Generation of a Magnetic Resonance Fingerprinting Pulse Sequence

Abstract

This specification discloses medical systems (100, 300). Execution of machine-executable instructions (120) causes a computing system to repeatedly generate random input vectors, receive an MRF pulse sequence generated in response to inputting the random input vectors into a GAN generator neural network, and add the generated MRF pulse sequence to an MRF pulse sequence database (130) (204). By execution of the machine-executable instructions, the computing system inputs each generated MRF pulse sequence in the MRF pulse sequence database into a trained scoring algorithm (124) to assign one or more score values to each generated MRF pulse sequence, and receives (210) an MRF pulse sequence selected from the MRF pulse sequence database by applying a predetermined criterion to the one or more score values of each generated MRF pulse sequence in the MRF pulse sequence database.

Description

Technical Field

[0001] The present invention relates to magnetic resonance imaging, and more particularly to magnetic resonance fingerprinting.

Background Art

[0002] Magnetic resonance fingerprinting (MRF) is a technique in which a large number of RF pulses are applied over time such that the signals from different materials or tissues have a unique contribution to the measured MR signal. The signal depends on not only the physical parameters of the tissue (T1, T2, etc.) but also the parameters of the magnetic fields such as B0 and B1. These are encoded in various combinations, resulting in perhaps tens of thousands, or even hundreds of thousands, of entries in a magnetic resonance fingerprinting dictionary.

Summary of the Invention

Problems to be Solved by the Invention

[0003] In most applications, MRF is used to determine highly resolved quantitative maps of T1, T2, or other physical tissue parameters. Optimization of the sequence is important because it is necessary to ensure that small changes in T1, T2, or other parameters are uniquely reflected in measurable changes in the signal shape.

[0004] The discussion in the paper "Machine learning for rapid Magnetic resonance fingerprinting tissue property quantification" by Hamilton, Jesse I., and Nicole Seiberlich (IEEE 108.1(2019):69-8) is that magnetic resonance fingerprinting (MRF) is a magnetic resonance imaging method that can simultaneously provide quantitative maps of multiple tissue properties from a single rapid acquisition. The tissue property maps are generated by matching the complex signal evolution collected by the scanner to a dictionary of signals derived using Bloch equation simulations. Further, it is disclosed that in some situations, the processes of dictionary generation and signal matching are time-consuming, which may reduce the usefulness of this technology. In recent years, several groups have proposed using machine learning to accelerate the extraction of quantitative maps from MRF data. In this paper, an overview of research combining MRF and machine learning is provided, along with an original study demonstrating how machine learning can speed up the dictionary generation of cardiac MRF (cMRF). The preprint "Game of Learning Bloch Equation Simulations fc MR Fingerprinting" by M. Yang et al., Arxiv.org, Cornell University Library, 5 April 2020, is related to GAN-based generation of an MRF dictionary (not the acquisition sequence) from random noise inputs.

Means for Solving the Problem

[0005] The present invention provides a medical system, a computer program, and a method in the independent claims. Embodiments are given in the dependent claims.

[0006] The above-mentioned paper by Hamilton et al. points out the time-consuming process of generating a dictionary for MRF. Generating a pulse sequence for executing MRF may also be time-consuming. It may be necessary to repeatedly modify the MRF pulse sequence and then calculate its effectiveness using Bloch simulation. Executing the optimization of the MRF pulse sequence can be computationally intensive. Embodiments can provide means for generating the MRF pulse sequence more rapidly. A Generative Adversarial Network (GAN) can be trained to generate an appropriate MRF pulse sequence. A GAN is formed by a GAN generator neural network and a GAN discriminator pair. This pair is trained with a GAN discriminator that is trained to distinguish between a true MRF pulse sequence and a false MRF pulse sequence. In this context, the GAN discriminator is trained to recognize an MRF pulse sequence that provides good information encoding ability. The GAN generator neural network is trained to generate a generated MRF pulse sequence in response to receiving a random input vector. The dimensionality of the random input vector can be selected. The GAN generator neural network is trained such that the generated MRF pulse sequence can fool the GAN discriminator. This ultimately achieves an increase in the GAN discriminator and improves its ability to identify an appropriate MRF pulse sequence.

[0007] The present invention as set forth in the set of claims 1 to 15 relates to a combination of GAN-based generation of a magnetic resonance fingerprinting (MRF) acquisition sequence with a trained score algorithm for generating a score value associated with the generated MRF acquisition sequence. Generative Adversarial Networks (GAN) in the present invention have the ability to synthesize a realistic new dataset from a given set in order to generate additional MRF sequences from a ground truth set of known good MRF acquisition sequences. A GAN comprises a generator and a discriminator network. The generator generates data based on whether the discriminator's flag is a real number. The discriminator assigns a correctness flag to the data based on learning from real data. The training of the two networks is performed alternately so that each can better generate and distinguish at the same pace (ideally) and gradually. The present invention achieves a faster generation of MRF acquisition sequences that can be generated on-the-fly or on-demand. The present invention avoids the need for very time-consuming Bloch simulations. The generator is supplied with noise to generate a random sequence definition, i.e., to return a new MRF sequence in response to a noise realization input. The discriminator is trained to distinguish such "fake" MRF sequences from the "true" MRF sequences provided for training. By repeatedly optimizing the weights of the two networks, the generator learns to generate MRF sequences having characteristics similar to those of the known ground truth set of MRF sequences. To quantify the similarity of the generated MRF sequences to their ground truth counterparts, the score values are due to the individual generated MRF acquisition sequences, which represent the encoding ability of the MRF sequences. The generator is supplied with noise and a random sequence definition is created. The discriminator is trained to distinguish such "fake" MRF sequences from the "true" MRF sequences provided for training. By repeatedly optimizing the weights of the two networks, the generator learns to generate MRF sequences having characteristics similar to those of the known ones.The discriminator is trained to distinguish such "false or spurious" MRF sequences from the "true" MRF sequences provided for training. By iteratively optimizing the weights of the two networks, the generator learns to generate MRF sequences with characteristics similar to those known. According to the present invention, the generated MRF sequences are selected based on their score values. The score value of the MRF acquisition sequence represents the encoding ability of the MR acquisition sequence, i.e., the discrimination level for each parameter setting in the MRF acquisition sequence. According to the present invention, the score value is returned in response to the generated MRF acquisition sequence by a trained scoring algorithm trained to predict the encoding ability. The score value may be implemented as the inner product between pairs of MRF sequences and, optionally, averaged over the parameter space of the imaging parameters that vary over the MRF sequence. Noise is supplied to the generator and a random sequence definition is created. These scores provide a metric for encoding ability and thus select the generated MRF acquisition sequences that have good discrimination for the parameter settings. The score value is known per se from the paper "Towards predicting encoding capability of MR fingerprinting sequences" by K. Sommer et al. in Magnetic Resonance Imaging 41(2017)7 - 14. Thus, the present invention functions to generate MRF acquisition sequences based on a ground truth set of MRF acquisition sequences, and the optimal solution among the generated acquisition sequences is selected based on the score value representing the encoding ability of the generated MRF acquisition sequences. In this way, the GAN can be further trained during the operation of the GAN. As a result of the selection of the MRF acquisition sequences, the MRF dictionary only needs to be calculated, for example, by Bloch simulation for the selected MRF acquisition sequences, and for this purpose, good encoding ability is established by their scoring values.

[0008] Embodiments of the present invention also include a trained scoring algorithm trained to output one or more score values in response to the reception of an MRF pulse sequence. The trained scoring algorithm replaces the use of a Bloch simulator to evaluate the effectiveness of a particular MRF pulse sequence.

[0009] To generate a new MRF pulse sequence, random input vectors are repeatedly generated and input into a GAN generator neural network. This can be used to generate a number of generated MRF pulse sequences. These can be stored, for example, in a file folder or an MRF pulse sequence database. As used herein, a database is a system for storing and retrieving data. This can range from a relational database to a system for finding stored files.

[0010] The trained scoring algorithm is then used to screen an MRF pulse sequence database. Each generated MRF pulse sequence can be input into the trained scoring algorithm to obtain one or more score values that can be compared to a predetermined criterion. For example, one or more score values can be calculated when the generated pulse sequence is stored in the MRF pulse sequence database, or one or more score values can be calculated on the fly when searching for a selected MRF pulse sequence.

[0011] The advantage of generating an MRF pulse sequence selected in this way can be that it can be very fast. The GAN generator neural network can generate a number of generated MRF pulse sequences very quickly, and the trained scoring algorithm can be used to quickly screen the MRF pulse sequence database. It can, for example, enable the generation of MRF pulse sequences on the fly or on demand.

[0012] A hybrid approach may also be used. For example, a trained scoring algorithm may be used to identify a subset of MRF pulse sequences from an MRF pulse sequence database that meets the screening criteria. Then, for members of the subset of MRF pulse sequences, numerically calculated score values may be calculated to select the MRF pulse sequence that most closely matches the requirements specified by the screening criteria. The numerically calculated score values can be calculated using a Bloch simulator. The advantage of this approach is that the number of numerical simulations is significantly reduced.

[0013] In one aspect, the present invention provides a medical system. The medical system includes a memory that stores several things. The memory stores machine-executable instructions, a GAN generator neural network, and a trained scoring algorithm. The GAN generator neural network is configured to output a generated magnetic resonance fingerprinting (MRF) pulse sequence in response to receiving a random input vector. The trained scoring algorithm is configured to output one or more score values in response to receiving the generated MRF pulse sequence as an input.

[0014] The medical system further includes a computing system. Execution of the machine-executable instructions causes the computing system to repeatedly generate a random input vector. Execution of the machine-executable instructions further causes the computing system to repeatedly receive the generated MRF pulse sequence in response to inputting the random input vector into the GAN generator neural network.

[0015] The execution of the machine-executable instructions further causes the computing system to repeatedly add the generated MRF pulse sequence to the MRF pulse sequence database. In response to receiving a random input vector, the GAN generator neural network outputs a valid or generated MRF pulse sequence. Typically, constructing a highly optimized MRF pulse sequence is very computationally intensive. By having a GAN trained to generate a GAN in response to receiving a random input vector, a large number of generated MRF pulse sequences can be constructed in a very short period of time. This large number of generated MRF pulse sequences can then be stored in a database system.

[0016] The execution of the machine-executable instructions further causes the computing system to input each generated MRF pulse sequence in the MRF pulse sequence database into a trained scoring algorithm to assign one or more score values to each generated MRF pulse sequence. The execution of the machine-executable instructions further causes the computing system to select an MRF pulse sequence selected from the MRF pulse sequence database by applying a predetermined criterion to one or more score values of each generated MRF pulse sequence in the MRF pulse sequence database. The GAN generator neural network can be used to generate a large number of MRF pulse sequences. However, for a particular application or use, it may not be obvious which is the best MRF pulse sequence to select from the MRF pulse sequence database.

[0017] To perform this in a very rapid manner, a trained scoring algorithm outputs one or more score values when an MRF pulse sequence is input thereto. These various score values can have different values associated with different image weighting or image quality aspects. Next, passing through the MRF pulse sequence database, one or more score values of each pulse sequence within the MRF pulse sequence database can be looked at or compared to a predetermined criterion. For example, the predetermined criterion can emphasize one particular image weighting factor such as T1 or T2. In other examples, several different factors can be considered simultaneously and a weighting between them can be used. In any case, the use of these two machine learning modules, the GAN generator neural network and the trained scoring algorithm, can enable a very rapid selection of a selected MRF pulse sequence from the MRF pulse sequence database.

[0018] In another embodiment, the execution of the machine-executable instructions further causes the computing system to compute a magnetic resonance fingerprinting (MRF) dictionary having a predetermined dictionary entry for a selected MRF pulse sequence by performing a Bloch equation simulation controlled by the selected MRF pulse sequence for each of the predetermined dictionary entries. In this embodiment, an MRF dictionary for the selected MRF pulse sequence is prepared. This can be beneficial as an appropriate MRF dictionary may be required to obtain useful information from the k-space data acquired with the selected MRF pulse sequence.

[0019] The MR signals in the dictionary can be calculated by modeling the response of each substance using the well-known Bloch equations for the pulse sequence. These predicted MR signal values can then be compared to the measured MR signals. Each of the substances in the dictionary can potentially make a positive contribution to the measured MR signal. Comparison over all or many measurement times enables accurate deconvolution of the composition of the regions contributing to the measured magnetic resonance signal with respect to the substances in the magnetic resonance fingerprinting dictionary.

[0020] In another embodiment, the medical system further comprises a magnetic resonance imaging system. Execution of the machine-executable instructions further causes the computing system to acquire MRF k-space data by controlling the magnetic resonance imaging system using a selected MRF pulse sequence. Execution of the machine-executable instructions further causes the computing system to reconstruct one or more magnetic resonance fingerprinting images using the MRF k-space data and the MRF dictionary according to a magnetic resonance fingerprinting protocol. This embodiment can be beneficial as one or more MRF images are reconstructed from the k-space data acquired using a selected MRF pulse sequence and an MRF dictionary customized therefor. This can provide magnetic resonance fingerprinting images with improved quality.

[0021] Acquisition of the MRF k-space data can involve adjustments to the selected MRF pulse sequence in some examples. For example, typically, the gradients are adjusted to control the field of view.

[0022] In another embodiment, the GAN generator neural network is further configured to receive one or more control switches as input. The one or more control switches may be values input to the input of the GAN generator neural network. Execution of the machine-executable instructions further causes the computing system to receive one or more control switches. Execution of the machine-executable instructions further causes the computing system to input one or more control switches to the GAN generator neural network using a random input vector. Execution of the machine-executable instructions further causes the computing system to store one or more control switches having the generated MRF pulse sequence in the MRF pulse sequence database. The selected MRF pulse sequence is selected at least in part using the selected set of control switches.

[0023] The control switch can be used, for example, to identify data about the subject, details about the acquisition, or even the anatomical region. Thus, this can enable a single GAN neural network to be used for various situations. For example, when different anatomical regions are being imaged, one solution is to have multiple specialized GAN generator neural networks. Another solution is to use control switches, and then during training, the control switches are also used so that the GAN generator is more flexible and its behavior can be controlled and configured.

[0024] In another embodiment, one or more control switches comprise an anatomical region.

[0025] In another embodiment, one or more control switches include a clinical question. The clinical question may be, for example, the type of examination being performed.

[0026] In another embodiment, one or more control switches include timing limitations related to the acquisition of k-space data. This can be useful in situations where there are repetitive movements such as heartbeats or breathing.

[0027] In another embodiment, one or more control switches include a duration of apnea.

[0028] In another embodiment, selecting an MRF pulse sequence selected from an MRF pulse sequence database by applying a predetermined criterion to one or more score values of each generated MRF pulse sequence in the MRF pulse sequence database includes calculating a composite score using one or more score values and selecting the generated MRF pulse sequence having the highest composite score. For example, one or more score values may comprise scores for different things, such as T1 weighting, T2 weighting, or detection of a particular tissue type. The composite score may be a way of balancing all these different factors and using them to select the selected MRF pulse sequence.

[0029] In another embodiment, execution of the machine-executable instructions further causes the computing system to receive user input. The user input specifies the composition of the composite score. The selected MRF pulse sequence is selected in response to receipt of the user input. For example, there may be a user interface that allows the user to select which of one or more of the score values are important. This can then be used to select the selected MRF pulse sequence.

[0030] In another embodiment, selecting an MRF pulse sequence selected from an MRF pulse sequence database includes selecting a subset of MRF pulse sequences from an MRF pulse sequence database that meets screening criteria. Selecting further includes calculating a set of numerically calculated score values for each of the subset of MRF pulse sequences. Selecting further includes selecting an MRF pulse sequence selected from the subset of MRF pulse sequences by comparing a predetermined criterion with the set of numerically calculated score values for each of the subset of MRF pulse sequences. A particular MRF pulse sequence can be evaluated by performing a simulation. For example, Bloch simulation can be used to determine how effective an MRF pulse sequence is in providing data that can be used for T1-weighted images. Since a trained scoring algorithm is a trained or machine learning component, it can also perform numerical determination of one or more score values, and thus may have limitations. For example, the database can have a large number of MRF pulse sequences. A trained scoring algorithm can be used to first screen and reduce the number of MRF pulse sequences that need to be searched. The MRF pulse sequences identified by the trained scoring algorithm can then be refined by performing numerical calculation of score values for each of the subset of MRF pulse sequences.

[0031] In another embodiment, the score value includes a T1 accuracy score.

[0032] In another embodiment, the score value includes a T2 accuracy score.

[0033] In another embodiment, the score value includes a T2-star accuracy score.

[0034] In another embodiment, one or more score values include a proton density accuracy score.

[0035] In another embodiment, the execution of the machine-executable instructions further causes the computing system to receive a training MRF pulse sequence and, in accordance with the GAN training protocol, use the training MRF pulse sequence to train a GAN generator neural network and a GAN discriminator neural network. In a GAN, two different neural network components are trained in cooperation. There is a GAN generator neural network and a GAN discriminator neural network. As described above, the GAN generator neural network receives a random input vector and outputs an MRF pulse sequence in response thereto. The GAN discriminator network is trained to identify whether the MRF pulse sequence is real or not. The GAN generator network is trained using the GAN discriminator network, and in turn, the GAN discriminator neural network is trained using the training MRF pulse sequence generated by the GAN generator neural network and an error or fake MRF pulse sequence.

[0036] In another embodiment, the execution of the machine-executable instructions further causes the computing system to receive a previous MRF pulse sequence. The execution of the machine-executable instructions further causes the computing system to calculate a set of numerically evaluated score values for each of one or more score values of the previous MRF pulse sequence. This can include, for example, setting a set of Bloch equations and numerically solving one or more score values. The execution of the machine-executable instructions further causes the computing system to use the previous MRF pulse sequence and the set of numerically evaluated score values to train a trained scoring algorithm. The exact training method depends on the type of algorithm or machine learning module used to implement the trained scoring algorithm.

[0037] Training of a GAN generator neural network and a GAN discriminator neural network (600) using a training MRF pulse sequence according to a GAN training protocol can be performed by a medical system in which a trained scoring algorithm is deployed for inference. The training may also be performed separately from, or remotely from, the medical system in which the trained scoring algorithm is deployed. Further, the GAN network may be remotely trained for use in an initially trained GAN network deployed in the medical system of the present invention, and thereafter, the deployed GAN operation, in particular, the trained scoring algorithm, may receive continued training while being deployed.

[0038] In another embodiment, the trained scoring algorithm is implemented as a decision tree. In another embodiment, the trained scoring algorithm is implemented as a support vector machine.

[0039] In another embodiment, the trained scoring algorithm is implemented as a multi-layer perceptron neural network. In another embodiment, the trained scoring algorithm is implemented as a ResNet neural network.

[0040] In another embodiment, the trained scoring algorithm is implemented as a convolutional neural network.

[0041] In another embodiment, the trained scoring algorithm is implemented as a U-Net neural network.

[0042] In another aspect, the present invention provides a computer program configured for execution of instructions by a computing system. Execution of the machine-executable instructions causes the computing system to repeatedly generate random input vectors. Execution of the machine-executable instructions further causes the computing system to repeatedly receive a generated MRF pulse sequence in response to inputting the random input vector into a GAN generator neural network. The GAN generator neural network is configured to output a generated MRF pulse sequence in response to receiving the random input vector.

[0043] Execution of the machine-executable instructions further causes the computing system to add the generated MRF pulse sequence to an MRF pulse sequence database.

[0044] Execution of the machine-executable instructions further causes the computing system to input each generated MRF pulse sequence in the MRF pulse sequence database into a training score algorithm to assign one or more score values to each generated MRF pulse sequence. The trained scoring algorithm is configured to output one or more score values in response to receiving the generated MRF pulse sequence as input. Execution of the machine-executable instructions further causes the computing system to receive an MRF pulse sequence selected from the MRF pulse sequence database by applying a predetermined criterion to the one or more score values of each generated MRF pulse sequence in the MRF pulse sequence database.

[0045] In another aspect, the present invention provides a method. The method includes repeatedly generating a random input vector. The method further includes repeatedly receiving a generated MRF pulse sequence in response to inputting the random input vector into a GAN generator neural network. The GAN generator neural network is configured to output a generated MRF pulse sequence in response to receiving the random input vector. The method further includes repeatedly adding the generated MRF pulse sequence to an MRF pulse sequence database.

[0046] The method further includes inputting each generated MRF pulse sequence in the MRF pulse sequence database into a trained scoring algorithm to assign one or more score values to each generated MRF pulse sequence. The trained scoring algorithm is configured to output one or more score values in response to receiving the generated MRF pulse sequence as an input. The method further includes receiving an MRF pulse sequence selected from the MRF pulse sequence database by applying a predetermined criterion to the one or more score values of each generated MRF pulse sequence in the MRF pulse sequence database.

[0047] It is understood that one or more of the foregoing embodiments of the present invention can be combined as long as the combined embodiments are not mutually exclusive.

[0048] As will be understood by those skilled in the art, aspects of the present invention may be embodied as an apparatus, method, or computer program product. Accordingly, aspects of the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment (including firmware, resident software, microcode, etc.), or an embodiment combining software aspects and hardware aspects that may generally be referred to herein as a “circuit,” “module,” or “system,” and furthermore, aspects of the present invention may take the form of a computer program product embodied in one or more computer-readable media having computer-executable code embodied thereon.

[0049] Any combination of one or more computer-readable media may be utilized. The computer-readable media may be a computer-readable signal media or a computer-readable storage media. As used herein, "computer-readable storage media" includes any tangible storage media that can store instructions executable by a processor of a computing device or a computing system. The computer-readable storage media may sometimes be referred to as computer-readable non-transitory storage media. The computer-readable storage media may also sometimes be referred to as tangible computer-readable media. In some embodiments, the computer-readable storage media may also be capable of storing data that can be accessed by a computing system of a computing device. Examples of computer-readable storage media include, but are not limited to, floppy disk (®) disks, magnetic hard disk drives, solid state hard disks, flash memory, USB thumb drives, random access memory (RAM), read only memory (ROM), optical disks, magneto-optical disks, and register files of a computing system. Examples of optical disks include compact disks (CDs) and digital versatile disks (DVDs), such as CD-ROM, CD-RW, CD-R, DVD-ROM, DVD-RW, or DVD-R disks. The term computer-readable storage media also refers to various types of recording media that can be accessed by a computer device via a network or a communication link. For example, data may be retrieved via a modem, via the Internet, or via a local area network. The computer-executable code embodied on the computer-readable media may be transmitted using any suitable media, including but not limited to wireless, wireline, fiber optic cable, RF, etc., or any suitable combination of the foregoing.

[0050] A computer-readable signal medium may include a propagated data signal embodying computer-executable code therein, for example, at baseband or as part of a carrier wave. Such a propagated signal may take any of a variety of forms including, but not limited to, electromagnetic, optical, or any suitable combination thereof. A computer-readable signal medium may be any computer-readable medium that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device, rather than a computer-readable storage medium.

[0051] "Computer memory" or "memory" is an example of a computer-readable storage medium. Computer memory is any memory directly accessible to a computing system. "Computer storage device" or "storage" is a further example of a computer-readable storage medium. A computer storage device is any non-volatile memory computer-readable storage medium. In some embodiments, a computer storage device may be a computer memory, and vice versa.

[0052] As used herein, "computing system" includes electronic components capable of executing a program, machine-executable instructions, or computer-executable code. References to a computing system that include examples of a computing system should, in some cases, be construed as including two or more computing systems or processing cores. A computing system may be, for example, a multi-core processor. A computing system may also refer to a set of computing systems that are distributed within a single computer system or across multiple computer systems. The term computing system should probably also be construed as referring to a set or network of computing devices, each having a processor or computing system. Machine-executable code or instructions may be within the same computing device or may be executed by multiple computing systems or processors distributed across multiple computing devices.

[0053] Machine-executable instructions or computer-executable code may comprise instructions or programs that cause a processor or other computing system to execute aspects of the present invention. Computer-executable code for performing the operations for aspects of the present invention may be written in any combination of one or more programming languages, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as the "C" programming language or similar programming languages, and may be compiled into machine-executable instructions. In some cases, the computer-executable code may be in the form of a high-level language or a pre-compiled form and may be used in conjunction with an interpreter that generates machine-executable instructions on the fly. In other cases, the machine-executable instructions or computer-executable code may be in the form of a program for a programmable logic gate array.

[0054] The computer-executable code can be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer via any type of network including a local area network (LAN) or a wide area network (WAN), or the connection may be made to an external computer (e.g., via the Internet using an Internet service provider).

[0055] Aspects of the present invention will be described with reference to the flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block or portion of the blocks in the flowchart illustrations, diagrams, and / or block diagrams may be implemented by computer program instructions in the form of computer-executable code, where applicable. Further, note that combinations of blocks in different flowchart illustrations, diagrams, and / or block diagrams may be combined if not mutually exclusive. These computer program instructions may be provided to a computing system of a general purpose computer, a special purpose computer, or other programmable data processing apparatus to create a machine that implements means for performing the functions / operations specified in the block or blocks of the flowchart and / or block diagram via the computing system of the computer or other programmable data processing apparatus.

[0056] These machine-executable instructions or computer program instructions may also be stored in a computer-readable medium that can direct a computer, other programmable data processing apparatus, or other device to function in a particular manner, such that the instructions stored in the computer-readable medium produce a manufacture including instructions for performing the functions / operations specified in the block or blocks of the flowchart and / or block diagram.

[0057] The machine-executable instructions or computer program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer to implement a computer-implemented process that provides instructions executed on the computer or other programmable apparatus to implement the functions / operations specified in the flowchart and / or block diagram block or blocks. As used herein, a "user interface" is an interface that enables a user or operator to interact with a computer or computer system. A "user interface" may also be referred to as a "human interface device." A user interface can provide information or data to an operator and / or receive information or data from an operator. A user interface can enable input from an operator to be received by a computer and can provide output from the computer to the user. In other words, a user interface can enable an operator to control or operate a computer, and the interface can enable the computer to indicate the effects of the operator's control or operation. The display of data or information on a display or graphical user interface is an example of providing information to an operator. The reception of data via a keyboard, mouse, trackball, touchpad, pointing stick, graphics tablet, joystick, gamepad, webcam, headset, pedal, wired glove, remote control, and accelerometer are all examples of user interface components that enable the reception of information or data from an operator. As used herein, "hardware interface" encompasses an interface that enables a computing system of a computer system to interact with and / or control an external computing device and / or apparatus. The hardware interface may enable the computer system to transmit control signals or instructions to the external computing device and / or apparatus. The hardware interface may also enable the computing system to exchange data with the external computing device and / or apparatus. Examples of hardware interfaces include, but are not limited to, Universal Serial Bus, IEEE 1394 port, parallel port, IEEE 1284 port, serial port, RS232 port, IEEE488 port, Bluetooth connection, wireless local area network connection, TCP / IP connection, Ethernet connection, control voltage interface, MIDI interface, analog input interface, and digital input interface.

[0058] As used herein, "display" or "display device" encompasses an output device or user interface adapted to display an image or data. The display can output visual data, audio data, and / or tactile data. Examples of displays include, but are not limited to, computer monitors, television screens, touchscreens, tactile electronic displays, braille screens, cathode ray tubes (liquid), storage tubes, bistable displays, electronic paper, vector displays, flat panel displays, vacuum fluorescent displays (VF), light emitting diode (LED) displays, electroluminescent displays (ELD), plasma display panels (PDP), liquid crystal displays (LCD), organic light emitting diode displays (OLED), projectors, and head-mounted displays.

[0059] k-space data is defined herein as the recorded measurements of the high-frequency signals emitted by atomic spins using an antenna of a magnetic resonance apparatus during a magnetic resonance imaging scan. Magnetic resonance data is an example of tomographic image data.

[0060] A magnetic resonance imaging (MRI) image or an MR image is defined herein as a reconstructed two-dimensional or three-dimensional visualization of anatomical data contained within magnetic resonance imaging data. This visualization can be performed using a computer.

[0061] In the following, preferred embodiments of the present invention will be described by way of example with reference to the drawings.

Brief Description of the Drawings

[0062]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Modes for Carrying Out the Invention

[0063] Elements with the same number in these figures are equivalent elements or perform the same function. The foregoing elements are not necessarily discussed in later figures if their functions are equivalent.

[0064] Figure 1 shows an example of a medical system 100. The medical system 100 includes a computer 102 having a computing system 104. The computer 102 can represent one or more computers located at one or more positions. The computing system 104 can also represent one or more computing systems such as one or more computing cores. The computing system 104 is shown as communicating with an optional hardware interface 106 and an optional user interface 108. The hardware interface 106 can enable the computing system 104 to communicate control, send control, and receive data from other components of the medical system 100 if they exist. The optional user interface 108 can also enable an operator to control and operate the medical system 100.

[0065] The computing system 104 is shown as further communicating with a memory 110. The memory 110 is intended to represent various types of memory accessible by the computing system 104. The memory 110 may be a non-transitory storage medium. The memory 110 is shown as including machine-executable instructions 120. The machine-executable instructions 120 enable the computing system 104 to perform basic tasks such as numerical calculations and / or image processing. The machine-executable instructions 120 can also enable the computing system 104 to control other components or devices via the hardware interface 106. The memory 110 is further shown as including a GAN generator neural network 122 and a trained scoring algorithm 124. The memory 110 is further shown as including a random input vector 126 and a generated MRF pulse sequence 128.

[0066] The random input vector 126 can be input into the GAN generator neural network 122 to generate or output the generated MRF pulse sequence 128. By generating a number of random input vectors 126, a number of generated MRF pulse sequences 128 can be generated. The memory 110 is further shown as including an MRF pulse sequence database 130. For example, then, a number of generated MRF pulse sequences 128 can be stored in the MRF pulse sequence database 130. The memory 110 is further shown as including one or more score values 132 generated for the MRF pulse sequence 128 by inputting the generated MRF pulse sequence 128 into the trained scoring algorithm 124. One or more score values 132 can be present for each of the generated MRF pulse sequences 128 within the MRF pulse sequence database 130. This can be done at different times. For example, after the generated MRF pulse sequence 128 is generated by the GAN generator neural network 122, one or more score values 132 can be generated and stored simultaneously in the MRF pulse sequence database 130. In other examples, one or more score values 132 can be generated on the fly when the MRF pulse sequence database 130 is searched. In both cases, there can be a predetermined criterion 134 used to select the MRF pulse sequence 136 selected from the MRF pulse sequence database 130. The predetermined criterion 134 can be means for evaluating one or more score values 132 for each of the MRF pulse sequence database 130.

[0067] Figure 2 shows a flowchart illustrating a method of operating the medical system 100 of FIG. 1. First, at step 200, a random input vector 126 is generated. Next, at step 202, in response to inputting the random input vector 126 into the GAN generator neural network 122, a generated MRF pulse sequence 128 is received. At step 204, the generated MRF pulse sequence 128 is added to the MRF pulse sequence database 130. In this example, one or more score values 132 have not yet been generated and are not yet placed in the MRF pulse sequence database 130. However, at this point, after step 202 is executed, the generation of one or more score values 132 for a particular generated MRF pulse sequence 128 can be calculated by inputting the generated MRF pulse sequence 128 into a trained scoring algorithm 124. In this case, when the generated MRF pulse sequence 128 is added to the MRF pulse sequence database 130, one or more score values 132 can be stored and associated with the record of this generated MRF pulse sequence 128 simultaneously.

[0068] The method then proceeds to box 206, which is a decision box. In this case, the question is, "Has the algorithm for generating the MRF pulse sequence ended?" If the answer is no, the method returns to step 200, another random input vector 126 is generated, and the process repeats itself. If the answer is yes, the method proceeds to box 208. In box 208, each generated MRF pulse sequence in the MRF pulse sequence database is input into a trained scoring algorithm to assign one or more score values to each generated MRF pulse sequence. Next, and finally, at step 210, a selected MRF pulse sequence 136 is received from the MRF pulse sequence database 130 by applying a predetermined criterion to one or more score values 132 of each generated MRF pulse sequence 128 in the MRF pulse sequence database 130.

[0069] Figure 3 shows a further example of a medical system 300. The medical system 300 shown in Figure 3 is similar to the medical system 100 shown in Figure 1, but differs in that it further includes a magnetic resonance imaging system 302 controlled by a computing system 104.

[0070] The magnetic resonance imaging system 302 includes a magnet 304. The magnet 304 is a superconducting cylindrical magnet having a type bore 306. The use of different types of magnets is also possible. For example, it is also possible to use both a split cylindrical magnet and a so-called open magnet. The split cylindrical magnet is similar to a standard cylindrical magnet except that the cryostat is divided into two sections to allow access to the isoplanes of the magnet, and such a magnet can be used, for example, in combination with charged particle beam therapy. The open magnet has two magnet sections, one on top of the other, with a space large enough to receive a subject, i.e., an arrangement of two section regions similar to the region of a Helmholtz coil. The open magnet is popular because there is less confinement of the subject. Inside the cryostat of the cylindrical magnet is an assembly of superconducting coils.

[0071] Inside the bore 306 of the cylindrical magnet 304 is an imaging zone 308 where the magnetic field is strong and uniform enough to perform magnetic resonance imaging. A field of view 309 is shown within the imaging zone 308. K-space data typically acquired for the field of view 309. The region of interest may be the same as the field of view 309 or a sub-volume of the field of view 309. The subject 318 is shown as being supported by a subject support 320 such that at least a portion of the subject 318 is within the imaging zone 308 and the field of view 309.

[0072] Within the bore 306 of the magnet, there is also a set of magnetic field gradient coils 310 used for acquiring preliminary k-space data to spatially encode the magnetic spins within the imaging zone 308 of the magnet 304. The magnetic field gradient coils 310 are connected to a magnetic field gradient coil power supply 312. The magnetic field gradient coils 310 are intended to be representative. Typically, the magnetic field gradient coils 310 include three separate coil sets for spatially encoding in three orthogonal spatial directions. The magnetic field gradient power supply supplies current to the magnetic field gradient coils. The current supplied to the magnetic field gradient coils 310 is controlled as a function of time and can be ramped or pulsed.

[0073] Adjacent to the imaging zone 308, there is a high-frequency coil 314 for manipulating the orientation of the magnetic spins within the imaging zone 308 and for receiving wireless transmissions from the spins within the imaging zone 308. The high-frequency antenna can include a plurality of coil elements. The high-frequency antenna may also be referred to as a channel or an antenna. The high-frequency coil 314 is connected to a high-frequency transceiver 316. The high-frequency coil 314 and the high-frequency transceiver 316 can be replaced by separate transmit and receive coils, as well as separate transmitters and receivers. It is understood that the high-frequency coil 314 and the high-frequency transceiver 316 are representative. The high-frequency coil 314 is also intended to represent a dedicated transmit antenna and a dedicated receive antenna. Similarly, the transceiver 316 can also represent separate transmitters and receivers. The high-frequency coil 314 can also have a plurality of receive / transmit elements, and the high-frequency transceiver 316 can have a plurality of receive / transmit channels.

[0074] The transceiver 316 and the gradient controller 312 are shown as being connected to the hardware interface 106 of the computer system 102. Both of these components, as well as other components such as the subject support that supplies position data, can supply sensor data 126.

[0075] Memory 110 is further shown as including a Bloch simulator. Memory 110 is further shown as including an MRF dictionary 332. The MRF dictionary 332 can be constructed by controlling the Bloch simulator 330 using a selected MRF pulse sequence 136. Memory 110 is further shown as including MRF k-space data 334 obtained by controlling the magnetic resonance imaging system 302 with a selected MRF pulse sequence 136. Memory 110 is further shown as including an MRF image 336 reconstructed from the MRF k-space data 334 using the MRF dictionary 332.

[0076] Figure 4 shows a flowchart illustrating a method of operating the medical system 300 of FIG. 3. The method shown in FIG. 4 is similar to the method shown in FIG. 2, with additional steps being performed. The method shown in FIG. 4 includes steps 200 through 210 shown in FIG. 2. After step 210, step 400 is further performed. In step 400, the MRF dictionary 332 is calculated by controlling the Bloch simulator 330 using a selected MRF pulse sequence 136. Next, in step 402, the MRF k-space data 334 is obtained by controlling the magnetic resonance imaging system 302 using a selected MRF pulse sequence 136. Finally, in step 404, one or more MRF images 336 are reconstructed using the MRF k-space data 334 and the MRF dictionary 332.

[0077] Magnetic resonance fingerprinting (MRF) is a relatively new method for quantitative MRI. Multiple quantitative parameters are simultaneously encoded in the transient signal evolution by using a series of RF pulses having varying characteristics (flip angle, phase, timing, spoiling, etc.). An MRF sequence typically consists of hundreds of RF pulses, resulting in a large parameter space of hundreds or thousands of dimensions. Thus, finding an optimized MRF sequence is a difficult task.

[0078] Generative Adversarial Networks (GAN) is a relatively new AI technology that is very good at synthesizing realistic new datasets from a given set. There are impressive examples on the web that introduce cases of creating comic characters or photos.

[0079] GAN consists of a generator and a discriminator network. The generator generates data based on whether the discriminator's flag is a real number or not. The discriminator sets a correct / incorrect flag for the data based on learning from real data. The training of the two networks is done alternately so that each can better generate and distinguish at the same pace (ideally) and gradually.

[0080] An example can facilitate the synthesis of an optimized MRF sequence from a set of known ones by using a Generative Adversarial Network (GAN). An example may include one or more of the following features. A set of known useful MR fingerprinting sequence parameters. A GAN that synthesizes new sequences. An AI algorithm that scores the quality of a fingerprinting sequence. A method of selecting a GAN-synthesized MRF sequence based on an AI prediction of a quality score.

[0081] An example can use a GAN to synthesize an optimized fingerprinting sequence from a set of known good sequences. It is necessary to define a metric for the quality of the MRF sequence. Typically, such a metric considers how well different regions within a quantitative parameter space can be distinguished. The exact definition of the quality metric is not important for this implementation.

[0082] By design, the GAN synthesis sequences are qualified as suitable fingerprinting sequences. The question is whether they are also good people. To determine how good the generated sequences are, it is proposed to implement another AI network trained to score the sequences. This second network can be trained from existing fingerprint sequences that have been scored.

[0083] Using this framework, an essentially infinite number of sequences can be generated and scored. The sequences with the best scores can be actually tried. If they are proven to be good, they can be integrated back into the discriminator training dataset.

[0084] Scoring the MRF sequences can be a very computationally intensive task. Typically, to determine the coding efficiency in different regions of the parameter space, it is necessary to calculate a large number of signals for different parameters (such as T1, T2, B1, B0, etc.). If a large number of MRF sequences need to be evaluated, this procedure can take a significant amount of time.

[0085] Examples can use AI algorithms (such as neural networks, trained scoring algorithms 124) for quality scoring (providing one or more score values 132). In this way, scoring can be realized much more quickly in order to quickly search a large number of MRF sequences to find the optimal one.

[0086] The training process of this algorithm is shown in Figure 5. For several MRF sequence definitions, quality scores are calculated. The AI algorithm is trained on the MRF sequence definitions as features and their respective quality scores as labels. After training, the AI algorithm can predict the quality scores of unknown MRF sequence definitions.

[0087] FIG. 5 shows a flowchart illustrating the training of a trained scoring algorithm 124, also referred to as a quality scoring AI algorithm. Using supervised learning, an MRF sequence definition 500 is input into a quality score calculation 502 and the quality scoring algorithm 124. The result of the quality score calculation 502 is also input into the quality scoring AI algorithm 124. The MRF sequence definition 500 provides features, and the quality score calculation 502 provides labels for supervised learning. The pulse sequence is basically a timing diagram. The MRF sequence definition 500 can be used to define the parameters of the timing sequence used to perform MRF acquisition.

[0088] FIG. 6 shows the training of a GAN generator neural network 122, or simply referred to as a GAN generator. The GAN generator 122 is paired with a GAN discriminator 600. A noise vector 126 is input into the GAN generator 122 to generate a generated MRF pulse sequence 128. The GAN discriminator 600 is trained to learn the difference between the MRF sequence definition 500 and the generated MRF pulse sequence 128. By alternately training the GAN generator 122 and the GAN discriminator 600, they both function better and better until the GAN generator 122 can generate an appropriate MRF pulse sequence.

[0089] A Generative Adversarial Network (GAN) is trained to generate an MRF sequence similar to a known MRF sequence with good encoding quality. A GAN consists of a discriminator network and a generator network. As shown in FIG. 6, both are trained in an iterative procedure. Noise is supplied to the generator, creating a random sequence definition. The discriminator is trained to distinguish such a "fake" MRF sequence from the "true" MRF sequence provided for training. By repeatedly optimizing the weights of the two networks, the generator learns to generate an MRF sequence with characteristics similar to the known ones.

[0090] Figure 7 shows one use of the GAN generator neural network 122 during deployment. In this flowchart, a random input vector 126 is input into the GAN generator 122. This then provides a generated MRF pulse sequence 128 that is input into the trained scoring algorithm 124. The trained scoring algorithm 124 then provides one or more score values 132. The one or more score values 132 can be used, for example, as a predetermined criterion 134 for selecting whether the generated MRF pulse sequence 128 is to be used.

[0091] Next, the trained network (GAN generator and quality scoring algorithm) is used in a sequential setup as shown in Figure 7. In the first step, the GAN generator generates a new MRF sequence from a noise input. In the second step, the quality scoring AI algorithm predicts the quality score of this sequence. After repeating this process many times, the MRF sequence with the best quality score is selected as the optimized MRF sequence.

[0092] Figure 8 shows an extension of the trained scoring algorithm 124 shown in Figure 4. In the example shown in Figure 7, as an additional feature, one or more control switches 800 are input. The one or more control switches 800 are used to encode, for example, the anatomical region to be imaged, details about the subject, and possibly clinical questions. This enables the trained scoring algorithm 124 to be configured or adjusted for specific acquisition conditions.

[0093] FIG. 9 shows a modification of the GAN generator neural network 122 shown in FIG. 5. In this example, in addition to the random input vector 126, one or more control switches 700 are also input to the GAN generator. By adding one or more control switches 800 to the GAN generator 122 during training, it becomes possible to control or configure the behavior of the GAN generator. As described above, one or more control switches 800 can be used to encode the anatomical structure to be imaged, the clinical question, or details about the subject to be imaged, etc.

[0094] Although the present invention has been illustrated and described in detail in the drawings and the foregoing description, such illustrations and descriptions should be regarded as illustrative or exemplary and not restrictive, and the present invention is not limited to the disclosed embodiments.

[0095] Other modifications to the disclosed embodiments can be understood and achieved by those skilled in the art when implementing the claimed invention from a study of the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite article "a" or "an" does not exclude a plurality. A single processor or other unit can fulfill the functions of several items listed in the claims. The mere fact that certain means are recited in mutually different dependent claims does not indicate that a combination of these means cannot be used advantageously. A computer program can be stored / distributed on a suitable medium such as an optical storage medium or a solid-state medium supplied together with or as part of other hardware, but can also be distributed in other forms such as via the Internet or other wired or wireless telecommunications systems. Any reference signs in the claims should not be construed as limiting the scope.

Explanation of Signs

[0096] 100 Medical system 102 Computer 104 Computing system 106 Hardware Interface 108 User Interface 110 Memory 120 Machine-Executable Instructions 122 GAN Generator Neural Network 124 Trained Scoring Algorithm 126 Random Input Vector 128 Generated MRF Pulse Sequence 130 MRF Pulse Sequence Database 132 One or More Score Values 134 Predetermined Criterion 136 Selected MRF Pulse Sequence 200 Generate Random Input Vector 202 Receive Generated MRF Pulse Sequence in Response to Inputting Random Input Vector into GAN Generator Neural Network 204 Add Generated MRF Pulse Sequence to MRF Pulse Sequence Database 208 Input Each Generated MRF Pulse Sequence in MRF Pulse Sequence Database into Trained Scoring Algorithm to Assign One or More Score Values to Each Generated MRF Pulse Sequence 210 Receive Selected MRF Pulse Sequence from MRF Pulse Sequence Database by Applying Predetermined Criterion to One or More Score Values of Each Generated MRF Pulse Sequence in MRF Pulse Sequence Database 300 Medical System 302 Magnetic Resonance Imaging Apparatus 304 Magnet 306 Bore of Magnet 308 Imaging Zone 309 Field of View 310 Magnetic Field Gradient Coil 312 Magnetic Field Gradient Coil Power Supply 314 Radio Frequency Coil 316 Transceiver 318 Subject 320 Subject support unit 330 Bloch simulator 332 MRF dictionary 334 MRF k-space data 336 MRF image 400 Calculating the MRF dictionary with a predetermined dictionary entry for the selected MRF pulse sequence by performing Bloch equation simulations controlled by the selected MRF pulse sequence for each of the predetermined dictionary entries 402 Acquiring MRF k-space data by controlling the magnetic resonance imaging system (402) 404 Reconstructing one or more MRF images using the MRF k-space data and the MRF dictionary (404) 500 MRF sequence definition 502 Quality score calculation 600 GAN discriminator 800 One or more control switches

Claims

1. A medical system, comprising a memory storing machine-executable instructions, a GAN generator neural network, and a trained scoring algorithm, wherein the GAN generator neural network is configured to output a generated MRF pulse sequence in response to receiving a random input vector, and the trained scoring algorithm is configured to output one or more score values in response to receiving the generated MRF pulse sequence as an input, a computing system, wherein the execution of the machine-executable instructions causes the computing system to generate a random input vector, to receive the generated MRF pulse sequence in response to inputting the random input vector into the GAN generator neural network, to add the generated MRF pulse sequence to an MRF pulse sequence database, and to repeatedly execute, wherein the execution of the machine-executable instructions causes the computing system to input each generated MRF pulse sequence in the MRF pulse sequence database into the trained scoring algorithm to assign the one or more score values to each generated MRF pulse sequence, and to receive a selected MRF pulse sequence from the MRF pulse sequence database by applying a predetermined criterion to the one or more score values of each generated MRF pulse sequence in the MRF pulse sequence database, a computing system. The medical system further comprising.

2. The medical system according to claim 1, wherein the execution of the machine-executable instructions further causes the computing system to calculate an MRF dictionary using a predetermined dictionary entry for the selected MRF pulse sequence by performing a Bloch equation simulation controlled by the selected MRF pulse sequence for each of the predetermined dictionary entries.

3. The medical system further comprises a magnetic resonance imaging system, and the execution of the machine-executable instructions causes the computing system Controlling the magnetic resonance imaging system using the selected MRF pulse sequence to obtain MRF k-space data; Reconstructing one or more MRF images using the MRF k-space data and the MRF dictionary; The medical system according to claim 2, which causes the above to be executed.

4. The GAN generator neural network is further configured to receive one or more control switches as inputs, and the execution of the machine-executable instructions further causes the computing system to Receive one or more control switches; Input the one or more control switches into the GAN generator neural network using the random input vector; Storing one or more control switches in the MRF pulse sequence database using the generated MRF pulse sequence, wherein the selected MRF pulse sequence is at least partially selected using a selected set of control switches; The medical system according to any one of claims 1 to 3, which causes the above to be executed.

5. The medical system according to claim 4, wherein the one or more control switches comprise any one of an anatomical region, a clinical question, a timing constraint, a breath-hold duration, and combinations thereof.

6. The step of selecting the selected MRF pulse sequence from the MRF pulse sequence database by applying a predetermined criterion to one or more score values of each generated MRF pulse sequence in the MRF pulse sequence database comprises: Calculating a composite score using the one or more score values; Selecting the generated MRF pulse sequence with the highest composite score as the selected MRF pulse sequence; The medical system according to any one of claims 1 to 5, which has the above.

7. The execution of the machine-executable instructions further causes the computing system to Receive a user input, wherein the user input specifies the composition of the composite score, and the selected MRF pulse sequence is selected in response to receiving the user input; The medical system according to claim 6, which causes the above to be executed.

8. The step of selecting the selected MRF pulse sequence from the MRF pulse sequence database comprises: selecting a subset of MRF pulse sequences from the MRF pulse sequence database that meet the screening criteria; calculating a set of numerically calculated score values for each of the subset of MRF pulse sequences; selecting the selected MRF pulse sequence from the subset of MRF pulse sequences by comparing the predetermined criteria with the set of numerically calculated score values for each of the subset of MRF pulse sequences The medical system according to any one of claims 1 to 5, comprising:

9. The one or more score values have any one of a T1 accuracy score, a T2 accuracy score, a T2* accuracy score, a proton density accuracy score, and combinations thereof. The medical system according to any one of claims 1 to 8.

10. The execution of the machine-executable instructions further causes the computing system to receive a training MRF pulse sequence; train a GAN generator neural network and a GAN discriminator neural network using the training MRF pulse sequence according to a GAN training protocol The medical system according to any one of claims 1 to 9, comprising:

11. The execution of the machine-executable instructions further causes the computing system to receive a previous MRF pulse sequence; calculate a set of score values numerically evaluated for each of one or more score values for the previous MRF pulse sequence; train the trained scoring algorithm using the previous MRF pulse sequence and the set of numerically evaluated score values The medical system according to any one of claims 1 to 10, comprising:

12. The trained scoring algorithm is implemented as any one of a decision tree, a support vector machine, a multi-layer perceptron neural network, a ResNET neural network, a convolutional neural network, and a U-Net neural network. The medical system according to any one of claims 1 to 11.

13. A computer program comprising machine-executable instructions for execution by a computing system, wherein execution of the machine-executable instructions causes the computing system to generate a random input vector; receive a generated MRF pulse sequence in response to inputting the random input vector into a GAN generator neural network, wherein the GAN generator neural network is configured to output the generated MRF pulse sequence in response to receiving the random input vector; add the generated MRF pulse sequence to an MRF pulse sequence database; and repeatedly execute, wherein execution of the machine-executable instructions causes the computing system to input each generated MRF pulse sequence in the MRF pulse sequence database into a trained scoring algorithm to assign one or more score values to each generated MRF pulse sequence, wherein the trained scoring algorithm is configured to output the one or more score values in response to receiving the generated MRF pulse sequence as input; receive a selected MRF pulse sequence from the MRF pulse sequence database by applying a predetermined criterion to the one or more score values of each generated MRF pulse sequence in the MRF pulse sequence database; and execute. A computer program.

14. A method, the method comprising: generating a random input vector; receiving a generated MRF pulse sequence in response to inputting the random input vector into a GAN generator neural network, wherein the GAN generator neural network is configured to output the generated MRF pulse sequence in response to receiving the random input vector; adding the generated MRF pulse sequence to an MRF pulse sequence database; and repeatedly executing; wherein the method Inputting each generated MRF pulse sequence in the MRF pulse sequence database into a trained scoring algorithm to assign one or more score values to each generated MRF pulse sequence, wherein the trained scoring algorithm is configured to output the one or more score values in response to receiving the generated MRF pulse sequence as input; Receiving a selected MRF pulse sequence from the MRF pulse sequence database by applying a predetermined criterion to the one or more score values of each generated MRF pulse sequence in the MRF pulse sequence database; A method further comprising. Claim 15 A computer-implemented training method for a scoring algorithm for a GAN generator neural network configured to output a generated MRF pulse sequence in response to receiving a random input vector, wherein the trained scoring algorithm outputs one or more score values in response to receiving the generated MRF pulse sequence as input and is configured to train the GAN generator neural network and the GAN discriminator neural network using the training MRF pulse sequence according to a GAN training protocol.

Images