Breast lesion multi-modal magnetic resonance imaging system based on multi-nuclear atom information fusion

Abstract

The present application relates to the technical field of magnetic resonance imaging, and discloses a breast lesion multi-modal magnetic resonance imaging system based on multi-nuclear atom information fusion, which comprises: synchronous acquisition of hydrogen nucleus, sodium nucleus and phosphorus nucleus signals and conversion into a quantitative concentration field, which is mapped to a hydrogen nucleus spatial system after a resolution matching operation; subsequently, a non-lesion reference background is extracted to generate a sodium abnormal distribution field and a phosphorus abnormal distribution field, and an initial lesion profile and its directional extension weight are generated; then, the abnormal layer center positions of the two abnormal distribution fields are calculated along the outer normal to obtain local layer position eccentricity, and the directional extension weight is used for weighted calculation to combine into a boundary correction amount; finally, a target lesion three-dimensional surface is reconstructed based on the boundary correction amount, and a medical image model is output by fusing the local layer position eccentricity on the surface. The present application effectively solves the multi-nuclear metabolic layer position eccentricity problem in the invasion front, and comprehensively improves the precision of lesion three-dimensional profile boundary reconstruction and metabolic distribution calibration.

Description

Technical Field

[0001] This invention relates to the field of magnetic resonance imaging technology, and more specifically, to a multimodal magnetic resonance imaging system for breast lesions based on multinuclear atomic information fusion. Background Technology

[0002] In clinical medical imaging, breast lesions typically extend along the duct-lobule axis, rather than expanding isotropically. Simultaneously, their invasive front is often accompanied by changes in spinous processes, collagen arrangement, and tumor-stromal interface remodeling. Peripheral edema and the outer stromal shell are objectively present in breast MRI. Because different nuclei have inherently different biological meanings—for example, the distribution of sodium nuclei (23Na) is influenced by both intracellular ion imbalance and the peripheral microenvironment, while the metabolic information of phosphorus nuclei (31P) is more focused on cell membrane turnover and surviving cell populations—existing multimodal MRI systems typically use simple signal subtraction or brightness superposition when processing multi-nucleus information. Since breast lesions have extremely long, curved, and branching boundaries, and the spatial resolution of multi-nucleus images is often coarse, traditional image registration schemes are prone to misfitting when processing high- and low-resolution images, resulting in large voxels being rigidly attached to the long, extended boundaries. More critically, existing fusion methods severely neglect the issue of multi-nucleus metabolic layer eccentricity existing from the invasion front to the peripheral edema shell. If the imaging system only addresses image coordinate registration without resolving this layer eccentricity, the final 3D model will still forcibly flatten the true spatial distribution of the sodium nucleus being off-center and the phosphorus nucleus being off-center into a false concentric boundary. This processing method undermines the expression target of precise tumor boundary and metabolically active pseudo-color annotation, resulting in the final 3D image failing to accurately reflect the complex and spatially offset true pathological state inside and outside the lesion. Summary of the Invention

[0003] This invention provides a multimodal magnetic resonance imaging system for breast lesions based on multinuclear atomic information fusion, which solves the technical problems mentioned in the background art.

[0004] A multimodal magnetic resonance imaging system for breast lesions based on multinuclear atomic information fusion includes: A quantitative acquisition module is used to simultaneously acquire hydrogen nucleus signals, sodium nucleus signals, and phosphorus nucleus signals, and convert the hydrogen nucleus signals, sodium nucleus signals, and phosphorus nucleus signals into a quantitative concentration field; An image registration module is used to perform resolution matching on the quantitative concentration field and map the matched quantitative concentration field to the hydrogen nucleus space system; An abnormal field construction module is used to extract a non-lesion reference background based on the hydrogen nucleus space system to generate an abnormal sodium distribution field and an abnormal phosphorus distribution field. The morphological analysis module is used to generate an initial lesion contour and extract the directional extension weights of the initial lesion contour; The eccentricity quantization module is used to calculate the abnormal layer center positions of the sodium abnormal distribution field and the phosphorus abnormal distribution field along the outer normal of the initial lesion contour, respectively, to obtain the local layer eccentricity difference, and to perform weighted calculation on the local layer eccentricity difference using the direction extension weight. The boundary reshaping module is used to combine the local layer eccentricity difference with the directional extension weight into a boundary correction amount, and reconstruct the three-dimensional surface of the target lesion based on the boundary correction amount. The image output module is used to fuse the local layer eccentricity difference on the three-dimensional surface of the target lesion and output a medical image model.

[0005] The beneficial effects of this invention include: by constructing a quantitative concentration field containing multi-nuclear atomic information and quantifying the eccentric differences in the distribution of sodium and phosphorus nuclei metabolic layers along the lesion extension direction, it precisely guides the reshaping of the three-dimensional surface of the target lesion, effectively overcoming the defect of traditional multimodal fusion methods that easily flatten and confuse internal and external abnormalities. This invention not only significantly improves the fidelity and boundary clarity of the three-dimensional spatial contour reconstruction of breast lesions, but also intuitively displays the spatial gradient evolution of metabolic activity through holographic pseudo-color mapping, providing a high-precision, high-spatial-fidelity multimodal magnetic resonance imaging scheme for clinical use. Attached Figure Description

[0006] Figure 1 This is a flowchart of the multimodal magnetic resonance imaging system for breast lesions based on multinuclear atomic information fusion of the present invention. Figure 2 This is a schematic diagram of a specific implementation scenario of the present invention. Detailed Implementation

[0007] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0008] like Figure 1 As shown, a multimodal magnetic resonance imaging system for breast lesions based on multinuclear atomic information fusion includes: A quantitative acquisition module is used to simultaneously acquire hydrogen nucleus signals, sodium nucleus signals, and phosphorus nucleus signals, and convert the hydrogen nucleus signals, sodium nucleus signals, and phosphorus nucleus signals into a quantitative concentration field; An image registration module is used to perform resolution matching on the quantitative concentration field and map the matched quantitative concentration field to the hydrogen nucleus space system; An abnormal field construction module is used to extract a non-lesion reference background based on the hydrogen nucleus space system to generate an abnormal sodium distribution field and an abnormal phosphorus distribution field. The morphological analysis module is used to generate an initial lesion contour and extract the directional extension weights of the initial lesion contour; The eccentricity quantization module is used to calculate the abnormal layer center positions of the sodium abnormal distribution field and the phosphorus abnormal distribution field along the outer normal of the initial lesion contour, respectively, to obtain the local layer eccentricity difference, and to perform weighted calculation on the local layer eccentricity difference using the direction extension weight. The boundary reshaping module is used to combine the local layer eccentricity difference with the directional extension weight into a boundary correction amount, and reconstruct the three-dimensional surface of the target lesion based on the boundary correction amount. The image output module is used to fuse the local layer eccentricity difference on the three-dimensional surface of the target lesion and output a medical image model.

[0009] Preferably, the process involves simultaneously acquiring hydrogen nucleus signals, sodium nucleus signals, and phosphorus nucleus signals, and converting these signals into a quantitative concentration field, including: The normalized intensity field of hydrogen nuclei is calculated using the following formula: ; The quantitative concentration field of sodium nuclei is calculated using the following formula: ; The quantitative concentration field of phosphate monoester is calculated using the following formula: ; The quantitative concentration field of phosphodiester is calculated using the following formula: ; The quantitative concentration field is composed of the hydrogen nucleus normalized intensity field, the sodium nucleus quantitative concentration field, the phosphate monoester quantitative concentration field, and the phosphate diester quantitative concentration field. in, Represents spatial coordinates, Indicates the original signal strength of the hydrogen nucleus. This represents the 95th percentile of hydrogen nuclear strength. Represents the normalized intensity field of hydrogen nuclei; This represents the original signal amplitude of the sodium nucleus. This represents the sodium nucleus quantitative calibration coefficient. This represents the sodium nucleus sensitivity correction factor. This represents the sodium core relaxation compensation factor. Represents the quantitative concentration field of sodium nuclei; This indicates the peak area of ​​the phosphate monoester. Indicates the quantitative calibration coefficient of phosphorus nuclei. This represents the phosphorus nucleus sensitivity correction factor. This indicates the phosphate monoester relaxation compensation factor. Represents the quantitative concentration field of phosphate monoesters; This represents the peak area of ​​the phosphate diester. This represents the phosphodiester relaxation compensator. This represents the quantitative concentration field of phosphate diester.

[0010] Spatial coordinates are three-dimensional geometric parameters that describe the position of each voxel in the breast region during magnetic resonance imaging.

[0011] The raw signal intensity of hydrogen nuclei is the signal amplitude of hydrogen nuclei at various voxel locations in the breast, obtained directly from magnetic resonance scanning. It can be acquired by combining a multi-frequency radio frequency coil with a high-resolution anatomical scanning sequence of hydrogen nuclei.

[0012] The 95th percentile of hydrogen nucleus intensity is the value at the 95th percentile after statistically sorting all voxel values ​​of the original hydrogen nucleus signal intensity, and is used for normalization processing of hydrogen nucleus signals.

[0013] The hydrogen nucleus normalized intensity field is a dimensionless, standardized signal field formed at each voxel location in the mammary gland after dividing the original hydrogen nucleus signal intensity by the ninety-fifth percentile of the hydrogen nucleus intensity.

[0014] The original signal amplitude of the sodium nucleus is the signal amplitude of the sodium nucleus at various voxel locations in the breast directly obtained by magnetic resonance scanning. It can be acquired by combining a multi-frequency radio frequency coil with a dedicated sodium nucleus scanning sequence.

[0015] The sodium nucleus quantitative calibration coefficient is a proportionality coefficient used to convert the original sodium nucleus signal amplitude into a quantitative concentration. It is preferably a value obtained by back-calculation using a reference tube of sodium chloride solution with a known concentration. The value is determined based on the fact that this coefficient needs to eliminate the quantitative influence of scanning hardware and sequence on the sodium nucleus signal, so that the signal amplitude corresponds linearly with the actual concentration.

[0016] The sodium nucleus sensitivity correction factor is a coefficient used to correct for differences in the receiving sensitivity of the sodium nucleus radio frequency coil at different locations in the breast. It is preferably normalized to a value of 1 at the center of the reference tube, because it is necessary to eliminate the interference of uneven coil spatial sensitivity on the quantitative analysis of the sodium nucleus signal and ensure the consistency of signal correction at different locations.

[0017] The sodium nucleus relaxation compensation factor is a coefficient used to compensate for the attenuation of the sodium nucleus signal due to the relaxation effect. It is preferably set to 1 when acquiring a single echo signal, and the value is calculated by estimating the relaxation time from the echo ratio when acquiring multiple echo signals. This is because the relaxation time cannot be robustly estimated with a single echo signal. Setting it to 1 can define an equivalent concentration field, and the value can be accurately compensated for the relaxation attenuation of the signal after calculation with multiple echo signals.

[0018] The sodium nucleus quantitative concentration field is the sodium nucleus concentration distribution field formed at various voxel locations in the mammary gland after the original sodium nucleus signal amplitude has been calibrated, sensitivity corrected, and relaxation compensated. The unit is millimoles per liter.

[0019] The peak area of ​​phosphate monoester is the peak area value obtained by fitting the resonance peak corresponding to phosphate monoester in the magnetic resonance phosphorus nucleus spectrum scan. It can be obtained by acquiring phosphorus nucleus spectrum or spectral image scan sequence through a multi-frequency radio frequency coil and then using a spectrum fitting algorithm.

[0020] The phosphorus nucleus quantitative calibration coefficient is a proportionality coefficient used to convert the phosphorus nucleus-related peak area into a quantitative concentration. It is preferably a value obtained by back-calculation using a reference tube of phosphate solution with a known concentration. This coefficient needs to eliminate the quantitative influence of scanning hardware and sequence on the phosphorus nucleus signal so that the peak area corresponds linearly with the actual concentration.

[0021] The phosphorus nucleus sensitivity correction factor is a coefficient used to correct the difference in the receiving sensitivity of the phosphorus nucleus radio frequency coil at different locations in the breast. It is preferably normalized to a value of 1 at the center of the reference tube, because it is necessary to eliminate the interference of the uneven spatial sensitivity of the coil on the quantitative measurement of the phosphorus nucleus signal and ensure the consistency of signal correction at different locations.

[0022] The phosphate monoester relaxation compensation factor is a coefficient used to compensate for the attenuation of the phosphate monoester signal due to the relaxation effect. It is preferred to set it to 1 when acquiring single echoes. When acquiring multiple echoes, the value is calculated by estimating the relaxation time from the echo ratio. This is because the relaxation time cannot be robustly estimated with single echoes. Setting it to 1 can define an equivalent concentration field. After calculation with multiple echoes, it can accurately compensate for the relaxation attenuation of the signal.

[0023] The quantitative concentration field of phosphate monoester is the concentration distribution field of phosphate monoester formed at various voxel positions in the mammary gland after calibration, sensitivity correction, and relaxation compensation of the peak area of ​​phosphate monoester. The unit is millimoles per liter.

[0024] The peak area of ​​phosphate diester is the peak area value obtained by fitting the resonance peak corresponding to phosphate diester in magnetic resonance phosphorus nucleus spectrum scanning. It can be obtained by acquiring phosphorus nucleus spectrum or spectrum scanning sequence through multi-frequency radio frequency coil combined with phosphorus nucleus spectrum or spectrum scanning sequence and then using spectrum fitting algorithm.

[0025] The phosphodiester relaxation compensation factor is a coefficient used to compensate for the attenuation of the phosphodiester signal due to the relaxation effect. It is preferably set to 1 for single-echo acquisition and calculated from the relaxation time estimated by the echo ratio for multi-echo acquisition. This is because the relaxation time cannot be robustly estimated with a single echo. Setting it to 1 can define an equivalent concentration field, and after calculation with multiple echoes, it can accurately compensate for the signal relaxation attenuation.

[0026] The quantitative concentration field of phosphodiester is the phosphodiester concentration distribution field formed at various voxel locations in the mammary gland after calibration, sensitivity correction, and relaxation compensation of the phosphodiester peak area. The unit is millimoles per liter.

[0027] In detail, robust normalization of the hydrogen nucleus signal is achieved through the 95th percentile, avoiding the problem of extreme value interference that traditional minimum-maximum normalization is susceptible to. At the same time, a sensitivity correction factor is introduced to eliminate the influence of coil spatial reception differences on quantification. The phosphorus nucleus signal is split into phosphate monoester and phosphate diester for separate quantification to match their different membrane turnover biological characteristics. A relaxation compensation factor is also included to account for attenuation interference caused by signal relaxation effect. Finally, a multi-nucleus quantification system is constructed, which includes a hydrogen nucleus normalized intensity field and a sodium nucleus phosphate monoester and phosphate diester quantitative concentration field. This system can effectively distinguish the biological differences of the nuclei themselves and the technical biases caused by scanning hardware and sequence. For example, the traditional method of directly superimposing the original brightness signals of sodium and phosphorus nuclei may lead to the signal being misjudged as abnormal due to the low reception sensitivity of the coil at the edge of the lesion. However, this scheme can accurately restore the actual concentration of nuclei in different locations of the breast after sensitivity correction. Moreover, all signal acquisition and quantitative conversion are completed in the same position and coordinate frame, ensuring the spatial consistency of the multi-nucleus quantitative field.

[0028] In detail, the specific calibration method for the quantitative calibration coefficients of sodium and phosphorus nuclei involves fixing reference tubes of sodium chloride solution and phosphate solution of known concentration outside the bilateral breast coils. After completing the magnetic resonance scan, the original signal amplitude of sodium nuclei and the related peak area of ​​phosphorus nuclei in the reference tubes are obtained. Then, the calibration coefficients are calculated by inverse calculation formulas using the sensitivity correction factor and relaxation compensation factor corresponding to the reference tubes. The method for obtaining the sensitivity correction factors of sodium and phosphorus nuclei is as follows: for magnetic resonance equipment with B1 mapping function, the coil sensitivity distribution is directly measured; for equipment without this function, a uniform phantom is used to obtain the receiver sensitivity template, and the template is normalized to the breast shape. In both methods, the correction factor is normalized to the center value of 1 in the reference tube. The relaxation compensation factor for sodium nuclei phosphate monoester and phosphate diester is calculated by estimating the relaxation time of the corresponding nucleus from the signal ratio of different echoes in dual-echo or multi-echo acquisition mode, and then using an exponential formula. The compensation value is calculated using the formula. In single-echo acquisition mode, the compensation factor is directly set to 1, and the resulting concentration field is defined as the equivalent concentration field. The acquisition sequence for hydrogen, sodium, and phosphorus nuclei is as follows: first, high-resolution anatomical scanning of hydrogen nuclei is performed, followed by simultaneous scanning of sodium nuclei and spectral or image scanning of phosphorus nuclei. The acquisition position is prone with both breasts naturally suspended, using the chest wall fixation plate as a rigid reference. All three-nucleus scan data are taken with the chest wall plane as the z-axis zero point, the line connecting the left and right nipples as the x-axis, and the head-to-foot direction as the y-axis to achieve coordinate unification. The statistical range of the 95th percentile of hydrogen nucleus intensity is all voxels in the effective imaging area of ​​the breast, excluding invalid voxels from air and scanning background. For example, if the effective imaging area of ​​the breast in a certain case contains 1000 voxels, after sorting the original hydrogen nucleus signal intensity in ascending order, the value at the 950th position is the 95th percentile. Normalizing with this value can effectively avoid the interference of extreme values ​​on the standardization results of hydrogen nucleus signals.

[0029] Preferably, resolution matching is performed on the quantitative concentration field, and the matched quantitative concentration field is mapped to the hydrogen nucleus space system, including: The denoised quantitative concentration field is calculated using the following formula: ; The required standard deviation for additional convolution can be calculated using the following formula: ; The matched quantitative concentration field is calculated using the following formula: ; in, and Represents spatial coordinates, Indicates the imaging volume. This represents the quantitative concentration field. Indicates the spatial kernel scale. Indicates the intensity kernel scale, This represents the normalized intensity field of the hydrogen nuclei. This represents the quantitative concentration field after noise reduction. Indicates the uniform standard deviation of the target. Indicates the effective standard deviation of the prokaryotes. Indicates the standard deviation required for additional convolution; This represents a Gaussian kernel with the required standard deviation for the additional convolution as a parameter. This represents the convolution operation. Represents the inverse mapping of deformation transformation. This represents the matched quantitative concentration field.

[0030] The imaging volume is the effective imaging space of the breast region in a magnetic resonance imaging scan, encompassing all voxels within that range.

[0031] Spatial coordinates are three-dimensional geometric parameters used for integration operations to determine the position of voxels within the breast imaging volume, and are defined in the same way as the coordinates in the hydrogen nucleus space system.

[0032] Spatial kernel scale is a scale parameter used in exponential kernel filtering denoising to measure the degree of influence of spatial distance on the denoising effect. It is preferably 1.5 times the side length of the hydrogen nucleus voxel. This value can ensure effective noise filtering while preserving the slender and curved boundary geometric features of breast lesions and avoid structural loss caused by excessive smoothing.

[0033] The intensity kernel scale is a scale parameter used in exponential kernel filtering denoising to measure the influence of the difference in hydrogen kernel normalized intensity on the denoising effect. It is preferably 0.10, since the hydrogen kernel normalized intensity field has been normalized to the numerical range of 0 to 1. This value can accurately match the weight allocation of intensity differences and achieve multi-kernel collaborative denoising that preserves the edges.

[0034] The denoised quantitative concentration field is the nuclear seed concentration distribution field formed at the location of each voxel in the breast after the quantitative concentration field has been denoised by exponential kernel filtering. It preserves the edge features of the lesion and the noise is effectively suppressed.

[0035] The target uniform standard deviation is a target standard deviation used to unify the ambiguity of the effective point spread function of different kernel types. It is preferably the standard deviation corresponding to the worst effective point spread function among all multi-kernel channels. This value can ensure that the resolution of all kernel types is matched to the lowest level, avoiding registration errors and voxel misalignment caused by the mismatch between high and low resolutions.

[0036] The effective standard deviation of prokaryotes is the original effective Gaussian point diffusion function standard deviation corresponding to the quantitative concentration field of each prokaryote, reflecting the spatial resolution characteristics of the original scan data of the prokaryotes.

[0037] The standard deviation required for additional convolution is the standard deviation required for additional Gaussian convolution of the quantitative concentration field of each nuclear species to achieve resolution matching. It is calculated from the target uniform standard deviation and the original effective standard deviation of the nuclear species.

[0038] The Gaussian kernel with the standard deviation required for additional convolution as a parameter is a Gaussian filter kernel used to perform convolution operations on the denoised quantitative concentration field. Its standard deviation is the standard deviation required for additional convolution, and it is used to adjust the kernel resolution to the target level.

[0039] Deformation transformation inverse mapping is a transformation relationship that converts the voxel coordinates of each nuclear space to the coordinates of the hydrogen nuclear space system. It is preferably an inverse mapping that combines rigid transformation with B-spline non-rigid transformation, because rigid transformation can ensure accurate alignment of coordinates of hard tissues such as the chest wall, while non-rigid transformation can adapt to the local deformation characteristics of breast soft tissue, thus achieving accurate mapping of multi-nucleus space.

[0040] The matched quantitative concentration field is the nuclear seed concentration distribution field formed in the hydrogen nucleus space system after resolution matching and deformation transformation of the denoised quantitative concentration field. The spatial resolution is uniform and the coordinates are consistent with the hydrogen nucleus space.

[0041] The matched sodium nucleus quantitative concentration field is the sodium nucleus concentration distribution field formed in the hydrogen nucleus space system after resolution matching and spatial mapping of the sodium nucleus quantitative concentration field, with the unit being millimoles per liter.

[0042] The matched quantitative concentration field of phosphate monoester is the phosphate monoester concentration distribution field formed in the hydrogen nucleus space system after resolution matching and spatial mapping, with the unit being millimoles per liter.

[0043] The matched quantitative concentration field of phosphate diester is the phosphate diester concentration distribution field formed in the hydrogen nucleus space system after resolution matching and spatial mapping, with the unit being millimoles per liter.

[0044] In detail, an exponential kernel filtering denoising model is constructed using the hydrogen nucleus normalized intensity field as the anatomical guidance parameter and combined with spatial distance parameters. This model ensures that the denoising process closely matches the actual anatomical structure of the breast, accurately preserving the slender boundaries and spinous process features of lesions. Furthermore, before coordinate registration, the standard deviation required for additional convolution is calculated, and a Gaussian convolution operation is performed to achieve equivalent resolution matching for different nuclear species. This avoids the voxel distortion problem caused by directly interpolating and amplifying low-resolution nuclear species signals. Finally, a deformation transformation inverse mapping is used to complete the coordinate transformation from the nuclear species space to the hydrogen nucleus space system, forming a standardized process of denoising, resolution matching, and spatial mapping. The process involves performing a complete set of operations on the quantitative concentration fields of sodium nuclei, monophosphate, and diphosphate to ensure the spatial consistency and resolution uniformity of each nucleus concentration field in the hydrogen nucleus space system. For example, the traditional method directly interpolates and amplifies the low-resolution sodium nucleus signal and registers it to the high-resolution hydrogen nucleus space, which will cause the slender, spinous boundaries of the breast lesion to be completely covered by the amplified macrocarpet, resulting in the loss of boundary features. However, this method first unifies the resolution of all nuclei to the worst level and then performs spatial mapping, which can accurately restore the slender boundaries and branching structures of the lesion and effectively block the misalignment effect of multi-nucleus macrocarpets to the fine boundaries of the breast lesion.

[0045] In detail, the spatial kernel scale is fixed at 1.5 times the side length of the hydrogen nucleus voxel, and the intensity kernel scale is fixed at 0.10. The target unified standard deviation is directly selected from the standard deviation corresponding to the worst effective point diffusion function among all multi-core channels, without additional adjustment. The deformation transformation inverse mapping transformation model consists of rigid transformation and B-spline non-rigid transformation. First, rigid alignment is performed on the chest wall and the entire breast, and then local B-spline non-rigid correction is performed within the volume of the suspended breast outside the chest wall, ensuring that the hard tissue coordinates are not deformed and adapting to the local deformation of soft tissue. The effective standard deviation of the original kernel is back-calculated from the transition width of 10% to 90% of the edge of the same scanning reference tube. The specific calculation method is to divide the transition width by 2.563, which is a fixed conversion factor between the transition width and the standard deviation of the Gaussian point diffusion function edge. Exponential kernel filtering The integral and Gaussian convolution operations for denoising are rewritten as voxel-level summation operations to achieve discretization. The inverse deformation transformation mapping achieves discretization by performing cubic linear interpolation on the voxel coordinates, which can be directly performed on the discrete voxel data of magnetic resonance imaging. The verification standard for the effectiveness of resolution matching and spatial mapping is that the coordinate deviation of each nuclear species at clearly defined anatomical landmarks such as the chest wall and nipple after matching does not exceed 1 voxel, and the spatial overlap rate of the breast lesion area is not less than 95%. For example, if the side length of the hydrogen nuclear voxel in a certain case is 1 mm, its spatial nuclear scale is taken as 1.5 mm. The transition width from 10% to 90% of the edge of the sodium nuclear reference tube is 3 mm, and its effective standard deviation of the pronuclear species is calculated to be 1.17 mm. With such standardized values ​​and operation methods, resolution matching and spatial mapping are implemented.

[0046] Preferably, a non-lesion reference background is extracted based on the hydrogen nucleus space system to generate an abnormal sodium distribution field and an abnormal phosphorus distribution field, including: The matched sodium nucleus quantitative concentration field, the matched phosphate monoester quantitative concentration field, and the matched phosphate diester quantitative concentration field are extracted from the matched quantitative concentration field; The sodium anomaly distribution field is calculated using the following formula: ; The phosphorus anomaly distribution field is calculated using the following formula: ; The unified metabolic driving field is calculated using the following formula: ; in, Represents spatial coordinates, This represents the matched sodium nucleus quantitative concentration field. This represents the median value of the sodium nucleus reference background. This represents the anomalous distribution field of sodium; This represents the quantitative concentration field of the matched phosphate monoester. This represents the quantitative concentration field of the matched phosphodiester. Represents a zero-limit decimal. This indicates the phosphorus-to-nucleus ratio as a reference to the background median value. This indicates the abnormal phosphorus distribution field; Indicates the sodium abnormal weight, Indicates the sodium anomaly scale factor. Indicates the phosphorus anomaly weight. Indicates the phosphorus anomaly scale factor. This represents a unified metabolic driving field.

[0047] The median value of the sodium nucleus reference background is the median value of the quantitative concentration field of the sodium nucleus after matching in the ipsilateral non-lesion fibrogland tissue reference area. It is used to remove the interference of normal tissue background in the sodium nucleus signal.

[0048] The sodium anomaly distribution field is the sodium nucleus anomaly signal distribution field obtained by subtracting the sodium nucleus reference background median value from the matched sodium nucleus quantitative concentration field and taking the maximum value of the result with zero, with the unit being millimoles per liter.

[0049] The zero-prevention minimum is a minimum value set to avoid the denominator being zero in the calculation of the ratio of monophosphate to diphosphate. It is preferably 0.05 times the median value of the quantitative concentration field of diphosphate after matching in the reference area of ​​ipsilateral non-lesion fibrogland tissue. This value can completely avoid the denominator being zero while minimizing the interference on the calculation results of the phosphorus nucleus concentration ratio field and not changing the true distribution characteristics of phosphorus nucleus abnormalities.

[0050] The adjusted quantitative concentration field of phosphate diester is the phosphate diester concentration distribution field obtained by adding the zero-limiting minimum to the matched quantitative concentration field of phosphate diester, and the unit is millimoles per liter.

[0051] The phosphorus nucleus concentration ratio field is a dimensionless ratio distribution field obtained by dividing the matched quantitative concentration field of phosphate monoester by the adjusted quantitative concentration field of phosphate diester, reflecting the abnormal characteristics of phosphorus nucleus membrane turnover.

[0052] The median reference background value of the phosphorus nucleus ratio is the median value of the phosphorus nucleus concentration ratio field in the ipsilateral non-lesion fibrogland tissue reference area, used to remove interference from the normal tissue background in the phosphorus nucleus ratio.

[0053] The phosphorus anomaly distribution field is the phosphorus nucleus anomaly signal distribution field obtained by subtracting the phosphorus nucleus ratio reference background median from the phosphorus nucleus ratio field and taking the maximum value of the result with zero. It is dimensionless.

[0054] The sodium anomalous scale factor is a scale parameter used to normalize the sodium anomalous distribution field. It is determined by the statistical value of the sodium anomalous distribution field in the initial body of the lesion.

[0055] Sodium anomaly weight is a weighting parameter used to allocate the contribution of the sodium anomaly distribution field to the unified metabolic driving field. It is preferably 0.5. This value can make the contributions of sodium nucleus anomaly signal and phosphorus nucleus anomaly signal in the unified metabolic driving field balanced, without biasing the anomalous information of a single nucleus.

[0056] The phosphorus anomaly scale factor is a scale parameter used to normalize the phosphorus anomaly distribution field. It is determined by the statistical value of the phosphorus anomaly distribution field in the initial body of the lesion.

[0057] The phosphorus anomaly weight is a weight parameter used to allocate the contribution ratio of the phosphorus anomaly distribution field in the unified metabolic driving field. It is preferably 0.5. This value can make the phosphorus nucleus anomaly signal and the sodium nucleus anomaly signal contribute equally in the unified metabolic driving field, and the sum of the sodium anomaly weight and the phosphorus anomaly weight is 1, which ensures the dimensional consistency of the driving field.

[0058] The unified metabolic driving field is a single metabolic abnormality distribution field obtained by normalizing and weighting the sodium abnormality distribution field and the phosphorus abnormality distribution field. It is dimensionless.

[0059] In detail, using ipsilateral non-lesion fibroadenoma tissue as a reference background and the median as the statistical measure, this approach avoids the problem of the whole breast mean being easily interfered with by the normal high signal of dense fibroadenoma tissue. Furthermore, the median value enhances robustness to outliers. Considering the biological characteristics of phosphate nucleus membrane turnover, instead of constructing an anomaly field using the absolute concentration of a single phosphate nucleus component, a phosphate nucleus concentration ratio field is constructed using the ratio of monophosphate to adjusted diphosphate, which better reflects the abnormal changes in phosphate nucleus membrane turnover. Additionally, a zero-limit prevention method is introduced to address the issue of zero denominators in ratio calculations. Furthermore, it addresses the anomaly distribution field of sodium and... After being normalized by a scale factor, the abnormal phosphorus distribution fields are fused into a unified metabolic driving field with equal weights, thereby standardizing and coordinating multi-nucleus abnormal signals. For example, the traditional method uses the average value of the entire breast as the sodium nucleus reference background, which may misjudge the normal high sodium signal of dense fibroglandular tissue outside the lesion as sodium nucleus abnormality, resulting in an incorrect expansion of the spatial range of the abnormal field. However, this scheme uses ipsilateral non-lesion fibroglandular tissue as the reference background, which can accurately identify sodium nucleus abnormalities only in the lesion area. Moreover, the unified metabolic driving field fused with equal weights can simultaneously reflect the ion imbalance abnormality of sodium nucleus and the membrane turnover abnormality of phosphorus nucleus.

[0060] In detail, the automatic selection process for the ipsilateral non-lesion fibroadenoma reference area is as follows: first, the entire breast is segmented into adipose and fibroadenoma tissues; then, areas within 15 mm beyond the lesion candidate area are removed; next, areas within 5 mm of the skin surface and within 10 mm of the post-nipple region are removed; finally, the largest connected region in the remaining fibroadenoma tissue is selected as the reference area. The statistical range of the median of the sodium nucleus reference background and the median of the phosphorus nucleus ratio reference background are all voxels of the automatically selected ipsilateral non-lesion fibroadenoma reference area; the zero-limit is calculated as 0.05 times the median of the matched phosphodiester quantitative concentration field of the ipsilateral non-lesion fibroadenoma reference area; if the median is 0, the smallest positive value among all orthophosphodiester voxels is taken; sodium abnormality scale... The scale factor is taken as the 95th percentile of the sodium abnormality distribution field in the initial body of the lesion, and the scale factor of the phosphorus abnormality is taken as the 95th percentile of the phosphorus abnormality distribution field in the initial body of the lesion. The weights of sodium abnormality and phosphorus abnormality are both fixed at 0.5, and their sum is 1. The effectiveness criterion for the unified metabolic driving field is that the mean voxel value of the driving field in the lesion area is significantly higher than that in the ipsilateral non-lesion fibrogland reference area, and the difference between the two values ​​is statistically tested with a P value of less than 0.05. For example, in a certain case, the median value of the quantitative concentration field of phosphodiester in the ipsilateral non-lesion fibrogland reference area is 2 mmol / L, and its zero minimum value is taken as 0.1 mmol / L. With such standardized selection rules and calculation methods, the construction and implementation of the multinucleus abnormality field and the unified metabolic driving field are realized.

[0061] Preferably, generating an initial lesion contour and extracting the directional extension weights of the initial lesion contour includes: The initial lesion contour is obtained by extracting the set of coordinate points with zero signed distance field values ​​for the initial contour using the following formula: ; The local covariance matrix is ​​calculated using the following formula: ; The directional extension weight is calculated using the following formula: ; in, Represents spatial coordinates, This represents the signed distance field of the initial contour. This represents the initial lesion outline; This represents a surface point on the initial lesion contour. Indicates the point number. Indicates the number of neighboring sampling points. Represents the coordinates of sampling points in the neighborhood of the surface. Represents the mean coordinates of the neighborhood points. Represents the transpose matrix notation. Represents the local covariance matrix; This represents the first covariance eigenvalue. This represents the eigenvalue of the second covariance. Indicates a small positive number with zero-prevention characteristics. This indicates the directional extension weight.

[0062] The initial lesion mask is a binary mask of the breast lesion region generated based on a unified metabolic driving field, representing the initial spatial extent of the lesion.

[0063] The initial contour signed distance field is the field distribution obtained after performing distance transformation operations on the initial lesion mask, and it represents the signed spatial distance from each voxel of the breast to the initial lesion contour.

[0064] The initial lesion contour is the set of three-dimensional voxel coordinates with a signed distance field value of zero, which represents the initial three-dimensional geometric boundary of the breast lesion.

[0065] Surface points are arbitrary three-dimensional geometric points on the initial lesion contour, serving as the basic building blocks of the contour.

[0066] The surface neighborhood sampling point coordinates are the coordinates of three-dimensional geometric points within the neighborhood range selected around each surface point on the initial lesion contour.

[0067] The mean coordinates of the neighborhood points are the average of the coordinates of all the sampling points in the neighborhood corresponding to each surface point, and are the center coordinates of the local neighborhood.

[0068] The number of neighborhood sampling points is the number of surface neighborhood sampling points selected around each surface point.

[0069] The local covariance matrix is ​​a matrix calculated based on the difference between the coordinates of the sampling points in the neighborhood of the surface and the coordinates of the mean coordinates of the neighborhood points. It represents the geometric distribution characteristics of the neighborhood of the surface points.

[0070] The first covariance parameter is a larger eigenvalue obtained after decomposing the local covariance matrix, which characterizes the degree of dispersion of the neighborhood geometric distribution in the main direction.

[0071] The second covariance parameter is a smaller eigenvalue obtained after decomposing the local covariance matrix, which characterizes the degree of dispersion of the neighborhood geometric distribution in this direction.

[0072] The zero-prevention minimum positive number is a minimum value set to avoid the case where the denominator is zero in the calculation of the directional extension weight. It is preferably 0.01 times the first covariance parameter. This value can completely avoid the problem of the denominator being zero, and because the value is extremely small, it will not have a substantial impact on the calculation result of the directional extension weight, and can accurately preserve the quantitative characteristics of the duct spinous process direction.

[0073] The directional extension weight is a value calculated based on the first and second covariance parameters, representing the extent of extension of each surface point on the initial lesion contour along the duct-like spinous process direction.

[0074] In detail, an initial lesion mask is generated based on a unified metabolic-driven field that integrates multi-core anomaly information. This allows the initial lesion extent to better reflect the true metabolic characteristics of breast lesions. A signed distance field is then obtained through distance transformation, and its set of zero-value points defines the initial lesion contour, replacing the traditional binary mask for direct contour extraction. This results in a smoother, more continuous 3D contour with higher geometric accuracy. Furthermore, curvature normal analysis is abandoned. Instead, the first and second covariance parameters are extracted by calculating the local covariance matrix of the surface point neighborhood. These parameters are then combined with positive values ​​to prevent zero minima to calculate directional extension weights. The unique characteristics of breast lesions are quantified using eigenvalue ratios. The method features tubular, spinous extension characteristics and calculates the local covariance matrix and directional extension weight for each surface point on the initial lesion contour separately, achieving directional-specific quantification of the contour rather than global weighting of the entire lesion contour. For example, traditional methods that generate lesion masks based on a single hydrogen nucleus signal are prone to missing metabolically abnormal spinous regions, and contours extracted directly from binary masks are prone to breakage and jaggedness. Curvature analysis cannot accurately quantify the direction of slender ductal spinous processes. However, this method can accurately define the initial lesion range, generate a continuous three-dimensional contour, and allow surface points along the direction of ductal spinous processes to obtain higher directional extension weights.

[0075] In detail, the initial lesion mask generation process involves selecting the maximum value point in a unified metabolic driving field as the seed point, performing regional growth on a hydrogen nucleus normalized intensity field, and using the growth criteria that the difference in hydrogen nucleus normalized intensity between adjacent voxels does not exceed the local 90th percentile gradient threshold and the Euclidean distance from the seed point does not exceed 25 mm. After growth, the mask undergoes three-dimensional closure operation and hole filling processing. The geodesic sampling radius of the surface neighborhood sampling points is 4 mm; if the number of neighborhood sampling points is less than 20, the sampling radius is increased to 6 mm. The calculation of the local covariance matrix requires first removing the normal component of the surface points, projecting the coordinates of all surface neighborhood sampling points onto the tangent plane of the surface point, and then solving the matrix. Zero-minimum positive numbers are calculated using 0.01 times the first covariance factor. For variance parameter calculation, if the first covariance parameter is 0, the directional extension weight of the surface point is directly reset to 1. The directional extension weight does not need to be normalized or thresholded; the original calculated value can be retained. The initial contour signed distance field is obtained by performing a three-dimensional signed distance transformation on the initial lesion mask. The voxel step size of the transformation is consistent with the voxel step size of the hydrogen nucleus scan. For example, if the first covariance parameter of a surface point is 5 square millimeters, its corresponding minimum positive number is taken as 0.05 square millimeters. If a surface point is sampled with a radius of 4 millimeters and only 18 neighborhood points are obtained, its sampling radius is adjusted to 6 millimeters. With such standardized operation steps and numerical settings, the initial lesion contour is generated and the directional extension weight is extracted.

[0076] Preferably, the abnormal layer center positions of the sodium abnormal distribution field and the phosphorus abnormal distribution field are calculated along the outer normal of the initial lesion contour, respectively, to obtain the local layer eccentricity difference, and the local layer eccentricity difference is weighted using the directional extension weight, including: The location of the center of the sodium anomaly layer can be calculated using the following formula: ; The location of the center of the phosphorus anomaly layer can be calculated using the following formula: ; The local stratum eccentricity difference is calculated using the following formula: ; The weighted eccentric metabolic shell difference of the ductal spinous process is calculated using the following formula: ; in, This represents a surface point on the initial lesion contour. Indicates the signed distance along the normal direction. Indicates the maximum sampling distance towards the inside of the lesion. Indicates the maximum sampling distance to the outside of the lesion. Indicates the external normal direction of a unit. This indicates the anomalous distribution field of sodium. Denotes the distance differential element. Indicates the location of the center of the sodium anomaly layer; This indicates the abnormal phosphorus distribution field. Indicates the location of the center of the phosphorus anomaly layer; This indicates the local stratum eccentricity difference; This represents the initial lesion outline. This indicates the directional extension weight. Represents the differential element of area. This indicates the weighted eccentric metabolic shell difference of the ductal spinous process.

[0077] The maximum sampling distance to the inside of the lesion is the maximum distance threshold for sampling from the outside of the initial lesion outline to the inside of the lesion, preferably 4 mm. This value can completely cover the spatial range of the inner layer of the breast lesion's surviving and invasive layer, and conforms to the physiological structural characteristics of the inner tissue.

[0078] The maximum sampling distance to the outside of the lesion is the maximum distance threshold for sampling to the outside of the lesion along the normal direction of the initial lesion outline, preferably 6 mm. This value can completely cover the spatial range of the outer edematous stroma layer of the breast lesion and is adapted to the pathological distribution characteristics of the outer tissue.

[0079] The signed distance along the outer normal is the spatial distance value per unit outer normal along the surface point of the initial lesion contour. The value is negative for the inner side of the lesion and positive for the outer side of the lesion.

[0080] The unit outward normal is the unit normal vector of each surface point on the initial lesion contour pointing outward from the lesion, representing the spatial direction of the normal of the surface point.

[0081] The distance differential element is a tiny distance increment when sampling along the unit outward normal.

[0082] The center position of the sodium anomaly layer is the normal centroid position of the sodium anomaly distribution field calculated along the unit outward normal, representing the overall spatial position of the sodium core anomaly layer, in millimeters.

[0083] The center position of the phosphorus anomaly layer is the position of the normal centroid of the phosphorus anomaly distribution field calculated along the unit outward normal, which characterizes the overall spatial position of the phosphorus nucleus anomaly layer, in millimeters.

[0084] Local eccentricity difference is the value of subtracting the center position of the phosphorus anomaly layer from the center position of the sodium anomaly layer. It characterizes the degree of normal eccentricity between the sodium nucleus and phosphorus nucleus anomaly layers at a single surface point, and the unit is millimeters.

[0085] The area differential element is a tiny increase in the area of ​​the initial lesion contour surface.

[0086] The weighted eccentric metabolic shell difference of the ductal spinous process is the area integral of the product of the directional extension weight and the local layer eccentricity difference on the initial lesion contour, divided by the area integral of the directional extension weight on the initial lesion contour. It represents the overall degree of eccentricity of the sodium and phosphorus nuclei abnormal layers in the direction of the ductal spinous process of the breast lesion, and the unit is millimeters.

[0087] In detail, targeting the two-shell structure of breast lesions, normal sampling intervals are established towards the inner and outer sides of the lesion. The normal centroid of the abnormal sodium and phosphorus distribution field, i.e., the center position of the abnormal layer, is solved by integral calculation along the outer normal of the initial lesion contour. This can accurately reflect the overall spatial distribution of the abnormal layer rather than local single-point features. Then, the difference between the two is used to construct the local layer eccentricity difference, directly quantifying the multi-nucleus layer eccentricity of a single surface point. Finally, using the directional extension weight as the weighting coefficient, the global single ductal spinous process weighted eccentricity metabolic shell difference feature is obtained by surface integral calculation. This amplifies the eccentricity feature in the direction of the ductal spinous process, making the feature results more consistent with the structural characteristics of breast lesions extending along the ductal spinous process. For example, the traditional method only compares the maximum value of sodium and phosphorus signals of a certain voxel of the lesion, which ignores the overall spatial offset of the abnormal layer and misjudges the local signal difference as layer eccentricity. However, this scheme can accurately capture the overall distribution characteristics of sodium nuclei on the outer layer and phosphorus nuclei on the inner layer by calculating the normal centroid. The weighted global feature can highlight the eccentricity difference in the direction of the ductal spinous process.

[0088] In detail, the maximum sampling distance to the inner side of the lesion is fixed at 4 mm, and the maximum sampling distance to the outer side of the lesion is fixed at 6 mm. The sampling step size along the outer normal is fixed at 0.5 mm. This step size can ensure sampling accuracy and avoid excessive computation. In the calculation of the center position of the sodium-phosphorus anomaly layer, if the integral value of the denominator, i.e., the anomaly field, is less than 1% of the maximum value of all sampled values ​​of the surface point, then the point is marked as an invalid point. The local layer eccentricity difference of the invalid point is replaced by the median of the valid points in its one-ring neighborhood. The discrete implementation of the surface integral is to divide the initial lesion contour into multiple triangular elements, and use the area of ​​the element as the weight. The calculated values ​​of all sampled points in the element are averaged and then used in the summation operation. The unit outer normal is calculated from the gradient of the signed distance field of the initial contour. After normalizing the gradient value, the unit outward normal vector can be obtained. The clinical effectiveness of the weighted eccentric metabolic shell difference of the ductal spinous process can be judged by setting a threshold based on the training set. In this single embodiment, a threshold of 0.5 mm is temporarily set. If the threshold is greater than this value, it is judged to be a significant multi-core layer eccentricity. The discrete implementation of the normal integral is to divide the sampling interval into several sampling points with a step size of 0.5 mm. The integral operation is rewritten as a weighted summation operation of the sampling points, with the weight being the sampling step size. For example, a surface point is sampled along the outward normal with a step size of 0.5 mm. The inner 4 mm contains 8 sampling points, and the outer 6 mm contains 12 sampling points, for a total of 20 sampling points participating in the integral calculation. With such standardized numerical settings and operation methods, the eccentricity feature is quantitatively calculated.

[0089] Preferably, the local layer eccentricity difference and the directional extension weight are combined into a boundary correction amount, and the three-dimensional surface of the target lesion is reconstructed based on the boundary correction amount, including: The normalized directional extension weights are calculated using the following formula: ; The eccentricity difference of the truncated layer is calculated using the following formula: ; The corrected range field is calculated using the following formula: ; The set of coordinate points with zero corrected distance field values ​​is extracted using the following formula to obtain the three-dimensional surface of the target lesion: ; in, This represents a surface point on the initial lesion contour. This indicates the directional extension weight. Indicates the weights extended in the normalization direction; This indicates the local stratum eccentricity difference. Indicates the maximum sampling distance towards the inside of the lesion. Indicates the maximum sampling distance to the outside of the lesion. This represents the truncation function. This indicates the difference in eccentricity of the truncated stratigraphic level; Represents spatial coordinates, This represents the signed distance field of the initial contour. Indicates the boundary correction gain. This represents the projection of spatial coordinates onto the nearest point of the initial lesion contour. Indicates the corrected range field. This represents the three-dimensional surface of the target lesion.

[0090] The normalized directional extension weight is a value obtained after standardization of the directional extension weight, which realizes the normalized control of the duct spinous process directional weight and is dimensionless.

[0091] The truncated horizon eccentricity difference is a value obtained by restricting the local horizon eccentricity difference to a specified normal sampling interval, avoiding excessive eccentricity values ​​without observational basis from interfering with boundary reshaping. The unit is millimeters.

[0092] The nearest point mapping projection is the mapping result of each voxel coordinate in three-dimensional space to the nearest geometric point on the initial lesion contour, realizing the mapping of local surface quantities to voxel space.

[0093] Boundary correction gain is a coefficient used to adjust the amount of boundary correction, preferably 0.6. This value can balance the accuracy and robustness of boundary reshaping, avoiding both insufficient correction that fails to reflect the eccentricity of the slice and excessive correction that leads to distortion of the lesion contour.

[0094] The corrected distance field is the field distribution obtained after adjusting the initial signed distance field of the contour by the boundary correction amount. It represents the signed spatial distance from each voxel to the corrected lesion contour, and the unit is millimeters.

[0095] The three-dimensional surface of the target lesion is a set of three-dimensional voxel coordinate points with a corrected distance field value of zero, which represents the precise three-dimensional geometric boundary of the breast lesion after being reshaped by multinucleus metabolic eccentricity features.

[0096] In detail, the directional extension weights are first normalized to make the weighting of the ductal spinous process direction more controllable. Then, the local slice eccentricity difference is truncated to the actual normal sampling interval to eliminate eccentric values ​​without observational data support, avoiding invalid boundary reshaping based on phantom values. At the same time, the nearest point mapping projection is introduced to accurately map the local correction parameters on the initial lesion contour to the three-dimensional voxel space, achieving voxel-level local boundary correction. Finally, the product of the initial contour signed distance field minus the boundary correction gain, the normalized directional extension weights, and the truncated slice eccentricity difference is constructed. By modifying the distance field, the boundary reshaping is driven solely by the ductal spinous process direction characteristics and multi-nucleus layer eccentricity characteristics of the breast lesion, achieving direction-specific local boundary reshaping. For example, traditional global expansion causes the lesion to expand outward in all directions, failing to reflect the specific eccentricity of the ductal spinous process direction. However, this scheme allows surface points in the ductal spinous process direction to have a higher weight due to the normalized direction extension, resulting in a more significant boundary correction. Non-ductal spinous process directions are only slightly corrected, so that the three-dimensional surface of the reshaping target lesion completely conforms to the real structure and metabolic eccentricity characteristics of the breast lesion.

[0097] In detail, the boundary correction gain is fixed at 0.6. This value can be directly used in the single implementation, and it can also be used as an empirical constant for the training set when the sample size is sufficient. The nearest point mapping projection is achieved through the gradient of the initial contour signed distance field. The calculation method is to subtract the value of the initial contour signed distance field from the voxel coordinates and multiply it by the normalized gradient value. Since the gradient magnitude of the signed distance field near the contour is approximately 1, this calculation method can achieve stable nearest point mapping. The technical basis for truncating the local slice eccentricity difference to the negative value of the maximum sampling distance to the inside of the lesion and the maximum sampling distance to the outside of the lesion is that this interval is the effective range of actual normal sampling. Slice eccentricity outside this range has no actual scanning. Data support is provided, but boundary reshaping based on these numerical values ​​lacks pathological basis. After extracting the zero isosurface from the corrected distance field to obtain the three-dimensional surface of the target lesion, a mesh repair operation needs to be performed. First, isolated connected domains with a volume of less than 5 cubic millimeters are deleted, then the main connected domains are smoothed twice using Laplacian smoothing, and finally, volume compensation is performed to ensure that the total volume difference of the lesion before and after smoothing does not exceed 2%. The geometric consistency verification standard between the three-dimensional surface of the target lesion and the initial lesion contour is that the matching degree between the boundary offset in the direction of the ductal spinous process and the local layer eccentricity difference is not less than 90%, and the volume change rate in the direction of the non-ductal spinous process does not exceed 5%. With such standardized numerical settings and operation steps, lesion boundary reshaping and three-dimensional surface reconstruction are achieved.

[0098] Preferably, the local tomographic eccentricity difference is fused onto the three-dimensional surface of the target lesion to output a medical imaging model, including: The local holographic activity scalar field is calculated using the following formula: ; The medical imaging model is constructed using the following formula: ; in, Represents spatial coordinates, This indicates the anomalous distribution field of sodium. This indicates the 95th percentile of sodium abnormalities. The weighting of sodium anomalies is indicated; This indicates the abnormal phosphorus distribution field. This indicates the 95th percentile of the phosphorus anomaly. The weighting of phosphorus anomalies is indicated; This indicates the eccentricity difference of the truncated layer. This represents the projection of spatial coordinates onto the nearest point on the three-dimensional surface of the target lesion. This represents the function for finding the maximum value. This represents the 95th percentile of the positive eccentricity difference. The weights are displayed to indicate the eccentricity difference. Represents a local holographic activity scalar field; This represents the three-dimensional surface of the target lesion. This represents the local holographic activity scalar field mapped onto the three-dimensional surface of the target lesion. This refers to the medical imaging model.

[0099] The positive truncation eccentricity difference is the value obtained by taking the maximum value of the truncation eccentricity difference and zero, and only retains the outward eccentricity information of the sodium nucleus anomaly layer relative to the phosphorus nucleus anomaly layer. The unit is millimeters.

[0100] The 95th percentile of sodium anomaly is the value at the 95th percentile after statistically sorting the voxel values ​​of the sodium anomaly distribution field within a specified statistical range. It is used for normalization processing of the sodium anomaly distribution field.

[0101] The sodium anomaly display weight is a weight parameter used to allocate the proportion of sodium anomaly distribution field's display contribution in the local holographic activity scalar field. It is preferably 0.35. This value allows the sodium nucleus anomaly signal, phosphorus nucleus anomaly signal, and layer eccentricity difference to contribute evenly to the holographic activity, without biasing towards a single information dimension.

[0102] The first display component is a value obtained by normalizing the sodium anomaly distribution field and multiplying it by the sodium anomaly display weight. It is a component of the local holographic activity scalar field and is dimensionless.

[0103] The 95th percentile of phosphorus anomalies is the value at the 95th percentile after statistically sorting the voxel values ​​of a phosphorus anomaly distribution field within a specified statistical range. It is used for normalization processing of the phosphorus anomaly distribution field.

[0104] The phosphorus anomaly display weight is a weight parameter used to allocate the proportion of phosphorus anomaly distribution field in the local holographic activity scalar field. It is preferably 0.35. This value matches the sodium anomaly display weight, which can make the proportion of multinucleus metabolic abnormality signals in the holographic activity balanced, while preserving a reasonable proportion for the layer eccentricity difference.

[0105] The second display component is a value obtained by normalizing the phosphorus anomaly distribution field and multiplying it by the phosphorus anomaly display weight. It is a component of the local holographic activity scalar field and is dimensionless.

[0106] The nearest point projection of the target surface is the mapping result of the coordinates of each voxel in three-dimensional space to the nearest geometric point on the three-dimensional surface of the target lesion, realizing the accurate mapping of the layer eccentricity information to the three-dimensional surface of the target lesion.

[0107] The 95th percentile of the positive eccentricity difference is the value at the 95th percentile after statistically sorting the positive truncation eccentricity difference within a specified statistical range. It is used for normalization of the positive truncation eccentricity difference.

[0108] The eccentricity difference display weight is a weight parameter used to allocate the proportion of positively truncated eccentricity difference in the display contribution of the local holographic activity scalar field. It is preferably 0.30. This value, together with the sodium and phosphorus anomaly display weight, is 1. It can highlight the structural information of the stratum eccentricity while ensuring that the metabolic signal is dominant, and realize the fusion of metabolic and structural information.

[0109] The third display component is the value obtained by normalizing the positively truncated eccentricity difference and multiplying it by the eccentricity difference display weight. It is a component of the local holographic activity scalar field and is dimensionless.

[0110] The local holographic activity scalar field is a scalar field obtained by summing the first, second, and third display components. It integrates the triple information of sodium nucleus anomaly, phosphorus nucleus anomaly, and stratigraphic eccentricity, and is dimensionless.

[0111] Medical imaging models are fusion models constructed by mapping a local holographic activity scalar field onto the three-dimensional surface of the target lesion, and include the precise geometric boundaries of the lesion and the spatial distribution of holographic information.

[0112] In detail, the metabolic abnormality signals of sodium and phosphorus nuclei are fused together, and the positive truncation eccentricity difference, which characterizes the lesion's structural features, is incorporated into the fusion system to construct a local holographic activity scalar field containing triple information. The three types of information are normalized using their respective 95th percentiles to ensure a balanced contribution of information of different dimensions during fusion. Simultaneously, precise spatial matching between the eccentricity information and the three-dimensional surface of the target lesion is achieved through the nearest point projection of the target surface, avoiding spatial misalignment of information mapping. The final output medical image model is not a simple color overlay of a three-dimensional image, but rather a holographic activity scalar field combined with a reconstructed... The model, which integrates precise geometric boundaries, can not only intuitively display the spatial distribution of metabolic activity, but also reflect the structural differences caused by tonal eccentricity. For example, the traditional method only uses red and blue to represent the signal intensity of the sodium nucleus and phosphorus nucleus, respectively, which cannot reflect the eccentric structure of the sodium nucleus being on the outer layer and the phosphorus nucleus being on the inner layer. However, in the medical imaging model of this solution, the ductal spinous process direction has a larger eccentricity difference, a higher holographic activity value, and a more significant gradient change in color. This allows doctors to simultaneously observe the geometric boundaries, metabolic intensity, and tonal eccentricity characteristics of the lesion, providing more comprehensive information for clinical diagnosis.

[0113] In detail, the statistical ranges for the 95th percentile of sodium abnormality, the 95th percentile of phosphorus abnormality, and the 95th percentile of positive eccentricity are all voxels within the corrected lesion, excluding normal tissue voxels outside the lesion; the display weights for sodium abnormality and phosphorus abnormality are fixed at 0.35, and the display weight for eccentricity is fixed at 0.30, with a sum of 1, requiring no additional adjustment; the positive truncation eccentricity is mapped from the initial lesion contour to the target lesion's 3D surface by finding the nearest point on the target lesion's 3D surface on the initial lesion contour and directly inheriting the positive truncation eccentricity value of that nearest point; the mapping method from the local holographic activity scalar field to the target lesion's 3D surface uses vertex interpolation, interpolating the scalar field values ​​in voxel space to each vertex of the target lesion's 3D surface; the output format of the medical imaging model is with vertex labels. The PLY or OBJ mesh of the quantity field contains 3D coordinates, normal, local holographic activity value, and pseudo-color RGB value for each vertex. The pseudo-color mapping rule is that the lower the local holographic activity value, the bluer the color, and the higher the value, the redder the color. Linear color gradient mapping is used, and the value range is 0 to 1. The clinical display method of the medical image model supports 3D rotation, zoom in, zoom out, and local sectioning. It can display geometric boundaries separately or overlay holographic activity pseudo-color. It can also highlight different activity value ranges with color. For example, after statistical analysis, the voxel values ​​of the sodium abnormality distribution field in the corrected lesion of a certain case are 95% of the positions with a value of 8 mmol / L. This value is used as the 95th percentile of sodium abnormality for normalization. With such standardized numerical settings and operation steps, holographic activity construction and medical image model output are realized.

[0114] like Figure 2As shown, Figure 2 The process includes: the patient lies supine on an MRI scanner, and hydrogen, sodium, and phosphorus nucleus signals are simultaneously acquired through a multi-nucleus breast coil. After processing, the signals are uniformly mapped to the hydrogen nucleus space. Within this space, multi-nucleus metabolic information is combined to generate an initial lesion outline. Subsequently, the system extracts the abnormal distribution intensity of sodium nuclei and the abnormal turnover intensity of phosphorus nuclei membranes in the sodium and phosphorus abnormal fields, respectively, and solves the local layer eccentricity difference of the two along the lesion normal. Finally, the geometric boundary, metabolic intensity, and layer eccentricity triple features are fused to construct a three-dimensional holographic image model containing local eccentric shell information, presenting the precise geometric boundary and deep pathological features of the breast lesion.

[0115] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A multimodal magnetic resonance imaging system for breast lesions based on multi-nuclear atomic information fusion, characterized in that, include: A quantitative acquisition module is used to simultaneously acquire hydrogen nucleus signals, sodium nucleus signals, and phosphorus nucleus signals, and convert the hydrogen nucleus signals, sodium nucleus signals, and phosphorus nucleus signals into a quantitative concentration field; An image registration module is used to perform resolution matching on the quantitative concentration field and map the matched quantitative concentration field to the hydrogen nucleus space system; An abnormal field construction module is used to extract a non-lesion reference background based on the hydrogen nucleus space system to generate an abnormal sodium distribution field and an abnormal phosphorus distribution field. The morphological analysis module is used to generate an initial lesion contour and extract the directional extension weights of the initial lesion contour; The eccentricity quantization module is used to calculate the abnormal layer center positions of the sodium abnormal distribution field and the phosphorus abnormal distribution field along the outer normal of the initial lesion contour, respectively, to obtain the local layer eccentricity difference, and to perform weighted calculation on the local layer eccentricity difference using the direction extension weight. The boundary reshaping module is used to combine the local layer eccentricity difference with the directional extension weight into a boundary correction amount, and reconstruct the three-dimensional surface of the target lesion based on the boundary correction amount. The image output module is used to fuse the local layer eccentricity difference on the three-dimensional surface of the target lesion and output a medical image model.

2. The multimodal magnetic resonance imaging system for breast lesions based on multinuclear atomic information fusion according to claim 1, characterized in that, Simultaneously acquiring hydrogen nucleus signals, sodium nucleus signals, and phosphorus nucleus signals, and converting the hydrogen nucleus signals, sodium nucleus signals, and phosphorus nucleus signals into a quantitative concentration field, including: The original hydrogen nucleus signal intensity was divided using the 95th percentile of the hydrogen nucleus intensity to obtain the normalized intensity field of the hydrogen nucleus. Multiply the original sodium nucleus signal amplitude by the sodium nucleus quantitative calibration coefficient and the sodium nucleus relaxation compensation factor, and divide by the sodium nucleus sensitivity correction factor to obtain the sodium nucleus quantitative concentration field; Multiply the peak area of ​​phosphate monoester by the phosphorus nucleus quantitative calibration coefficient and the phosphate monoester relaxation compensation factor, and divide by the phosphorus nucleus sensitivity correction factor to obtain the quantitative concentration field of phosphate monoester. The phosphate diester peak area is multiplied by the phosphorus nucleus quantitative calibration coefficient and the phosphate diester relaxation compensation factor, and then divided by the phosphorus nucleus sensitivity correction factor to obtain the phosphate diester quantitative concentration field. The quantitative concentration field is composed of the hydrogen nucleus normalized intensity field, the sodium nucleus quantitative concentration field, the phosphate monoester quantitative concentration field, and the phosphate diester quantitative concentration field.

3. The multimodal magnetic resonance imaging system for breast lesions based on multinuclear atomic information fusion according to claim 2, characterized in that, Performing resolution matching on the quantitative concentration field and mapping the matched quantitative concentration field to the hydrogen nucleus space system includes: Using the hydrogen nucleus normalized intensity field as the intensity guiding parameter, combined with the spatial distance parameter, the quantitative concentration field is subjected to exponential kernel filtering for denoising to obtain the denoised quantitative concentration field. Calculate the difference between the square of the target uniform standard deviation and the square of the effective standard deviation of the proto-nuclear species, and take the square root of the difference to obtain the standard deviation required for the additional convolution; The denoised quantitative concentration field is convolved using a Gaussian kernel with the standard deviation required for the additional convolution as a parameter. The denoised quantitative concentration field after the convolution operation is transformed by inverse deformation transformation to obtain the matched quantitative concentration field mapped to the hydrogen nucleus space system. The matched quantitative concentration field includes the matched sodium nucleus quantitative concentration field, the matched phosphate monoester quantitative concentration field, and the matched phosphate diester quantitative concentration field.

4. The multimodal magnetic resonance imaging system for breast lesions based on multinuclear atomic information fusion according to claim 3, characterized in that, Based on the aforementioned hydrogen nucleus space system, a non-lesion reference background is extracted to generate an abnormal sodium distribution field and an abnormal phosphorus distribution field, including: Extract the matched sodium nucleus quantitative concentration field, the matched phosphate monoester quantitative concentration field, and the matched phosphate diester quantitative concentration field from the matched quantitative concentration field; Subtract the sodium nucleus reference background median value from the matched sodium nucleus quantitative concentration field, and perform a maximum value acquisition operation with zero to obtain the sodium abnormal distribution field. Add a zero-preventing minima to the matched phosphate diester quantitative concentration field to obtain the adjusted phosphate diester quantitative concentration field. Divide the matched phosphate monoester quantitative concentration field by the adjusted phosphate diester quantitative concentration field to obtain the phosphorus nucleus concentration ratio field. Subtract the reference background median value of phosphorus nucleus ratio from the phosphorus nucleus concentration ratio field, and perform a maximum value acquisition operation with zero to obtain the phosphorus anomaly distribution field. Divide the sodium anomalous distribution field by the sodium anomalous scale factor and multiply by the sodium anomalous weight, and add the phosphorus anomalous distribution field divided by the phosphorus anomalous scale factor and multiplied by the phosphorus anomalous weight to obtain the unified metabolic driving field.

5. The multimodal magnetic resonance imaging system for breast lesions based on multinuclear atomic information fusion according to claim 4, characterized in that, Generate an initial lesion contour and extract the directional extension weights of the initial lesion contour, including: An initial lesion mask is generated based on the unified metabolic driving field, and a distance transformation operation is performed on the initial lesion mask to obtain an initial contour signed distance field. Extract the set of three-dimensional voxel coordinate points with a signed distance field value of zero from the initial contour to obtain the initial lesion contour; Obtain the coordinates of the surface neighborhood sampling points and the mean coordinates of the neighborhood points of the surface points on the initial lesion contour; The difference matrix is ​​obtained by subtracting the mean coordinates of the neighboring sampling points from the coordinates of the neighboring points. The difference matrix is ​​then multiplied by the transpose of the difference matrix and summed. Finally, it is divided by the number of neighboring sampling points to obtain the local covariance matrix. Extract the first covariance parameter and the second covariance parameter of the local covariance matrix; Divide the first covariance parameter by the sum of the second covariance parameter and the minimum positive number to prevent zero, and take the square root of the result of the division operation to obtain the directional extension weight.

6. The multimodal magnetic resonance imaging system for breast lesions based on multinuclear atomic information fusion according to claim 5, characterized in that, The abnormal layer center positions of the sodium abnormal distribution field and the phosphorus abnormal distribution field are calculated along the outer normal of the initial lesion contour, respectively, to obtain the local layer eccentricity difference. The local layer eccentricity difference is then weighted using the directional extension weight, including: Establish the normal sampling interval between the negative value of the maximum sampling distance to the inside of the lesion and the maximum sampling distance to the outside of the lesion; Within the normal sampling interval, the signed distance along the outer normal is multiplied and integrated with the sodium anomaly distribution field distributed along the outer normal, and the product integral is divided by the integral of the sodium anomaly distribution field distributed along the outer normal to obtain the center position of the sodium anomaly layer. Within the normal sampling interval, the signed distance along the outer normal is multiplied and integrated with the phosphorus anomaly distribution field distributed along the outer normal, and this product integral is divided by the integral of the phosphorus anomaly distribution field distributed along the outer normal to obtain the center position of the phosphorus anomaly layer. Subtracting the center position of the phosphorus anomaly layer from the center position of the sodium anomaly layer yields the local stratum eccentricity difference. Multiply the directional extension weight by the local layer eccentricity difference and perform area integration on the initial lesion contour. Divide this area integration by the area integration of the directional extension weight on the initial lesion contour to obtain the ductal spinous process weighted eccentric metabolic shell difference.

7. The multimodal magnetic resonance imaging system for breast lesions based on multinuclear atomic information fusion according to claim 6, characterized in that, The local layer eccentricity difference and the directional extension weight are combined into a boundary correction amount, and the three-dimensional surface of the target lesion is reconstructed based on the boundary correction amount, including: Subtract a constant from the directional extension weight and divide the result by the directional extension weight to obtain the normalized directional extension weight. The local layer eccentricity difference is limited to the negative value of the maximum sampling distance to the inside of the lesion and the maximum sampling distance to the outside of the lesion to obtain the cut-off layer eccentricity difference; Calculate the nearest point mapping projection from the coordinates of each voxel in three-dimensional space to the initial lesion contour; The corrected distance field is obtained by subtracting the boundary correction gain, the normalized direction extension weight mapped to the nearest point projection, and the truncated layer eccentricity difference mapped to the nearest point projection from the initial contour signed distance field. Extract the set of three-dimensional voxel coordinate points where the corrected distance field value is zero to obtain the three-dimensional surface of the target lesion.

8. The multimodal magnetic resonance imaging system for breast lesions based on multinuclear atomic information fusion according to claim 7, characterized in that, The local torsional eccentricity difference is fused onto the three-dimensional surface of the target lesion to output a medical imaging model, including: The positive truncation eccentricity difference is obtained by taking the maximum value of the truncation eccentricity difference and zero. Divide the sodium anomaly distribution field by the ninety-fifth percentile of sodium anomaly and multiply by the sodium anomaly display weight to obtain the first display component; Divide the phosphorus anomaly distribution field by the ninety-fifth percentile of phosphorus anomaly and multiply it by the phosphorus anomaly display weight to obtain the second display component. Calculate the projection of each voxel coordinate in the three-dimensional space onto the nearest point on the target surface of the target lesion; The positive truncation eccentricity difference mapped to the projection of the nearest point on the target surface is divided by the ninety-fifth percentile of the positive eccentricity difference and multiplied by the eccentricity difference display weight to obtain the third display component; The first display component, the second display component, and the third display component are summed to obtain a local holographic activity scalar field; The local holographic activity scalar field is mapped onto the three-dimensional surface of the target lesion to jointly construct the medical imaging model.

Images