# US 11,304,619 B2 · Generating setting parameters for an MRI sequence > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 11,304,619 B2 - Document ID: US11304619B2 - Status: Granted - Kind: B2 - Application number: 16/515,173 - Filed: 2019-07-18 - Published: 2020-01-23 - Granted: 2022-04-19 - Earliest claimed: 2018-07-18 - Assignees: Siemens Healthcare GmbH - Applicants: Siemens Healthcare GmbH - Inventors: Michael Zenge - Web page: https://patentgrep.com/patent/US11304619B2 ## CPC - A61B 5/055: involving electronic [EMR] or nuclear [NMR] magnetic resonance, e.g. magnetic resonance imaging - A61B 5/7267: {involving training the classification device} - A61B 5/7271: {Specific aspects of physiological measurement analysis (specific diagnostics methods using bioelectric or biomagnetic signals A61B5/316)} - G01R 33/50: based on the determination of relaxation times {, e.g. T1 measurement by IR sequences; T2 measurement by multiple-echo sequences} - G06N 3/02: Neural networks - G06N 3/045: Combinations of networks - G06N 3/0455: Auto-encoder networks; Encoder-decoder networks - G06N 3/0475: Generative networks - G06N 3/0499: Feedforward networks - G06N 3/084: Backpropagation, e.g. using gradient descent - G06N 3/09: Supervised learning - G06N 3/094: Adversarial learning - G16H 30/00: ICT specially adapted for the handling or processing of medical images (computed tomography A61B6/03) - G16H 40/63: for local operation - G01R 33/543: {Control of the operation of the MR system, e.g. setting of acquisition parameters prior to or during MR data acquisition, dynamic shimming, use of one or more scout images for scan plane prescription (G01R33/546 takes precedence)} - G06N 3/047: Probabilistic or stochastic networks - A61B: DIAGNOSIS; SURGERY; IDENTIFICATION (usefulness limited to only animals A61D) - G01R: MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES (indicating correct tuning of resonant circuits H03J3/12) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS - G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA ## Abstract Techniques are described for generating setting parameters for an MRI sequence to be performed, wherein a conditional generative artificial neural network is used that has been trained as part of a conditional generative adversarial net. The techniques also include establishing a conditional generative artificial neural network. The conditioning of the conditional generative artificial neural network can be used to specify a specific medical application or question to obtain setting parameters for an MRI apparatus that are optimized with respect to said application or question. ## Claims ### Claim 1 A system for generating setting parameters for a magnetic resonance imaging (MRI) sequence to be performed in accordance with an MRI scanner, comprising: an input interface configured to receive input data having a data structure corresponding to an input vector indicating a percentage of one or more types of MRI sequences to be performed; and a conditional generative artificial neural network configured to be trained as part of a conditional generative adversarial net to generate setting parameters for the MRI sequence to be performed based on the input data received from the input interface relating to the MRI sequence, the conditional generative adversarial net comprising the conditional generative artificial neural network as a generative part and a conditional discriminative artificial neural network as a discriminative part, wherein the conditional generative artificial neural network is provided with conditions to be fulfilled with respect to the percentage of the one or more types of MRI sequences to be performed so as to generate setting parameters for one T1-weighted MRI sequence, one T2-weighted MRI sequence, and one proton-density-weighted (PD-weighted) MRI sequence, wherein the conditions to be fulfilled are provided to the conditional generative artificial neural network as a concatenation of a condition vector with the input vector, and wherein the condition vector has a data structure comprising non-zero entries that respectively indicate a percentage of the one or more types of MRI sequences to be performed. ### Claim 2 (depends on 1) The system as claimed in claim 1, wherein the input data comprises or indicates at least one of the following percentages: a percentage with which the MRI sequence to be performed is to be T1-weighted, a percentage with which the MRI sequence to be performed is to be T2-weighted, and a percentage with which the MRI sequence to be performed is to be PD-weighted. ### Claim 3 (depends on 2) The system of claim 2, wherein the input data includes a plurality of input vectors, each respective one of the plurality of input vectors having respective entries corresponding to (i) a percentage with which the MRI sequence to be performed is to be T1-weighted, (ii) a percentage with which the MRI sequence to be performed is to be T2-weighted, and (iii) a percentage with which the MRI sequence to be performed is to be PD-weighted. ### Claim 4 (depends on 3) The system of claim 3, wherein the conditional generative artificial neural network is provided with the conditions to be fulfilled by concatenating a respective one of a plurality of condition vectors with a respective one of the plurality of input vectors. ### Claim 5 (depends on 4) The system of claim 4, wherein the plurality of condition vectors each have respective entries identifying which one of the T1-weighted MRI sequence, the T2-weighted MRI sequence, or the PD-weighted MRI sequence that an entry of a respectively concatenated one of the plurality of input vectors corresponds. ### Claim 6 (depends on 1) The system as claimed in claim 1, wherein the setting parameters comprise a repetition time and an echo time for the MRI sequence to be performed. ### Claim 7 (depends on 1) The system as claimed in claim 1, wherein the setting parameters comprise a turbo factor and/or a bandwidth for the MRI sequence to be performed. ### Claim 8 (depends on 1) The system as claimed in claim 1, wherein the conditional generative artificial neural network is additionally provided with conditions to be fulfilled with respect to at least one of the following conditions: a magnetic field strength for the MRI sequence to be performed, information regarding a location and/or an alignment of a sectional plane through a body on which the MRI sequence is to be performed, information regarding a location and/or a body part of a body on which the MRI sequence is to be performed, a body-fat percentage, and artifact suppression. ### Claim 9 (depends on 1) The system as claimed in claim 1, wherein the conditional generative artificial neural network comprises an input layer, an output layer, and an intermediate hidden layer. ### Claim 10 (depends on 9) The system as claimed in claim 9, wherein the conditional generative artificial neural network further comprises at least one hidden layer, and wherein the input layer and the at least one hidden layer form at least one part of an autoencoder network. ### Claim 11 (depends on 1) The system as claimed in claim 1, wherein the conditional generative artificial neural network includes a multilayer perceptron. ### Claim 12 A computer-implemented method for establishing a system for generating setting parameters for a magnetic resonance imaging (MRI) sequence to be performed in accordance with an MRI scanner, the method comprising: operating an MRI scanner to generate training data comprising input data relating to the MRI sequence to be performed, the input data having a data structure corresponding to an input vector including corresponding setting parameters as labels and indicating a percentage of one or more types of MRI sequences to be performed; operating the MRI scanner to generate a conditional generative artificial neural network and a conditional discriminative artificial neural network; operating the MRI scanner to train the conditional generative artificial neural network using the training data, the conditional generative artificial neural network being trained such that the setting parameters for MRI sequences to be performed are generated based on the input data, the conditional generative artificial neural network being provided with conditions to be fulfilled with respect to the percentage of the one or more types of MRI sequences to be performed such that setting parameters are generated for, respectively, at least one T1-weighted MRI sequence, one T2-weighted MRI sequence, and one proton-density-weighted MRI sequence, wherein the conditions to be fulfilled are provided to the conditional generative artificial neural network as a concatenation of a condition vector with the input vector, and wherein the condition vector has a data structure comprising non-zero entries that respectively indicate a percentage of the one or more types of MRI sequences to be performed; operating the MRI scanner to train the conditional discriminative artificial neural network using the training data and the setting parameters generated by the conditional generative artificial neural network, the trained conditional discriminative artificial neural network being configured to discriminate, for each respective one of the setting parameters from among the generated setting parameters, whether each setting parameter is associated with the training data or is associated with the setting parameters generated by the conditional generative artificial neural network, the conditional generative artificial neural network being further trained until the conditional discriminative artificial neural network is unable to discriminate the training data from the setting parameters generated by the conditional generative artificial neural network; and operating the MRI scanner to perform MRI sequences in accordance with the received input data and the setting parameters generated using the trained conditional generative artificial neural network. ### Claim 13 (depends on 12) The method as claimed in claim 12, wherein the training of the conditional generative artificial neural network and/or the training of the conditional discriminative artificial neural network is performed using a loss function based on the least squares method. ### Claim 14 A computer-implemented method for establishing a system for generating setting parameters for a magnetic resonance imaging (MRI) sequence to be performed in accordance with an MRI scanner, the method comprising: operating an MRI scanner to receive input data in the form of a data structure corresponding to an input vector indicating a percentage of one or more types of MRI sequences to be performed; and operating an MRI scanner to generate, using a conditional generative artificial neural network that has been trained as part of a conditional generative adversarial net, setting parameters for the MRI sequence to be performed based on the received input data, wherein the conditional generative adversarial net comprises the conditional generative artificial neural network as a generative part and a conditional discriminative artificial neural network as a discriminative part; and wherein the conditional generative artificial neural network is provided with conditions to be fulfilled with respect to the percentage of the one or more types of MRI sequences to be performed such that setting parameters are generated for, respectively, at least one T1-weighted MRI sequence, one T2-weighted MRI sequence, and one proton-density-weighted MRI sequence, wherein the conditions to be fulfilled are provided to the conditional generative artificial neural network as a concatenation of a condition vector with the input vector, and wherein the condition vector has a data structure comprising non-zero entries that respectively indicate a percentage of the one or more types of MRI sequences to be performed. ### Claim 15 (depends on 14) The method as claimed in claim 14, wherein the setting parameters comprise a repetition time, an echo time, a turbo factor and/or a bandwidth for the MRI sequence to be performed. ### Claim 16 (depends on 14) The method as claimed in claim 14, wherein the conditional generative artificial neural network is additionally provided with conditions to be fulfilled with respect to: a magnetic field strength for the MRI sequence to be performed, information regarding a location and/or an alignment of a sectional plane through a body on which the MRI sequence is to be performed, information regarding a location and/or a body part of a body on which the MRI sequence is to be performed, a body-fat percentage, and an artifact suppression. ### Claim 17 A non-transitory computer readable medium loaded onto a control computer of a magnetic resonance imaging (MRI) scanner for generating setting parameters for a magnetic resonance imaging (MRI) sequence to be performed, the non-transitory computer readable medium having instructions stored thereon that, when executed by the control computer, cause the MRI scanner to: receive input data in the form of a data structure corresponding to an input vector indicating a percentage of one or more types of MRI sequences to be performed; and generate, using a trained conditional generative artificial neural network that has been trained as part of a conditional generative adversarial net, setting parameters for the MRI sequence to be performed based on the received input data, wherein the conditional generative adversarial net comprises the conditional generative artificial neural network as a generative part and a conditional discriminative artificial neural network as a discriminative part; and wherein the conditional generative artificial neural network is provided with conditions to be fulfilled with respect to the percentage of the one or more types of MRI sequences to be performed such that setting parameters are generated for, respectively, at least one T1-weighted MRI sequence, one T2-weighted MRI sequence, and one proton-density-weighted MRI sequence, wherein the conditions to be fulfilled are provided to the conditional generative artificial neural network as a concatenation of a condition vector with the input vector, and wherein the condition vector has a data structure comprising non-zero entries that respectively indicate a percentage of the one or more types of MRI sequences to be performed. ### Claim 18 (depends on 17) The non-transitory computer readable medium as claimed in claim 17, wherein the setting parameters comprise a repetition time, an echo time, a turbo factor and/or a bandwidth for the MRI sequence to be performed. ### Claim 19 (depends on 17) The non-transitory computer readable medium as claimed in claim 17, wherein the conditional generative artificial neural network is additionally conditioned by at least one of the following conditions: a magnetic field strength for the MRI sequence to be performed, information regarding a location and/or an alignment of a sectional plane through a body on which the MRI sequence is to be performed, information regarding a location and/or a body part of a body on which the MRI sequence is to be performed, a body-fat percentage, and an artifact suppression.