown in FIGS. 4, 5, and 6. In the case of the machine learning model 521-1 in FIG. 4, mask data and k-space data including signal loss are used as input data, and an MR image with reduced signal loss is used as output data. In the case of the machine learning model 521-2 in FIG. 5, mask data and k-space data including signal loss are used as input data, and k-space data with reduced signal loss is used as output data. In the case of the machine learning model 521-3 in FIG. 6, mask data and an MR image including signal loss are used as input data, and an MR image with reduced signal loss is used as output data.

Below, as an example of the machine learning model 521, the machine learning model 521-1 shown in FIG. 4, in which parameters are learned to input k-space data and mask data and output an MR image, will be described, and the machine learning process will be explained. explain.

FIG. 8 is a diagram schematically showing processing by the learning sample generation function 611. As shown in FIG. 8, full k-space data RDF is acquired by the processing circuit 61. Full k-space data is k-space data obtained by collecting data for all candidate trajectories with a finite number of elements. The full k-space data RDF is collected in advance by the magnetic resonance imaging apparatus 1 or the like.

As shown in FIG. 8, in the learning sample generation function 611, the processing circuit 61 performs thinning processing on the full k-space data RDF, and creates a virtual k Spatial data RD1 to RDn are generated. For example, the processing circuit 61 calculates k-space data corresponding to each trajectory combination by simulation based on the full k-space data RDF. In FIG. 8, the virtual k-space data RD1 to RDn illustrate the k-space data regarding three processing target trajectories, but the number of processing target trajectories constituting the virtual k-space data RD1 to RDn is not limited to the full number. Any number of candidate trajectories may be used as long as it is less than the number of candidate trajectories configuring the k-space data RDF. The virtual k-space data RD1 to RDn are used as input data for the machine learning model 521-1.

When the thinning process is performed, the processing circuit 61 performs Fourier transform on each of the virtual k-space data RD1 to RDn to generate virtual MR images RI1 to RIn corresponding to the virtual k-space data RD1 to RDn. The virtual MR images RI1 to RIn are used as correct MR images for the machine learning model 521-1.

Furthermore, the processing circuit 61 performs mask data conversion processing on each of the virtual k-space data RD1 to RDn to generate mask data MD1 to MDn corresponding to the virtual k-space data RD1 to RDn. Specifically, the processing circuit 61 first specifies trajectory combinations of each of the virtual k-space data RD1 to RDn. For each candidate trajectory, the processing circuit 61 assigns a numerical value "1" to a trajectory included in the trajectory combination, and assigns a numerical value "0" to a trajectory not included in the trajectory combination. As a result, mask data MD1 to MDn are generated. Mask data MD1 to MDn are used as input data for machine learning model 521-1.

Mask data MD1 to MDn, k-space data RD1 to RDn, and MR images RI1 to RIn are associated for each locus combination. Mask data MD1 to MDn, k-space data RD1 to RDn, and MR images RI1 to RIn for each trajectory combination are treated as learning samples. Learning samples for each trajectory combination are stored in the memory 62.

This completes the processing by the learning sample generation function 611. Next, learning processing by the learning function 612 is performed. In the learning function 612, the processing circuit 61 performs forward propagation processing by applying k-space data RD1 to RDn and mask data MD1 to MDn to a machine learning model for each trajectory combination, and generates an MR image (hereinafter referred to as an estimated MR image). ) is output. Next, the processing circuit 61 applies the difference (error) between the estimated MR result and the correct MR image to the machine learning model, performs backpropagation processing, and calculates a gradient vector. Next, the processing circuit 61 updates parameters such as a weighted matrix and bias of the machine learning model based on the gradient vector. A trained machine learning model 521-1 is generated by repeating the forward propagation process, backpropagation process, and parameter update process while changing the learning samples.

The trained machine learning model 521-1 is stored in the memory 62. Further, the trained machine learning model 521-1 is transmitted to the magnetic resonance imaging apparatus 1 via the communication interface 64.

As described above, the type of k-space filling method according to this embodiment may be a spiral method. In this case as well, mask data can be generated as in the radial method. A candidate trajectory related to the spiral method draws a spiral starting from the center of k-space and ending at an arbitrary point. Similar to the radial method, the processing circuit 51 selects an arbitrary candidate trajectory as a collection target trajectory from a set including a plurality of candidate trajectories, and generates mask data corresponding to the selected collection target trajectory. Note that the candidate trajectories that can be generated depend on the hardware performance of the gradient coil, so it is necessary to prepare learning data for each target hardware. or. A set of candidate trajectories for the spiral method may be determined according to a low gradient coil so as to be compatible with a plurality of hardware.

In the above description, it is assumed that the medical information processing device 50 is incorporated in the magnetic resonance imaging device 1. However, this embodiment is not limited to this.

FIG. 9 is a diagram showing another configuration of the medical information processing device 50 according to this embodiment. As shown in FIG. 9, the medical information processing device 50 is a single computer independent of a modality device such as a magnetic resonance imaging device. The medical information processing device 50 in FIG. 9 includes a processing circuit 51, a memory 52, a display 53, an input interface 54, and a communication interface 55, like the medical information processing device 50 in FIG. The functions of each component are the same as those in FIG. 1, so description thereof will be omitted.

Note that the medical information processing device 50 in FIG. 9 can process raw data collected by any modality device. For example, the raw data may be sinogram data collected by an X-ray computed tomography device.

As described above, the medical information processing device 50 according to this embodiment includes the processing circuit 51. The processing circuit 51 implements at least a mask data acquisition function 512, a raw data acquisition function 513, and an image generation function 514. In the mask data acquisition function 512, the processing circuit 51 acquires mask data in which numerical values corresponding to the number of acquisition times and/or the acquisition direction in target imaging are assigned to a plurality of data acquisition locus candidates having a finite number of elements. In the raw data acquisition function 513, the processing circuit 51 acquires raw data collected during object imaging of the subject P. In the image generation function 514, the processing circuit 51 performs reconstruction processing using the machine learning model 521 on the acquired mask data and the acquired raw data to generate a medical image regarding the subject P.

According to the above configuration, in addition to the k-space data, the machine learning model 521 uses mask data indicating the collection target trajectory corresponding to the k-space data as an input, which is compared to the case where only the k-space data is input. As a result, the quality of the output medical image can be improved. Furthermore, since the candidate trajectories are limited to trajectories with a predetermined finite number of elements, there is no need to calculate the acquisition target trajectories for each MR imaging. In connection with this, it is also possible to easily match the data collection trajectory used for learning data and the data collection trajectory used for MR imaging.

In the above embodiment, the input of the machine learning model is based on raw data or medical data such as medical images, and multiple data collection trajectory candidates with a finite number of elements, depending on the number of times of acquisition and/or the direction of acquisition in target imaging. It is assumed that the mask data is assigned a numerical value. However, this embodiment is not limited to this. It is possible to extend the concept of mask data to numerical data that digitizes medical data collection conditions. The medical data collection conditions are a concept that includes not only the imaging protocol for medical imaging but also the data processing conditions for raw data and the image processing conditions for medical images. The imaging protocol for medical imaging includes not only the data acquisition trajectory described above, but also the type of pulse sequence, frame number, and type of k-space filling trajectory.

Furthermore, in the above embodiment, the machine learning model is used for DNN reconstruction from raw data to medical images. However, this embodiment is not limited to this. The output of the machine learning model may be any data that can be used for medical diagnosis. For example, machine learning models can restore missing parts of raw data, restore missing parts of medical images, generate segmentation images with anatomical tissues, etc., and output the results of identifying anatomical tissues, etc. But that's fine.

FIG. 10 is a diagram schematically showing the numerical data 93. Numerical data 93 shown in FIG. 10 is data in which the type of pulse sequence is expressed numerically. As shown in FIG. 10, the numerical data 93 includes a finite number of elements 94, which is two or more. The number of elements corresponds to the number of candidate conditions included in the category to which the collection condition to be quantified belongs. For example, as shown in FIG. 10, when the category of collection conditions to be quantified is the type of pulse sequence, candidate conditions include, for example, FSE (Fast Spin Echo), FE (Field Echo), and EPI (Echo Planar). There are three types: (Imaging). In this case, the numerical data 93 is expressed as (FSE, FE, EPI).

Each element 94 corresponds to a numerical value depending on whether or not the acquisition condition for the target imaging adopts a plurality of candidate conditions. For example, as the value of the element 94, either a first value indicating that the candidate condition is adopted or a second value indicating that the candidate condition is not adopted is assigned. In this case, the numerical data 93 is called a one-hot vector. The first value and the second value may be any integer or natural number as long as they are different values, but for example, the first value is set to "1" and the second value is set to "0". do.

For example, as shown in FIG. 10, when the acquisition condition for target imaging adopts FSE, the value of the element 84 corresponding to FSE is "1", the value of the element 84 corresponding to FE is "0", and it corresponds to EPI. The value of element 84 is set to "0". Note that the number of elements may be any finite or infinite number. In other words, the number of candidate conditions may be any number. For example, it is also possible to set the number of elements, ie, the number of candidate conditions, to 1 million or more.

FIG. 11 is a diagram schematically showing input of the numerical data 93-1 in FIG. 10 to the machine learning model. As shown in the upper part of FIG. 11, when FSE is adopted, the value of the element corresponding to FSE of numerical data 93-1 is "1", the value of the element corresponding to FE is "0", and the value of the element corresponding to EPI is "1". The value of the element is set to "0". A combination of the numerical data 93-1 and the input medical image 101 is input to the machine learning model. As shown in the middle part of FIG. 11, when FE is adopted, the value of the element corresponding to FE of numerical data 93-1 is "0", the value of the element corresponding to FE is "1", and the value of the element corresponding to FE is "1", which corresponds to EPI. The value of the element is set to "0". A combination of the numerical data 93-1 and the input medical image 101 is input to the machine learning model. As shown in the lower part of FIG. 11, when EPI is adopted, the value of the element corresponding to EPI of numerical data 93-1 is "0", and the value of the element corresponding to FE is "0", which corresponds to EPI. The value of the element is set to "1". A combination of the numerical data 93-1 and the input medical image 101 is input to the machine learning model. By adding the numerical data 93-1, which digitizes the type of pulse sequence, to the input of the machine learning model, output data can be generated taking into account not only the medical image or raw data but also the type of pulse sequence. becomes possible. Therefore, the accuracy of the output data of the machine learning model is improved.

FIG. 12 is a diagram schematically showing input of other numerical data 93-2 to the machine learning model. Numerical data 93-2 shown in FIG. 12 digitizes the collection condition "golden angle frame number". The golden angle frame number is a frame number of video imaging using the radial method in which the collection angle of the data collection trajectory is set according to the rules of the golden angle, and the frame number is associated with the golden angle of the reference data collection trajectory regarding the frame. It is something that When the frame is composed of a plurality of data collection trajectories, the reference data collection trajectory is a data collection trajectory in any order such as the first, last, or middle. In this case, the numerical data 93-2 is expressed as (golden angle frame #1, golden angle frame #2, . . . ). The number of elements of the numerical data 93-2 matches or corresponds to the number of frame number candidates.

As shown in the upper part of FIG. 12, when the input medical image 101 is golden angle frame #1, the value of the element corresponding to golden angle frame #1 in the numerical data is "1", and the value of the element corresponding to other frames is "1". The value is set to "0". A combination of the numerical data 93-2 and the input medical image 101 is input to the machine learning model. As shown in the middle part of FIG. 12, when the input medical image 101 is golden angle frame #2, the value of the element corresponding to golden angle frame #2 in the numerical data is "1", and the value of the element corresponding to other frames is "1". The value is set to "0". A combination of the numerical data 93-2 and the input medical image 101 is input to the machine learning model. In this way, by adding numerical data that digitizes golden angle frame numbers to the input of a machine learning model, it is possible to generate output data taking into account not only medical images or raw data but also golden angle frame numbers. Become. Therefore, the accuracy of the output data of the machine learning model is improved.

Note that the collection angle and collection order of the data collection locus of the radial method in video imaging are not limited to those using golden angles, but may also be those using inverted bit patterns or sequential patterns. In this case, the number of elements of the numerical data matches or corresponds to the number of candidates for the inverted bit pattern or sequential pattern. The value of each element is set depending on whether the inverted bit pattern or sequential pattern regarding the input medical image adopts the inverted bit pattern or sequential pattern corresponding to the element.

FIG. 13 is a diagram schematically showing input of other numerical data 93-3 to the machine learning model. Numerical data 93-3 shown in FIG. 13 digitizes the collection condition "pseudo-random pattern number." The pseudo-random pattern number is a number of a random number pattern for video imaging using the pseudo-random Cartesian method, and is a number in which a frame number is associated with a random number pattern regarding the frame. Note that the pseudo-random Cartesian method refers to a Cartesian method in which a data collection trajectory is set according to a random number pattern. In this case, the numerical data 93-3 is expressed as (pseudo-random pattern #1, pseudo-random pattern #2, . . . ). The number of elements of the numerical data 93-3 matches or corresponds to the number of candidates of the random number pattern.

As shown in the upper part of FIG. 13, when the input medical image 101 is pseudo-random pattern #1, the value of the element corresponding to pseudo-random pattern #1 of numerical data 93-3 is "1", and the value of the element corresponding to pseudo-random pattern #1 of numerical data 93-3 is "1"; The value of the element corresponding to is set to "0". A combination of the numerical data 93-3 and the input medical image 101 is input to the machine learning model. As shown in the middle part of FIG. 13, when the input medical image 101 is pseudo-random pattern #2, the value of the element corresponding to pseudo-random pattern #2 of numerical data 93-3 is "1", and other pseudo-random patterns The value of the element corresponding to is set to "0". A combination of the numerical data 93-3 and the input medical image 101 is input to the machine learning model. In this way, by adding numerical data that digitizes pseudorandom pattern numbers to the input of a machine learning model, it is possible to generate output data taking into account not only medical images or raw data but also pseudorandom pattern numbers. Become. Therefore, the accuracy of the output data of the machine learning model is improved.

The acquisition conditions for video imaging to be quantified are not limited to those described above, and may be, for example, frame numbers defined by the cardiac phase. The frame number defined by the cardiac phase is, for example, a number that defines the cardiac phase of the frame, assuming that from an R wave to the next R wave is 100%.

The acquisition condition to be quantified may be the magnification rate of parallel imaging. In this case, the number of elements of numerical data is defined by the number of speed factor candidates. The value of the element has a value depending on whether the magnification rate corresponding to the element is employed in object imaging. Artifacts caused by parallel imaging have characteristics that depend on the magnification rate. Therefore, by adding numerical data that quantifies the multiplication rate of parallel imaging to the input of the machine learning model, the machine learning model can detect features according to the multiplication rate, which in turn improves the accuracy of the output data of the machine learning model. will improve.

The collection conditions to be quantified are not limited to those described above, and may be, for example, data processing conditions for raw data or image processing conditions for medical images. The data processing conditions or image processing conditions include, for example, the number of repetitions of repeated calculations. The repeated calculations are calculations such as noise removal processing, edge enhancement processing, smoothing processing, etc. that are repeatedly applied to raw data or medical images, and the number of repetitions is the number of times the calculations are repeated. Alternatively, when reconstructing a medical image from raw data, it is the number of times the calculation is repeated when the calculation is performed alternately to reduce the reconstruction error between the raw data and the reconstructed image. The calculation is performed using a filter, a DNN, or the like. The number of elements of the numerical data is defined as the number of candidates for the number of repetitions, and the value of the element has a value depending on whether or not the target operation is repeated by the number of repetitions corresponding to the element. In this way, by adding medical data after repeated calculations and numerical data that digitizes the number of repeated calculations to the input of a machine learning model, the output data can be calculated taking into account not only the medical data but also the number of repetitions. It becomes possible to generate. Therefore, the accuracy of the output data of the machine learning model is improved. Note that the number of repetitions may be a one-hot vector. Alternatively, the first few times (for example, three times) may be given in the form of a one-hot vector, and the vector may be given as the number of repetitions exceeding that number.

The numerical data regarding the data processing conditions or the image processing conditions may be numerical data indicating whether or not a plurality of types of data processing or image processing are adopted. In this case, the number of elements of the numerical data is the number of candidates for data processing or image processing, and the value of each element is set to a value depending on whether individual data processing or image processing is adopted. For example, the numerical data may be expressed as (noise reduction, edge enhancement, segment processing), etc.

The number of elements of the above numerical data was assumed to match the number of candidate conditions. However, this embodiment is not limited to this.

FIG. 14 is a diagram schematically showing other numerical data 93-4. As shown in FIG. 14, the numerical data 93-4 includes an element 94 corresponding to a candidate condition and an element 95 corresponding to not falling under any of the candidate conditions. For example, the value of element 95 is set to "0" when any candidate condition is met, and is set to "1" when any candidate condition is met. For example, as shown in FIG. 14, if the numerical data 93-4 is a numerical representation of a pulse sequence, and the actual pulse sequence of MR imaging is neither FSE, FE, nor EPI, the value of element 95 is " 1”. By providing the element 95 in the numerical data 93-4, it is possible to clearly indicate to the machine learning model that the pulse sequence of actual MR imaging does not correspond to any of the candidate conditions.

It is assumed that the above numerical data is a numerical representation of one type of collection condition, such as the type of pulse sequence or frame number. However, this embodiment is not limited to this. For example, the numerical data may represent the pulse sequence type and frame number as a single column vector. In this case, the numerical data is specifically expressed by (FSE, FE, EPI, frame number 1, frame number 2, etc.).

Furthermore, in the above description, the numerical data element is a one-hot vector that expresses in binary terms whether or not the candidate condition is adopted. However, the element may be a vector that expresses a multilevel concept such as the strength, direction, and number of times of the candidate condition using three or more values.

Next, the network structure of the machine learning model according to this embodiment will be explained.

FIG. 15 is a diagram schematically showing the network structure of the machine learning model 521-4 according to this embodiment. As shown in FIG. 15, the machine learning model 521-4 receives, as an example, a combination of a medical image #1 with noise and numerical data #2 that digitizes the collection conditions regarding the medical image, and Assume that the DNN is a DNN whose parameters have been trained to output a medical image that does not contain any images. It is assumed that medical image #1 and numerical data #2 are each one set.

Machine learning model 521-4 has CNN layer 523. The CNN layer 523 is configured to multiply each of the medical image #1 and the numerical data #2 by different weights for each channel and add them together. For example, a plurality of filters having the same number of channels are applied to the medical image #1, and a plurality of filters having the same number of channels are applied to the numerical data #2. The filtered medical image and numerical data are added for each channel by an adder 203 and converted into added data. The added data is converted into bias data by adding a constant value for each channel by the bias 204. Bias data for each channel is output from the CNN layer 523.

The network structure of the CNN layer 523 described above is an example, and various modifications are possible. For example, the CNN layer 523 may be provided with multiple stages of filters 201 and 202 instead of one stage of filters 201 and 202. For example, the size of the filter 201 may be set to 5×5, the size of the filter 202 may be set to 1×1, etc., but the sizes of the filters 201 and 202 may be any size.

Following the CNN layer 523, a postlayer 525 is provided. The post layer 525 performs calculations on bias data for multiple channels from the CNN layer 523, and outputs a medical image without noise as the output of the machine learning model 521-4. The postlayer 525 has at least one fully connected layer and an output layer, but in addition to these layers, it may have any layer such as one or more CNN layers, pooling layers, normalization layers, etc.

It is assumed that the input to the CNN layer 523 shown in FIG. 15 is one set of medical images and one set of numerical data. However, this embodiment is not limited to this. The input to the CNN layer may be multiple sets of medical images and the same set of numerical data.

FIG. 16 is a diagram schematically showing the network structure of another machine learning model 521-5 according to this embodiment. As shown in FIG. 16, the machine learning model 521-5 includes, as an example, a medical image #1 with noise, a medical image #2 with noise, and numerical data #3 that digitizes the collection conditions regarding the medical image #1. It is assumed that the DNN is a DNN whose parameters have been learned so as to input a combination with numerical data #4, which is a numerical representation of acquisition conditions regarding medical image #2, and output a medical image without noise. Medical image #2 is a medical image derived from medical image #1, such as a copy of medical image #1 or a medical image obtained by performing arbitrary image processing on medical image #1. Further, medical image 2 may be another medical image collected under the same acquisition conditions as medical image #1, or may be a medical image collected under different acquisition conditions.

Machine learning model 521-5 has CNN layer 527. The CNN layer 527 is configured to multiply each of medical image #1, medical image #2, numerical data #3, and numerical data #4 by different weights for each channel and add them. For example, multiple filters 211 with the same number of channels are applied to medical image #1, multiple filters 212 with the same number of channels are applied to medical image #2, and multiple filters 213 with the same number of channels are applied to numerical data #3. is applied, and a plurality of filters 214 having the same number of channels are applied to numerical data #4. The filtered medical image and numerical data are added for each channel by an adder 215 and converted into added data. The added data is converted into bias data by adding a constant value for each channel by a bias 216. Bias data for each channel is output from the CNN layer 527.

The network structure of the CNN layer 527 described above is an example, and various modifications are possible. For example, the CNN layer 527 may be provided with multi-stage filters 211, 212, 213, and 214 instead of single-stage filters 211, 212, 213, and 214.

Following the CNN layer 527, a postlayer 529 is provided. The post layer 529 performs calculations on bias data for multiple channels from the CNN layer 527, and outputs a medical image without noise as the output of the machine learning model 521-5. The output medical image is a medical image corresponding to medical image #1, such as an image obtained by removing noise from medical image #1. The postlayer 529 has at least one fully connected layer and an output layer, but in addition to these layers, it may have any layer such as one or more CNN layers, pooling layers, normalization layers, etc.

Note that in the case of a machine learning model having multiple CNN layers, numerical data does not necessarily have to be input to each CNN layer.

FIG. 17 is a diagram schematically showing input and output of a machine learning model 521-6 having a plurality of CNN layers 531. Machine learning model 521-6 shown in FIG. 17 illustratively has a series of three CNN layers 531-1, CNN layer 531-2, and CNN layer 531-3. A postlayer 533 is provided after the CNN 531-3. CNN 531-1 receives input medical images, CNN layer 531-2 receives output data of CNN layer 531-1, and CNN layer 531-3 receives output data of CNN layer 531-2. The postlayer 533 receives the output data from the CNN layer 531-3, and outputs, for example, a medical image from which noise has been removed from the input medical image #1 as the output of the machine learning model 521-6. The network structures of the CNN layer 531-1, CNN layer 531-2, and CNN layer 531-3 may be the same or different, and may be designed as desired.

As shown in FIG. 17, numerical data #2 does not need to be input for all three CNN layers 531-1, CNN layer 531-2, and CNN layer 531-3; for example, the CNN closest to the input side Numerical data #2 may not be input to the layer 531-1, but numerical data #2 may be input to the subsequent CNN layer 531-2 and CNN layer 531-3. The present invention is not limited to this, but numerical data #2 may be input to the CNN layer 531-1 closest to the input side, and numerical data #2 may not be input to the subsequent CNN layer 531-2 and CNN layer 531-3. . Also, even if numerical data #2 is input to any one CNN layer of CNN layer 531-1, CNN layer 531-2, and CNN layer 531-3, and numerical data #2 is not input to the other CNN layer, good. Of course, numerical data #2 may be input to all of the CNN layer 531-1, CNN layer 531-2, and CNN layer 531-3.

In the above embodiment, the machine learning model outputs a medical image without noise. However, the output of the machine learning model is not limited to this.

For example, the machine learning model may output a segmentation result of the medical image from the medical image and numerical data. For example, in response to input of a T1-weighted image or a T2-weighted image, an image in which resolving tissues and lesion areas are segmented is output as a segmentation result. The lesion area is, for example, an image area that appears to be cerebral infarction, ischemia, or cancer. The identification result of the medical image may be output from the medical image and numerical data. For example, in response to an input of a T1-weighted image, a T2-weighted image, or a combination of a T1-weighted image and a T2-weighted image, the probability of cerebral infarction is output as the identification result. A segmentation result of the medical image may be output from the medical image and numerical data. For example, in response to an input of a T1-weighted image, a T2-weighted image, or a combination of a T1-weighted image and a T2-weighted image, an image in which an image region that is similar to a cerebral infarction is segmented is output as a segmentation result.

Even when learning is performed based on medical images of different image types, such as T1-weighted images and T2-weighted images, some of the parameters of the machine learning model are standardized. Therefore, the machine learning model can obtain versatility for inputting medical images of different image types while suppressing reduction in learning efficiency.

Furthermore, the machine learning model may output super-resolution data of the medical data from the medical data and numerical data. Super-resolution data is medical data that has a higher spatial resolution than input medical data. It is possible to generate the machine learning model by making the DNN learn based on input data including medical data and numerical data and super-resolution data that is teacher data.

For example, when k-space data is collected using the Half-Fourier method with limited frequencies, part of the k-space data is missing and the spatial resolution is reduced. The machine learns a DNN based on input data including k-space data and numerical data collected by the half-Fourier method, and k-space data (teacher data) collected from MR imaging without frequency restrictions. A learning model can be generated. By using the machine learning model, it becomes possible to output k-space data (super-resolution data) in which missing data portions are restored from k-space data collected by the half-Fourier method.

The machine learning model shown above can be generated by the model learning device 6 using supervised machine learning. Learning samples are prepared by collecting medical data under various collection conditions. Specifically, the input data of the learning sample includes input medical data collected under a certain collection condition and numerical data obtained by quantifying the collection condition. The output data of the learning sample includes output medical data that corresponds to the medical data and is in accordance with the purpose of the machine learning model. For example, if the purpose of the machine learning model is denoising, the output data is medical data with reduced noise compared to the input medical data.

In the learning function 612, the processing circuit 61 applies input medical data and numerical data to a machine learning model, performs forward propagation processing, and outputs output medical data. Next, the processing circuit 61 applies the difference (error) between the estimated output medical data and the correct output medical data to the machine learning model, performs backpropagation processing, and calculates a gradient vector. Next, the processing circuit 61 updates parameters such as a weighted matrix and bias of the machine learning model based on the gradient vector. A trained machine learning model is generated by repeating the forward propagation process, backpropagation process, and parameter update process while changing the learning samples.

As described above, the medical information processing device 50 includes the processing circuit 51. The processing circuit 51 acquires medical data regarding the subject, acquires numerical data that digitizes the collection conditions of the medical data, applies the machine learning model 521 to input data including the numerical data and the medical data, and applies the machine learning model 521 to the input data including the numerical data and the medical data. Generate output data based on medical data.

According to the above configuration, the machine learning model 521 inputs not only medical data but also numerical data that digitizes the collection conditions corresponding to the medical data. , the accuracy of output data can be improved. The number of elements of the numerical data corresponds to the number of predetermined candidate conditions, and the value of each element is a value depending on whether the candidate condition is adopted or not. That is, since the rules for quantifying collection conditions are determined in advance, the machine learning model can determine collection conditions with high accuracy during learning and application. This is also a factor in improving the accuracy of output data.

According to at least one embodiment described above, the accuracy of output data using machine learning can be improved.

The term "processor" used in the above description refers to, for example, a CPU, GPU, or Application Specific Integrated Circuit (ASIC), or a programmable logic device (for example, a Simple Programmable Logic Device). :SPLD), Complex Programmable Logic Device (CPLD), and Field Programmable Gate Array (FPGA). The processor implements each of the functions 511-516 by reading and executing programs stored in the memory circuit. Note that instead of storing the program in the memory circuit, the program may be directly incorporated into the circuit of the processor. In this case, the processor implements each of the functions 511-516 by reading and executing a program built into the circuit. Further, instead of executing a program, each of the functions 511 to 516 corresponding to the program may be realized by a combination of logic circuits. Note that each processor of this embodiment is not limited to being configured as a single circuit for each processor, but may also be configured as a single processor by combining multiple independent circuits to realize its functions. good. Furthermore, multiple components in FIGS. 1, 3, and 7 may be integrated into one processor to implement its functions.

Although several embodiments of the invention have been described, these embodiments are presented by way of example and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, and changes can be made without departing from the gist of the invention. These embodiments and their modifications are included within the scope and gist of the invention as well as within the scope of the invention described in the claims and its equivalents.

1 Magnetic resonance imaging device
11 mounts
13 berths
131 Top plate
133 Base
21 gradient magnetic field power supply
23 transmitting circuit
25 receiving circuit
27 Bed drive device
29 Sequence control circuit
41 Static magnetic field magnet
43 Gradient magnetic field coil
45 Transmission coil
47 Receiving coil
50 medical information processing equipment
51 Processing circuit 52 Memory
53 display
54 Input interface 55 Communication interface 511 Imaging protocol setting function
512 mask data acquisition function
513 raw data acquisition function
514 image generation function
515 image processing function
516 display control function
521 Machine learning model
P Subject

Claims (27)

a first acquisition unit that acquires medical data regarding the subject;
a second acquisition unit that acquires numerical data digitizing the medical data collection conditions;
Applying a machine learning model to input data including the numerical data and the medical data,
a generation unit that generates output data obtained by processing the medical data;
Equipped with
The medical data is raw data collected by a medical image diagnostic device or a medical image based on the raw data,
The acquisition conditions include an imaging protocol for medical imaging by the medical image diagnostic apparatus, data processing conditions for the raw data, and/or image processing conditions for the medical image,
The numerical data includes a finite number of elements of 2 or more,
The number of elements corresponds to the number of the plurality of candidate conditions corresponding to the collection condition, and corresponds to the fact that the first element corresponding to the plurality of candidate conditions does not correspond to any of the plurality of candidate conditions. and a second element that
The value of the element corresponds to a numerical value according to the acceptance or rejection of the collection condition with respect to the plurality of candidate conditions and/or the value.
Medical information processing equipment. a first acquisition unit that acquires medical data regarding the subject;
a second acquisition unit that acquires numerical data digitizing the medical data collection conditions;
Applying a machine learning model to input data including the numerical data and the medical data,
a generation unit that generates output data obtained by processing the medical data;
Equipped with
The medical data is raw data collected by a medical image diagnostic device or a medical image based on the raw data,
The medical image diagnostic device is a magnetic resonance imaging device,
The acquisition conditions include the type of pulse sequence, the frame number, the type of k-space filling trajectory, the number of repeated operations, and/or the multiplication rate of parallel imaging.
Medical information processing equipment. a first acquisition unit that acquires medical data regarding the subject;
a second acquisition unit that acquires numerical data digitizing the medical data collection conditions;
Applying a machine learning model to input data including the numerical data and the medical data,
a generation unit that generates output data obtained by processing the medical data;
Equipped with
The medical data is raw data collected by a medical image diagnostic device or a medical image based on the raw data,
The acquisition conditions include an imaging protocol for medical imaging by the medical image diagnostic apparatus, data processing conditions for the raw data, and/or image processing conditions for the medical image,
The second acquisition unit acquires the numerical data that digitizes the collection conditions of medical imaging for the subject,
The first acquisition unit acquires, as the medical data, the raw data collected by the medical imaging according to the collection conditions,
The generation unit applies the machine learning model to input data including the numerical data and the raw data or the medical image based on the raw data, and generates the output data based on the raw data or the medical image. death,
The medical image diagnostic device is a magnetic resonance imaging device,
Medical information processing equipment. a first acquisition unit that acquires medical data regarding the subject;
a second acquisition unit that acquires numerical data digitizing the medical data collection conditions;
Applying a machine learning model to input data including the numerical data and the medical data,
a generation unit that generates output data obtained by processing the medical data;
Equipped with
The medical data is raw data collected by a medical image diagnostic device or a medical image based on the raw data,
The acquisition conditions include an imaging protocol for medical imaging by the medical image diagnostic apparatus, data processing conditions for the raw data, and/or image processing conditions for the medical image,
The second acquisition unit acquires the numerical data that digitizes the collection conditions of medical imaging for the subject,
The first acquisition unit acquires, as the medical data, the raw data collected by the medical imaging according to the collection conditions,
The generation unit generates the medical image regarding the subject by performing a reconstruction process on the input data including the numerical data and the raw data using the machine learning model.
Medical information processing equipment. 4. The medical information processing apparatus according to claim 3 , wherein the machine learning model receives the numerical data and the medical data and is trained using teacher data. The medical information processing apparatus according to claim 1 , wherein the value of the element has either a first value indicating that the collection condition is adopted or a second value indicating that the collection condition is not adopted. The medical information processing apparatus according to claim 3 , wherein the generation unit generates data used for medical diagnosis regarding the subject as the output data. 4. The medical information processing apparatus according to claim 3 , wherein the machine learning model is configured to multiply each of the numerical data and the medical data by different weights for each channel and add them. The collection condition is the number of times of collection and/or the direction of collection for each data collection trajectory,
The numerical data is mask data in which a numerical value corresponding to the number of acquisition times and/or the acquisition direction in target imaging is assigned to a plurality of data acquisition trajectory candidates with a finite number of elements.
The medical information processing device according to claim 3 . The medical information processing apparatus according to claim 9 , wherein the second acquisition unit determines the number of times of collection and/or the direction of collection for the plurality of data collection trajectory candidates according to a user's instruction or automatically. The medical information processing apparatus according to claim 9 , wherein the second acquisition unit determines the number of times of collection and/or the direction of collection for the plurality of data collection trajectory candidates according to a predetermined rule based on the imaging time in the medical imaging. . 10. The medical information processing apparatus according to claim 9 , wherein the numerical values include a first numerical value indicating that data collection is to be performed and a second numerical value indicating that data collection is not to be performed. 10. The medical information processing apparatus according to claim 9 , wherein each of the plurality of data collection trajectory candidates is a data collection trajectory in k-space related to a radial method, and passes through substantially the center of k-space in a straight line. The medical information processing apparatus according to claim 13 , wherein an angular interval between the plurality of data collection trajectory candidates is set to a predetermined angle. 14. The medical information processing apparatus according to claim 13 , wherein the plurality of data collection trajectory candidates are arranged at approximately equal intervals at an angle based on an angle obtained by dividing 360 degrees by the number of elements. 10. The medical information processing apparatus according to claim 9 , wherein each of the plurality of data collection trajectory candidates is a data collection trajectory in k-space related to spiral collection, and passes through k-space in a spiral shape. The generation unit applies the machine learning model to the mask data and the raw data, generates denoised raw data as the output data, and performs a Fourier transform on the denoised raw data to generate the medical image. The medical information processing apparatus according to claim 9 , which generates. The generation unit performs a Fourier transform on the raw data to generate a temporary medical image, applies the machine learning model to the mask data and the temporary medical image, and converts the denoised medical image into the output data. 10. The medical information processing apparatus according to claim 9 , wherein the medical information processing apparatus outputs as follows. The medical information processing apparatus according to claim 3 , further comprising an imaging unit that performs magnetic resonance imaging on the subject and collects k-space data as the medical data. obtaining medical data regarding the subject;
acquiring numerical data that quantifies the collection conditions of the medical data;
applying a machine learning model to input data including the numerical data and the medical data to generate output data obtained by processing the medical data;
Equipped with
The medical data is raw data collected by a medical image diagnostic device or a medical image based on the raw data,
The acquisition conditions include an imaging protocol for medical imaging by the medical image diagnostic apparatus, data processing conditions for the raw data, and/or image processing conditions for the medical image,
The numerical data includes a finite number of elements of 2 or more,
The number of elements corresponds to the number of the plurality of candidate conditions corresponding to the collection condition, and corresponds to the fact that the first element corresponding to the plurality of candidate conditions does not correspond to any of the plurality of candidate conditions. and a second element that
The value of the element corresponds to a numerical value according to the acceptance or rejection of the collection condition with respect to the plurality of candidate conditions and/or the value.
Medical information processing method. obtaining medical data regarding the subject;
acquiring numerical data that quantifies the collection conditions of the medical data;
applying a machine learning model to input data including the numerical data and the medical data to generate output data obtained by processing the medical data;
Equipped with
The medical data is raw data collected by a medical image diagnostic device or a medical image based on the raw data,
The acquisition conditions include an imaging protocol for medical imaging by the medical image diagnostic apparatus, data processing conditions for the raw data, and/or image processing conditions for the medical image,
The medical image diagnostic device is a magnetic resonance imaging device,
The acquisition conditions include the type of pulse sequence, the frame number, the type of k-space filling trajectory, the number of repeated operations, and/or the multiplication rate of parallel imaging.
Medical information processing method. a first acquisition step of acquiring medical data regarding the subject;
a second acquisition step of acquiring numerical data quantifying the medical data collection conditions;
a generation step of applying a machine learning model to input data including the numerical data and the medical data to generate output data obtained by processing the medical data;
Equipped with
The medical data is raw data collected by a medical image diagnostic device or a medical image based on the raw data,
The acquisition conditions include an imaging protocol for medical imaging by the medical image diagnostic apparatus, data processing conditions for the raw data, and/or image processing conditions for the medical image,
The second acquisition step acquires the numerical data that digitizes the collection conditions of medical imaging for the subject;
The first acquisition step acquires, as the medical data, the raw data collected by the medical imaging according to the collection conditions,
The generation step applies the machine learning model to input data including the numerical data and the raw data or the medical image based on the raw data to generate the output data based on the raw data or the medical image. death,
The medical image diagnostic device is a magnetic resonance imaging device,
Medical information processing method. a first acquisition step of acquiring medical data regarding the subject;
a second acquisition step of acquiring numerical data quantifying the medical data collection conditions;
a generation step of applying a machine learning model to input data including the numerical data and the medical data to generate output data obtained by processing the medical data;
Equipped with
The medical data is raw data collected by a medical image diagnostic device or a medical image based on the raw data,
The acquisition conditions include an imaging protocol for medical imaging by the medical image diagnostic apparatus, data processing conditions for the raw data, and/or image processing conditions for the medical image,
The second acquisition step acquires the numerical data that digitizes the collection conditions of medical imaging for the subject;
The first acquisition step acquires, as the medical data, the raw data collected by the medical imaging according to the collection conditions,
The generation step includes performing a reconstruction process on the input data including the numerical data and the raw data using the machine learning model to generate the medical image regarding the subject.
Medical information processing method. to the computer,
A function to obtain medical data regarding the subject,
a function of acquiring numerical data that quantifies the collection conditions of the medical data;
A function of applying a machine learning model to input data including the numerical data and the medical data to generate output data obtained by processing the medical data;
Realize,
The medical data is raw data collected by a medical image diagnostic device or a medical image based on the raw data,
The acquisition conditions include an imaging protocol for medical imaging by the medical image diagnostic apparatus, data processing conditions for the raw data, and/or image processing conditions for the medical image,
The numerical data includes a finite number of elements of 2 or more,
The number of elements corresponds to the number of the plurality of candidate conditions corresponding to the collection condition, and corresponds to the fact that the first element corresponding to the plurality of candidate conditions does not correspond to any of the plurality of candidate conditions. and a second element that
The value of the element corresponds to a numerical value according to the acceptance or rejection of the collection condition with respect to the plurality of candidate conditions and/or the value.
Medical information processing program. to the computer,
A function to obtain medical data regarding the subject,
a function of acquiring numerical data that quantifies the collection conditions of the medical data;
A function of applying a machine learning model to input data including the numerical data and the medical data to generate output data obtained by processing the medical data;
Realize,
The medical data is raw data collected by a medical image diagnostic device or a medical image based on the raw data,
The acquisition conditions include an imaging protocol for medical imaging by the medical image diagnostic apparatus, data processing conditions for the raw data, and/or image processing conditions for the medical image,
The medical image diagnostic device is a magnetic resonance imaging device,
The acquisition conditions include the type of pulse sequence, the frame number, the type of k-space filling trajectory, the number of repeated operations, and/or the multiplication rate of parallel imaging.
Medical information processing program. to the computer,
a first acquisition function that acquires medical data regarding the subject;
a second acquisition function that acquires numerical data that digitizes the medical data collection conditions;
a generation function that applies a machine learning model to input data including the numerical data and the medical data to generate output data obtained by processing the medical data;
Realize,
The medical data is raw data collected by a medical image diagnostic device or a medical image based on the raw data,
The acquisition conditions include an imaging protocol for medical imaging by the medical image diagnostic apparatus, data processing conditions for the raw data, and/or image processing conditions for the medical image,
The second acquisition function acquires the numerical data that digitizes the collection conditions of medical imaging for the subject;
The first acquisition function acquires, as the medical data, the raw data collected by the medical imaging according to the collection conditions,
The generation function applies the machine learning model to input data including the numerical data and the raw data or the medical image based on the raw data to generate the output data based on the raw data or the medical image. death,
The medical image diagnostic device is a magnetic resonance imaging device,
Medical information processing program. to the computer,
a first acquisition function that acquires medical data regarding the subject;
a second acquisition function that acquires numerical data that digitizes the medical data collection conditions;
a generation function that applies a machine learning model to input data including the numerical data and the medical data to generate output data obtained by processing the medical data;
Realize,
The medical data is raw data collected by a medical image diagnostic device or a medical image based on the raw data,
The acquisition conditions include an imaging protocol for medical imaging by the medical image diagnostic apparatus, data processing conditions for the raw data, and/or image processing conditions for the medical image,
The second acquisition function acquires the numerical data that digitizes the collection conditions of medical imaging for the subject;
The first acquisition function acquires, as the medical data, the raw data collected by the medical imaging according to the collection conditions,
The generation function performs a reconstruction process on the input data including the numerical data and the raw data using the machine learning model to generate the medical image regarding the subject.
Medical information processing program.