Magnetic resonance channel sensitivity map estimation method and computer device

Abstract

The embodiment of the specification provides a magnetic resonance channel sensitivity map estimation method and computer equipment, and relates to the technical field of magnetic resonance imaging. The method comprises the following steps: acquiring multi-channel magnetic resonance reference echo data of multiple repeated acquisitions; for the reference echo data of each repeated acquisition, a local convolution kernel is constructed respectively, and an effective kernel base is screened through singular value decomposition to obtain an effective kernel subspace corresponding to the reference echo data of each repeated acquisition; the effective kernel subspaces corresponding to the reference echo data of each repeated acquisition are fused to obtain a fused kernel subspace; and based on the fused kernel subspace, feature decomposition is performed in an image domain to obtain a channel sensitivity map. The robustness and accuracy of the magnetic resonance channel sensitivity estimation are improved.

Description

Technical Field

[0001] The embodiments described in this specification relate to the field of magnetic resonance imaging technology, specifically to a magnetic resonance channel sensitivity map estimation method and computer equipment. Background Technology

[0002] In parallel magnetic resonance imaging, the channel sensitivity map is a key prior information for image reconstruction, and its accuracy directly affects the quality of the reconstructed image.

[0003] In related technologies, sensitivity estimation methods are mainly divided into two categories. One category is based on a single automatic calibration signal region, such as ESPIRiT. However, this type of method is highly dependent on the quality of the single data and is easily affected by noise and motion. The other category involves averaging multiple repeatedly acquired reference data point by point before performing sensitivity estimation. However, when there are phase differences or slight motions between different repeated data, direct averaging can lead to blurring of the calibration signal region and reduce estimation accuracy.

[0004] Therefore, there is an urgent need to provide a magnetic resonance channel sensitivity map estimation method that can make full use of multiple repeated reference data while avoiding the ambiguity introduced by direct averaging, so as to obtain a more robust and accurate channel sensitivity map. Summary of the Invention

[0005] In view of this, various embodiments of this specification aim to provide a magnetic resonance channel sensitivity map estimation method and computer device to improve the robustness and accuracy of magnetic resonance channel sensitivity estimation.

[0006] This specification provides a method for estimating a magnetic resonance channel sensitivity map, comprising: acquiring multi-channel magnetic resonance reference echo data acquired multiple times; constructing local convolution kernels for each repeatedly acquired reference echo data, and filtering effective kernel bases through singular value decomposition to obtain an effective kernel subspace corresponding to each repeatedly acquired reference echo data; fusing the effective kernel subspaces corresponding to each repeatedly acquired reference echo data to obtain a fused kernel subspace; and performing feature decomposition in the image domain based on the fused kernel space to obtain a channel sensitivity map.

[0007] In some implementations, for each repeatedly acquired reference echo data, a local convolution kernel is constructed, and effective kernel bases are selected through singular value decomposition to obtain an effective kernel subspace corresponding to each repeatedly acquired reference echo data. This includes: extracting an autocalibration signal region from each repeatedly acquired reference echo data; using a local sliding window approach, expanding the autocalibration signal region into multiple local data blocks, and vectorizing each local data block; stacking all the vectorized local data blocks corresponding to the sliding window positions row-wise to construct a local data block matrix; performing singular value decomposition on the local data block matrix to obtain a right singular vector matrix; and selecting right singular vectors that meet the conditions according to a preset singular value threshold as effective kernel bases, with the effective kernel bases constituting the effective kernel subspace.

[0008] In some implementations, fusing the effective kernel subspaces corresponding to the reference echo data acquired in each repeated acquisition includes: directly splicing the effective kernel subspaces corresponding to the reference echo data acquired in each repeated acquisition along the kernel basis dimension to obtain the fused kernel subspace.

[0009] In some implementations, feature decomposition is performed in the image domain based on the fusion kernel space to obtain an effective region mask; the method further includes: normalizing the channel sensitivity map based on the effective region mask and outputting the normalized channel sensitivity map.

[0010] In some implementations, feature decomposition is performed in the image domain based on the fusion kernel space, including: constructing an image domain channel consistency calibration operator based on the fusion kernel space; performing feature decomposition on the image domain channel consistency calibration operator at each spatial location to obtain feature values ​​and corresponding feature vectors; selecting feature vectors at each spatial location that satisfy preset feature value conditions as unnormalized channel sensitivity vectors for that spatial location to obtain an unnormalized channel sensitivity map; and determining each spatial location according to a preset feature value threshold to generate an effective region mask.

[0011] In some implementations, selecting a feature vector that satisfies a preset feature value condition at each spatial location includes: selecting the feature vector corresponding to the maximum feature value at each spatial location as the unnormalized channel sensitivity vector of the spatial location.

[0012] In some implementations, the channel sensitivity map is normalized by: for each spatial location, dividing the sensitivity value of each receiving channel by the square root of the sum of the squares of the sensitivity values ​​of all receiving channels, and setting the area outside the effective area mask to zero.

[0013] In some implementations, at least one of the following is a fixed value, an adaptively determined value, or a value determined by looking up a table: the size of the automatic calibration signal region, the window size of the local sliding window, the singular value threshold, and the eigenvalue threshold preset for the eigenvalue decomposition.

[0014] In some implementations, the repeatedly acquired multichannel magnetic resonance reference echo data is reference echo data acquired multiple times in a diffuse magnetic resonance imaging sequence by averaging.

[0015] In some embodiments, before fusing the effective nucleus space corresponding to the reference echo data acquired in each repeated acquisition, the method further includes: removing abnormal data or performing weighted preprocessing on the data based on the quality evaluation index of the reference echo data acquired in each repeated acquisition.

[0016] In some implementations, fusing the effective kernel subspaces corresponding to the reference echo data acquired in each repeated acquisition includes: assigning different fusion weights to the effective kernel subspaces corresponding to the reference echo data acquired in different repeated acquisitions, performing weighted splicing fusion, and obtaining the fused kernel subspace.

[0017] In some implementations, fusing the effective nucleus space corresponding to the reference echo data acquired in each repeated acquisition includes: grouping the reference echo data acquired in each repeated acquisition according to the acquisition category, first fusing within the group to obtain the group-intra-group fused nucleus space, and then fusing the group-intra-group fused nucleus spaces a second time to obtain the fused nucleus space.

[0018] In some implementations, when there are multiple sets of feature vectors that satisfy preset feature value conditions, multiple sets of channel sensitivity maps are output.

[0019] This specification provides a computer device, including: a memory storing a computer program; and a processor that, when executing the computer program, implements the magnetic resonance channel sensitivity map estimation method described in any of the above embodiments.

[0020] In the various implementation methods provided in this specification, firstly, multi-channel magnetic resonance reference echo data acquired multiple times is obtained. Then, for each repeated acquisition of the reference echo data, a local convolution kernel is constructed, and effective kernel bases are selected through singular value decomposition to obtain the effective kernel subspace corresponding to each repeated acquisition of the reference echo data. Next, the effective kernel subspaces corresponding to each repeated acquisition of the reference echo data are fused to obtain a fused kernel subspace. Finally, based on the fused kernel subspace, feature decomposition is performed in the image domain to obtain the channel sensitivity map. This approach fully utilizes the multiple repeated reference data while avoiding the ambiguity introduced by direct averaging, thereby obtaining a more robust and accurate channel sensitivity map and improving the robustness and accuracy of magnetic resonance channel sensitivity estimation. Attached Figure Description

[0021] Figure 1 This is a scenario example of the magnetic resonance channel sensitivity map estimation method provided in the embodiments of this specification; Figure 2 This is a schematic flowchart of the magnetic resonance channel sensitivity map estimation method provided in the embodiments of this specification; Figure 3 This is a schematic diagram illustrating the effect of the magnetic resonance channel sensitivity map estimation method provided in the embodiments of this specification; Figure 4 This is a schematic diagram of the computer device provided in the embodiments of this specification. Detailed Implementation

[0022] To enable those skilled in the art to better understand the solutions described in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of them. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this specification.

[0023] This specification provides an example application scenario for a magnetic resonance channel sensitivity map estimation method. Please refer to [link to relevant documentation]. Figure 1 This application scenario can include a terminal 110 and a magnetic resonance imaging system 120. The terminal 110 can be a computer or PC with image processing capabilities; for example, the terminal 110 can be, but is not limited to, a personal computer, laptop, or tablet. The magnetic resonance imaging system 120 can be a medical scanning device capable of performing magnetic resonance sequence scanning, and can acquire image echo data and reference echo data from the patient or subject.

[0024] For example, terminal 110 can be network-connected to magnetic resonance imaging system 120 so that terminal 110 can acquire data collected by magnetic resonance imaging system 120.

[0025] For example, the application scenario may also include a server 130, which may be network connected to the terminal 110 and the magnetic resonance imaging system 120 respectively, so that the server 130 can directly obtain data collected by the magnetic resonance imaging system 120 or obtain data collected by the magnetic resonance imaging system 120 from the terminal 110, or enable the terminal 110 to obtain data collected by the magnetic resonance imaging system 120 from the server 130.

[0026] In this scenario example, terminal 110 or server 130 can estimate the magnetic resonance channel sensitivity map based on the fusion of multiple repeated reference echo subspaces. For example, firstly, multi-channel reference echo data acquired multiple times is obtained; then, the autocalibrated signal region is extracted from each repeated reference echo data; next, a local convolution kernel is constructed for each repeated reference echo data, and its effective kernel subspace is obtained by filtering through singular value decomposition; then, the effective kernel subspaces corresponding to multiple repeated reference echo data are fused to obtain a fused kernel subspace; next, image domain feature decomposition is performed on the fused kernel subspace to obtain the channel sensitivity map and an effective region mask; finally, the obtained channel sensitivity map is normalized based on the effective region mask to output the final channel sensitivity map. In this way, a more robust and accurate channel sensitivity map can be obtained, improving the robustness and accuracy of magnetic resonance channel sensitivity estimation.

[0027] This specification provides a method for estimating magnetic resonance channel sensitivity maps. Please refer to [link to relevant documentation]. Figure 2 , Figure 2 This is a flowchart illustrating a magnetic resonance channel sensitivity map estimation method provided in this specification. This embodiment provides the method operation steps as shown in the flowchart, but based on conventional or non-inventive methods, more or fewer operation steps may be included. The order of steps listed in the embodiment is merely one possible execution order among many, and does not represent the only possible execution order. In actual system or server product execution, the method can be executed sequentially as shown in the embodiment or in parallel (e.g., in a parallel processor or multi-threaded processing environment). Specifically, as follows... Figure 2 As shown, the magnetic resonance channel sensitivity map estimation method may include the following steps.

[0028] Step S210: Acquire multi-channel magnetic resonance reference echo data acquired multiple times.

[0029] Repeated acquisition refers to performing two or more independent scans or independent data acquisition operations on the same imaging object and the same imaging layer using the same imaging sequence parameters.

[0030] For example, the multi-channel magnetic resonance reference echo data acquired repeatedly can be the reference echo data acquired multiple times in a diffusion magnetic resonance imaging sequence by averaging. For example, abdominal diffusion magnetic resonance imaging.

[0031] As an example, in diffuse magnetic resonance imaging, to improve the image signal-to-noise ratio or meet the averaging requirements, the scanning sequence can be set to an averaging of 6 times. Each acquisition obtains a complete set of reference echo data. These repeatedly acquired reference echo data have the same spatial coding direction and field of view, and reflect the spatial response characteristics of the same set of receiving coils.

[0032] Multi-channel magnetic resonance reference echo data is frequency-domain spatial data acquired during magnetic resonance scanning by an array of radio frequency receiving coils containing multiple independent receiving units. Each receiving channel independently receives the magnetic resonance signal and forms one channel of K-space data. For example, if the coil array contains 28 receiving channels, the reference echo data acquired in a single acquisition is a 28-channel K-space data matrix. Reference echo data is full-sampled or low-resolution frequency-domain data acquired for calibration purposes, used to estimate the spatial sensitivity distribution of each receiving channel.

[0033] Reference echo data differs from image echo data used for final imaging. Reference echo data is acquired for calibration purposes and typically uses full sampling or low-resolution scanning at high acceleration to obtain sufficient calibration information while shortening acquisition time. Image echo data, on the other hand, is acquired for clinical diagnostic purposes and usually has higher spatial resolution.

[0034] Step S220: For each repeated acquisition of reference echo data, construct local convolution kernels and filter effective kernel bases through singular value decomposition to obtain the effective kernel subspace corresponding to each repeated acquisition of reference echo data.

[0035] Specifically, when constructing local convolutional kernels for each repeatedly acquired reference echo data, each repeatedly acquired reference echo data can be treated as an independent processing object. A separate convolutional kernel matrix is ​​constructed for each acquired reference echo data to characterize the correlation between local channels. In this way, the local channel correlation structure in each repeatedly acquired reference echo data can be explicitly expressed in matrix form, providing a data foundation for subsequent subspace extraction.

[0036] Specifically, Singular Value Decomposition (SVD) is a matrix factorization operation used to decompose a convolution kernel matrix into a product of a left singular vector matrix, a singular value diagonal matrix, and a right singular vector matrix. The singular value diagonal matrix can also be directly called the singular value matrix. Through SVD, the principal components that contribute the most energy to the convolution kernel matrix can be identified.

[0037] Specifically, when selecting effective kernel bases, a preset singular value threshold can be used to select right singular vectors with singular values ​​greater than or equal to the threshold from all right singular vectors obtained from singular value decomposition. For example, the singular value threshold can be set to 0.1, thus retaining all right singular vectors with singular values ​​greater than or equal to 0.1. In this way, kernel bases corresponding to noise and redundant components can be removed through selection, retaining only the main kernel base components with concentrated signal energy and clear physical meaning.

[0038] Specifically, the effective kernel subspace is a linear subspace spanned by the retained effective kernel basis after filtering. Its mathematical representation can be a matrix composed of column vectors of the effective kernel basis, i.e., the effective kernel subspace matrix. Each time the reference echo data is repeatedly acquired, after local convolution kernel construction and singular value decomposition filtering, a corresponding effective kernel subspace is obtained. This effective kernel subspace contains the most important local channel-related structural information in the reference echo data of that repeated acquisition, and noise and redundant components have been removed.

[0039] Step S230: The effective nucleus spaces corresponding to the reference echo data acquired in each repeated acquisition are fused to obtain the fused nucleus space.

[0040] By fusing the effective kernel subspaces corresponding to the reference echo data acquired in each repeated acquisition, a unified fused kernel subspace can be obtained. For example, the fusion operation can include, but is not limited to, directly concatenating the effective kernel subspace matrices along the column direction, weighted concatenation, or grouped fusion followed by concatenation.

[0041] The fused kernel space is a unified kernel space obtained by fusing the effective kernel spaces corresponding to each repeated acquisition data set. Its column count is the sum of the number of effective kernel bases from each acquisition. This fused kernel space also contains the main local correlation structures from different repeated acquisition data sets, which are used to solve the subsequent channel sensitivity map.

[0042] The fusion operation employs kernel-based dimensional stitching rather than data-level averaging, which differs from existing techniques that first average the reference echo data point by point before processing. Through kernel-based stitching fusion, even if there are phase differences or slight motions between different repeated acquisition data, the calibration signal region will not be blurred due to direct averaging, and the effective local correlation structure in each repeated acquisition data is preserved in the fusion kernel space.

[0043] Step S240: Based on the fusion kernel subspace, perform feature decomposition in the image domain to obtain the channel sensitivity map.

[0044] For example, an image domain channel consistency calibration operator can be constructed using the fusion kernel subspace as input. Then, feature decomposition is performed on this calibration operator to map the frequency domain local correlation structure information contained in the fusion kernel subspace to the image domain. The inter-channel coherence information at each spatial location is extracted through feature decomposition. As an example, image domain feature decomposition can be a pixel-by-pixel operation, meaning that the calibration operator is constructed and feature decomposition is performed at each spatial location. Performing feature decomposition on this calibration operator yields eigenvalues ​​and corresponding eigenvectors. The eigenvalues ​​reflect the principal component intensity of the signal at that spatial location, and the eigenvectors reflect the sensitivity distribution of each channel at that spatial location.

[0045] The channel sensitivity map is a spatial sensitivity distribution map of each receiving channel in a multi-channel receiving coil array at various spatial locations within the imaging field of view. It represents the difference in signal reception capability of each channel at different spatial locations in image form. The channel sensitivity map can be used for subsequent parallel imaging reconstruction, for example, as prior information in SENSE or GRAPPA-type reconstruction algorithms. In this embodiment, the sensitivity value of each channel at each spatial location can be obtained by selecting the eigenvector corresponding to the maximum eigenvalue after image domain feature decomposition, thus forming a channel sensitivity map covering the entire field of view.

[0046] In the above embodiments, by making full use of the multi-channel magnetic resonance reference echo data acquired repeatedly to construct effective subspaces and perform subspace fusion, and by solving for features based on the unified fused subspace, ambiguity introduced by direct averaging can be avoided, resulting in more robust and accurate channel sensitivity maps, and improving the robustness and accuracy of magnetic resonance channel sensitivity estimation.

[0047] In some implementations, step S220 involves constructing local convolution kernels for each repeatedly acquired reference echo data and filtering effective kernel bases through singular value decomposition to obtain the effective kernel subspace corresponding to each repeatedly acquired reference echo data. This may include the following steps S310-S350.

[0048] Step S310: Extract the automatic calibration signal region from the reference echo data acquired in each repeated acquisition.

[0049] Specifically, for each repeated acquisition of reference echo data, a fixed-size sub-region can be extracted from the central region of its K-space data matrix as an automatic calibration signal region.

[0050] The automatic calibration signal region is the low-frequency signal portion of the central region of K-space, containing the main energy and reliable phase information of the magnetic resonance signal. It can be used in parallel imaging to estimate the correlation and sensitivity distribution between channels.

[0051] Each time the reference echo data is repeatedly acquired, the automatically calibrated signal region is truncated at the same location and size to ensure that the kernel space constructed in subsequent iterations has a consistent dimensional structure and fusion compatibility. In this way, reliable, high signal-to-noise ratio low-frequency signal components can be extracted from the reference echo data acquired in each iteration, while unreliable data dominated by noise in the peripheral high-frequency regions can be eliminated, providing high-quality input data for subsequent local convolution kernel construction.

[0052] Step S320: Using a local sliding window method, the automatic calibration signal region is expanded into multiple local data blocks, and each local data block is vectorized.

[0053] Local sliding windowing is a data block extraction operation that slides a fixed-size window across the autocalibrated signal region with a preset step size, traversing all possible locations within the region and extracting a local data block within the window's coverage area at each sliding window position. This progressive window movement divides the overall data into multiple overlapping or non-overlapping local sub-blocks, facilitating subsequent modeling and analysis of each sub-block. By dividing the overall autocalibrated signal region into multiple local sub-blocks, the subsequently constructed local convolutional kernels can reflect signal correlations within local neighborhoods in K-space, rather than global average correlations, thereby improving the spatial resolution and local fidelity of sensitivity map estimation.

[0054] Each local data block contains K-space data values ​​for all channels at the corresponding sliding window position, and its dimension is determined by the sliding window size and the number of channels. The local data blocks extracted at different sliding window positions correspond to each other in spatial location and together cover the entire automatic calibration signal area.

[0055] For example, the sliding window's movement step size can be set to 1, meaning the window moves one data point each time along the readout direction or phase encoding direction. In this case, there is a significant overlap between the data in adjacent windows, which can preserve local spatial information to the greatest extent. For instance, when the size of the automatic calibration signal region is 32×32 and the size of the local sliding window is 8×8, the number of sliding positions of the window in both the readout direction and the phase encoding direction is (32-8+1)=25. Therefore, the total number of sliding window positions is 25×25=625.

[0056] Vectorizing each local data block means converting the multidimensional local data block extracted at each sliding window position into a one-dimensional vector. For example, it can be expanded into a one-dimensional column vector or row vector in a specific order, so that data from different spatial locations and different channels can participate in subsequent matrix construction and matrix decomposition in a unified vector form.

[0057] Step S330: Stack all the vectorized local data blocks corresponding to the sliding window positions in rows to construct a local data block matrix.

[0058] In some cases, converting multidimensional local data blocks into one-dimensional vector form allows local data blocks at multiple sliding window positions to be stacked into a two-dimensional matrix.

[0059] Row stacking refers to concatenating multiple row vectors with the same dimensions vertically to form a matrix structure where the number of rows equals the number of vectors and the number of columns equals the length of the vectors.

[0060] Specifically, the vectorized local data blocks obtained at all sliding window positions can be used as row vectors, arranged and stacked sequentially according to the row direction to form a two-dimensional matrix. This two-dimensional matrix is ​​the local data block matrix, or in other words, the local data block matrix corresponding to the currently repeatedly acquired reference echo data is used as the matrix representation of the local convolution kernel.

[0061] The number of rows in the local data block matrix equals the total number of local sliding window positions, and the number of columns equals the product of the sliding window size and the number of channels. Each element is a complex number representing the K-space signal value at the corresponding position. The column vectors of the local data block matrix constitute the original kernel basis.

[0062] Step S340: Perform singular value decomposition on the local data block matrix to obtain the right singular vector matrix.

[0063] In some cases, singular value decomposition (SVD) can decompose any complex matrix into the product of a left singular vector matrix, a singular value matrix, and a right singular vector matrix. SVD can separate the signal and noise components in the original matrix along the singular value dimension.

[0064] By performing singular value decomposition on the local data block matrix to obtain the right singular vector matrix, the local data block matrix can be decomposed into signal components and noise components, which facilitates the identification and retention of effective kernel basis corresponding to the signal components through threshold filtering.

[0065] The space spanned by the right singular vectors corresponds to the row space of the local data block matrix. Each column vector represents the main changes in the feature dimensions of the K-space data, and effective kernel basis can be selected from these right singular vectors.

[0066] Step S350: Based on the preset singular value threshold, select right singular vectors that meet the conditions as effective kernel basis, and construct an effective kernel subspace from the effective kernel basis.

[0067] Singularity thresholds can be used to distinguish between the principal components of a signal and noise components. Their values ​​can be obtained through fixed value setting, adaptive calculation, or table lookup.

[0068] Specifically, each right singular vector value can be compared with a singular value threshold. Right singular vectors with singular values ​​greater than or equal to the threshold are retained as signal components, while those with singular values ​​less than the threshold are discarded as noise components. The retained right singular vectors are used as the effective kernel basis. The subspace spanned by these right singular vectors represents the main signal components in the reference echo data of this repeated acquisition, while excluding the interference of noise components. For example, when the preset singular value threshold is 0.1, all right singular vectors greater than or equal to 0.1 are retained, while those less than 0.1 are discarded.

[0069] The selected effective kernel bases are used as a set of basis vectors. The linear space spanned by these basis vectors is the effective kernel subspace corresponding to the reference echo data of this repeated acquisition. The column vectors of the effective kernel subspace are the selected effective kernel bases. The number of rows is equal to the length of the vectorized local data block, and the number of columns is equal to the number of effective kernel bases retained in the reference echo data of this repeated acquisition.

[0070] In the above implementation, by independently performing a complete automatic calibration signal region extraction, sliding window expansion, vectorization, stacking matrix construction, singular value decomposition, and threshold screening process on each repeatedly acquired reference echo data, the effective kernel subspace corresponding to each repeated acquisition is obtained. Each reference echo data is processed independently and does not affect each other, so that the personalized information in each repeatedly acquired reference echo data is fully preserved before subsequent fusion. This can provide multiple high-quality subspace inputs containing complementary information for subsequent subspace fusion, and can avoid the signal ambiguity and phase cancellation problems introduced by directly averaging at the original data level.

[0071] In some implementations, step S230, fusing the effective nucleus space corresponding to each repeated acquisition of the reference echo data, may include: The effective nucleus spaces corresponding to the reference echo data from each repeated acquisition are directly concatenated along the nucleus basis dimension to obtain the fused nucleus space. For example, assuming there are 6 repeated acquisitions, and the effective nucleus spaces obtained in each acquisition are V1, V2, ..., V6, the fused nucleus space using the direct concatenation method is [V1, V2, V3, V4, V5, V6]. This integrates the complementary nucleus basis information from the reference echo data from each repeated acquisition while retaining the effective components from each acquisition.

[0072] In some implementations, step S240, which involves feature decomposition in the image domain based on the fusion kernel subspace, may include the following steps S410-S440.

[0073] Step S410: Construct an image domain channel consistency calibration operator based on the fusion kernel space.

[0074] In some cases, through matrix operations or operator mapping, mathematical operators that impose consistency constraints on multi-channel data in the image domain can be constructed, thereby performing eigenvalue decomposition and solving for the channel sensitivity map.

[0075] Among them, the channel consistency calibration operator can describe the linear transformation of signal consistency between different receiving channels at each spatial location in the image domain. It is represented as a square matrix at each spatial location. The eigenvector and eigenvalue of the square matrix reflect the linear combination method and coherence of the signals of different channels at that spatial location, respectively.

[0076] For example, the channel consistency calibration operator can be constructed using the ESPIRiT method, which involves performing a specific linear transformation and matrix rearrangement on the fusion kernel subspace to map the kernel basis information from the feature domain or frequency domain where the local sliding window is located to a spatial location grid in the image domain, so that each spatial location corresponds to a square matrix. For example, when the number of received channels is 28, each spatial location corresponds to a 28×28 square matrix, which is the channel consistency calibration operator at that spatial location.

[0077] Step S420: Perform eigenvalue decomposition on the image domain channel consistency calibration operator at each spatial location to obtain eigenvalues ​​and corresponding eigenvectors.

[0078] Spatial location refers to the independent grid points in the image domain divided according to the readout direction and phase encoding direction, that is, the location of each pixel or voxel in the imaging space. These locations together constitute the complete imaging field of view.

[0079] Feature decomposition is the process of performing feature decomposition operations on the channel consistency calibration operator at each pixel or voxel location in the image domain.

[0080] Specifically, for each spatial location within the imaging field of view, there exists a corresponding image domain channel consistency calibration operator. Eigenvalue decomposition is performed on this consistency calibration operator to obtain multiple eigenvalues ​​and corresponding eigenvectors at that spatial location. In this way, the correlation structure between channels at each spatial location can be quantified into eigenvalues ​​and eigenvectors. The magnitude of the eigenvalue reflects the principal component intensity of the signal at that spatial location, while the eigenvector represents the projection coefficients of each channel onto the principal component of the signal. The eigenvector is the raw data required for subsequent extraction of the channel sensitivity map and also reflects the sensitivity distribution of each channel at that spatial location.

[0081] Step S430: Select the feature vector that satisfies the preset feature value condition at each spatial location as the unnormalized channel sensitivity vector at that spatial location, and obtain the unnormalized channel sensitivity map.

[0082] Specifically, for each spatial location, one or more feature vectors can be selected from the multiple feature vectors obtained by feature decomposition according to preset feature value conditions as the channel sensitivity vector for that spatial location.

[0083] Among them, the preset feature value conditions can be pre-set feature value determination rules used to distinguish between the principal components of the signal and the noise components. These rules can take various forms such as fixed thresholds, relative proportion thresholds, or adaptive thresholds.

[0084] By arranging the channel sensitivity vectors selected at all spatial locations according to their corresponding spatial locations, a complete image domain data, namely a channel sensitivity map, can be formed.

[0085] In the above embodiments, the robustness of the fused kernel space is utilized to enable the channel sensitivity map obtained through eigenvalue decomposition to have higher quality and stronger resistance to abnormal interference across the entire image range. Step S440: Based on the preset feature value threshold, determine each spatial location and generate an effective region mask.

[0086] The eigenvalue threshold is a parameter used to distinguish between effective imaging areas (i.e., areas where real body tissue is located) and invalid areas (i.e., background air areas). Its value can be determined by setting a fixed value, adaptive calculation, or looking up empirical values ​​in a table.

[0087] Specifically, a binary classification can be performed on each spatial location by comparing it with a feature value threshold. This allows for the determination of whether each spatial location is valid or invalid. In this way, each spatial location can be identified as either a valid imaging location or a background location.

[0088] The binary classification results obtained from determining each spatial location, when arranged according to their corresponding spatial locations, form a binary image with the same spatial dimension as the imaging space, i.e., an effective region mask. The effective region mask is a binary matrix of the same size as the image domain space, where elements are either 1 or 0. A value of 1 represents the effective imaging region, while a value of 0 represents the background region, which should be excluded from subsequent processing. This provides spatial constraints for subsequent sensitivity map normalization and image reconstruction.

[0089] In some implementations, after step S440, or after determining each spatial location according to a preset feature value threshold and generating an effective region mask, the magnetic resonance channel sensitivity map estimation method may further include: normalizing the channel sensitivity map based on the effective region mask and outputting the normalized channel sensitivity map.

[0090] In some implementations, step S430, selecting a feature vector that satisfies a preset feature value condition at each spatial location, may include: selecting the feature vector corresponding to the maximum feature value at each spatial location as the unnormalized channel sensitivity vector of the spatial location.

[0091] In some cases, within the imaged object, signals from different receiving channels are coherent, and the eigenvalues ​​corresponding to the principal components of the signal are significantly larger than those corresponding to the noise components. However, in the background region, due to the absence of coherent signals, the magnitudes of the eigenvalues ​​are relatively close, and there is no prominent maximum eigenvalue. For example, at a pixel location inside a human liver, the maximum eigenvalue may be significantly larger than other eigenvalues. Therefore, selecting the eigenvector corresponding to the maximum eigenvalue as the sensitivity vector at that location can effectively capture the true channel sensitivity distribution.

[0092] For each spatial location, the eigenvector corresponding to the largest eigenvalue is selected as the channel sensitivity vector for that spatial location. This allows for the identification and selection of the principal signal components that truly reflect the spatial response characteristics of the channel from multiple feature patterns, eliminating the interference of noise components on sensitivity estimation.

[0093] In some implementations, the channel sensitivity map is normalized, including: For each spatial location, the sensitivity value of each receiving channel is divided by the square root of the sum of the squares of the sensitivity values ​​of all receiving channels, and the area outside the effective area mask is set to zero.

[0094] In some implementations, at least one of the following is a fixed value, an adaptively determined value, or a value determined by looking up a table: the size of the automatically calibrated signal region, the window size of the local sliding window, the singular value threshold, and the eigenvalue threshold preset for eigenvalue decomposition.

[0095] In some implementations, before fusing the effective nucleus space of the reference echo data corresponding to each repeated acquisition in step S130, the magnetic resonance channel sensitivity map estimation method may further include: removing abnormal data or performing weighted preprocessing on the data based on the quality evaluation index of the reference echo data of each repeated acquisition.

[0096] In some implementations, step S130, fusing the effective nucleus spaces corresponding to the reference echo data acquired in each repeated acquisition, may include: assigning different fusion weights to the effective nucleus spaces corresponding to the reference echo data acquired in different repeated acquisitions, performing weighted splicing fusion, and obtaining a fused nucleus space.

[0097] In some implementations, step S130, fusing the effective nucleus spaces corresponding to each repeated acquisition of reference echo data, may include: grouping the reference echo data of each repeated acquisition according to the acquisition category, first fusing within the group to obtain the group-intra-group fused nucleus space, and then performing a second fusion of the group-intra-group fused nucleus spaces to obtain the fused nucleus space.

[0098] In some implementations, when there are multiple sets of feature vectors that satisfy preset feature value conditions, multiple sets of channel sensitivity maps are output.

[0099] This specification provides a method for estimating magnetic resonance channel sensitivity maps, which may include the following steps S510-S560.

[0100] Step S510: Acquire multi-channel magnetic resonance reference echo data acquired multiple times.

[0101] The reference echo data acquired repeatedly can be data obtained by a magnetic resonance imaging system using a diffuse magnetic resonance sequence scan, with an average number of scans of Na, where Na is a positive integer greater than 1.

[0102] The k-space image data obtained from each repeated scan, also known as image echo or image echo data, is denoted as... Where the subscript 'a' indicates the number of repeated samplings; Represents the complex field; RO, PE, , representing the frequency coding dimension, phase coding dimension, and number of channels of the image, respectively. Image data in the frequency domain. It is undersampled, and an artifact-free image needs to be reconstructed. Image data is acquired in each repeated scan. Then, a reference echo with full sampling, low resolution, and the same imaging field of view (FOV) and image echo can be acquired immediately. Reference echo, also known as reference echo data or reference data.

[0103] Therefore, for the Na repeatedly acquired image data to be reconstructed There are multi-channel frequency domain reference echo data acquired repeatedly (Na times). .

[0104] Step S520: Extract the automatic calibration signal region from the reference echo data acquired in each repeated acquisition.

[0105] For each repeated reference echo data Extract its central region to form a region of fixed size. This is referred to as the automatic calibration signal or automatic calibration signal region. The size of the extracted central region is Nx × Ny.

[0106] Step S530: For the automatic calibration signal region of the reference echo data acquired in each repeated acquisition, construct local convolution kernels respectively, and filter effective kernel basis through singular value decomposition to obtain the effective kernel subspace corresponding to the reference echo data acquired in each repeated acquisition.

[0107] Automatic calibration signal region for the a-th repetition of the reference echo data Expand a partial sliding window. Let the size of the partial sliding window be... Then, the local data block or local multi-channel data block located at position (i, j) can be represented as: ; in, , Vectorizing each local data block yields: ; Here, vec represents the vectorization operation. The vectors corresponding to all sliding window positions are stacked row-wise to obtain the a-th repeated reference echo data. The corresponding local data block matrix: ; in, This represents the Hankel expansion operator based on a local sliding window. For local data block matrices Perform singular value decomposition: ; in: , q is the index of the singular value, the total number of singular values ​​is Q, and diag represents the operation of taking the diagonal elements.

[0108] Selecting an effective nucleus space based on a singular value threshold, let the singular value threshold be... , If the coefficient is , then retain the condition that satisfies . The singular values. The corresponding number of effective cores is Where # represents counting. Therefore, the effective nucleus subspace corresponding to the a-th repeated reference echo data is: ; in, express The q-th right singular vector in the middle.

[0109] Step S540: The effective nucleus spaces corresponding to the reference echo data acquired in each repeated acquisition are fused to obtain the fused nucleus space.

[0110] After performing the local sliding window expansion and singular value decomposition on all repeating reference echo data respectively, multiple effective nucleus spaces can be obtained: ; Where Na represents the number of times the reference echo data was repeatedly acquired. The effective kernel spaces corresponding to each repeated reference echo data are concatenated along the kernel basis dimension to obtain the fused kernel space: ; Right now: .in: .

[0111] This fused nucleus subspace contains the main local correlation structures from different repeating reference echo data, which can be used to solve subsequent channel sensitivity maps.

[0112] Step S550: Based on the fusion kernel space, perform feature decomposition in the image domain to obtain the channel sensitivity map and the effective region mask.

[0113] Based on the fusion nucleus space obtained in step S540 An image domain channel consistency calibration operator can be constructed using the standard algorithm ESPIRIT for channel sensitivity estimation. Where x represents the spatial location in the image domain. Eigenvalue decomposition is performed on the image domain calibration operator at each spatial location x: ; in, This represents the m-th eigenvalue at position x. This represents the eigenvector corresponding to the eigenvalue.

[0114] The eigenvector whose eigenvalues ​​satisfy a preset eigenvalue condition is selected as the unnormalized channel sensitivity vector for that spatial location. Preferably, the eigenvector corresponding to the largest eigenvalue is selected. .

[0115] in: argmax m Indicates finding the independent variable m This makes the function reach its maximum value.

[0116] Simultaneously, an effective region mask is generated based on the feature value threshold: ; in, To preset the feature value threshold, This is an indicator function; it outputs 1 if the condition is met, and 0 otherwise. This yields the unnormalized channel sensitivity map. And the effective area mask M.

[0117] Step S560: Based on the effective region mask, normalize the channel sensitivity map and output the normalized channel sensitivity map.

[0118] The unnormalized channel sensitivity map obtained in step S550 Perform normalization. Let the unnormalized sensitivity value of the c-th channel at spatial location x be [value missing]. The normalized channel sensitivity map is then: ; in, Indicates the number of channels. To prevent extremely small positive numbers with a denominator of zero, This is the effective region mask obtained in step S550. The final output is the normalized channel sensitivity map: For subsequent multichannel diffused magnetic resonance data Image reconstruction.

[0119] The following explanation uses abdominal diffusion magnetic resonance imaging as an example.

[0120] 1. Acquire or read multiple repeated reference echo data The input reference echo data is in the form of a four-dimensional array: The first and second dimensions represent the readout direction and phase encoding direction, respectively, both with a size of 48; the third dimension represents the 28 receiving channels; and the fourth dimension represents the resampling dimension, with a total of 6 resampling iterations. Therefore, the a-th resampling reference echo data can be represented as: .

[0121] 2. Extract the automatic calibration signal area From each repeated reference echo data By extracting a (32×32) area of ​​the automatic calibration signal from the center, we obtain: The calibration data corresponding to all repeated reference echo data are as follows: The automatic calibration signal regions of each repeating reference echo data adopt a consistent position and size to ensure the fusionability of subsequent nuclei.

[0122] 3. Kernel space construction of single-repeated reference echo data Automatic calibration signal region for the a-th repeated reference echo data Perform an 8×8 local sliding window expansion. Since the size of the automatic calibration signal region is 32×32, the number of sliding window positions is: (32-8+1)×(32-8+1)=25×25=625. The length of each local multi-channel data block after vectorization is: 8×8×28=1792.

[0123] Therefore, the local data block matrix corresponding to the a-th repeated reference echo data is: .

[0124] in, This represents the local sliding window Hankel expansion operator. For Perform singular value decomposition: Retain the right singular vectors that satisfy the singular value threshold condition as effective kernel basis: The effective nucleus space of the a-th repeated reference echo data is obtained as follows: Where Ra represents the number of valid kernel bases retained in this repeated reference echo data. In this example data, When Ra can be set to 0.1, it can also be set to 100.

[0125] In this way, a kernel space can be constructed separately for each repeated reference echo data, instead of averaging the original frequency domain data first, thus preserving the local channel correlation structure in each repeated data.

[0126] 4. Construction of integrated nucleus space After performing the local sliding window expansion and singular value decomposition on all 6 repeated reference echo data, 6 effective nucleus subspaces are obtained: ; The effective kernel spaces corresponding to each repeated reference echo data are concatenated along the kernel basis dimension to obtain the fused kernel space: ; in: .

[0127] This fused nucleus subspace contains the main local correlation structures from different repeating reference echo data, which can be used to solve subsequent channel sensitivity maps.

[0128] 5. Solving the channel sensitivity map For fusion nucleus space An image domain channel consistency calibration operator is constructed according to the standard algorithm ESPIRIT for channel sensitivity estimation. Where x represents the spatial location in the image domain. Eigenvalue decomposition is performed on the image domain calibration operator at each spatial location x: ; in, This represents the m-th eigenvalue at position x. This represents the eigenvector corresponding to the eigenvalue.

[0129] The eigenvector corresponding to the largest eigenvalue is selected as the unnormalized channel sensitivity vector for that spatial location: .

[0130] in: ; Simultaneously, an effective region mask is generated based on the feature value threshold: ; in, This is the eigenvalue threshold, which is set to 0.05 in this example. This is the indicator function. From this, the unnormalized channel sensitivity plot is obtained. And the effective area mask M.

[0131] 6. Output normalized channel sensitivity map Sensitivity map of unnormalized channels Perform normalization. Let the unnormalized sensitivity value of the c-th channel at spatial location x be [value missing]. The normalized channel sensitivity map is then: ; in, Indicates the number of channels. To prevent extremely small positive numbers with a denominator of zero, This is the effective region mask. The final output is the normalized channel sensitivity map: For subsequent multichannel diffused magnetic resonance data Image reconstruction.

[0132] Please see Figure 3 , Figure 3This diagram illustrates a comparison between the magnetic resonance channel sensitivity map estimation method described in this specification and existing channel sensitivity map estimation methods based on a single automatic calibration signal region. Figure 3 The document includes images a, b, c, and d. Images a and c are the channel sensitivity and reconstructed diffusion images obtained using the embodiments described in this specification, respectively. Images b and d are the channel sensitivity and reconstructed diffusion images obtained using existing channel sensitivity map estimation methods based on a single autocalibrated signal region, respectively. Images c and d use the same reconstruction algorithm, differing only in the channel sensitivity they employ.

[0133] It can be observed that the channel sensitivity map obtained by the implementation method in this specification ( Figure 3 Image a) in the image is of better quality, and the reconstructed image ( Figure 3 Image c) has fewer artifacts.

[0134] This specification provides a magnetic resonance channel sensitivity map estimation device, which includes a reference echo acquisition module, a nucleus space construction module, and a feature decomposition module.

[0135] The reference echo acquisition module is used to acquire multi-channel magnetic resonance reference echo data acquired multiple times. The kernel space construction module is used to construct local convolution kernels for each repeated acquisition of reference echo data, and to filter effective kernel bases through singular value decomposition to obtain the effective kernel space corresponding to each repeated acquisition of reference echo data; the effective kernel spaces corresponding to each repeated acquisition of reference echo data are fused to obtain the fused kernel space. The feature decomposition module is used to perform feature decomposition in the image domain based on the fusion kernel subspace to obtain the channel sensitivity map.

[0136] The specific functions and effects of the magnetic resonance channel sensitivity map estimation device can be explained by referring to other embodiments in this specification, and will not be repeated here. Each module in the magnetic resonance channel sensitivity map estimation device can be implemented entirely or partially through software, hardware, or a combination thereof. Each module can be embedded in the processor of a computer device in hardware form or independent of it, or it can be 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.

[0137] This specification also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a computer, implements the magnetic resonance channel sensitivity map estimation method in any of the above embodiments.

[0138] This specification also provides a computer program product containing instructions that, when executed by a computer, cause the computer to implement the magnetic resonance channel sensitivity map estimation method in any of the above embodiments.

[0139] This specification also provides a computer device, including a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, implements the magnetic resonance channel sensitivity map estimation method in any of the above embodiments.

[0140] In some implementations, please refer to Figure 4 The computer device can be a terminal, and its internal structure diagram can be as follows: Figure 4 As shown, the computer device includes a processor, memory, and a communication interface connected via a system bus. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium 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 medium. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a magnetic resonance channel sensitivity map estimation method.

[0141] It is understood that the specific examples in this document are only intended to help those skilled in the art better understand the embodiments described herein, and are not intended to limit the scope of the invention.

[0142] It is understood that in the various embodiments described in this specification, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments described in this specification.

[0143] It is understood that the various implementation methods described in this specification can be implemented individually or in combination, and the implementation methods in this specification are not limited in this respect.

[0144] Unless otherwise stated, all technical and scientific terms used in the embodiments of this specification have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of this specification. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items. The singular forms "a," "the," and "the" as used in the embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0145] It is understood that the processor in the embodiments of this specification can be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method embodiments can be completed by integrated logic circuits in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this specification. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this specification can be directly implemented by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above methods.

[0146] It is understood that the memory in the embodiments of this specification may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory may be random access memory (RAM). It should be noted that the memory in the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0147] Those skilled in the art will recognize that the units and algorithm steps of the various examples 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 implementation should not be considered beyond the scope of this specification.

[0148] Those skilled in the art will clearly 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 aforementioned method implementations, and will not be repeated here.

[0149] The above description is merely a specific embodiment of this specification, but the scope of protection of this invention 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 specification should be included within the scope of protection of this specification. Therefore, the scope of protection of this invention should be determined by the scope of the claims.

Claims

1. A method for estimating the sensitivity map of a magnetic resonance channel, characterized in that, include: Acquire multi-channel magnetic resonance reference echo data acquired multiple times; For each repeated acquisition of the reference echo data, a local convolution kernel is constructed, and effective kernel basis is selected by singular value decomposition to obtain the effective kernel subspace corresponding to each repeated acquisition of the reference echo data. The effective kernel space corresponding to the reference echo data acquired in each repeated acquisition is fused to obtain the fused kernel space; Based on the fusion kernel subspace, feature decomposition is performed in the image domain to obtain the channel sensitivity map.

2. The magnetic resonance channel sensitivity map estimation method according to claim 1, characterized in that, For each repeatedly acquired reference echo data, a local convolution kernel is constructed, and effective kernel bases are selected through singular value decomposition to obtain the effective kernel subspace corresponding to each repeatedly acquired reference echo data, including: Extract the automatic calibration signal region from the reference echo data acquired in each repeated acquisition; The automatic calibration signal region is expanded into multiple local data blocks using a partial sliding window approach, and each local data block is vectorized. Stack the vectorized local data blocks corresponding to all sliding window positions row by row to construct a local data block matrix; Singular value decomposition is performed on the local data block matrix to obtain the right singular vector matrix; Based on a preset singularity threshold, right singular vectors that meet the conditions are selected as effective kernel bases, and the effective kernel bases constitute the effective kernel subspace.

3. The magnetic resonance channel sensitivity map estimation method according to claim 1, characterized in that, The effective kernel space corresponding to the reference echo data acquired in each repeated acquisition is fused, including: The effective nucleus spaces corresponding to the reference echo data collected in each repeated acquisition are directly spliced ​​in the nucleus basis dimension to obtain the fused nucleus space.

4. The magnetic resonance channel sensitivity map estimation method according to claim 1, characterized in that, Based on the fusion kernel space, feature decomposition is performed in the image domain, and an effective region mask is also obtained. The method further includes: Based on the effective region mask, the channel sensitivity map is normalized, and the normalized channel sensitivity map is output.

5. The magnetic resonance channel sensitivity map estimation method according to claim 1, characterized in that, Based on the fusion kernel subspace, feature decomposition is performed in the image domain, including: An image domain channel consistency calibration operator is constructed based on the fusion kernel subspace; For each spatial location, the image domain channel consistency calibration operator is decomposed to obtain eigenvalues ​​and corresponding eigenvectors; Select the feature vector that satisfies the preset feature value condition at each spatial location as the unnormalized channel sensitivity vector at that spatial location to obtain the unnormalized channel sensitivity map. Based on the preset feature value threshold, each spatial location is determined to generate an effective region mask.

6. The magnetic resonance channel sensitivity map estimation method according to claim 5, characterized in that, Selecting a feature vector that satisfies a preset feature value condition at each spatial location includes: selecting the feature vector corresponding to the maximum feature value at each spatial location as the unnormalized channel sensitivity vector of the spatial location.

7. The magnetic resonance channel sensitivity map estimation method according to claim 6, characterized in that, The channel sensitivity map is normalized, including: For each spatial location, the sensitivity value of each receiving channel is divided by the square root of the sum of the squares of the sensitivity values ​​of all receiving channels, and the area outside the effective area mask is set to zero.

8. The magnetic resonance channel sensitivity map estimation method according to claim 2, characterized in that, The size of the automatic calibration signal region, the window size of the local sliding window, the singular value threshold, and the eigenvalue threshold preset for the eigenvalue decomposition are at least one of the following: a fixed value, an adaptively determined value, or a value determined by looking up a table.

9. The method for estimating magnetic resonance channel sensitivity maps according to any one of claims 1 to 7, characterized in that, The multi-channel magnetic resonance reference echo data acquired repeatedly is the reference echo data acquired multiple times in the diffusion magnetic resonance imaging sequence.

10. The method for estimating magnetic resonance channel sensitivity maps according to any one of claims 1 to 7, characterized in that, Before fusing the effective nucleus space corresponding to each repeated acquisition of the reference echo data, the method further includes: Based on the quality evaluation index of the reference echo data collected in each repeated acquisition, abnormal data are removed or the data is preprocessed with weighting.

11. The method for estimating magnetic resonance channel sensitivity maps according to any one of claims 1 to 7, characterized in that, The process of fusing the effective kernel subspace corresponding to the reference echo data acquired in each repeated acquisition includes: Different fusion weights are assigned to the effective nucleus space corresponding to the reference echo data acquired in different repeated acquisitions, and weighted splicing and fusion are performed to obtain the fused nucleus space.

12. The method for estimating magnetic resonance channel sensitivity maps according to any one of claims 1 to 7, characterized in that, The process of fusing the effective kernel subspace corresponding to the reference echo data acquired in each repeated acquisition includes: The reference echo data collected in each repeated acquisition are grouped according to the acquisition category. First, the data is fused within the group to obtain the fusion nucleus space within the group. Then, the fusion nucleus spaces of each group are fused a second time to obtain the fusion nucleus space.

13. The method for estimating magnetic resonance channel sensitivity maps according to any one of claims 1 to 7, characterized in that, When there are multiple sets of feature vectors that satisfy the preset feature value conditions, multiple sets of channel sensitivity maps are output.

14. A computer device, characterized in that, include: Memory, which stores computer programs; The processor, when executing the computer program, implements the magnetic resonance channel sensitivity map estimation method as described in any one of claims 1 to 13.

Images