Bias field prediction method and system based on underlying physical field and coil sensitivity, medium, program product and terminal

Abstract

This application provides a bias field prediction method, system, medium, program product, and terminal based on fundamental physical fields and coil sensitivity. This application constructs a multi-branch deep learning network and trains it based on the input image set and supervision label set to obtain a trained multi-branch deep learning network. It acquires real-time input images and anatomical imaging images of the patient to be examined, inputs the real-time input images into the trained multi-branch deep learning network, and outputs corresponding fundamental physical field prediction results and coil sensitivity prediction results. Based on the fundamental physical field prediction results and coil sensitivity prediction results, it generates a corresponding target bias field for the anatomical imaging image, thereby achieving high-precision correction of the inhomogeneities of the magnetic resonance imaging equipment itself under conventional scanning conditions.

Description

Technical Field

[0001] This application relates to the field of bias field prediction technology, and in particular to bias field prediction methods, systems, media, program products and terminals based on fundamental physical fields and coil sensitivity. Background Technology

[0002] Magnetic resonance imaging (MRI), as an important medical imaging technique, is affected by various factors in terms of image uniformity. For a single MRI image, grayscale non-uniformity mainly stems from two aspects. One is related to the MRI equipment itself, such as static field non-uniformity, gradient field eddy currents, and non-uniformity in radio frequency signal transmission and reception. The other is related to factors inherent to the imaging object, including the object's shape and position, its orientation within the magnet, and the object's magnetic permeability and electrolyte properties.

[0003] To address the inhomogeneities inherent in MRI equipment, the industry commonly employs body coil images to correct subsequent anatomical images acquired by surface coils. This method presupposes that the body coil images are sufficiently homogeneous. However, due to inhomogeneities in the B1 and B0 fields, eddy currents, and other issues, signal intensity deviations are inherent, particularly in large-area images such as those of the abdomen and lumbar spine. Using such inherently biased body coil images as a correction benchmark propagates and amplifies these errors in all subsequent anatomical images, failing to fundamentally solve the problem. Another assumption is that the obtained bias field is uniformly and continuously varying. However, if the patient has large lesions, the obtained bias field is often not smooth enough. Introducing numerous smoothing algorithms may alter the original distribution.

[0004] To obtain uniform volume coil images, existing techniques employ fine B1-field and B0-field calibrations, as well as other related calibrations. However, these methods require additional, lengthy scans, making them impractical for routine clinical examinations. Another approach is to use standard atlases for registration correction; however, significant differences exist between individual brain anatomy and standard templates, and forced registration introduces errors due to anatomical mismatches, especially for diseased brain tissue, where registration results are even worse. Current techniques suffer from the following drawbacks:

[0005] Therefore, there is an urgent need in this field for a technical solution that can generate a bias field quickly and accurately under conventional scanning conditions, so as to achieve high-precision correction of the inhomogeneity of the magnetic resonance equipment itself. Summary of the Invention

[0006] In view of the shortcomings of the prior art described above, the purpose of this application is to provide a bias field prediction method, system, medium, program product and terminal based on the fundamental physical field and coil sensitivity, so as to solve the problems of dependence on volume coil image and non-smooth and inaccurate bias field estimation in bias field prediction.

[0007] To achieve the above and other related objectives, a first aspect of this application provides a bias field prediction method based on fundamental physical fields and coil sensitivity. The method includes: acquiring images from the same body part of several patients using a magnetic resonance imaging (MRI) device to obtain several acquired images; performing channel stitching on each acquired image to generate an input image, thus obtaining a set of input images; wherein each acquired image includes a primary body coil image and multiple surface coil images; performing prospective field map calibration on the MRI device to obtain a calibrated MRI device; and acquiring images from the same body part of several patients using the calibrated MRI device to obtain several calibration bodies. Coil images are used to calculate several supervision labels based on several calibration coil images and corresponding acquired images, thus obtaining a supervision label set. A multi-branch deep learning network is constructed and trained based on the input image set and supervision label set to obtain a trained multi-branch deep learning network. Real-time input images and anatomical imaging images of the patient to be examined are acquired, and the real-time input images are input into the trained multi-branch deep learning network to output corresponding basic physical field prediction results and coil sensitivity prediction results. Based on the basic physical field prediction results and coil sensitivity prediction results, a corresponding target bias field is generated for the anatomical imaging image.

[0008] In some embodiments of the first aspect of this application, a plurality of supervision tags are calculated based on a plurality of calibration coil images and corresponding acquired images; wherein, the process of generating each supervision tag includes: calculating a basic physical field tag based on the current calibration coil image and the original coil image in the corresponding acquired image; calculating a correction bias field based on the current calibration coil image and a plurality of surface coil images in the corresponding acquired image; calculating a coil sensitivity tag based on the correction bias field and the basic physical field tag; and combining the calculated basic physical field tag and the coil sensitivity tag to form a supervision tag.

[0009] In some embodiments of the first aspect of this application, the structure of the multi-branch deep learning network includes: a shared encoder and two parallel branch decoders; the two parallel branch decoders are a basic physics prediction branch and a coil sensitivity prediction branch, respectively.

[0010] In some embodiments of the first aspect of this application, the process of training the multi-branch deep learning network based on the input image set and the supervision label set to obtain a trained multi-branch deep learning network includes: using data from the input image set as input data for the shared encoder in the multi-branch deep learning network; inputting the basic physical field labels from the supervision label set into a first loss function corresponding to the basic physical field prediction branch; inputting the coil sensitivity labels from the supervision label set into a second loss function corresponding to the coil sensitivity prediction branch; and jointly training the multi-branch deep learning network based on the first loss function and the second loss function until a preset training termination condition is met to obtain a trained multi-branch deep learning network.

[0011] In some embodiments of the first aspect of this application, a corresponding target bias field is generated for the anatomical imaging image based on the basic physical field prediction results and the coil sensitivity prediction results. The process includes: obtaining imaging configuration information of the anatomical imaging image, the imaging configuration information including a set of surface coil channels and corresponding channel merging rules; selecting a subset of coil sensitivity predictions corresponding to the set of surface coil channels from the coil sensitivity prediction results, and combining the subset of coil sensitivity predictions according to the channel merging rules to obtain a combined coil sensitivity result; and calculating the target bias field corresponding to the anatomical imaging image based on the combined coil sensitivity result and the basic physical field prediction results.

[0012] In some embodiments of the first aspect of this application, the prospective field map calibration includes: center frequency calibration, global shimming and higher-order shimming of the B0 field, transmit and receive calibration of the B1 field, and gradient eddy current compensation calibration.

[0013] To achieve the above and other related objectives, a second aspect of this application provides a bias field prediction system based on fundamental physical fields and coil sensitivity. The system includes: an input construction module, used to acquire images from the same body part of several patients using a magnetic resonance imaging (MRI) device to obtain several acquired images; channel stitching is performed on each acquired image to generate an input image, resulting in a set of input images; wherein each acquired image includes a primitive body coil image and multiple surface coil images; and a supervision construction module, used to perform prospective field map calibration on the MRI device to obtain a calibrated MRI device; and to acquire images from the same body part of several patients using the calibrated MRI device to obtain several calibration bodies. The system comprises several modules: a coil image module, a supervision label set, and a model training module. The model training module constructs a multi-branch deep learning network and trains it based on the input image set and the supervision label set. The bias field synthesis module acquires the patient's model input image and anatomical imaging image, inputs the model input image into the trained multi-branch deep learning network, obtains corresponding basic physical field prediction results and coil sensitivity prediction results, and generates a corresponding target bias field for the anatomical imaging image based on these prediction results.

[0014] To achieve the above and other related objectives, a third aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned bias field prediction method based on fundamental physical fields and coil sensitivity.

[0015] To achieve the above and other related objectives, a fourth aspect of this application provides a computer program product comprising computer program code that, when executed on a computer, causes the computer to implement the bias field prediction method based on fundamental physical fields and coil sensitivity.

[0016] To achieve the above and other related objectives, a fifth aspect of this application provides an electronic terminal, including a memory, a processor, and a computer program stored in the memory; the processor executes the computer program to implement the bias field prediction method based on fundamental physical fields and coil sensitivity.

[0017] As described above, the bias field prediction method, system, medium, program product, and terminal based on fundamental physical fields and coil sensitivity of this application have the following beneficial effects: By constructing a multi-branch deep learning network with "physical factor decoupling," the bias field is decoupled from the fundamental physical field and coil sensitivity in terms of physical origin. This decoupling allows the multi-branch deep learning network to learn the intrinsic laws of these two independent physical quantities, rather than superficial image artifacts, thus exhibiting strong robustness to unseen patient data and images with large lesions. By flexibly combining the two predicted physical components, accurate bias fields can be synthesized in real time for any existing or potentially new coil combinations, solving the compatibility problem between coil configurations. Attached Figure Description

[0018] Figure 1 The diagram shown is a flowchart illustrating a bias field prediction method based on fundamental physical fields and coil sensitivity in one embodiment of this application.

[0019] Figure 2 The diagram shown is a structural schematic of a multi-branch deep learning network in one embodiment of this application.

[0020] Figure 3 The diagram shown is a schematic representation of a bias field prediction system based on fundamental physical fields and coil sensitivity, according to an embodiment of this application.

[0021] Figure 4 The diagram shown is a structural schematic of an electronic terminal according to an embodiment of this application. Detailed Implementation

[0022] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0023] In this application, unless otherwise expressly specified and limited, the terms "installation," "connection," "linking," "fixing," and "holding" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0024] It should be noted that, in the embodiments of this application, the words "exemplary" or "for example" indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0025] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0026] In the embodiments of this application, the terms "first" and "second" are used to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and the terms "first" and "second" do not necessarily imply that they are different.

[0027] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for explaining the present invention and are not intended to limit the invention.

[0028] Before providing a further detailed description of the present invention, the nouns and terms used in the embodiments of the present invention are explained, and the nouns and terms used in the embodiments of the present invention are subject to the following interpretations:

[0029] <1> B0 field: This is a stable and uniform static magnetic field generated by the main magnet in a magnetic resonance imaging device. It is used to cause energy level splitting of hydrogen nuclei in the human body and form the basis of magnetization. It is the physical basis of magnetic resonance imaging.

[0030] <2> B1 field: This is an alternating radio frequency magnetic field generated by the radio frequency coil in a magnetic resonance device. It is used to excite hydrogen nuclei to resonate under the action of the B0 field and generate a detectable magnetic resonance signal.

[0031] <3> Bias field: refers to the spatial distribution field in magnetic resonance imaging where the image intensity changes slowly due to factors such as non-uniform B1 field and differences in coil sensitivity, which can cause inconsistent brightness of the same tissue in the image.

[0032] <4> Physical field: refers to the spatial distribution field that affects the imaging signal during magnetic resonance imaging, formed by hardware such as magnets, radio frequency systems, and gradient systems, as well as electromagnetic effects.

[0033] <5> T1 images are magnetic resonance images obtained by using the difference in longitudinal relaxation time (T1) of tissues as the main contrast mechanism. They typically show brighter fat signals and darker water signals, and are suitable for displaying anatomical details.

[0034] <6> T2 images are magnetic resonance images obtained by using the difference in transverse relaxation time (T2) of tissues as the main contrast mechanism. They typically show brighter signals in tissues with high water content and are suitable for displaying pathological changes such as lesions and edema.

[0035] <7> Radio frequency pulse: An electromagnetic pulse of a specific frequency and duration emitted by a radio frequency coil in magnetic resonance imaging, used to excite hydrogen nuclei to resonate under the influence of a B0 field and change their magnetization direction.

[0036] <8> Magnetic resonance: refers to the physical phenomenon in which atomic nuclei with magnetic moments undergo energy level transitions and produce a resonance response under specific radio frequency excitation conditions in the presence of an applied static magnetic field. It is the foundation of magnetic resonance imaging technology.

[0037] To facilitate understanding of the embodiments of this application, firstly, in conjunction with Figure 1 Detailed explanation. Figure 1 This document illustrates a flowchart of a bias field prediction method based on fundamental physical fields and coil sensitivity, as described in an embodiment of the present invention. The bias field prediction method based on fundamental physical fields and coil sensitivity in this embodiment mainly includes the following steps:

[0038] Step S11: Based on the magnetic resonance imaging device, the same body part of several patients is acquired to obtain several acquired images. Each acquired image is stitched together to generate an input image, and several input images are obtained to form an input image set. Each acquired image includes a raw body coil image and multiple surface coil images.

[0039] Specifically, the magnetic resonance imaging (MRI) device applies a stable static magnetic field to the space where the human body is located, causing the hydrogen nuclei in the body to align their spins. Radio frequency pulses are then applied to excite nuclear magnetic resonance (NMR) signals, which are spatially encoded using a gradient magnetic field. Echo signals are acquired by a receiving coil and reconstructed to obtain an MRI image. The receiving coil includes a body coil and a surface coil.

[0040] The spatial distribution of signal intensity in magnetic resonance imaging (MRI) images is not only related to the MRI characteristics of human tissue, but also influenced by the fundamental physical field distribution of the MRI equipment itself and the spatial sensitivity differences of the surface coils. Specifically, the inhomogeneity of the static magnetic field (B0) and the radio frequency field (B1) constitutes the fundamental physical field influence related to the imaging sequence and the human body's position, while the surface coils exhibit a non-uniform spatial distribution of sensitivity due to differences in geometry and arrangement. The combined effect of these factors results in a pervasive spatially varying intensity inhomogeneity in MRI images, known as the bias field.

[0041] In this application, the "same body part" of several patients refers to the same anatomical location in different patients within the field of view of magnetic resonance imaging, such as the head, legs, and hands. This is used to characterize the spatial variations of the fundamental physical field and coil sensitivity during magnetic resonance imaging. Since both the fundamental physical field and coil sensitivity in magnetic resonance imaging change with spatial location, data collected from only a single human body location is insufficient to reflect the overall spatial distribution characteristics. By collecting data from several human body locations, a training dataset with spatial diversity can be constructed, thereby improving the generalization ability of the trained model to different imaging locations.

[0042] In this embodiment, the receiving coil of the magnetic resonance device includes a body coil and multiple surface coils. The body coil also functions to transmit radio frequency pulse signals. During a single acquisition, the body coil transmits radio frequency pulse signals, and the body coil and multiple surface coils act as receiving coils to receive the magnetic resonance echo signals, thereby reconstructing an image of the body coil and images of the multiple surface coils, i.e., the acquired images.

[0043] Furthermore, the volume coil image and multiple surface coil images in the acquired image are concatenated and combined along the channel dimension while maintaining the same image matrix size and pixel values, to form the input image for training the deep learning model. For example, suppose a certain acquired image contains one volume coil image and n surface coil images, where the volume coil image is I... QBC The n surface coil images are [I Local_1 ,I local_2, I local_3 …, I local_n The result obtained by splicing and combining according to the channel dimension is [I] QBC ,I Local_1 ,I local_2, I local_3 …, I local_n ].

[0044] Furthermore, the above acquisition and channel stitching process is repeated on the same body part of different patients to obtain multiple input images, which together constitute an input image set. The stitching combination is only used to construct the multi-channel input of the deep learning model. Its purpose is to completely preserve the original spatial distribution information and coil-related physical features contained in the volume coil image and each surface coil image for the encoder, without fusing, weighting, reconstructing, or merging the images. This allows the subsequent multi-branch deep learning network to directly learn the intrinsic relationship between the basic physical field and the coil sensitivity based on the original multi-channel signals, thereby achieving effective decoupling from different physical causes of the bias field. The same body part includes, but is not limited to, various parts of the human body, such as the head, legs, waist, etc. No specific limitation is made here.

[0045] Step S12: Perform prospective field map calibration on the magnetic resonance device to obtain a calibrated magnetic resonance device. Based on the calibrated magnetic resonance device, acquire images of the same body part of several patients to obtain several calibration coil images. Calculate several supervision labels based on the several calibration coil images and the corresponding acquired images, and obtain a supervision label set accordingly.

[0046] In one embodiment of this application, the prospective field map calibration includes, but is not limited to: center frequency calibration, global shimming and higher-order shimming of the B0 field, transmission and reception calibration of the B1 field, gradient eddy current compensation calibration, etc.

[0047] Specifically, prospective field map calibration refers to the process of pre-adjusting and compensating the imaging-related physical field distribution and system parameters of the magnetic resonance imaging equipment before acquisition. Its purpose is to reduce the imaging deviations introduced by equipment state drift, physical field inhomogeneity and system non-ideal characteristics, thereby improving the consistency and repeatability of imaging results under different acquisition conditions and providing a reliable reference for subsequent supervision data construction.

[0048] In the prospective field map calibration, center frequency calibration adjusts the radio frequency center frequency of the magnetic resonance system by measuring the resonant frequency distribution of the scanned region to reduce frequency shift caused by magnetic field drift or load changes. Global and higher-order shimming in the B0 field uses shimming coils to adjust the main magnetic field globally and in higher orders to reduce spatial non-uniformity of the main magnetic field within the imaging region, improving phase consistency and imaging stability of the magnetic resonance signal. Transmit and receive calibration in the B1 field calibrates the radio frequency transmit and receive fields to correct the distribution of radio frequency power, pulse amplitude, and receiver sensitivity, thereby reducing the impact of radio frequency field non-uniformity on imaging intensity. Gradient eddy current compensation calibration corrects the eddy current effect generated during gradient switching through gradient system calibration and eddy current compensation, reducing geometric distortion and signal strength deviation caused by gradient non-ideality. Through these calibration steps, a calibrated magnetic resonance device is obtained, used to acquire images of the calibration coil. The prospective field map calibration methods described above are merely illustrative and do not constitute a limitation on the scope of protection of this invention. Any other calibration methods that can improve the stability of magnetic resonance imaging and enhance the reliability of supervisory data are applicable to the technical solutions of this invention.

[0049] It should be noted that the acquired images from the original magnetic resonance imaging (MRI) device serve as input data for subsequent deep learning network training. Therefore, it is necessary to construct reference supervision data corresponding to the acquired images for supervision and constraints during network training. To this end, prospective field map calibration is performed on the original MRI device to reduce imaging biases introduced by factors such as main magnetic field inhomogeneity, radio frequency field instability, and gradient system non-ideality. This ensures that the calibrated MRI device meets preset standards in terms of imaging consistency and fundamental physical field stability, serving as a reference device for constructing supervision labels. Based on the calibration coil images acquired by the calibrated MRI device, and the physical field-related data and coil characteristic-related data calculated based on these images, standard supervision data is constructed for supervised training of the deep learning network.

[0050] In one embodiment of this application, several supervision tags are calculated based on several calibration coil images and corresponding acquired images; wherein, the process of generating each supervision tag includes: calculating a basic physical field tag based on the current calibration coil image and the original coil image in the corresponding acquired image; calculating a correction bias field based on the current calibration coil image and several surface coil images in the corresponding acquired image; calculating a coil sensitivity tag based on the correction bias field and the basic physical field tag; and combining the calculated basic physical field tag and the coil sensitivity tag to form a supervision tag.

[0051] Furthermore, the basic physical field label is calculated based on the current calibration coil image and the original coil image in the corresponding acquired image. The calculation process is shown in Formula 1:

[0052] GT Physics = I QBC_uniform / I QBC ;(Formula 1)

[0053] Among them, GT Physics Basic physics field label, I QBC_uniform For the calibration coil image, I QBC This is an image of the original coil.

[0054] It should be noted that the fundamental physical field label is calculated by comparing the calibration volume coil image acquired by the MRI equipment after prospective field map calibration with the corresponding volume coil image acquired by the original MRI equipment. This label reflects the overall signal intensity variation trend introduced by factors such as main magnetic field inhomogeneity, spatial distribution of the radio frequency transmission field, and system drift. Since the volume coil has a large coverage area and relatively smooth spatial sensitivity distribution in the MRI system, the fundamental physical field label obtained based on the volume coil image can accurately characterize the imaging physical field influence independent of the specific surface coil sensitivity, serving as a supervision label for the fundamental physical field prediction branch in subsequent deep learning networks.

[0055] Furthermore, the calculation process for obtaining the correction bias field based on the current calibration coil image and several surface coil images in the corresponding acquired image is shown in Formula 2:

[0056] Bias Channel_i = I QBC_uniform / I Local_i ;(Formula 2)

[0057] Among them, Bias Channel_i I represents the correction bias field of the i-th surface coil image among several surface coil images. QBC_uniform For the calibration coil image, which belongs to the same acquired image as several surface coil images, I Local_i Let i represent the i-th surface coil image among several surface coil images.

[0058] The calculation process for obtaining the coil sensitivity tag based on the correction bias field and the basic physical field tag is shown in Formula 3:

[0059] GT Coil_i = Bias Channel_i / GT Physics ;(Formula 3)

[0060] Among them, GT Coil_iBias represents the sensitivity label of the i-th coil in a set of surface coil images. Channel_i Let GT represent the correction bias field of the i-th surface coil image among several surface coil images. Physics Labels for basic physical fields.

[0061] Furthermore, each monitoring tag contains a physical field tag and several corresponding coil sensitivity tags.

[0062] Step S13: Construct a multi-branch deep learning network and train the multi-branch deep learning network based on the input image set and the supervision label set to obtain the trained multi-branch deep learning network.

[0063] In one embodiment of this application, the structure of the multi-branch deep learning network includes: a shared encoder and two parallel branch decoders; the two parallel branch decoders are a basic physics prediction branch and a coil sensitivity prediction branch, respectively.

[0064] like Figure 2 As shown, the encoder consists of multiple stacked convolutional and pooling layers, serving as the core of a multi-branch deep learning network. It is responsible for automatically and efficiently extracting highly discriminative deep features from the input image set. These features simultaneously contain the fundamental physical field and coil sensitivity information within each input image.

[0065] In one embodiment of this application, the process of training the multi-branch deep learning network based on the input image set and the supervision label set to obtain a trained multi-branch deep learning network includes: using the data in the input image set as the input data of the shared encoder in the multi-branch deep learning network; inputting the basic physical field labels in the supervision label set into the first loss function corresponding to the basic physical field prediction branch; inputting the coil sensitivity labels in the supervision label set into the second loss function corresponding to the coil sensitivity prediction branch; and jointly training the multi-branch deep learning network based on the first loss function and the second loss function until a preset training termination condition is met to obtain a trained multi-branch deep learning network.

[0066] Specifically, the input images in the input image set are used as input data for the shared encoder, which performs feature encoding on the input image composed of the volume coil image and multiple surface coil images to extract multi-level feature representations related to the imaging physical environment and coil configuration.

[0067] Furthermore, the basic physics prediction branch, acting as the first decoder, takes the aforementioned features as input and, through multi-layer convolution operations and feature mapping reconstruction, outputs a basic physics prediction map corresponding to the spatial distribution of the input image. This map characterizes the spatial distribution characteristics of the basic physics field under the current imaging conditions. The basic physics field labels from the supervised label set are input into the first loss function corresponding to the basic physics prediction branch. By calculating the difference between the basic physics prediction result and the basic physics field labels, the output result of the basic physics prediction branch is subject to supervisory constraints. The first loss function is shown below:

[0068] Loss physicss =L1 Loss (P physics GT physics ); (Formula 4)

[0069] Loss physicss It is the loss of the fundamental physics prediction branch, L1. Loss It is the L1 loss function, P physics It is the basic physics prediction map output by the first decoder, GT Physics Labels for basic physical fields.

[0070] The first loss function calculates the difference between the predicted basic physics field output by the basic physics field prediction branch and the corresponding basic physics field label to obtain the loss value used for backpropagation, thereby guiding the network parameters of the basic physics field prediction branch to be updated in the direction of reducing prediction error.

[0071] Furthermore, the coil sensitivity prediction branch, acting as the second decoder, also uses the feature representation output by the shared encoder as input. Through a fully convolutional network structure, it outputs coil sensitivity prediction maps corresponding one-to-one with multiple surface coil channels, characterizing the spatial distribution of the receiving sensitivity of each surface coil. This coil sensitivity is only related to the coil geometry, essentially decoupling the two physical causes of the bias field. During the training of the multi-branch deep learning network, the coil sensitivity labels from the supervision label set are input into the second loss function corresponding to the coil sensitivity prediction branch. By calculating the difference between the coil sensitivity prediction result and the coil sensitivity label, the output result of the coil sensitivity prediction branch is subject to supervision and constraint. The second loss function is shown below:

[0072] Loss coil =∑ i=1toN [L1 Loss (P coil_i GT coil_i )]; (Formula 5)

[0073] Among them, Loss coilIt is the loss of the channel coil sensitivity prediction branch, L1 Loss It is the L1 loss function, P coil_i It is the i-th of the multiple coil sensitivity prediction branches output by the second decoder, GT coil_i Is with P coil_i The i-th of the multiple coil sensitivity labels.

[0074] The second loss function calculates the difference between the predicted coil sensitivity distribution output by the coil sensitivity prediction branch and the corresponding coil sensitivity label to obtain the loss value used for backpropagation, thereby guiding the network parameters of the coil sensitivity prediction branch to be updated in the direction of reducing prediction error.

[0075] In the training process of the multi-branch deep learning network of this application, the first loss function and the second loss function are used to measure the difference between the output result of the corresponding prediction branch and its supervision label, respectively. To adapt to different imaging scenarios and different training requirements, the specific loss form used in the first loss function and the second loss function is not limited to a certain fixed loss function, but can be selected or combined according to the actual application needs.

[0076] Furthermore, a total loss function is designed to simultaneously monitor the outputs of both branches, as shown below:

[0077] Total Loss =α*Loss physicss +β*Loss coil ;(Formula 6)

[0078] Total Loss For the total loss, Loss physicss It is the loss of the fundamental physics prediction branch, Loss coil It is the loss of the coil sensitivity prediction branch, and α and β are weighting coefficients used to adjust the relative importance of the basic physics prediction task and the coil sensitivity prediction task in the overall training process.

[0079] By introducing the aforementioned weight coefficients, the optimization strength of the two prediction branches can be flexibly balanced according to different imaging scenarios, training stages, or data characteristics, thereby preventing the loss of one branch from dominating the overall training process. The weight coefficients α and β can be preset constants or adaptively adjusted during training based on changes in loss; this application does not impose any limitations on this. By minimizing the total loss function, the multi-branch deep learning network can simultaneously consider the prediction accuracy of the basic physical field and the prediction accuracy of the coil sensitivity within the same training framework, thereby improving the accuracy and stability of subsequent bias field synthesis results. During network training, the total loss function is used as a unified optimization objective, and the network parameters in the multi-branch deep learning network are iteratively updated through backpropagation. As the number of training rounds increases, the network output gradually approaches the corresponding supervision label.

[0080] When the training process meets the preset training termination conditions, further updates to the network parameters are stopped, thereby obtaining a trained multi-branch deep learning network. The preset training termination conditions can be set according to actual application needs, such as, but not limited to: reaching a preset upper limit for the number of training epochs, the loss function value converging to a preset threshold, the loss change between adjacent training epochs being less than a preset change amount, or the validation set performance no longer improving.

[0081] Step S14: Acquire the real-time input image and anatomical imaging image of the patient to be examined, input the real-time input image into the trained multi-branch deep learning network, and output the corresponding basic physical field prediction result and coil sensitivity prediction result; generate the corresponding target bias field for the anatomical imaging image based on the basic physical field prediction result and coil sensitivity prediction result.

[0082] When the multi-branch deep learning network is obtained after training, there is no need to calibrate the magnetic resonance equipment. The target body part of the patient to be examined is acquired based on the original magnetic resonance equipment to obtain a body coil image and multiple surface coil sub-channel images. Under the premise that the image matrix size and pixel value remain unchanged, the body coil image and the multiple surface coil sub-channel images are stitched together according to the channel dimension to form the patient's model input image, which is used to input into the trained multi-branch deep learning network.

[0083] In one embodiment of this application, a target bias field is generated for the anatomical imaging image based on the basic physical field prediction results and the coil sensitivity prediction results. The process includes: obtaining imaging configuration information of the anatomical imaging image, the imaging configuration information including a set of surface coil channels and corresponding channel merging rules; selecting a subset of coil sensitivity predictions corresponding to the set of surface coil channels from the coil sensitivity prediction results, and combining the subset of coil sensitivity predictions according to the channel merging rules to obtain a combined coil sensitivity result; and calculating the target bias field corresponding to the anatomical imaging image based on the combined coil sensitivity result and the basic physical field prediction results.

[0084] In actual clinical imaging, the anatomical imaging images used for diagnosis are not directly equivalent to the original images of a single surface coil channel or body coil channel. Instead, they are fused images obtained by selecting a set of surface coil channels from multiple surface coil channels based on preset imaging configuration information, and then combining the images of the selected surface coil channels according to preset channel merging rules. For example, different anatomical imaging sequences (such as T1-weighted imaging, T2-weighted imaging, fat suppression imaging, etc.) may correspond to different surface coil channel selection methods and different channel merging strategies. The channel merging rules may include, but are not limited to: amplitude sum-of-squares merging, weighted merging based on preset weights, adaptive merging based on coil sensitivity, phase consistency merging, noise-weighted merging, or any combination of the above merging methods.

[0085] In order to ensure that the target bias field can accurately reflect the imaging conditions of the anatomical imaging image, after obtaining the basic physical field prediction results and the coil sensitivity prediction results, this application selects a subset of surface coil channels corresponding to the anatomical imaging image from the coil sensitivity prediction results based on the imaging configuration information of the anatomical imaging image, and combines the coil sensitivity according to the channel merging rules, and then calculates the target bias field that matches the anatomical imaging image together with the basic physical field prediction results.

[0086] In this way, the generated target bias field can maintain the same spatial distribution characteristics as the anatomical imaging image, thus providing a more accurate basis for subsequent anatomical imaging image bias correction.

[0087] like Figure 3 The diagram illustrates a schematic representation of a bias field prediction system based on fundamental physical fields and coil sensitivity, according to an embodiment of the present invention. The system 300 in this embodiment includes the following modules: an input construction module 301, a supervision construction module 302, a model training module 303, and a bias field synthesis module 304.

[0088] The input construction module 301 is used to acquire several human body positions. Based on the original magnetic resonance device, the acquisition of several human body positions is performed to obtain several acquisition images containing one original body coil image and several surface coil images. Each acquisition image is channel-stitched to obtain several input images, and an input image set is obtained accordingly.

[0089] The supervision construction module 302 is used to perform prospective field map calibration on the original magnetic resonance device to obtain a calibrated magnetic resonance device, to acquire several human body positions based on the calibrated magnetic resonance device to obtain several calibration body coil images, to calculate several supervision tags containing basic physical field tags and coil sensitivity tags based on the several calibration body coil images and the corresponding acquired images, and to obtain a supervision tag set accordingly.

[0090] The model training module 303 is used to construct a multi-branch deep learning network and train the multi-branch deep learning network based on the input image set and the supervision label set to obtain the trained multi-branch deep learning network.

[0091] The bias field synthesis module 304 is used to acquire the patient's model input image and anatomical imaging image, input the model input image into a trained multi-branch deep learning network, obtain the corresponding basic physical field prediction results and coil sensitivity prediction results, and generate a corresponding target bias field for the anatomical imaging image based on the basic physical field prediction results and coil sensitivity prediction results. It should be understood that the specific processes by which each module performs the above-mentioned steps have been described in detail in the above method embodiments, and for the sake of brevity, will not be repeated here.

[0092] It should also be understood that the module division in the embodiments of this application is illustrative and only represents a logical functional division; in actual implementation, there may be other division methods. Furthermore, the functional modules in the various embodiments of this application can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0093] Figure 4 This is a schematic block diagram of the electronic terminal provided in an embodiment of this application. Figure 4 As shown, the electronic terminal includes at least one processor 401, a memory 402, at least one network interface 403, and a user interface 405. The various components in the device are coupled together via a bus system 404. It is understood that the bus system 404 is used to implement communication between these components. In addition to a data bus, the bus system 404 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in… Figure 4The general will label all buses as bus systems.

[0094] The user interface 405 may include a monitor, keyboard, mouse, trackball, clicker, button, touchpad, or touch screen.

[0095] It is understood that memory 402 can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM) or programmable read-only memory (PROM), which serves as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM) and synchronous static random access memory (SSRAM). The memories described in the embodiments of this invention are intended to include, but are not limited to, these and any other suitable categories of memory.

[0096] In this embodiment of the invention, the memory 402 is used to store various types of data to support the operation of the electronic terminal 400. Examples of this data include: any executable program for operation on the electronic terminal 400, such as the operating system 4021 and application programs 4022; the operating system 4021 contains various system programs, such as the framework layer, core library layer, driver layer, etc., for implementing various basic services and handling hardware-based tasks. The application program 4022 may contain various applications, such as media players, browsers, etc., for implementing various application services. The implementation of the automatic farmland irrigation method based on region division provided in this embodiment of the invention can be included in the application program 4022.

[0097] The methods disclosed in the above embodiments of the present invention can be applied to processor 401, or implemented by processor 401. Processor 401 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in processor 401 or by instructions in the form of software. The processor 401 may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 401 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. General-purpose processor 401 may be a microprocessor or any conventional processor, etc. The steps of the accessory optimization method provided in the embodiments of the present invention can be directly reflected as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software module may be located in a storage medium, which is located in a memory. The processor reads the information in the memory and combines it with its hardware to complete the steps of the aforementioned method.

[0098] In an exemplary embodiment, the electronic terminal 400 may be used to execute the aforementioned method by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), or complex programmable logic devices (CPLDs).

[0099] According to the method provided in the embodiments of this application, this application also provides a computer program product, which includes: computer program code, which, when run on a computer, causes the computer to execute a bias field prediction method based on fundamental physical fields and coil sensitivity in any of the embodiments shown.

[0100] According to the method provided in the embodiments of this application, this application also provides a computer-readable storage medium storing program code that, when run on a computer, causes the computer to execute a bias field prediction method based on fundamental physical fields and coil sensitivity according to any of the embodiments shown.

[0101] As used in this specification, the terms "component," "module," "system," etc., are used to refer to computer-related entities, hardware, firmware, combinations of hardware and software, software, or software in execution. For example, a component can be, but is not limited to, a process running on a processor, a processor, an object, an executable file, an execution thread, a program, and / or a computer. As illustrated, applications running on computing devices and computing devices can both be components. One or more components may reside in a process and / or an execution thread, and components may be located on a single computer and / or distributed among two or more computers. Furthermore, these components can be executed from various computer-readable media on which various data structures are stored. Components can communicate, for example, via local and / or remote processes based on signals having one or more data packets (e.g., data from two components interacting with another component between a local system, a distributed system, and / or a network, such as the Internet interacting with other systems via signals).

[0102] Those skilled in the art will recognize that the various illustrative logical blocks and steps described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0103] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0104] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0105] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0106] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0107] In the above embodiments, the functions of each functional unit can be implemented entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. A computer program product includes one or more computer instructions (programs). When the computer program instructions (programs) are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., high-density digital video discs, DVDs), or semiconductor media (e.g., solid-state disks, SSDs, etc.).

[0108] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0109] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0110] In summary, this application provides a bias field prediction method, system, medium, program product, and terminal based on fundamental physical fields and coil sensitivity. This application constructs a multi-branch deep learning network and trains it based on the input image set and supervision label set to obtain a trained multi-branch deep learning network. It acquires real-time input images and anatomical imaging images of the patient to be examined, inputs the real-time input images into the trained multi-branch deep learning network, and outputs corresponding fundamental physical field prediction results and coil sensitivity prediction results. Based on the fundamental physical field prediction results and coil sensitivity prediction results, it generates a corresponding target bias field for the anatomical imaging image, thereby achieving high-precision correction of the inhomogeneities of the magnetic resonance imaging equipment itself under conventional scanning conditions.

[0111] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. A bias field prediction method based on fundamental physical fields and coil sensitivity, characterized in that, include: Based on the magnetic resonance imaging (MRI) device, several images are acquired from the same body part of several patients. Each acquired image is then stitched together to generate an input image, resulting in a set of input images. Each acquired image includes a raw body coil image and multiple surface coil images. The magnetic resonance imaging device is prospectively calibrated to obtain a calibrated magnetic resonance imaging device. Based on the calibrated magnetic resonance imaging device, the same body part of several patients is acquired to obtain several calibration coil images. Based on the several calibration coil images and the corresponding acquired images, several supervision labels are calculated to obtain a supervision label set. A multi-branch deep learning network is constructed, and the multi-branch deep learning network is trained based on the input image set and the supervision label set to obtain the trained multi-branch deep learning network. The system acquires real-time input images and anatomical imaging images of the patient to be examined. The real-time input images are then input into a trained multi-branch deep learning network, which outputs corresponding basic physical field prediction results and coil sensitivity prediction results. Based on the basic physical field prediction results and coil sensitivity prediction results, a corresponding target bias field is generated for the anatomical imaging images.

2. The bias field prediction method based on fundamental physical fields and coil sensitivity according to claim 1, characterized in that, Several supervision labels are calculated based on several calibration coil images and corresponding acquired images; the process of generating each supervision label includes: The basic physical field label is obtained by calculating based on the current calibration coil image and the original coil image in the corresponding acquired image; The correction bias field is calculated based on the current calibration coil image and several surface coil images in the corresponding acquired images; The coil sensitivity tag is calculated based on the correction bias field and the basic physical field tag. The monitoring tag is formed by combining the basic physical field tag and the coil sensitivity tag obtained from the calculation.

3. The bias field prediction method based on fundamental physical fields and coil sensitivity according to claim 1, characterized in that, The structure of the multi-branch deep learning network includes: a shared encoder and two parallel branch decoders; the two parallel branch decoders are a basic physics prediction branch and a coil sensitivity prediction branch, respectively.

4. The bias field prediction method based on fundamental physical fields and coil sensitivity according to claim 3, characterized in that, The process of training the multi-branch deep learning network based on the input image set and the supervision label set to obtain the trained multi-branch deep learning network includes: The data in the input image set is used as the input data for the shared encoder in the multi-branch deep learning network; Input the basic physics field labels from the supervised label set into the first loss function corresponding to the basic physics field prediction branch; Input the coil sensitivity labels from the supervised label set into the second loss function corresponding to the coil sensitivity prediction branch; The multi-branch deep learning network is jointly trained based on the first loss function and the second loss function until a preset training termination condition is met, thereby obtaining a trained multi-branch deep learning network.

5. The bias field prediction method based on fundamental physical fields and coil sensitivity according to claim 1, characterized in that, Based on the predicted results of the fundamental physical field and the coil sensitivity prediction results, a corresponding target bias field is generated for the anatomical imaging image. The process includes: The imaging configuration information of the anatomical imaging image is obtained, and the imaging configuration information includes the surface coil channel set and the corresponding channel merging rules; Select a subset of coil sensitivity predictions corresponding to the set of surface coil channels from the coil sensitivity prediction results, and combine the subset of coil sensitivity predictions according to the channel merging rules to obtain a combined coil sensitivity result; The target bias field corresponding to the anatomical imaging image is obtained by calculating based on the combined coil sensitivity results and the basic physical field prediction results.

6. The bias field prediction method based on fundamental physical fields and coil sensitivity according to claim 1, characterized in that, The forward-looking field map calibration includes: center frequency calibration, global and higher-order shimming of the B0 field, transmission and reception calibration of the B1 field, and gradient eddy current compensation calibration.

7. A bias field prediction system based on fundamental physical fields and coil sensitivity, characterized in that, include: The input construction module is used to acquire images of the same body part from several patients using a magnetic resonance imaging device to obtain several acquired images. Each acquired image is then stitched together to generate an input image, resulting in a set of input images. Each acquired image includes a raw body coil image and multiple surface coil images. The supervision construction module is used to perform prospective field map calibration on the magnetic resonance device to obtain a calibrated magnetic resonance device. Based on the calibrated magnetic resonance device, the same body part of several patients is acquired to obtain several calibration coil images. Based on the several calibration coil images and the corresponding acquired images, several supervision labels are calculated to obtain a supervision label set. The model training module is used to construct a multi-branch deep learning network and train the multi-branch deep learning network based on the input image set and the supervision label set to obtain the trained multi-branch deep learning network. The bias field synthesis module is used to acquire real-time input images and anatomical imaging images of the patient to be examined, input the real-time input images into a trained multi-branch deep learning network, and output corresponding basic physical field prediction results and coil sensitivity prediction results; based on the basic physical field prediction results and coil sensitivity prediction results, a corresponding target bias field is generated for the anatomical imaging image.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the bias field prediction method based on the fundamental physical field and coil sensitivity as described in any one of claims 1 to 6.

9. A computer program product, characterized in that, The computer program product includes computer program code that, when run on a computer, causes the computer to implement the bias field prediction method based on fundamental physical fields and coil sensitivity as described in any one of claims 1 to 6.

10. An electronic terminal, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the bias field prediction method based on fundamental physical fields and coil sensitivity as described in any one of claims 1 to 6.

Images