Method for training electromagnetic interference prediction model, medical signal processing method and device

Abstract

The application relates to a training method of an electromagnetic interference (EMI) prediction model, a medical signal processing method and device, which comprises the following steps: when a new sample measurement signal exists in a target magnetic resonance device, training a local prediction model corresponding to the target magnetic resonance device through the new sample measurement signal to obtain updated local model information; sending the updated local model information to a central server, so that the central server adjusts global model information of a global prediction model corresponding to the central server according to the updated local model information, and sends the updated global model information to terminals corresponding to each magnetic resonance device, so that each terminal adjusts local model information corresponding to the terminal based on the updated global model information to obtain a new local prediction model. The method can improve the capability of the local prediction model of each terminal, so that more types of EMI signals can be suppressed, and the EMI suppression capability and effect of each local prediction model are improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a training method for an electromagnetic interference prediction model, a medical signal processing method, an apparatus, a computer device, a storage medium, and a computer program product. Background Technology

[0002] Suppressing EMI (Electromagnetic Interference) based on artificial intelligence algorithms can effectively reduce the requirements for shielding environment during magnetic resonance scanning, enabling magnetic resonance scanning to achieve lower construction costs and adapt to more complex application scenarios.

[0003] Existing artificial intelligence algorithms primarily train EMI signal prediction models using data pre-collected in certain hospitals or sites, or optimize these models online using data collected by the MRI equipment itself. However, this method, where a site-specific prediction model can only be optimized based on data collected at that site, struggles to achieve satisfactory EMI suppression. Summary of the Invention

[0004] Therefore, it is necessary to address the technical problem that the above-mentioned prediction model can only be optimized based on the data collected at the site, which makes it difficult to achieve a good EMI suppression effect. This requires providing a training method for an electromagnetic interference prediction model, a medical signal processing method, an apparatus, a computer device, a computer-readable storage medium, and a computer program product.

[0005] Firstly, this application provides a method for training an electromagnetic interference prediction model. The method includes:

[0006] When new sample measurement signals are available from the target MRI device, the local prediction model corresponding to the target MRI device is trained using the new sample measurement signals to obtain updated local model information; the target MRI device includes at least one of MRI devices deployed in multiple different regions; the local prediction model is used to predict interference signals in the MRI signals acquired by the target MRI device; the new sample measurement signals represent signals that have not participated in the training of the local prediction model corresponding to the target MRI device.

[0007] The updated local model information is sent to the central server. The central server is used to adjust the global model information of the global prediction model corresponding to the central server according to the updated local model information, and send the updated global model information to the terminals corresponding to the magnetic resonance devices in the multiple different regions, so that the terminals corresponding to the magnetic resonance devices in the multiple different regions can adjust the local model information of their respective local prediction models based on the updated global model information to obtain a new local prediction model.

[0008] In one embodiment, the method further includes:

[0009] For any given magnetic resonance imaging (MRI) device, acquire the magnetic resonance signal to be processed collected from the area where the MRI device is located;

[0010] The magnetic resonance signal to be processed is analyzed using the new local prediction model corresponding to the magnetic resonance device to obtain the predicted interference signal.

[0011] Based on the magnetic resonance signal to be processed and the predicted interference signal, the effective signal in the magnetic resonance signal to be processed is obtained.

[0012] In one embodiment, obtaining the effective signal from the magnetic resonance signal to be processed based on the magnetic resonance signal to be processed and the predicted interference signal includes:

[0013] The effective signal in the magnetic resonance signal to be processed is obtained by subtracting the predicted interference signal from the magnetic resonance signal to be processed.

[0014] In one embodiment, the updated local model information includes at least one of the model parameters and model gradient information of the local prediction model;

[0015] Before sending the updated local model information to the central server, the process also includes:

[0016] The updated local model information is encrypted to obtain encrypted local model information;

[0017] The encrypted local model information is sent to the central server.

[0018] In one embodiment, the local prediction model corresponding to each magnetic resonance device in the magnetic resonance devices deployed in multiple different regions includes a first local prediction model and a second local prediction model;

[0019] The first local prediction model is used to predict the actual interference signal of the magnetic resonance signal collected in the area where each magnetic resonance device is located.

[0020] The second local prediction model is used to update the local model information when a new sample measurement signal is available at each magnetic resonance device, and participates in the update of the global model information of the central server.

[0021] Secondly, this application also provides a method for training an electromagnetic interference prediction model. The method includes:

[0022] The system receives updated local model information from a target magnetic resonance imaging (MRI) device. The updated local model information is obtained by training a local prediction model corresponding to the target MRI device using new sample measurement signals corresponding to the target MRI device. The target MRI device includes at least one MRI device deployed in multiple different regions. The local prediction model is used to predict interference signals in the MRI signals acquired by the target MRI device. The new sample measurement signals represent signals that have not participated in the training of the local prediction model corresponding to the target MRI device.

[0023] Based on the updated local model information, adjust the global model information of the global prediction model corresponding to the central server;

[0024] The updated global model information is sent to the terminals corresponding to the magnetic resonance devices in the multiple different regions, so that the terminals corresponding to the magnetic resonance devices in the multiple different regions can adjust the local model information of their respective local prediction models based on the updated global model information to obtain new local prediction models.

[0025] In one embodiment, adjusting the global model information of the global prediction model corresponding to the central server based on the updated local model information includes:

[0026] In the case where the target magnetic resonance device includes a magnetic resonance device, the global model information of the global prediction model corresponding to the central server is replaced with the updated local model information.

[0027] In one embodiment, adjusting the global model information of the global prediction model corresponding to the central server based on the updated local model information further includes:

[0028] When the target magnetic resonance imaging (MRI) device includes at least two MRI devices, the weight coefficients corresponding to each of the at least two MRI devices are obtained; the weight coefficients characterize the degree of influence of the updated local model information corresponding to the at least two MRI devices on the global model information of the global prediction model;

[0029] Based on the weighting coefficients, the updated local model information corresponding to each of the at least two magnetic resonance devices is weighted and summed, and the global model information of the global prediction model corresponding to the central server is replaced with the model information obtained by the weighted summation.

[0030] In one embodiment, obtaining the weighting coefficients corresponding to each of the at least two magnetic resonance devices includes:

[0031] Obtain the amount of training data corresponding to each of the at least two magnetic resonance devices; the amount of training data represents the amount of training data used in the process of training the local prediction model corresponding to each magnetic resonance device.

[0032] Based on the positive correlation between the amount of training data and the weight coefficients, the weight coefficients corresponding to each of the at least two magnetic resonance devices are determined.

[0033] In one embodiment, obtaining the weighting coefficients corresponding to each of the at least two magnetic resonance devices further includes:

[0034] Obtain the prediction accuracy of the local prediction model corresponding to each of the at least two magnetic resonance devices;

[0035] Based on the negative correlation between prediction accuracy and weighting coefficient, the weighting coefficients corresponding to each of the at least two magnetic resonance devices are determined.

[0036] Thirdly, this application also provides a medical signal processing method applicable to medical scanning devices set up in the current area, characterized in that the medical scanning device has a corresponding terminal, the terminal is communicatively connected to a central server, the central server is also connected to terminals set up in other areas, and the terminal is equipped with a local prediction model.

[0037] The method includes:

[0038] The medical signals acquired by the medical scanning device are processed using a target prediction model to obtain a predicted interference signal; the medical signals are then corrected based on the predicted interference signal to obtain a corrected medical signal.

[0039] The target prediction model is obtained by adjusting the local model information of the local prediction model corresponding to the medical scanning device based on the updated global model information sent by the central server.

[0040] The updated global model information is obtained by adjusting the global model information of the global prediction model based on the updated local model information sent by the target medical scanning device by the central server; the updated local model information is obtained by training the local prediction model corresponding to the target medical scanning device with the new sample measurement signal corresponding to the target medical scanning device; the new sample measurement signal represents the signal that has not participated in the training of the local prediction model corresponding to the target medical scanning device.

[0041] In one embodiment, the medical signal contains interfering components, and the corrected medical signal has its interfering components suppressed relative to the medical signal.

[0042] Fourthly, this application also provides a training device for an electromagnetic interference prediction model. The device includes:

[0043] The model training module is used to train the local prediction model corresponding to the target magnetic resonance imaging (MRI) device using the new sample measurement signals when new sample measurement signals are available from the target MRI device, thereby obtaining updated local model information. The target MRI device includes at least one of MRI devices deployed in multiple different regions. The local prediction model is used to predict interference signals in the magnetic resonance signals acquired by the target MRI device. The new sample measurement signals represent signals that have not participated in the training of the local prediction model corresponding to the target MRI device.

[0044] The local information sending module is used to send the updated local model information to the central server; the central server is used to adjust the global model information of the global prediction model corresponding to the central server according to the updated local model information, and send the updated global model information to the terminals corresponding to the magnetic resonance devices in the multiple different regions, so that the terminals corresponding to the magnetic resonance devices in the multiple different regions can adjust the local model information of their respective local prediction models based on the updated global model information to obtain a new local prediction model.

[0045] Fifthly, this application also provides a training device for an electromagnetic interference prediction model. The device includes:

[0046] A local information receiving module is used to receive updated local model information sent by a target magnetic resonance imaging (MRI) device. The updated local model information is obtained by training a local prediction model corresponding to the target MRI device using new sample measurement signals corresponding to the target MRI device. The target MRI device includes at least one MRI device deployed in multiple different areas. The local prediction model is used to predict interference signals in the magnetic resonance signals acquired by the target MRI device. The new sample measurement signals represent signals that have not participated in the training of the local prediction model corresponding to the target MRI device.

[0047] The global information adjustment module is used to adjust the global model information of the global prediction model corresponding to the central server based on the updated local model information.

[0048] The global information sending module is used to send the updated global model information to the terminals corresponding to the magnetic resonance devices in the multiple different regions, so that the terminals corresponding to the magnetic resonance devices in the multiple different regions can adjust the local model information of their respective local prediction models based on the updated global model information to obtain new local prediction models.

[0049] Sixthly, this application also provides a medical signal processing device. The device includes:

[0050] The prediction module is used to process medical signals acquired by medical scanning equipment using a target prediction model to obtain predicted interference signals;

[0051] The correction module is used to correct the medical signal based on the predicted interference signal to obtain the corrected medical signal.

[0052] The target prediction model is obtained by adjusting the local model information of the local prediction model corresponding to the medical scanning device based on the updated global model information sent by the central server; the updated global model information is obtained by adjusting the global model information of the global prediction model based on the updated local model information sent by the target medical scanning device; and the updated local model information is obtained by training the local prediction model corresponding to the target medical scanning device using the new sample measurement signal corresponding to the target medical scanning device.

[0053] Seventhly, this application also provides a medical signal processing system, including multiple medical scanning devices in different areas, terminals corresponding to each medical scanning device, and a central server. The central server is communicatively connected to each terminal, and each terminal is equipped with a local prediction model, wherein:

[0054] When a new sample measurement signal is present in the medical scanning device, the medical scanning device trains its corresponding local prediction model using the new sample measurement signal to obtain updated local model information, and sends the updated local model information to the central server; the new sample measurement signal represents a signal that has not participated in the training of the local prediction model corresponding to the medical scanning device.

[0055] The central server is used to adjust the global model information of the global prediction model based on the updated local model information to obtain updated global model information, and send the updated global model information to the terminals corresponding to the medical scanning devices in the multiple different regions.

[0056] The terminal corresponding to the medical scanning device is used to adjust the local model information of the local prediction model on the terminal based on the updated global model information to obtain the target prediction model.

[0057] When the medical scanning device has a corresponding target local prediction model, the medical scanning device is used to process the medical signals collected by the medical scanning device using the target local prediction model to obtain a predicted interference signal; and to correct the medical signals according to the predicted interference signal to obtain a corrected medical signal.

[0058] Eighthly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps:

[0059] When new sample measurement signals are available from the target MRI device, the local prediction model corresponding to the target MRI device is trained using the new sample measurement signals to obtain updated local model information; the target MRI device includes at least one of MRI devices deployed in multiple different regions; the local prediction model is used to predict interference signals in the MRI signals acquired by the target MRI device; the new sample measurement signals represent signals that have not participated in the training of the local prediction model corresponding to the target MRI device.

[0060] The updated local model information is sent to the central server. The central server is used to adjust the global model information of the global prediction model corresponding to the central server according to the updated local model information, and send the updated global model information to the terminals corresponding to the magnetic resonance devices in the multiple different regions, so that the terminals corresponding to the magnetic resonance devices in the multiple different regions can adjust the local model information of their respective local prediction models based on the updated global model information to obtain a new local prediction model.

[0061] Ninthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, performs the following steps:

[0062] When new sample measurement signals are available from the target MRI device, the local prediction model corresponding to the target MRI device is trained using the new sample measurement signals to obtain updated local model information; the target MRI device includes at least one of MRI devices deployed in multiple different regions; the local prediction model is used to predict interference signals in the MRI signals acquired by the target MRI device; the new sample measurement signals represent signals that have not participated in the training of the local prediction model corresponding to the target MRI device.

[0063] The updated local model information is sent to the central server. The central server is used to adjust the global model information of the global prediction model corresponding to the central server according to the updated local model information, and send the updated global model information to the terminals corresponding to the magnetic resonance devices in the multiple different regions, so that the terminals corresponding to the magnetic resonance devices in the multiple different regions can adjust the local model information of their respective local prediction models based on the updated global model information to obtain a new local prediction model.

[0064] Tenthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, performs the following steps:

[0065] When new sample measurement signals are available from the target MRI device, the local prediction model corresponding to the target MRI device is trained using the new sample measurement signals to obtain updated local model information; the target MRI device includes at least one of MRI devices deployed in multiple different regions; the local prediction model is used to predict interference signals in the MRI signals acquired by the target MRI device; the new sample measurement signals represent signals that have not participated in the training of the local prediction model corresponding to the target MRI device.

[0066] The updated local model information is sent to the central server. The central server is used to adjust the global model information of the global prediction model corresponding to the central server according to the updated local model information, and send the updated global model information to the terminals corresponding to the magnetic resonance devices in the multiple different regions, so that the terminals corresponding to the magnetic resonance devices in the multiple different regions can adjust the local model information of their respective local prediction models based on the updated global model information to obtain a new local prediction model.

[0067] The aforementioned electromagnetic interference (EMI) prediction model training method, medical signal processing method, device, computer equipment, storage medium, and computer program product, when new sample measurement signals are present at the target MRI device, first train the local prediction model corresponding to the target MRI device using the new sample measurement signals. Then, the updated local model information is sent to a central server, which adjusts the global prediction model information based on the updated local model information and sends the updated global model information to the terminals corresponding to each MRI device for updating the local prediction models. This federated learning method allows each terminal to train and update its prediction model even without acquiring new sample measurement signals. Furthermore, only model information is transmitted between the terminal and the central server, enabling each terminal's local model to fully utilize sample measurement signals acquired from different sites and MRI devices without exchanging or transmitting raw data, thus optimizing the local prediction models on all participating terminals. While avoiding direct data exchange, this improves the capabilities of each terminal's local prediction model, enabling it to suppress more types of EMI signals and enhancing the EMI suppression capabilities and effectiveness of each local prediction model. Attached Figure Description

[0068] Figure 1 This is an application environment diagram of the training method for an electromagnetic interference prediction model in one embodiment;

[0069] Figure 2 This is a flowchart illustrating the training method for an electromagnetic interference prediction model in one embodiment.

[0070] Figure 2a This is a flowchart illustrating the training method for an electromagnetic interference prediction model in another embodiment.

[0071] Figure 2b This is an application environment diagram of the training method for the electromagnetic interference prediction model in another embodiment;

[0072] Figure 3 This is a flowchart illustrating the training method for an electromagnetic interference prediction model in another embodiment.

[0073] Figure 4 This is a flowchart illustrating a medical signal processing method in one embodiment;

[0074] Figure 5 This is a structural block diagram of a training device for an electromagnetic interference prediction model in one embodiment.

[0075] Figure 6 This is a structural block diagram of a training device for an electromagnetic interference prediction model in one embodiment.

[0076] Figure 7 This is a structural block diagram of a training device for an electromagnetic interference prediction model in one embodiment.

[0077] Figure 8 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0078] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0079] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0080] The training method for the electromagnetic interference prediction model provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, each terminal 102 maintains a local prediction model corresponding to a medical scanning device and its corresponding local model information. The central server 104 maintains a global prediction model and its corresponding global model information. The central server 104 communicates with each terminal 102 via a network. After the local model information of the local prediction model maintained on a terminal 102 is updated, the updated local model information of that terminal can be sent to the central server 104. The central server 104 adjusts the global model information of the global prediction model based on the updated local model information and sends the updated global model information to each terminal 102. Each terminal 102 then adjusts its corresponding local prediction model information based on the updated global model information to obtain a new local prediction model. The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. The central server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers. Medical scanning equipment can be a single-modal imaging device such as an ultrasound scanner, X-ray scanner, computed tomography (CT) scanner, magnetic resonance (MR) scanner, positron emission tomography (PET) scanner, or single-photon emission computed tomography (SPECT) scanner, or a multimodal imaging device such as a PET-MRI scanner, SPECT-MRI scanner, or PET-CT scanner.

[0081] In one embodiment, such as Figure 2 As shown, a training method for an electromagnetic interference prediction model is provided, which is then applied to... Figure 1 Taking terminal 102 and MR medical scanning equipment as an example, the following steps are included:

[0082] Step S210: When there are new sample measurement signals for the target magnetic resonance device, the local prediction model corresponding to the target magnetic resonance device is trained using the new sample measurement signals to obtain updated local model information; the target magnetic resonance device includes at least one of magnetic resonance devices deployed in multiple different areas; the local prediction model is used to predict interference signals in the magnetic resonance signals collected by the target magnetic resonance device; the new sample measurement signals represent signals that have not participated in the training of the local prediction model corresponding to the target magnetic resonance device.

[0083] The new sample measurement signals can be real-time acquired signals or signals historically stored on the target MRI machine, as long as they haven't been used in the training of the local prediction model corresponding to the target MRI machine. For example, each local MRI machine can acquire sample measurement signals as training data during the intervals between scans and store them locally. After the daily scan task is completed, the local prediction model is updated using the computing resources during idle periods and the previously stored training data, and the updated local model information is uploaded to the central server. After all local MRI machines have completed uploading their updated local model information for the day, the central server performs a global model information update and distributes it to each MRI machine before the start of the next day's scan, enabling each MRI machine to update its corresponding local prediction model, achieving daily federated learning. The model update cycle based on federated learning can be fixed (e.g., daily) or dynamic (training and uploading on demand). For example, for each round of training data acquired by each MRI machine, predictions are made using the existing local prediction model. If the accuracy is high, it indicates that the newly acquired training data has been overwritten by the previous training data. We could consider adaptively adjusting the strategy based on the prediction results. For example, if we don't need to train the local prediction model or update the local model information in this round, then we don't need to participate in this round of uploading.

[0084] A magnetic resonance imaging (MRI) device includes a receiving coil and an EMI (Electromagnetic Interference) coil. The EMI coil is a specially designed coil deployed inside or around the MRI device; it only acquires EMI signals and not imaging signals. The receiving coil can acquire EMI signals or MRI signals containing EMI components.

[0085] The magnetic resonance signal acquired by the magnetic resonance equipment can be represented as a signal pair consisting of the signal acquired by the receiving coil and the signal acquired by the EMI coil: {Sc, Se}. Se contains only the EMI signal component. Sc includes at least one of the imaging signal component Sci and the EMI interference component Sce, as shown in Table 1 below: When EMI interference is present, imaging occurs: Sc = Sci + Sce; when EMI interference is absent, imaging occurs: Sc = Sci; when EMI interference is present, imaging is not triggered, but the signal is acquired: Sc = Sce.

[0086] Table 1. Composition of signals acquired by the target magnetic resonance imaging device under different conditions.

[0087]

[0088] The local prediction model is used to predict EMI interference signals. The training data is the data from the third scenario in Table 1, where EMI interference exists but excitation and imaging are not performed. Signals are collected simultaneously using both the receiving coil and the EMI coil: the signal Se collected by the EMI coil and the signal Sc collected by the receiving coil both contain only EMI interference components. The prediction model trained in this way can predict the EMI interference component Sce in the conventional imaging signal using the signal Se collected by the EMI coil. Since Sc only contains EMI interference components, Sce = Sc, so Sc can be used as the standard for network training, i.e., Sce = f(Se). In this embodiment, the new sample measurement signal comes from the interval when the target MRI device is not executing a scanning sequence. This interval can be before, after, or during the non-acquisition period of the scanning sequence. During the aforementioned interval, the receiving coil collects signals from the environment where the target MRI device is located, and the EMI coil collects signals from the environment where the target MRI device is located; these two signals together form the new sample measurement signal.

[0089] Furthermore, when using the trained prediction model, the input data for the model is the second case shown in Table 1, that is, when there is EMI interference, magnetic resonance imaging is excited, and signals are collected by the receiving coil and the EMI coil at the same time: the signal Se collected by the EMI coil only contains the EMI interference component, and the signal Sc collected by the receiving coil contains both the imaging signal component and the EMI interference component.

[0090] Using a pre-trained model, the EMI interference component in the signal Sc acquired by the imaging coil can be predicted: Sce = f(Se), and the corresponding image signal component can be obtained: Sci = Sc - Sce. By removing the EMI signal component from the image signal, an image without EMI artifacts or an image with suppressed EMI artifacts can be reconstructed.

[0091] The new sample measurement signal represents a newly acquired signal that has not been used in the training of the prediction model.

[0092] The local model information may include the model parameters and gradient information of the local prediction model. The model parameters may include the weights, bias, and accuracy of the local prediction model.

[0093] In practice, the magnetic resonance device, the terminal, and the local prediction model can be in a one-to-one correspondence. That is, each area that needs EMI suppression can be equipped with a magnetic resonance device and a terminal. A local prediction model is deployed on the terminal. The local prediction model deployed on the terminal corresponding to the magnetic resonance device is trained by the sample measurement signal collected by the magnetic resonance device. The trained local prediction model is then used to predict the EMI interference signal in that area.

[0094] When new sample measurement signals are available for the target magnetic resonance imaging (MRI) device, and the number of new sample measurement signals reaches a certain value, the local prediction model deployed on the terminal corresponding to the target MRI device can be trained using the new sample measurement signals to obtain the trained local prediction model and the updated local model information corresponding to the target MRI device.

[0095] Step S220: Send the updated local model information to the central server; The central server is used to adjust the global model information of the global prediction model corresponding to the central server according to the updated local model information, and send the updated global model information to the terminals corresponding to the magnetic resonance devices in multiple different regions, so that the terminals corresponding to the magnetic resonance devices in multiple different regions can adjust the local model information of their respective local prediction models based on the updated global model information to obtain a new local prediction model.

[0096] In this model, the local prediction model corresponding to each MRI device shares some or all of the same network structure as the global prediction model corresponding to the central server. For example, the local prediction model corresponding to the MRI device and the global prediction model corresponding to the central server may have the same neural network input layer, intermediate layer, and output layer, differing only in the activation function representing the input and output. Alternatively, the local prediction model corresponding to the MRI device and the global prediction model corresponding to the central server may have the same neural network input layer, intermediate layer, and output layer, differing only in the cost function. Furthermore, some structures of the local prediction model corresponding to the MRI device may be identical to those of the global prediction model corresponding to the central server, and in addition, the local prediction model corresponding to the MRI device may possess unique structural features.

[0097] The global model information may include the model parameters and gradient information of the global prediction model.

[0098] In practice, after the local prediction model corresponding to the target MRI device is updated, the updated local model information can be sent to the central server through the terminal corresponding to the target MRI device. The central server can update the global model information of the corresponding global prediction model according to the received updated local model information, and send the updated global model information to the terminals corresponding to each MRI device. This allows each terminal to adjust the local model information of its corresponding local prediction model based on the updated global model information, thus obtaining a new local prediction model.

[0099] In the training method of the aforementioned electromagnetic interference prediction model, when new sample measurement signals are present at the target MRI equipment, the local prediction model corresponding to the target MRI equipment is first trained using these new sample measurement signals. Then, the updated local model information is sent to the central server. The central server adjusts the global prediction model information based on the updated local model information and sends the updated global model information to the terminals corresponding to each MRI equipment for updating their respective local prediction models. This federated learning method allows each terminal to train and update its prediction model even without acquiring new sample measurement signals. Furthermore, only model information is transmitted between the terminal and the central server. This allows each terminal's local model to fully utilize sample measurement signals acquired from different sites and MRI equipment without exchanging or transmitting raw data, thus optimizing the local prediction models on each participating terminal. While avoiding direct data exchange, this method improves the capabilities of each terminal's local prediction model, enabling it to suppress more types of EMI signals and enhancing the EMI suppression capabilities and effectiveness of each local prediction model.

[0100] In one exemplary embodiment, such as Figure 2a As shown, the effectiveness of the local prediction model can be determined before updating the electromagnetic interference prediction model:

[0101] First, new local training data is acquired, which may include EMI signals collected by the EMI coil and EMI interference components collected by the receiving coil.

[0102] Next, the EMI signal collected by the EMI coil in the local training data is input into the current local prediction model, and the prediction result is obtained. Inputting the local training data into the current local prediction model allows the acquisition of interference signals in the predicted sample measurement signals. Comparing the interference signals in the predicted sample measurement signals with the EMI interference components collected by the receiving coil determines the prediction effect. When the interference signals in the predicted sample measurement signals are close to the EMI interference components collected by the receiving coil, the prediction effect is considered good. Since the training data collected in this round has been covered by previous training data, the local model is not updated in this round, does not participate in the parameter upload, and the update operation of the current local prediction model ends. When the interference signals in the predicted sample measurement signals exceed a set threshold compared to the EMI interference components collected by the receiving coil, the prediction effect is considered poor, and the update of the electromagnetic interference prediction model continues according to steps S210 and S220. In this embodiment, the training of the electromagnetic interference prediction model includes updating the amount of local training data in this round, evaluating the prediction effect of the new prediction model on the local training data, determining the parameters of the new prediction model, and uploading the new prediction model parameters to the central server.

[0103] Optionally, the central server may update the global model information of the corresponding global prediction model based on the received updated local model information by averaging all the updated local model information; or by setting weighting coefficients according to the prediction performance of different local prediction models to update the global model information of the corresponding global prediction model; or by setting weighting coefficients according to the amount of training data of different local prediction models in this round to update the global model information of the corresponding global prediction model.

[0104] In one exemplary embodiment, the method further includes:

[0105] Step 230: For any magnetic resonance imaging (MRI) device, acquire the magnetic resonance signal to be processed collected from the area where the MRI device is located;

[0106] Step 240: Analyze the magnetic resonance signal to be processed using the new local prediction model corresponding to the magnetic resonance equipment to obtain the predicted interference signal;

[0107] Step 250: Based on the magnetic resonance signal to be processed and the predicted interference signal, obtain the effective signal in the magnetic resonance signal to be processed.

[0108] Furthermore, in an exemplary embodiment, in step 250 above, obtaining the effective signal in the magnetic resonance signal to be processed based on the magnetic resonance signal to be processed and the predicted interference signal includes: subtracting the magnetic resonance signal to be processed from the predicted interference signal to obtain the effective signal in the magnetic resonance signal to be processed.

[0109] Among them, the effective signal represents the imaging signal acquired by the receiving coil.

[0110] In practice, after updating the local prediction model and obtaining the new local prediction model, the actual process of using the local prediction model to predict EMI interference signals is as follows:

[0111] Taking magnetic resonance imaging (MRI) device A as an example, MRI device A acquires the MRI signal to be processed from its local area. The MRI signal to be processed is input into the new local prediction model corresponding to MRI device A. The new local prediction model processes and analyzes the MRI signal to be processed to obtain the predicted interference signal. Based on the MRI signal to be processed and the predicted interference signal, the effective signal in the MRI signal to be processed can be obtained.

[0112] More specifically, since the magnetic resonance signal is a superposition of the EMI interference signal and the effective signal, the effective signal in the magnetic resonance signal to be processed can be obtained by subtracting the predicted interference signal from the magnetic resonance signal to be processed.

[0113] In this embodiment, after training to obtain a new local prediction model, since the new local prediction model can predict the interference signal in the magnetic resonance signal, the effective signal in the magnetic resonance signal to be processed can be obtained by subtracting the predicted interference signal from the magnetic resonance signal to be processed, so as to facilitate subsequent suppression of the interference signal.

[0114] In one exemplary embodiment, the updated local model information mentioned above includes at least one of the model parameters and model gradient information of the local prediction model;

[0115] Before sending the updated local model information to the central server in step S220, the method further includes: encrypting the updated local model information to obtain encrypted local model information; and sending the encrypted local model information to the central server.

[0116] The model gradient information can represent the difference between the predicted interference signal and the interference signal collected by the EMI coil.

[0117] In practice, when transmitting model information between the terminal and the central server, the transmitted model information can be encrypted to improve data transmission security. More specifically, for the terminal that receives updated local model information, the updated local model information can be encrypted using a preset encryption method to obtain encrypted local model information. This encrypted local model information is then sent to the central server. Upon receiving the encrypted local model information, the central server decrypts it using a key and adjusts the global model information of the global prediction model based on the decrypted local model information.

[0118] In this embodiment, before sending the updated local model information to the central server, the updated local model information is encrypted and then sent to the central server to further improve the security of data transmission between the central server and the terminal.

[0119] In one embodiment, the local prediction model corresponding to each magnetic resonance device deployed in multiple different regions includes a first local prediction model and a second local prediction model.

[0120] The first local prediction model is used to predict the actual interference signal of the magnetic resonance signal collected in the area where each magnetic resonance device is located.

[0121] The second local prediction model is used to update the local model information when a new sample measurement signal is available at each magnetic resonance device, and does not participate in the update of the global model information on the central server.

[0122] Specifically, considering the unique EMI interference types posed by different regional environments, a proposal is made to maintain two local prediction models for each region's MRI equipment: a first local prediction model and a second local prediction model. The first local prediction model participates in federated learning, uploading updated local model information after self-training and receiving updates from the global prediction model. The second local prediction model, however, is trained only for EMI signals specific to the current environment of the local MRI equipment. It does not participate in federated learning, does not upload updated local model information to the central server, and does not affect the local prediction models on other MRI equipment. Specifically, the method for identifying EMI specific to the current environment of the local MRI equipment is as follows: if the accuracy of the collected training data tested on the first local prediction model is very low, it can be considered that the EMI is specific to the local MRI equipment, and the second local prediction model should be used for training and prediction.

[0123] like Figure 2b The diagram illustrates a training method for an electromagnetic interference prediction model according to another embodiment of this application. The sample measurement signal (private data in the diagram) can be divided into an environmental feature component and a non-environmental feature component using an adaptive data segmentation method. The environmental feature component is influenced by the specific environment in which the magnetic resonance imaging (MRI) device is located. A first local prediction model (local model in the diagram) is used to process the non-environmental feature component, and a second local prediction model (local private model in the diagram) is used to process the environmental feature component. The central server updates the global model information based solely on the first local prediction model of the MRI device in each region. The updated global model information is sent to each MRI device, allowing each MRI device to adjust its corresponding local model information for its first local prediction model based on the updated global model information. The second local prediction model of the MRI device in each region is updated separately.

[0124] Corresponding to the above, the magnetic resonance signal to be processed, acquired from the area where the magnetic resonance equipment is located, can be divided into a first feature part and a second feature part. The first feature part corresponds to the non-environmental feature part of the current area, and the second feature part corresponds to the environmental feature part of the current area. The first feature part is input into the first local prediction model of the current area to obtain the first prediction component; the second feature part is input into the second local prediction model of the current area to obtain the second prediction component; the superposition of the first prediction component and the second prediction component is the predicted interference signal.

[0125] In this embodiment, considering the unique EMI interference types posed by different environments in different regions, a method is proposed to maintain two local prediction models for the magnetic resonance equipment in each region. One model participates in federated learning, and the other model is used for predicting EMI signals specific to that region. This approach ensures the accuracy of prediction results while implementing federated learning. An adaptive data segmentation method is used to distinguish between environmental and non-environmental features, achieving interference prediction with good adaptive performance under different environments. This method demonstrates better performance in both accuracy and robustness, thereby improving the accuracy of interference signal prediction.

[0126] In one embodiment, such as Figure 3 As shown, a training method for an electromagnetic interference prediction model is provided, which is then applied to... Figure 1 Taking the central server 104 as an example, the following steps are included:

[0127] Step S310: Receive updated local model information sent by the target magnetic resonance device; the updated local model information is obtained by training the local prediction model corresponding to the target magnetic resonance device using the new sample measurement signal corresponding to the target magnetic resonance device; the target magnetic resonance device includes at least one magnetic resonance device deployed in multiple different areas, and the local prediction model is used to predict interference signals in the magnetic resonance signals collected by the target magnetic resonance device; the new sample measurement signal represents a signal that has not participated in the training of the local prediction model corresponding to the target magnetic resonance device.

[0128] Step S320: Adjust the global model information of the global prediction model corresponding to the central server based on the updated local model information.

[0129] Step S330: Send the updated global model information to the terminals corresponding to the magnetic resonance devices in multiple different regions, so that the terminals corresponding to the magnetic resonance devices in multiple different regions can adjust their local prediction model information based on the updated global model information to obtain a new local prediction model.

[0130] Among them, the local prediction model corresponding to each magnetic resonance device and the global prediction model corresponding to the central server have the same network structure locally or globally.

[0131] In the specific implementation, after the central server 104 receives the updated local model information sent by the target magnetic resonance device, before adjusting the global model information of the corresponding global prediction model of the central server based on the updated local model information, it can obtain the number of target magnetic resonance devices that sent the updated local model information. Based on the number of target magnetic resonance devices, it determines the corresponding adjustment method to adjust the global model information of the global prediction model, and then sends the updated global model information to the terminals corresponding to each magnetic resonance device, so that the terminals corresponding to each magnetic resonance device can adjust the local model information of their respective local prediction models based on the updated global model information to obtain a new local prediction model.

[0132] The federated learning-based electromagnetic interference prediction model training method provided in this embodiment enables each terminal to train and update the prediction model even without acquiring new sample measurement signals. Furthermore, only model information is transmitted between the terminal and the central server, achieving highly efficient collaborative model updates across the computing platforms of numerous terminals while protecting terminal data privacy.

[0133] In an exemplary embodiment, step S320 above, adjusting the global model information of the global prediction model corresponding to the central server based on the updated local model information, includes:

[0134] Step S3201: If the target magnetic resonance device includes a magnetic resonance device, replace the global model information of the global prediction model corresponding to the central server with the updated local model information.

[0135] In practice, when the target magnetic resonance device includes a single magnetic resonance device, since the network structure of the global prediction model is the same as that of the local prediction model, the updated local model information can be directly used to replace the global model information of the global prediction model corresponding to the central server.

[0136] In an exemplary embodiment, step S320 above, which adjusts the global model information of the global prediction model corresponding to the central server based on the updated local model information, further includes:

[0137] Step S3202: When the target magnetic resonance device includes at least two magnetic resonance devices, obtain the weight coefficients corresponding to each of the at least two magnetic resonance devices; the weight coefficients characterize the degree of influence of the updated local model information corresponding to the at least two magnetic resonance devices on the global model information of the global prediction model;

[0138] Step S3203: Based on the weighting coefficients, perform weighted summation on the updated local model information corresponding to each of the at least two magnetic resonance devices, and replace the global model information of the global prediction model corresponding to the central server with the model information obtained by weighted summation.

[0139] In specific implementation, when the target magnetic resonance imaging (MRI) device includes at least two MRI devices, in order to ensure that the updated global model information can take into account the actual situation of different regions, the updated local model information corresponding to each of the at least two MRI devices can be weighted according to the pre-set weight coefficients of each MRI device, so as to obtain at least two weighted local model information. The weighted local model information is then summed to obtain the weighted summation model information, and the global model information of the global prediction model corresponding to the central server is replaced with the weighted summation model information.

[0140] In the above embodiments, different adjustment methods are used depending on the number of target magnetic resonance devices to adjust the global model information of the global prediction model, so as to ensure the accuracy of the updated global model information.

[0141] Furthermore, in an exemplary embodiment, step S3202, obtaining the weight coefficients corresponding to each of the at least two magnetic resonance devices, includes: obtaining the amount of training data corresponding to each of the at least two magnetic resonance devices; the amount of training data represents the amount of training data used in the process of training the local prediction model corresponding to each magnetic resonance device; and determining the weight coefficients corresponding to each of the at least two magnetic resonance devices according to the positive correlation between the amount of training data and the weight coefficients.

[0142] Specifically, the weighting coefficients can be determined based on the amount of training data for each MRI device in this round. More specifically, the more training data there is, the larger the weighting coefficients will be; the less training data there is, the smaller the weighting coefficients will be.

[0143] For example, if the training data for MRI device 1 is 100 cases, the training data for MRI device 2 is 30 cases, and the training data for MRI device 3 is 70 cases, then the weights of each MRI device can be set to 0.5, 0.15, and 0.35, respectively.

[0144] In this embodiment, by establishing a positive correlation between the amount of training data and the weight coefficients, the weight coefficients corresponding to each magnetic resonance imaging (MRI) device are determined. This enables the allocation of importance of the updated local model information uploaded by each MRI device, thereby improving the accuracy of the global prediction model.

[0145] In another exemplary embodiment, step S3202, which involves obtaining the weight coefficients corresponding to each of the at least two magnetic resonance devices, further includes: obtaining the prediction accuracy of the local prediction model corresponding to each of the at least two magnetic resonance devices; and determining the weight coefficients corresponding to each of the at least two magnetic resonance devices based on the negative correlation between prediction accuracy and weight coefficients.

[0146] Specifically, the weighting coefficients can be determined based on the prediction accuracy of the local prediction models of each magnetic resonance device after this round of training. More specifically, the better the prediction accuracy, the larger the weighting coefficient; the worse the prediction accuracy, the smaller the weighting coefficient.

[0147] For example, before self-training, the prediction accuracies of magnetic resonance devices 1, 2, and 3 in predicting EMI signals in the local sample measurement signals they acquire are 0.7, 0.9, and 0.8, respectively. Then, the weights of the updated model parameters of magnetic resonance devices 1, 2, and 3 after self-training can be set to 0.5, 0.17, and 0.33.

[0148] In this embodiment, by determining the weight coefficient corresponding to each magnetic resonance imaging (MRI) device through the negative correlation between prediction accuracy and weight coefficient, the importance allocation of the updated local model information uploaded by each MRI device is realized, thereby improving the accuracy of the global prediction model.

[0149] In another embodiment, step S3202, which involves obtaining the weight coefficients corresponding to each of the at least two magnetic resonance devices, further includes: obtaining the amount of training data corresponding to each of the at least two magnetic resonance devices and the prediction accuracy of the local prediction model corresponding to each of the at least two magnetic resonance devices; and determining the weight coefficients corresponding to each of the at least two magnetic resonance devices based on the amount of training data and the prediction accuracy.

[0150] In practice, corresponding influence factors can be pre-assigned to the amount of training data and the prediction accuracy, for example, the influence factor for the amount of training data is 0.4 and the influence factor for the prediction accuracy is 0.6. Then, for each MRI device, a first initial weight is determined based on the amount of training data, and a second initial weight is determined based on the prediction accuracy. The first and second initial weights are weighted according to the influence factors of the training data and the prediction accuracy to obtain the final weight coefficient for that MRI device. Thus, the weight coefficients for each MRI device in at least two MRI devices are obtained.

[0151] In this embodiment, the weights of the updated local model information uploaded by each magnetic resonance imaging device are determined by the amount of training data and the prediction accuracy, which can further improve the accuracy of the global prediction model.

[0152] In one embodiment, such as Figure 4 As shown, a medical signal method is provided, applicable to medical scanning devices located in the current area. The medical scanning device has a corresponding terminal, which is communicatively connected to a central server. The central server is also connected to terminals located in other areas, and each terminal is equipped with a local prediction model. In this embodiment, the method includes the following steps:

[0153] Step S410: The medical signals collected by the medical scanning device are processed using the target prediction model to obtain the predicted interference signal;

[0154] Step S420: Correct the medical signal based on the predicted interference signal to obtain the corrected medical signal;

[0155] Specifically, the target prediction model is obtained by adjusting the local model information of the local prediction model corresponding to the medical scanning device based on the updated global model information sent by the central server; the updated global model information is obtained by adjusting the global model information of the global prediction model based on the updated local model information sent by the target medical scanning device; and the updated local model information is obtained by training the local prediction model corresponding to the target medical scanning device with the new sample measurement signals corresponding to the target medical scanning device, where the new sample measurement signals represent signals that have not participated in the training of the local prediction model corresponding to the target medical scanning device.

[0156] In one exemplary embodiment, the medical signal contains interfering components, and the interfering components of the corrected medical signal are suppressed relative to the medical signal.

[0157] In practice, after obtaining the trained target prediction model, the target prediction model can be applied to correct the medical signals acquired by the medical scanning device. Specifically, the medical signals acquired by the medical scanning device can be input into the target prediction model corresponding to the medical scanning device, and the target prediction model can analyze and process them to obtain the predicted interference signal. The predicted interference signal can be an EMI interference signal. By subtracting the predicted interference signal from the medical signal, an interference-free imaging signal can be obtained, thereby correcting the medical signal. The interference-free imaging signal obtained after subtracting the predicted interference signal is used as the corrected medical signal.

[0158] The medical signal processing method provided in this embodiment analyzes and processes the medical signals acquired by the medical scanning device through a target prediction model to obtain a predicted interference signal. The predicted interference signal is then subtracted from the medical signal to obtain an interference-free imaging signal, thereby suppressing the interference components of the medical signal and obtaining an accurate imaging signal.

[0159] In one exemplary embodiment, a medical signal processing system is also provided, including multiple medical scanning devices in different regions, terminals corresponding to each medical scanning device, and a central server. The central server is communicatively connected to each terminal, and each terminal is equipped with a local prediction model, wherein:

[0160] When a new sample measurement signal is present in the medical scanning device, the medical scanning device is used to train the local prediction model corresponding to the medical scanning device using the new sample measurement signal to obtain updated local model information, and then send the updated local model information to the central server; the new sample measurement signal represents a signal that has not participated in the training of the local prediction model corresponding to the medical scanning device.

[0161] The central server is used to adjust the global model information of the global prediction model based on the updated local model information to obtain updated global model information, and send the updated global model information to the terminals corresponding to the medical scanning devices in the multiple different regions.

[0162] The terminal corresponding to the medical scanning device is used to adjust the local model information of the local prediction model on the terminal based on the updated global model information to obtain the target prediction model.

[0163] When the medical scanning device has a corresponding target local prediction model, the medical scanning device is used to process the medical signals collected by the medical scanning device using the target local prediction model to obtain a predicted interference signal; and to correct the medical signals according to the predicted interference signal to obtain a corrected medical signal.

[0164] This method employs federated learning, enabling each terminal to train and update its predictive model even without acquiring new medical signals. Furthermore, only model information is transmitted between the terminals and the central server. This allows each terminal's local predictive model to fully utilize medical signals acquired from different locations and medical scanning devices without exchanging or transmitting raw data, thus optimizing the local predictive models on all participating terminals. While avoiding direct data exchange, this method enhances the capabilities of each terminal's local predictive model, enabling it to suppress more types of EMI signals and improving the EMI suppression capabilities and effectiveness of each local predictive model.

[0165] Understandably, given the diverse forms and types of EMI interference, pre-trained models may perform poorly in predicting certain untrained EMI types. Therefore, after a pre-trained model is deployed to a terminal, its parameters can still be updated by collecting model training data (without imaging but by collecting signals when EMI interference is present), in the hope of achieving better EMI suppression on the current data.

[0166] While self-training of local prediction models on each terminal can improve their performance to some extent, the additional training data and updated model information collected remain only on that terminal. To achieve collaborative training and model updates of data collected from various sites and terminals without exchanging data, this application proposes a federated learning approach to integrate the data and model information collected from each terminal.

[0167] Assume there are N terminals participating in federated learning, and each terminal maintains a local prediction model F. i (i = 1, ..., N), and simultaneously, a global prediction model F0 is maintained on the central server connecting these terminals. The network structure of the global prediction model and the local prediction model maintained on each terminal is the same, i.e., F0 and F i (i=1,…,N) all have the same network structure.

[0168] Each terminal's local prediction model, after multiple iterations from the same set of basic model parameters θ0, develops its own set of local prediction model parameters θ. i (i = 1, ..., N), in the next iteration, each terminal will collect a batch of training data D. i (i=1,…,N): When EMI interference is present, excitation and imaging are not performed, and signals are acquired simultaneously using the receiving coil and the EMI coil.

[0169] In the federated learning framework, these local training data D i They do not exchange or collect data from each other to the central server. Model F on each terminal... i All will be based on the data they collect, D i Train the model and obtain relevant information for model updates, such as changes in model parameters Δθ. i Or the intermediate result G of the gradient calculation in this round of training. i The updated local model information on each terminal will be encrypted and uploaded to the central server, and used to update the global model information θ0 in the global prediction model F0.

[0170] After the global prediction model F0 is updated, the updated global model information is encrypted and then distributed to each terminal for use in the local prediction model F0 on each terminal.i The parameter θ i This update concludes the current iteration.

[0171] This solution is based on a federated learning framework, where each terminal collects data D. i Collaborative updates of local prediction models on each terminal can be achieved without exchanging data or uploading it to a central server. This saves significant time and costs associated with data transmission while protecting user data privacy and security. Furthermore, through collaborative updates of prediction models, the F-value of local prediction models on each terminal is improved. i Performance in predicting and suppressing EMI interference.

[0172] It should be noted that, apart from each terminal maintaining a set of model information, all devices within the same hospital / institution can be interconnected within the local area network and share the original data and the same set of model information, since there are no privacy and security restrictions on data transmission.

[0173] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0174] Based on the same inventive concept, embodiments of this application also provide a training apparatus for an electromagnetic interference prediction model to implement the training method for the electromagnetic interference prediction model mentioned above, and a medical signal processing apparatus to implement the medical signal processing method mentioned above. The solutions provided by each apparatus are similar to the solutions described in the methods above; therefore, specific limitations in one or more apparatus embodiments provided below can be found in the limitations of the corresponding methods above, and will not be repeated here.

[0175] In one embodiment, such as Figure 5 As shown, a training device for an electromagnetic interference prediction model is provided, comprising: a model training module 510 and a local information transmission module 520, wherein:

[0176] The model training module 510 is used to train the local prediction model corresponding to the target magnetic resonance imaging (MRI) device using the new sample measurement signals when new sample measurement signals are available from the target MRI device, thereby obtaining updated local model information. The target MRI device includes at least one of MRI devices deployed in multiple different areas. The local prediction model is used to predict interference signals in the magnetic resonance signals acquired by the target MRI device. The new sample measurement signals represent signals that have not participated in the training of the local prediction model corresponding to the target MRI device.

[0177] The local information sending module 520 is used to send the updated local model information to the central server. The central server is used to adjust the global model information of the global prediction model corresponding to the central server according to the updated local model information, and send the updated global model information to the terminals corresponding to the magnetic resonance devices in multiple different regions. This allows the terminals corresponding to the magnetic resonance devices in multiple different regions to adjust the local model information of their respective local prediction models based on the updated global model information, thereby obtaining a new local prediction model.

[0178] In one embodiment, the above-mentioned apparatus further includes a signal processing module, configured to acquire, for any magnetic resonance device, a magnetic resonance signal to be processed collected from the area where the magnetic resonance device is located; analyze the magnetic resonance signal to be processed using a new local prediction model corresponding to the magnetic resonance device to obtain a predicted interference signal; and obtain the effective signal in the magnetic resonance signal to be processed based on the magnetic resonance signal to be processed and the predicted interference signal.

[0179] In one embodiment, the signal processing module is further configured to subtract the predicted interference signal from the magnetic resonance signal to be processed to obtain the effective signal in the magnetic resonance signal to be processed.

[0180] In one embodiment, the updated local model information includes at least one of the model parameters and model gradient information of the local prediction model; the above apparatus further includes an encryption module for encrypting the updated local model information to obtain encrypted local model information.

[0181] The local information sending module 520 is also used to send encrypted local model information to the central server.

[0182] In one embodiment, the local prediction model corresponding to each magnetic resonance imaging (MRI) device deployed in multiple different regions includes a first local prediction model and a second local prediction model. The first local prediction model is used to predict the actual interference signal of the MRI signal collected in the region where each MRI device is located. The second local prediction model is used to update the local model information when there is a new sample measurement signal for each MRI device, and participate in the update of the global model information of the central server.

[0183] In one embodiment, such as Figure 6 As shown, a training device for an electromagnetic interference prediction model is provided, comprising: a local information receiving module 610, a global information adjustment module 620, and a global information sending module 630, wherein:

[0184] The local information receiving module 610 is used to receive updated local model information sent by the target magnetic resonance imaging (MRI) device. The updated local model information is obtained by training the local prediction model corresponding to the target MRI device with new sample measurement signals corresponding to the target MRI device. The target MRI device includes at least one of MRI devices deployed in multiple different areas. The local prediction model is used to predict interference signals in the magnetic resonance signals collected by the target MRI device. The new sample measurement signals represent signals that have not participated in the training of the local prediction model corresponding to the target MRI device.

[0185] The global information adjustment module 620 is used to adjust the global model information of the global prediction model corresponding to the central server based on the updated local model information.

[0186] The global information sending module 630 is used to send updated global model information to the terminals corresponding to the magnetic resonance devices in multiple different regions, so that the terminals corresponding to the magnetic resonance devices in multiple different regions can adjust their local prediction model information based on the updated global model information to obtain a new local prediction model.

[0187] In one embodiment, the global information adjustment module 620 is further configured to replace the global model information of the global prediction model corresponding to the central server with the updated local model information when the target magnetic resonance device includes a magnetic resonance device.

[0188] In one embodiment, the global information adjustment module 620 is further configured to, when the target magnetic resonance device includes at least two magnetic resonance devices, obtain the weight coefficients corresponding to each of the at least two magnetic resonance devices; the weight coefficients characterize the degree of influence of the updated local model information corresponding to the at least two magnetic resonance devices on the global model information of the global prediction model; based on the weight coefficients, perform weighted summation processing on the updated local model information corresponding to each of the at least two magnetic resonance devices, and replace the global model information of the global prediction model corresponding to the central server with the model information obtained by weighted summation.

[0189] In one embodiment, the global information adjustment module 620 further includes a weight determination module, used to obtain the amount of training data corresponding to each of the at least two magnetic resonance devices; the amount of training data represents the amount of training data used in the process of training the local prediction model corresponding to each magnetic resonance device; and to determine the weight coefficients corresponding to each of the at least two magnetic resonance devices according to the positive correlation between the amount of training data and the weight coefficients.

[0190] In one embodiment, the weight determination module is further configured to obtain the prediction accuracy of the local prediction model corresponding to each of the at least two magnetic resonance devices; and determine the weight coefficient corresponding to each of the at least two magnetic resonance devices according to the negative correlation between prediction accuracy and weight coefficient.

[0191] In one embodiment, such as Figure 7 As shown, a training device for an electromagnetic interference prediction model is provided, comprising: a prediction module 710 and a correction module 720, wherein:

[0192] The prediction module 710 is used to process the medical signals collected by the medical scanning device using the target prediction model to obtain the predicted interference signal;

[0193] The correction module 720 is used to correct the medical signal based on the predicted interference signal to obtain the corrected medical signal.

[0194] Specifically, the target prediction model is obtained by adjusting the local model information of the local prediction model corresponding to the medical scanning device based on the updated global model information sent by the central server; the updated global model information is obtained by adjusting the global model information of the global prediction model based on the updated local model information sent by the target medical scanning device; the updated local model information is obtained by training the local prediction model corresponding to the target medical scanning device using the new sample measurement signals corresponding to the target medical scanning device; the new sample measurement signals represent signals that have not participated in the training of the local prediction model corresponding to the target medical scanning device.

[0195] In one embodiment, the medical signal contains interfering components, and the corrected medical signal has its interfering components suppressed relative to the medical signal.

[0196] The modules in the training device and medical processing device of the aforementioned electromagnetic interference prediction model can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0197] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 8 As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a training method for an electromagnetic interference prediction model. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0198] Those skilled in the art will understand that Figure 8The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0199] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0200] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0201] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0202] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0203] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0204] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A training method for an electromagnetic interference prediction model, characterized in that, The method includes: When new sample measurement signals are available from the target MRI device, the local prediction model corresponding to the target MRI device is trained using the new sample measurement signals to obtain updated local model information; the target MRI device includes at least one of MRI devices deployed in multiple different regions; the local prediction model is used to predict interference signals in the MRI signals acquired by the target MRI device; the new sample measurement signals represent signals that have not participated in the training of the local prediction model corresponding to the target MRI device. The updated local model information is sent to the central server. The central server is used to adjust the global model information of the global prediction model corresponding to the central server according to the updated local model information, and send the updated global model information to the terminals corresponding to the magnetic resonance devices in the multiple different regions, so that the terminals corresponding to the magnetic resonance devices in the multiple different regions can adjust the local model information of their respective local prediction models based on the updated global model information to obtain a new local prediction model. The local prediction model for each magnetic resonance imaging (MRI) device deployed in multiple different regions includes a first local prediction model and a second local prediction model. The first local prediction model participates in federated learning and is used to predict the actual interference signal of the MRI signals collected in the region where each MRI device is located. The second local prediction model is used to update the local model information when there is a new sample measurement signal for each MRI device, and does not participate in the update of the global model information of the central server.

2. The method according to claim 1, characterized in that, The method further includes: For any given magnetic resonance imaging (MRI) device, acquire the magnetic resonance signal to be processed collected from the area where the MRI device is located; The magnetic resonance signal to be processed is analyzed using the new local prediction model corresponding to the magnetic resonance device to obtain the predicted interference signal. Based on the magnetic resonance signal to be processed and the predicted interference signal, the effective signal in the magnetic resonance signal to be processed is obtained.

3. The method according to claim 2, characterized in that, The step of obtaining the effective signal from the magnetic resonance signal to be processed based on the magnetic resonance signal to be processed and the predicted interference signal includes: The effective signal in the magnetic resonance signal to be processed is obtained by subtracting the predicted interference signal from the magnetic resonance signal to be processed.

4. The method according to claim 1, characterized in that, The updated local model information includes at least one of the model parameters and model gradient information of the local prediction model; Before sending the updated local model information to the central server, the process also includes: The updated local model information is encrypted to obtain encrypted local model information; The encrypted local model information is sent to the central server.

5. A training method for an electromagnetic interference prediction model, characterized in that, The method includes: The system receives updated local model information from a target magnetic resonance imaging (MRI) device. The updated local model information is obtained by training a local prediction model corresponding to the target MRI device using new sample measurement signals corresponding to the target MRI device. The target MRI device includes at least one MRI device deployed in multiple different regions. The local prediction model is used to predict interference signals in the MRI signals acquired by the target MRI device. The new sample measurement signals represent signals that have not participated in the training of the local prediction model corresponding to the target MRI device. Based on the updated local model information, adjust the global model information of the global prediction model corresponding to the central server; The updated global model information is sent to the terminals corresponding to the magnetic resonance devices in the multiple different regions, so that the terminals corresponding to the magnetic resonance devices in the multiple different regions can adjust the local model information of their respective local prediction models based on the updated global model information to obtain new local prediction models. The local prediction model for each magnetic resonance imaging (MRI) device deployed in multiple different regions includes a first local prediction model and a second local prediction model. The first local prediction model participates in federated learning and is used to predict the actual interference signal of the MRI signals collected in the region where each MRI device is located. The second local prediction model is used to update the local model information when there is a new sample measurement signal for each MRI device, and does not participate in the update of the global model information of the central server.

6. The method according to claim 5, characterized in that, The step of adjusting the global model information of the global prediction model corresponding to the central server based on the updated local model information includes: In the case where the target magnetic resonance device includes a magnetic resonance device, the global model information of the global prediction model corresponding to the central server is replaced with the updated local model information.

7. The method according to claim 5, characterized in that, The step of adjusting the global model information of the global prediction model corresponding to the central server based on the updated local model information further includes: When the target magnetic resonance imaging (MRI) device includes at least two MRI devices, the weight coefficients corresponding to each of the at least two MRI devices are obtained; the weight coefficients characterize the degree of influence of the updated local model information corresponding to the at least two MRI devices on the global model information of the global prediction model; Based on the weighting coefficients, the updated local model information corresponding to each of the at least two magnetic resonance devices is weighted and summed, and the global model information of the global prediction model corresponding to the central server is replaced with the model information obtained by the weighted summation.

8. The method according to claim 7, characterized in that, The step of obtaining the weighting coefficients corresponding to each of the at least two magnetic resonance devices includes: Obtain the amount of training data corresponding to each of the at least two magnetic resonance devices; the amount of training data represents the amount of training data used in the process of training the local prediction model corresponding to each magnetic resonance device. Based on the positive correlation between the amount of training data and the weight coefficients, the weight coefficients corresponding to each of the at least two magnetic resonance devices are determined.

9. The method according to claim 7, characterized in that, The step of obtaining the weighting coefficients corresponding to each of the at least two magnetic resonance devices further includes: Obtain the prediction accuracy of the local prediction model corresponding to each of the at least two magnetic resonance devices; Based on the negative correlation between prediction accuracy and weighting coefficient, the weighting coefficients corresponding to each of the at least two magnetic resonance devices are determined.

10. A medical signal processing method, applicable to a medical scanning device located in a current area, characterized in that, The medical scanning device has a corresponding terminal, which is connected to a central server. The central server is also connected to terminals located in other areas, and each terminal is equipped with a local prediction model. The method includes: The medical signals acquired by the medical scanning device are processed using a target prediction model to obtain a predicted interference signal; the medical signals are then corrected based on the predicted interference signal to obtain a corrected medical signal. The target prediction model is obtained by adjusting the local model information of the local prediction model corresponding to the medical scanning device based on the updated global model information sent by the central server. The updated global model information is obtained by adjusting the global model information of the global prediction model based on the updated local model information sent by the target medical scanning device by the central server. The updated local model information is obtained by training the local prediction model corresponding to the target medical scanning device using the new sample measurement signal corresponding to the target medical scanning device. The target medical scanning device includes at least one medical scanning device deployed in multiple different regions. The local prediction model corresponding to each medical scanning device deployed in multiple different regions includes a first local prediction model and a second local prediction model. The first local prediction model participates in federated learning and is used to predict the actual interference signal of the magnetic resonance signal collected in the region where each medical scanning device is located. The second local prediction model is used to update the local model information when there is a new sample measurement signal for each medical scanning device, and does not participate in the update of the global model information of the central server. The new sample measurement signal represents a signal that has not participated in the training of the local prediction model corresponding to the target medical scanning device.

11. The method according to claim 10, characterized in that, The medical signal contains interfering components, and the corrected medical signal has its interfering components suppressed relative to the medical signal.

12. A training device for an electromagnetic interference prediction model, characterized in that, The device includes: The model training module is used to train the local prediction model corresponding to the target magnetic resonance imaging (MRI) device using the new sample measurement signals when new sample measurement signals are available from the target MRI device, thereby obtaining updated local model information. The target MRI device includes at least one of MRI devices deployed in multiple different regions. The local prediction model is used to predict interference signals in the magnetic resonance signals acquired by the target MRI device. The new sample measurement signals represent signals that have not participated in the training of the local prediction model corresponding to the target MRI device. The local information sending module is used to send the updated local model information to the central server; the central server is used to adjust the global model information of the global prediction model corresponding to the central server according to the updated local model information, and send the updated global model information to the terminals corresponding to each magnetic resonance device, so that the terminals corresponding to each magnetic resonance device can adjust the local model information of their respective local prediction models based on the updated global model information to obtain a new local prediction model. The local prediction model for each magnetic resonance imaging (MRI) device deployed in multiple different regions includes a first local prediction model and a second local prediction model. The first local prediction model participates in federated learning and is used to predict the actual interference signal of the MRI signals collected in the region where each MRI device is located. The second local prediction model is used to update the local model information when there is a new sample measurement signal for each MRI device, and does not participate in the update of the global model information of the central server.

13. A medical signal processing system, characterized in that, It includes multiple medical scanning devices in different areas, terminals corresponding to each medical scanning device, and a central server. The central server is communicatively connected to each terminal, and each terminal is equipped with a local prediction model, wherein: When a new sample measurement signal is present in the medical scanning device, the medical scanning device is used to train the local prediction model corresponding to the medical scanning device using the new sample measurement signal to obtain updated local model information, and then send the updated local model information to the central server; the new sample measurement signal represents a signal that has not participated in the training of the local prediction model corresponding to the medical scanning device. The central server is used to adjust the global model information of the global prediction model based on the updated local model information to obtain updated global model information, and send the updated global model information to the terminals corresponding to the medical scanning devices in the multiple different regions. The terminal corresponding to the medical scanning device is used to adjust the local model information of the local prediction model on the terminal based on the updated global model information to obtain the target prediction model. When the medical scanning device has a corresponding target local prediction model, the medical scanning device is used to process the medical signals collected by the medical scanning device using the target local prediction model to obtain a predicted interference signal; and to correct the medical signals according to the predicted interference signal to obtain a corrected medical signal. The local prediction model for each medical scanning device in the multiple medical scanning devices in different regions includes a first local prediction model and a second local prediction model. The first local prediction model participates in federated learning and is used to predict the actual interference signal of the magnetic resonance signal collected in the region where each medical scanning device is located. The second local prediction model is used to update the local model information when there is a new sample measurement signal in each medical scanning device, and does not participate in the update of the global model information of the central server.

Images