A breast lesion region identification system and method based on big data

Abstract

The application relates to the field of big data management and discloses a breast lesion region identification system and method based on big data, which is used for converting a medical image analysis problem into a biophysical process simulation and feature comparison problem.The breast lesion region identification method based on big data comprises the following steps: processing breast DCE-MRI time series data, generating driving source data, extracting image data at a specific time point as a steady-state driving source of a reaction-diffusion system, generating a morphology field representing tissue biochemical characteristics through numerical solution, and constructing a standard morphology field cloud library by using large-scale normal breast data.The application simulates a biophysical process by using a reaction-diffusion theory, avoids data dependency and a black box problem of a machine learning model, and improves the interpretability and accuracy of lesion identification.

Description

Technical Field

[0001] This invention relates to the field of big data management, and in particular to a system and method for identifying breast lesion areas based on big data. Background Technology

[0002] Breast cancer is one of the most common and deadliest malignant tumors among women worldwide. Early and accurate identification is crucial for improving patient survival and quality of life. Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI), due to its high soft tissue resolution and ability to reflect hemodynamic characteristics, has become an important imaging tool for the detection and diagnosis of breast lesions. However, existing breast lesion identification methods based on big data analysis have the following shortcomings in both theoretical foundation and practical application: The decision-making process of deep learning models lacks transparency and interpretability, making it difficult for doctors to understand why the model makes a specific diagnosis, thus reducing clinical credibility and acceptance.

[0003] Existing methods often learn feature representations that are statistically correlated rather than causally correlated. They cannot effectively capture the biophysical processes that lead to disease formation.

[0004] In cross-center validation with different devices, acquisition parameters, or population distributions, machine learning models often exhibit performance degradation, indicating that the learned feature representations may be overly dependent on the specific distribution of the training data.

[0005] Therefore, we propose a big data-based breast lesion area identification system and method to solve the above problems. Summary of the Invention

[0006] This invention provides a breast lesion region identification system and method based on big data, which transforms the medical image analysis problem into a biophysical process simulation and feature comparison problem, providing a research paradigm for breast lesion identification.

[0007] The first aspect of this invention provides a method for identifying breast lesion regions based on big data. This method includes: performing spatiotemporal registration and intensity normalization processing on breast magnetic resonance imaging data to generate driving source data; extracting image data from the driving source data to obtain a steady-state driving source for a reaction-diffusion system; inputting the steady-state driving source into the reaction-diffusion system for numerical solution, and outputting a spatial distribution map, denoted as a morphological field, based on the intensity distribution evolution of the steady-state driving source; processing breast magnetic resonance imaging data of normal breast tissue sequentially through the driving source data, the steady-state driving source, and the morphological field to form a standard morphological field cloud library; obtaining the morphological field to be tested based on the data of the breast region to be tested, calculating the difference between the morphological field to be tested and the standard morphological field cloud library to obtain a morphological distortion value, comparing the morphological distortion value with a preset value, and identifying lesion regions if the morphological distortion value exceeds the preset value.

[0008] Optionally, in a first implementation of the first aspect of the present invention, the method includes: processing breast magnetic resonance imaging data to eliminate positional bias and generate time-series image data; extracting the enhancement dynamics curves of pixels in the time-series image data, calculating the maximum enhancement intensity and minimum baseline intensity of the curves, performing normalization calculation based on the difference between the maximum and minimum, and generating normalized intensity data; and selecting image data from the normalized intensity data as steady-state driving source data for the reaction-diffusion system.

[0009] Optionally, in a second implementation of the first aspect of the present invention, the method includes: analyzing the temporal intensity curves of pixels in the driving source data, calculating the maximum rising slope and peak time of the curves, and generating a set of temporal dynamic feature parameters; based on the set of temporal dynamic feature parameters, analyzing and identifying characteristic time points, and determining an optimal time selection scheme; and according to the optimal time selection scheme, extracting image data from the driving source data as the steady-state driving source input of the reaction-diffusion system.

[0010] Optionally, in a third implementation of the first aspect of the present invention, the method includes: establishing a computational grid matching the steady-state driving source based on the dynamic parameters of the reaction-diffusion system, including the activator diffusion coefficient, the inhibitor diffusion coefficient, and the reaction rate constant; establishing an activator field and an inhibitor field on the computational grid, setting the activator field as the intensity distribution of the steady-state driving source and the inhibitor field as a uniform distribution; discretizing the reaction-diffusion equations, solving the diffusion term and the reaction term within a discrete time step to update the activator field and the inhibitor field; calculating the numerical residual of the activator field after each time step, and determining that a steady state has been reached when the numerical residual is less than a preset tolerance; and extracting the spatial distribution of the activator field in the steady state to generate a morphological field.

[0011] Optionally, in the fourth implementation of the first aspect of the present invention, the method includes: inputting breast magnetic resonance imaging data of normal breast tissue into driving source data for processing to generate a normal tissue driving source dataset; inputting the normal tissue driving source dataset into a steady-state driving source for processing to generate a normal tissue steady-state driving source set; inputting the normal tissue steady-state driving source set into a morphological field for processing to generate a normal morphological field set; performing feature extraction and encoding on the normal morphological field set to generate a morphological field feature descriptor set; storing the morphological field feature descriptor set and metadata, establishing an index structure, and constructing a standard morphological field cloud library.

[0012] Optionally, in the fifth implementation of the first aspect of the present invention, the method includes: inputting breast magnetic resonance imaging data of the breast region to be tested into driving source data for processing to generate driving source data to be tested; inputting the driving source data to be tested into a steady-state driving source for processing to generate a steady-state driving source to be tested; inputting the steady-state driving source into a morphological field for processing to generate a morphological field to be tested, and performing feature extraction and encoding on the morphological field to be tested to generate a feature descriptor of the morphological field to be tested; based on the feature descriptor of the morphological field to be tested, searching in a standard morphological field cloud library, calculating the difference between the feature descriptor of the morphological field to be tested and the feature descriptor of the normal morphological field, and generating a difference set; analyzing the difference set, calculating the average difference as a morphological distortion value, and comparing the morphological distortion value with a preset threshold to output a conclusion.

[0013] Optionally, in a sixth implementation of the first aspect of the present invention, the difference between the feature descriptors of the morphological field to be tested and the normal morphological field is calculated using Euclidean distance. As a measure of difference: in, and These are the values ​​of the test sample and a normal sample on the i-th feature dimension, respectively, where i is the index of the feature dimension. This represents the total dimension of the feature descriptor.

[0014] Optionally, in the seventh implementation of the first aspect of the present invention, the method further includes: selecting typical case data from historical clinical data, including DCE-MRI data of normal tissue and diseased tissue, to construct a validation test set; processing all data in the validation test set sequentially through driving source data, steady-state driving source, and morphological field to generate a validation morphological field set; calculating the difference between the morphological field in the validation morphological field set and the standard morphological field cloud library to obtain a validation distortion dataset; based on the validation distortion dataset and pathological diagnosis results, analyzing the morphological distortion threshold through receiver operating characteristic curves to generate a diagnostic threshold parameter; and using the diagnostic threshold parameter to test the validation test set to generate a system performance evaluation report.

[0015] The second aspect of this invention provides a breast lesion region identification system based on big data. The system includes: a processing module that performs spatiotemporal registration and intensity normalization processing on breast magnetic resonance imaging data to generate driving source data; an extraction module that extracts image data from the driving source data to obtain a steady-state driving source of a reaction-diffusion system; a calculation module that inputs the steady-state driving source into the reaction-diffusion system for numerical solution, and the system outputs a spatial distribution map, denoted as a morphological field, based on the intensity distribution evolution of the steady-state driving source; a construction module that processes breast magnetic resonance imaging data of normal breast tissue sequentially through the driving source data, the steady-state driving source, and the morphological field to form a standard morphological field cloud library; and an identification module that obtains the morphological field to be tested based on the data of the breast region to be tested, calculates the difference between the morphological field to be tested and the standard morphological field cloud library to obtain a morphological distortion value, compares the morphological distortion value with a preset value, and identifies lesion regions if the morphological distortion value exceeds the preset value.

[0016] The mechanism of this invention is as follows: a physical model is established that can simulate the dynamic process of contrast agents in tissues, and the imaging data is transformed into the driving input of the model. By utilizing the inherent physical properties of the model, microscopic and physiological differences are amplified and manifested into macroscopic and stable spatial pattern differences. These differences are quantified through pattern comparison to achieve objective and accurate lesion identification.

[0017] Beneficial effects: This invention simulates the underlying biophysical causal processes that cause image contrast. This fundamentally enhances the interpretability of the model. Doctors can understand the physical basis of the model's decisions, rather than facing an incomprehensible black box, thus increasing clinical trust and acceptance. This invention requires no training; it only needs to collect normal tissue data to construct a normal baseline cloud library, reducing the cost and difficulty of data preparation. Based on universal physical laws and normalization processing, its core model is insensitive to the aforementioned changes, thus exhibiting more stable performance when dealing with data from different medical institutions. The storage space required for the morphological field cloud library is much smaller than that of the original image database, and the recognition process is based on efficient feature comparison rather than complex network forward propagation, resulting in faster computation speed. As a nonlinear system, the response-diffusion system can amplify minute, early physiological changes that are imperceptible to the naked eye and traditional methods. These changes may not yet present obvious abnormalities in image grayscale, but they are sufficient to trigger pattern variations in the morphological field, thereby enabling earlier warning and diagnosis. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of an embodiment of a method for identifying breast lesion areas based on big data according to an embodiment of the present invention; Figure 2This is a schematic diagram of another embodiment of a method for identifying breast lesion areas based on big data according to an embodiment of the present invention; Figure 3 This is a schematic diagram of an embodiment of a breast lesion area identification system based on big data according to an embodiment of the present invention. Detailed Implementation

[0019] This invention provides a breast lesion region identification system and method based on big data, which transforms the medical image analysis problem into a biophysical process simulation and feature comparison problem, providing a research paradigm for breast lesion identification. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0020] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the method for identifying breast lesion areas based on big data in this invention includes: 101. Spatiotemporal registration and intensity normalization are performed on the raw time-series data of the input dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) of the breast to generate dimensionless driving source data.

[0021] It is understood that the executing entity of this invention can be a breast lesion area identification device based on big data, or it can be a terminal or a server; the specific implementation is not limited here. This embodiment of the invention will be described using a server as an example.

[0022] It should be noted that, in this embodiment, the raw time-series breast DCE-MRI data received by the server comes from a 3T MRI scanner, acquiring 3D image sequences at five time points (before contrast agent injection, and 1 minute, 3 minutes, 5 minutes, and 7 minutes after injection). The image resolution at each time point is 256×256×60 voxels, and the intensity value ranges from 0 to 4095 (corresponding to 12-bit depth). The following is a detailed description: The image at the first time point (before contrast agent injection) is chosen as the reference image because it typically involves minimal patient movement. Then, for each subsequent time point, an alignment algorithm based on mutual information is used. Specifically, the registration process optimizes the displacement field of each voxel to maximize the similarity between the target and reference images (using mutual information as the similarity metric). After registration, the image at each time point is spatially aligned with the reference image, eliminating spatial offsets caused by breathing, heartbeat, or slight patient movement. For example, a lesion area that was originally offset by 2 mm is correctly aligned to the reference position after registration.

[0023] After registration, the temporal images are intensity normalized to eliminate the influence of factors such as scanner gain and contrast agent concentration differences, generating dimensionless data. A global maximum intensity normalization method is used: the maximum intensity value in the entire registered temporal data is calculated (assumed to be 4095), and then the intensity value of each voxel at each time point is divided by this maximum value. In this way, all intensity values ​​are mapped to the [0,1] interval, forming dimensionless scalar values. The original intensity of a voxel at time point 2 is 2048, which becomes 0.5 after normalization. This processing ensures data consistency and facilitates comparison and analysis in subsequent steps.

[0024] Finally, the processed time-series data is output as the driving source data. This is a 4D array (spatial dimension 256×256×60, temporal dimension 5), where the value of each voxel-time point is a normalized dimensionless value (between 0 and 1). At coordinates (100, 100, 30), the time-series data may be [0.1, 0.3, 0.6, 0.5, 0.4], representing the enhancement mode at that location at different time points. This data is spatiotemporally consistent and dimensionless, and can be directly used for steady-state driving source extraction in step 102.

[0025] Through the above processing, the raw DCE-MRI data were transformed into standardized driving source data, providing reliable input for subsequent reaction-diffusion system simulations. The entire process was executed automatically on the server, taking approximately 10 minutes, ensuring efficiency and accuracy.

[0026] 102. Extract image data at specific time points from the driving source data and use it as a steady-state driving source for the reaction-diffusion system.

[0027] It should be noted that in this embodiment, the server has completed the processing in step 101 and obtained spatiotemporally registered and intensity-normalized driving source data. This data is a 4D array containing 5 time points (T0 before contrast agent injection, and T1 to T4 at 1 minute, 3 minutes, 5 minutes and 7 minutes after injection, respectively). The image size of each time point is 256×256×60 voxels, and all intensity values ​​are dimensionless scalars ranging from [0,1].

[0028] The specific process is as follows: Determining a specific time point: The reaction-diffusion system requires a stable input to simulate the process of reaching steady state. In DCE-MRI, at a certain time point after contrast agent injection, its concentration distribution in different tissues reaches a relatively stable state that reflects tissue perfusion and permeability characteristics. Through analysis of the kinetic curves of a large number of normal breast DCE-MRI data, it was found that the intensity distribution of the image data at 3 minutes after injection (T2 time point) best represents this relatively stable "input" state. Therefore, this embodiment selects the image data at T2 time point as the specific time point.

[0029] Steady-state driving source extraction: The server directly indexes and extracts all voxel intensity values ​​in the entire 3D space at time T2 from the driving source data (4D array). This operation generates a new 3D array with a size of 256×256×60. Each voxel value in this array is a dimensionless scalar representing the relative concentration of contrast agent at that spatial location at time T2.

[0030] For a specific data example, at spatial coordinates (120, 150, 25), the intensity value sequence at time points T0 to T4 might be [0.12, 0.45, 0.82, 0.78, 0.70]. The value extracted at time point T2 is 0.82. This 0.82 is a driving source intensity value. The entire extracted 3D image data constitutes the steady-state driving source of the reaction-diffusion system.

[0031] The extracted 3D array representing the intensity distribution at time T2 is output and passed to the next processing step (step 103). This data serves as the input (i.e., the driving source) to the reaction-diffusion system, and its stable spatial distribution will be used to drive the system's evolution over the simulation time to generate a morphological field characterizing tissue properties.

[0032] This step involves only indexing and extracting the data, without performing any numerical calculations or transformations. This ensures that the original physical meaning of the driving source data is preserved and provides accurate input for subsequent physical simulations. The entire process is completed instantaneously on the server.

[0033] 103. Input the steady-state driving source into the reaction-diffusion system with preset parameters for numerical solution. The system evolves according to the intensity distribution of the steady-state driving source during the simulation time, and finally outputs a stable spatial distribution map that characterizes the dynamic characteristics of the biochemical process inside the tissue, which is denoted as the morphological field.

[0034] It should be noted that in this embodiment, the server has extracted 3D image data at a specific time point (T2, 3 minutes after injection) from the driving source data as a steady-state driving source. This driving source is a 256×256×60 array, where each voxel value is a dimensionless intensity ranging from [0,1]. It now needs to be input into the reaction-diffusion system for evolution.

[0035] The specific process is as follows: This system uses a variant of the FitzHugh-Nagumo model to simulate biochemical reactions and diffusion processes within tissues. The system includes two key variables: the concentrations of the activator (u) and the inhibitor (v). All parameters have been dimensionless.

[0036] Preset parameters: the diffusion coefficient Du of the activator is 0.08, the diffusion coefficient Dv of the inhibitor is 0.04, and the reaction rate parameters a and b are 0.75 and 0.06 respectively. These parameter values ​​were determined after simulation and debugging on a large amount of normal tissue data to ensure stable convergence of the system. The simulation time step Δt is set to 0.02 dimensionless time units, and the total simulation time is 500 time steps (i.e., 10 dimensionless time units).

[0037] The intensity value of the steady-state driving source is used as the initial spatial distribution of the activator variable u. The initial value of the inhibitor variable v is uniformly set to 0.05 throughout the space. Subsequently, at each time step, the system is updated according to the reaction-diffusion partial differential equations. The explicit Euler method is used for numerical solution on a three-dimensional mesh, and the next state of each voxel is calculated from its current state and the states of its six adjacent voxels.

[0038] At spatial coordinates (120, 150, 25), the initial intensity of the driving source is u = 0.82. After 50 time steps, the value of u may oscillate to 0.25, and the value of v becomes 0.12. After 300 time steps, the amplitude of change of u and v decreases sharply. At 500 time steps, the system reaches a steady state, and the value of u at this point finally stabilizes at 0.18, and the value of v stabilizes at 0.15.

[0039] When the system reaches stability across the entire simulation domain (the rate of change of u and v values ​​on all voxels is less than 1), After that, the calculation is terminated. The final stable activator concentration distribution u(x,y,z) is output, which is the desired morphological field. This morphological field is a 3D array (256×256×60) with the same size as the input driving source, and its value represents the stable spatial pattern that is finally formed by the initial driving source under specific biochemical kinetic parameters and reflects the internal characteristics of the tissue.

[0040] 104. The DCE-MRI data from large-scale normal breast tissue are processed sequentially through the driver source data generation step, the steady-state driver source extraction step, and the morphological field generation step. All the generated normal morphological fields are centrally stored and managed to construct a standard morphological field cloud library.

[0041] It should be noted that this embodiment describes how the server constructs a benchmark database for comparison—the standard morphological field cloud library.

[0042] The server reads raw time-series data from a dataset of 1000 pathologically confirmed normal breast DCE-MRI cases. Each case contains 3D images (256×256×60 pixels) at 5 time points. This data is then fed in batches into an automated processing pipeline. This pipeline executes the following sequentially: For each data point, spatiotemporal registration and intensity normalization are performed to generate dimensionless driving source data. In one case, the original intensity sequence of a voxel [3500, 8000, 12000, 11000, 9500] is transformed into [0.29, 0.67, 1.00, 0.92, 0.79]. Extract the 3D image at time point 3 (T2) from the processed drive source data of each case as the steady-state drive source; Numerical solutions were performed on the reaction-diffusion system with preset parameters (Du=0.08, Dv=0.04) for each case, using a steady-state driving source. After 500 time steps of evolution, the system outputs a stable morphological field. This morphological field is a 256×256×60 floating-point array, where each value represents a stable activator concentration (a dimensionless scalar).

[0043] For each normal case morphological field generated, the server stores it as an independent data object in a dedicated cloud storage system. Each data object comes with key metadata, which is recorded in a management table of a relational database. Its main fields include: Morphological Field ID: A unique identifier; Corresponding Case ID: Used to trace the original data; Data Storage Path: Points to the location of the morphological field binary file in cloud storage; Statistical Information: Some basic statistics calculated for this morphological field, used for subsequent quick retrieval and preliminary screening.

[0044] Once all 1000 normal data points have been processed, a standard morphological field cloud library containing 1000 normal morphological fields has been constructed. Logically, this cloud library is a collection; physically, it consists of a database (managing metadata) and a cloud storage system (storing entity data). All morphological field data has been validated to ensure its numerical validity and stability.

[0045] 105. The DCE-MRI data of the breast region to be tested is processed sequentially through the driver source data generation step, the steady-state driver source extraction step, and the morphological field generation step to obtain the morphological field to be tested; the average difference between the morphological field to be tested and all normal morphological fields in the standard morphological field cloud library is calculated to obtain the morphological distortion value; the morphological distortion value is compared with a preset threshold. If it exceeds the threshold, the breast region to be tested is determined to be a lesion region.

[0046] It should be noted that this embodiment describes how the server uses the constructed standard morphological field cloud library to identify lesions in the breast region to be tested.

[0047] The server receives raw time-series data from a breast DCE-MRI scan for diagnosis. This data first undergoes the same processing pipeline as the cloud database: spatiotemporal registration and intensity normalization are performed to generate dimensionless driving source data; the image at time 3 (T2) is extracted as the steady-state driving source; using the same preset parameters as the cloud database case (Du=0.08, Dv=0.04, etc.), the steady-state driving source is input into the reaction-diffusion system for evolution. After 500 time steps of simulation, the system reaches a steady state and finally outputs a 256×256×60 morphological field to be tested. The value of each voxel in this morphological field is a dimensionless scalar.

[0048] The server reads all 1000 normal morphological fields one by one from the standard morphological field cloud library. For each normal morphological field in the cloud library, the root mean square error (RMSE) of all corresponding voxel values ​​between it and the morphological field to be tested is calculated. The error is calculated as follows: calculate the sum of squares of the differences between all corresponding voxel values ​​in the two morphological fields; divide the sum of squares by the total number of voxels (256*256*60); and take the square root of the result.

[0049] This RMSE value is a single difference measure, a dimensionless scalar that quantifies the overall deviation between the tested morphological field and a given normal morphological field. Next, the server calculates the arithmetic mean of the 1000 RMSE values ​​between the tested morphological field and all 1000 normal morphological fields in the cloud database. This final average is the morphological distortion value (D). The calculated D value can be 0.15.

[0050] The server compares the calculated morphological distortion value (D=0.15) with a pre-determined threshold (0.12) determined through extensive statistical analysis. If the morphological distortion value is greater than the threshold (i.e., 0.15>0.12), the breast region being tested is determined to be a lesion. Conversely, if the distortion value is lower than or equal to the threshold, it is determined to be a normal region.

[0051] Finally, the server will output a report containing the morphological distortion value and a judgment of "suspected lesion" or "no abnormality found" for the doctor's reference.

[0052] In this embodiment of the invention, objective and quantitative identification of breast lesions is achieved by systematically comparing the data to be tested with a standard morphological field cloud library. The morphological field generated using a reaction-diffusion model is used to deeply mine the dynamic information of tissue biochemical characteristics contained in DCE-MRI data. By calculating the average difference with massive amounts of normal data, individual differences are effectively eliminated, ensuring that the judgment is based on big data statistical patterns rather than subjective experience. This significantly improves the reliability and repeatability of the results, contributing to the accurate detection of early lesions and greatly improving diagnostic consistency.

[0053] Please see Figure 2 Another embodiment of the method for identifying breast lesion areas based on big data in this invention includes: 201. Spatiotemporal registration and intensity normalization are performed on the raw temporal data of the input dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) of the breast to generate dimensionless driving source data. Specifically, rigid registration is performed on the original DCE-MRI time-series data to eliminate spatial positional deviations caused by patient movement, generating registered time-series image data. The enhancement dynamics curve of each pixel in the registered time-series image data is extracted, and the maximum enhancement intensity and minimum baseline intensity of each curve are calculated. Normalization is performed based on the difference between the maximum enhancement intensity and the minimum baseline intensity of each pixel to generate normalized intensity data. Image data at specific time points that can reflect tissue perfusion characteristics during contrast agent perfusion are selected from the normalized intensity data as steady-state driving source data for the reaction-diffusion system.

[0054] It should be noted that in this embodiment, the raw time-series data of a patient's breast DCE-MRI is input. This data contains 5 time points (t=0, 1, 2, 3, 4 minutes), with an image spatial resolution of 256×256 pixels and an intensity value of arbitrary units (au), ranging from approximately 0-1000 a.u. The goal is to generate dimensionless driving source data.

[0055] First, rigid registration was performed on the original time-series data to eliminate spatial positional deviations caused by slight patient movements. A mutual information-based registration algorithm was used to align the images at each time point (t=1,2,3,4) with the baseline time point (t=0). After registration, the spatial coordinates of the images at all time points were consistent. Before registration, a certain vascular structure in the image at time point t=2 was offset by 2 pixels; after registration, this structure was aligned with the baseline. The registered time-series image data was stored as a 256×256×5 matrix, with the intensity values ​​remaining unchanged.

[0056] For each pixel in the registered data, extract the temporal intensity curve. Taking pixel (100, 100) as an example, its intensity value changes over time as follows: t0 = 100 a.u., t1 = 120 a.u., t2 = 150 a.u., t3 = 140 a.u., t4 = 130 a.u. Calculate the minimum baseline intensity: take the average of the first two time points (t0 and t1), i.e. au represents the baseline before perfusion. Calculate the maximum enhancement intensity: take the maximum value from the curve, i.e., 150 a.u.att². Then, perform normalization calculations: for the intensity value at each time point, apply the formula: This formula is dimensionless, and the output value is between 0 and 1. For pixel (100, 100), the difference (maximum enhancement intensity - minimum baseline intensity) = 40 a.u., therefore the normalized data is t0: (But it is usually truncated to 0), t1: t2: t3: t4: Perform this operation on all pixels to generate a normalized intensity data matrix (256×256×5) with a value range of [0,1].

[0057] From the normalized data, time points reflecting tissue perfusion characteristics during contrast agent perfusion were selected. Based on typical perfusion kinetics, image data at the peak time (t=2 minutes) were chosen because the enhancement intensity is greatest at this time point, clearly distinguishing the high-perfusion area (tumor) from normal tissue. Therefore, the full-frame normalized image (256×256 pixels) at t=2 minutes was extracted as the steady-state driving source data for the reaction-diffusion system. This data is dimensionless, with values ​​between 0 and 1, representing relative enhancement intensity.

[0058] Through the above steps, spatiotemporal registration and intensity normalization were completed, generating driving source data that can be used for subsequent processing.

[0059] 202. Extract image data at specific time points from the driving source data and use it as the steady-state driving source for the reaction-diffusion system; Specifically, the temporal intensity curve of each pixel in the driving source data is analyzed, the maximum rising slope and peak time of each curve are calculated, and a set of temporal dynamic characteristic parameters is generated. Based on the set of temporal dynamic characteristic parameters, the characteristic time points that best distinguish the perfusion characteristics of different tissues are identified by cluster analysis, and the optimal time selection scheme is determined. According to the optimal time selection scheme, the full-frame image data of the corresponding time points are extracted from the driving source data as the steady-state driving source input of the reaction-diffusion system.

[0060] It should be noted that, in this embodiment, the input is the normalized driving source data generated in step 201. This data is a 256×256 pixel three-dimensional matrix containing 5 time points (t=0,1,2,3,4 minutes), and the intensity value of each pixel is a dimensionless value ranging from 0 to 1.

[0061] The temporal intensity curve of each pixel in the driving source data is analyzed to calculate two key dynamic parameters: maximum rise slope and peak time.

[0062] Maximum slope calculation: The slope is calculated using the central difference method, the formula is as follows: Where I(t) is the intensity at time point t, and Δt is the time interval (1 minute in this example). The maximum rising slope is the maximum value among the slopes calculated for all consecutive time points. For example, if the intensity sequence of a pixel is [0.1, 0.4, 0.9, 0.8, 0.7], its slope sequence is calculated as [-, (0.4-0.1) / 1=0.3, (0.9-0.1) / 2=0.4, (0.8-0.4) / 2=0.2, (0.7-0.9) / 2=-0.1]. The maximum rising slope is 0.4 min. -1 (Dimensionless).

[0063] Peak time calculation refers to the time point at which the intensity curve reaches its maximum value. For the same pixel curve mentioned above, the maximum intensity of 0.9 occurs at t=2 minutes, therefore its peak time is 2 minutes. Perform the above calculation on all 256×256 pixels to generate a dataset containing 512 feature parameters (each pixel contains two parameters: slope and time).

[0064] Based on the feature parameter set of all pixels (maximum rising slope and peak time), the K-means clustering algorithm (preset K=2) is used to divide the pixels into two main clusters to distinguish between rapidly perfused tissue (potentially lesions) and normal / slowly perfused tissue.

[0065] The clustering results assume that the central feature of cluster C1 is (slope = 0.05min). -1 The peak time was 3.5 min, representing slowly enhancing normal tissue; the central characteristic of cluster C2 was (slope = 0.40 min).-1 (Peak time = 1.8 min), representing rapidly enhancing suspected lesion tissue.

[0066] Analysis of the typical time-intensity curves of the two clusters revealed that time point t=2 minutes (i.e., the third time point) is the critical moment for distinguishing the two types of tissue. At this time point, the average intensity of cluster C2 reaches its peak of approximately 0.95, while the average intensity of cluster C1 is only 0.35, showing the most significant difference. Therefore, the optimal time selection scheme is determined to be t=2 minutes.

[0067] Based on the optimal time selection scheme (t=2 minutes), full-frame two-dimensional image data at the corresponding time points are extracted from the original driving source data (256×256×5). This image is a 256×256 matrix, where each pixel value represents its normalized intensity at t=2 minutes. This image serves as the steady-state driving source input for the reaction-diffusion system, preserving the differential information of perfusion characteristics in different tissues to the greatest extent possible.

[0068] 203. Input the steady-state driving source into the reaction-diffusion system with preset parameters for numerical solution. The system evolves according to the intensity distribution of the steady-state driving source during the simulation time, and finally outputs a stable spatial distribution map that characterizes the dynamic characteristics of the biochemical process inside the tissue, which is denoted as the morphological field. Specifically, based on the characteristics of biological tissues, the dynamic parameters of the reaction-diffusion system are preset, including the activator diffusion coefficient, the inhibitor diffusion coefficient, and the reaction rate constant, and a discretized computational grid matching the spatial resolution of the steady-state driving source is established. The activator field and inhibitor field are initialized on the computational grid, with the initial distribution of the activator field set as the intensity distribution of the steady-state driving source, and the inhibitor field set as a uniform distribution. The reaction-diffusion equations are discretized using the finite difference method, and the diffusion and reaction terms are solved alternately within the discrete time step to achieve synchronous updates of the activator field and inhibitor field. The numerical residual of the activator field after each time step is calculated, and the system is considered to have reached a steady state when the numerical residual of multiple consecutive time steps is less than the preset tolerance. The spatial distribution of the activator field in the steady state is extracted and subjected to Gaussian filtering for noise reduction to generate the final morphological field characterizing the dynamic properties of the biochemical processes within the tissue.

[0069] It should be noted that the input in this embodiment is the steady-state driving source extracted in step 202, which is a dimensionless image of 256×256 pixels, whose pixel value ranges from 0 to 1, representing the tissue normalized enhancement intensity at a specific time point (t=2 minutes).

[0070] Based on the biophysical characteristics of breast tissue, the kinetic parameters of the reaction-diffusion system are pre-defined. a Set to 0.20 (pixels) 2 / unit time), inhibitor diffusion coefficient (D i Set to 0.60 (pixels) 2 The reaction rate constant (k) is set to 4.5 (1 / unit time). These parameters collectively determine the system's ability to generate Turing patterns. A 256×256 discrete computational grid is established that perfectly matches the spatial resolution of the input image, with a spatial step size Δx=Δy=1 pixel.

[0071] Two fields are initialized on the computational grid: an activator field (a) and an inhibitor field (b). The initial distribution of the activator field is directly set to the intensity distribution of the steady-state driving source. The initial activator concentration at pixel (100, 100) is a0(100, 100) = 0.95. The inhibitor field is initialized to a uniform distribution, with an initial concentration of 0.10 for all pixels.

[0072] The reaction-diffusion equations are discretized and solved using the explicit finite difference method. System evolution is performed in a time-step manner with a time step size of Δt = 0.01 (per unit time). Within each time step, the diffusion term is calculated first: the diffusion amount of activator and inhibitor at each pixel is calculated using a discrete Laplacian operator (five-point stencil). The reaction term is then calculated: a standard activator-inhibitor reaction model (a variant of the Gierer-Meinhardt model) is used, whose dimensionless form includes a pre-defined reaction rate constant k. By summing the increments of the diffusion and reaction terms, the activator and inhibitor fields across the entire grid are updated synchronously. After the first time step, the concentration in high-activator regions (a ≈ 0.95) may decrease slightly, while the concentration in surrounding regions may increase slightly due to diffusion.

[0073] Every 10 time steps (i.e., every 0.1 unit of simulation time), the numerical residual of the activator field is calculated. The residual is defined as the average of the absolute values ​​of the pixel density differences between the current field and the previous checkpoint field. The activator field is considered complete when the residuals calculated for 5 consecutive checkpoints (i.e., 0.5 units of simulation time) are all less than the preset tolerance of 1 × 10⁻⁶. -5 When the system reaches a stable state, the evolution terminates.

[0074] The spatial distribution of the activator field under steady-state conditions was extracted. This distribution exhibits a stable spatial pattern determined by both the initial driving source and system dynamics. Gaussian filtering (using a Gaussian kernel with σ=0.8 pixels) was applied to this distribution to suppress minor noise that might be introduced by numerical calculations, ultimately generating a smooth morphological field characterizing the dynamic properties within the tissue. This morphological field is a 256×256 matrix whose value range is no longer limited to [0,1], but retains relative spatial structure information, which can be used as input for subsequent analysis.

[0075] 204. The DCE-MRI data from large-scale normal breast tissue are processed sequentially through the driver source data generation step, the steady-state driver source extraction step, and the morphological field generation step. All the normal morphological fields produced are centrally stored and managed to construct a standard morphological field cloud library. Specifically, DCE-MRI data of large-scale normal breast tissue is input into the driving source data generation step for processing to generate a corresponding normal tissue driving source dataset; the normal tissue driving source dataset is input into the steady-state driving source extraction step for processing to generate a normal tissue steady-state driving source set; the normal tissue steady-state driving source set is input into the morphological field generation step for batch processing to generate a large-scale normal morphological field set; features are extracted and standardized for each morphological field in the normal morphological field set to generate a morphological field feature descriptor set in a unified format; the morphological field feature descriptor set is associated and stored with the corresponding metadata to establish a multi-level index structure based on feature similarity, thus completing the construction of a standard morphological field cloud library.

[0076] It should be noted that this embodiment aims to construct a standard morphological field cloud library, the input of which is DCE-MRI time-series data from 1000 cases of pathologically confirmed normal breast tissue. Each case's data consists of 5 time points and 256×256 pixel images, consistent with the aforementioned steps.

[0077] Using these 1000 normal DCE-MRI data as input, the processing steps 201 to 203 are executed in batches. For each data point, spatiotemporal registration and intensity normalization are performed (step 201). For case 001, the intensity curve of its pixel (150, 150) is processed and the normalized value becomes [0.05, 0.20, 0.45, 0.40, 0.35].

[0078] Based on the kinetic characteristics clustering analysis of all case-driven source data (step 202), the optimal time point was determined to be t=2 minutes. Accordingly, a full-frame image (256×256) at that time point was extracted from the driving source of each case as the steady-state driving source. The steady-state driving source of each case was input into a reaction-diffusion system with preset parameters (step 203). The system parameters were fixed as follows: activator diffusion coefficient 0.18, inhibitor diffusion coefficient 0.52, and reaction rate 4.2. After numerical solution reached stability, Gaussian filtering (σ=0.7) was applied to the activator field, ultimately generating a 256×256 morphological field for each case. This resulted in a set containing 1000 normal morphological fields.

[0079] For each element in the morphological field set, a feature vector characterizing its spatial distribution is extracted and standardized. Taking the morphological field of case 001 as an example, four Haralick features of its Gray-Level Co-occurrence Matrix (GLCM)—contrast, correlation, energy, and homogeneity—are calculated, with values ​​of 1.85, 0.92, 0.15, and 0.65, respectively. Simultaneously, the energy coefficients of its wavelet transform (Haar wavelet) in the second layer are calculated, resulting in a 20-dimensional sub-vector. These features are then concatenated to form a preliminary 24-dimensional feature vector.

[0080] Z-score normalization was applied to the feature vectors of all 1000 morphological fields, making the mean of each feature dimension 0 and the standard deviation 1. Ultimately, each morphological field was encoded as a 24-dimensional, dimensionless, normalized feature descriptor.

[0081] Feature descriptors are associated with metadata, and an index structure is established. Each morphological field feature descriptor is associated and stored with its corresponding metadata (case ID, patient age, acquisition device model, morphological field generation parameters, etc.). Principal component analysis (PCA) is used to reduce the dimensionality of 1000 24-dimensional feature descriptors, retaining the top 5 principal components (cumulative contribution rate of 90%). A KD-tree (K-Dimensionaltree) index structure is constructed on this 5-dimensional space to achieve fast nearest neighbor retrieval based on feature similarity.

[0082] At this point, a standard morphological field cloud library containing 1000 normal morphological field feature descriptors, metadata, and multi-level indexes has been completed. This cloud library will serve as the benchmark for determining whether the tissue under test is diseased in subsequent steps.

[0083] 205. The DCE-MRI data of the breast region to be tested is processed sequentially through the driver source data generation step, the steady-state driver source extraction step, and the morphological field generation step to obtain the morphological field to be tested; the average difference between the morphological field to be tested and all normal morphological fields in the standard morphological field cloud library is calculated to obtain the morphological distortion value; the morphological distortion value is compared with a preset threshold. If it exceeds the threshold, the breast region to be tested is determined to be a lesion region. Specifically, the DCE-MRI data of the breast region to be tested is input into the driver source data generation step for processing to generate the driver source data to be tested; the driver source data to be tested is input into the steady-state driver source extraction step for processing to generate the steady-state driver source to be tested; the steady-state driver source to be tested is input into the morphological field generation step for processing to generate the morphological field to be tested, and the morphological field to be tested is subjected to feature extraction and standardized encoding to generate the morphological field feature descriptor; based on the morphological field feature descriptor to be tested, a multi-level index search is performed in the standard morphological field cloud library, and the difference between the morphological field to be tested and each normal morphological field feature descriptor in the cloud library is calculated to generate a difference set; the difference set is statistically analyzed, and the average difference is calculated as the morphological distortion value. This morphological distortion value is compared with a clinically determined preset threshold, and the lesion area judgment conclusion is output based on the comparison result.

[0084] It should be noted that this embodiment aims to process DCE-MRI data of a breast region to be tested and use a pre-constructed standard morphological field cloud library (containing 1000 feature descriptors of normal morphological fields) to determine whether it is a lesion region. The input is a DCE-MRI data sample to be tested (5 time points, 256×256 pixels).

[0085] The data to be tested is processed sequentially through steps 201, 202, and 203 to generate its corresponding morphological field.

[0086] Perform spatiotemporal registration and intensity normalization (step 201). The original intensity curve of pixel (120,80) in the test data is [105,400,850,820,780]au. After calculation, its minimum baseline intensity is (105+400) / 2=252.5au, the maximum enhancement intensity is 850a.u., and the normalized curve is [0,0.25,1.0,0.95,0.88].

[0087] Based on the optimal time point (t=2 minutes) determined during the cloud library construction, the corresponding image is extracted from the driving source. Numerical solutions are performed using the exact same reaction-diffusion system parameters as those used during cloud library construction (activator diffusion coefficient 0.18, inhibitor diffusion coefficient 0.52, reaction rate 4.2) to generate the test morphological field. Features identical to those in the cloud library (24-dimensional Haralick and wavelet energy features) are extracted from this test morphological field, and Z-score standardization is performed using the mean and standard deviation calculated by the cloud library to generate a 24-dimensional feature descriptor for the test morphological field. The first feature (contrast) is assumed to have a value of 2.30.

[0088] Based on the feature descriptors of the morphological field to be tested, a search and dissimilarity calculation are performed in a standard morphological field cloud database. Using the KD-tree index of the cloud database, the 100 most similar normal samples to the descriptor to be tested are quickly found (nearest neighbor search). The Euclidean distance between the descriptor to be tested and each normal morphological field descriptor in the cloud database is calculated as the dissimilarity. The Euclidean distance formula is as follows: ,in and These are the values ​​of the test sample and a normal sample on the i-th feature, respectively. This distance is dimensionless. The difference between the test sample and normal sample A in the cloud database is calculated to be 1.75, and the difference between the test sample and sample B is 1.82.

[0089] Statistical analysis was performed on all calculated differences. The arithmetic mean of the differences between the tested morphological field and all 1000 normal morphological fields in the cloud database was calculated to obtain the morphological distortion value. This value was assumed to be 1.68. The calculated morphological distortion value of 1.68 was compared with a clinically validated preset threshold (1.50). Since 1.68 > 1.50, the tested breast region was determined to be a lesion region.

[0090] 206. It also includes optimizing the diagnostic threshold based on pathological validation data to improve the system's recognition performance, selecting typical case data confirmed by pathology from historical clinical data, including DCE-MRI data of normal tissue and different types of lesion tissue, and constructing a system validation test set; All data in the verification test set are processed sequentially through the driving source data generation step, the steady-state driving source extraction step, and the morphological field generation step to generate a verification morphological field set. Calculate the average difference between each morphological field in the verification morphological field set and all normal morphological fields in the standard morphological field cloud library to obtain the verification distortion dataset. Based on the validation distortion dataset and the corresponding pathological diagnosis results, the optimal morphological distortion threshold is determined by receiver operating characteristic curve analysis, and optimized diagnostic threshold parameters are generated. The validation test set was retested using the optimized diagnostic threshold parameters. The diagnostic accuracy, sensitivity, and specificity of the system were calculated, and a system performance evaluation report was generated. It should be noted that this embodiment aims to comprehensively verify and optimize the performance of the breast lesion recognition system. The input consists of a system verification test set and a standard morphological field cloud library (containing feature descriptors of 1000 normal morphological fields) constructed in the previous steps.

[0091] A validation set was constructed by selecting 200 DCE-MRI data points from historical clinical data that were confirmed by pathological gold standards. This set included 100 normal tissue samples and 100 lesion tissue samples (50 malignant and 50 benign). All data in this validation set were processed sequentially through the same workflow from steps 201 to 203 to generate a corresponding "validation morphological field" for each data point, ultimately resulting in a "validation morphological field set" containing 200 morphological fields.

[0092] For each morphological field in the validation morphological field set, calculate its average difference (morphological distortion value) with all 1000 normal morphological fields in the standard morphological field cloud library. Use the exact same procedure and formula as in step 205 (Euclidean distance of 24-dimensional Z-score normalized feature descriptors).

[0093] Calculations showed that a pathologically confirmed normal case (verification ID: N-045) had an average morphological field discrepancy of 1.15 compared to the cloud database. A pathologically confirmed malignant case (verification ID: M-087) had an average morphological field discrepancy of 2.85 compared to the cloud database. Ultimately, a "verification distortion dataset" containing 200 distortion values ​​was obtained, each corresponding to a known pathological label.

[0094] Based on these 200 pairs of "distortion value-pathological result" data, receiver operating characteristic (ROC) curves were plotted for threshold optimization analysis. Using morphological distortion as the diagnostic indicator and pathological results as the gold standard, the true positive rate (sensitivity) and false positive rate (1-specificity) at different critical thresholds were calculated. It was assumed that when the threshold was set to 1.60, the true positive rate was 92% (i.e., 92% of lesions were correctly identified), and the false positive rate was 10% (i.e., 10% of normal tissue was misdiagnosed as lesions). With the goal of improving diagnostic efficacy, the threshold corresponding to the maximum Youden index (sensitivity + specificity - 1) was selected. Calculations showed that the Youden index reached its maximum value of 0.86 at a threshold of 1.65 (at which point sensitivity was 90% and specificity was 96%). Therefore, the optimized diagnostic threshold parameter for the system was determined to be 1.65.

[0095] The validation test set was retested using the optimized new threshold (1.65), and key performance indicators were calculated. Performance calculation: In 200 validation cases, the diagnostic conclusions were consistent with the pathological results in 180 cases.

[0096] Accuracy = 180 / 200 = 90%; Sensitivity (lesion detection rate) = 90 / 100 = 90%; Specificity (normal exclusion rate) = 90 / 100 = 90%.

[0097] The final system performance evaluation report clearly states: "Based on independent validation set testing, at the optimal morphological distortion threshold of 1.65, the system has an accuracy of 90%, a sensitivity of 90%, and a specificity of 90% in identifying breast lesions."

[0098] In this embodiment of the invention, the beneficial effects are as follows: By constructing an independent test set containing pathologically confirmed cases and performing blinded validation, the reliability and unbiasedness of the performance evaluation results are ensured, proving that the method's overall accuracy of 90% has clinical reference value. Secondly, the optimal diagnostic threshold is automatically determined based on big data and ROC curve analysis, replacing the traditional subjective setting relying on expert experience. This standardizes and quantifies the decision-making process, effectively balancing sensitivity and specificity, and reducing the risk of misdiagnosis and missed diagnosis. Finally, this process generates a comprehensive performance evaluation report, providing not only a basis for optimizing algorithm parameters but also clear efficacy evidence for its subsequent clinical translation and application.

[0099] The above describes a method for identifying breast lesion areas based on big data in an embodiment of the present invention. The following describes a system for identifying breast lesion areas based on big data in an embodiment of the present invention. Please refer to [link to relevant documentation]. Figure 3 One embodiment of the breast lesion area identification system based on big data in this invention includes: 301. Processing module: Performs spatiotemporal registration and intensity normalization processing on breast magnetic resonance imaging (DCE-MRI) data to generate driving source data; 302. Extraction module: Extracts image data from the driving source data as a steady-state driving source for the reaction-diffusion system; 303. Calculation module: Inputs the steady-state driving source into the reaction-diffusion system for numerical solution. The system outputs a spatial distribution map, denoted as the morphological field, based on the intensity distribution evolution of the steady-state driving source; 304. Construction module: Processes the DCE-MRI data of normal breast tissue sequentially through the driving source data, steady-state driving source, and morphological field, stores and manages the generated morphological field, forming a standard morphological field cloud library; 305. Identification module: Processes the data of the breast region to be tested to obtain the morphological field to be tested, calculates the difference between the morphological field to be tested and the standard morphological field cloud library to obtain the morphological distortion value, compares the morphological distortion value with a preset value, and identifies the lesion region if the morphological distortion value exceeds the preset value.

[0100] In this embodiment of the invention, the beneficial effects are as follows: A reaction-diffusion model is used to simulate biophysical processes, generating a visualized morphological field and solving the black-box problem of machine learning. Rigorous spatiotemporal registration and intensity normalization preprocessing eliminate interference from equipment and operational differences. Physical simulation amplifies microscopic physiological differences, resulting in higher sensitivity for identifying early and atypical lesions. By constructing a standard morphological field cloud library and quantifying morphological distortion, efficient management, rapid retrieval, and objective and accurate diagnosis of massive amounts of data are achieved, improving diagnostic efficiency and standardization.

[0101] The present invention also provides a breast lesion area identification device based on big data. The breast lesion area identification device based on big data includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor performs the steps of the breast lesion area identification method based on big data in the above embodiments.

[0102] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the method for identifying breast lesion areas based on big data.

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

[0104] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0105] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying breast lesion areas based on big data, characterized in that, The method for identifying breast lesion areas based on big data includes: The breast magnetic resonance imaging data is processed to eliminate positional bias and generate time-series image data; The enhancement dynamics curves of pixels in the time-series image data are extracted, the maximum enhancement intensity and minimum baseline intensity of the curves are calculated, and normalization calculation is performed based on the difference between the maximum and minimum to generate normalized intensity data. Image data is selected from the normalized intensity data as the steady-state driving source data of the reaction-diffusion system; Based on the kinetic parameters of the reaction-diffusion system, including the activator diffusion coefficient, the inhibitor diffusion coefficient, and the reaction rate constant, a computational grid matching the steady-state driving source was established. An activator field and an inhibitor field are initialized on the computational grid, with the activator field set to the intensity distribution of the steady-state driving source and the inhibitor field set to a uniform distribution. The reaction-diffusion equations are discretized, and the diffusion and reaction terms are solved within the discrete time step to update the activator field and inhibitor field. Calculate the numerical residual of the activator field after each time step. When the numerical residual is less than the preset tolerance, it is determined that a steady state has been reached. Extract the spatial distribution of the activator field under stable conditions to generate a morphological field; The breast magnetic resonance imaging data of normal breast tissue is sequentially processed by the driving source data, steady-state driving source and morphological field to form a standard morphological field cloud library. Based on the data of the breast region to be tested, the morphological field to be tested is obtained. The difference between the morphological field to be tested and the standard morphological field cloud library is calculated to obtain the morphological distortion value. The morphological distortion value is compared with the preset value. If it exceeds the preset value, it is a lesion area.

2. The method for identifying breast lesion areas based on big data according to claim 1, characterized in that, include: Analyze the temporal intensity curves of pixels in the driving source data, calculate the maximum rising slope and peak time of the curves, and generate a set of temporal dynamic characteristic parameters; Based on the aforementioned time dynamics characteristic parameter set, characteristic time points are analyzed and identified to determine the optimal time selection scheme; Based on the optimal time selection scheme, image data is extracted from the driving source data and used as the steady-state driving source input for the reaction-diffusion system.

3. The method for identifying breast lesion areas based on big data according to claim 1, characterized in that, include: Magnetic resonance imaging data of normal breast tissue is input into the driving source data for processing to generate a normal tissue driving source dataset. The normal tissue driving source dataset is input into the steady-state driving source for processing to generate a normal tissue steady-state driving source set; The set of normal tissue steady-state driving sources is input into the morphological field for processing to generate a set of normal morphological fields. The normal morphological field set is subjected to feature extraction and encoding to generate a morphological field feature descriptor set; The set of morphological field feature descriptors and metadata are stored, an index structure is established, and a standard morphological field cloud library is constructed.

4. The method for identifying breast lesion areas based on big data according to claim 3, characterized in that, include: The breast magnetic resonance imaging data of the breast region to be tested is input into the driving source data for processing to generate the driving source data to be tested. The data of the driving source under test is input into the steady-state driving source for processing to generate the steady-state driving source under test; The steady-state driving source input morphological field is processed to generate the morphological field to be tested, and the features of the morphological field to be tested are extracted and encoded to generate the feature descriptor of the morphological field to be tested. Based on the feature descriptor of the morphological field to be tested, a search is performed in the standard morphological field cloud library to calculate the difference between the feature descriptor of the morphological field to be tested and the normal morphological field, and a set of difference is generated. The set of differences is analyzed, the average difference is calculated as the morphological distortion value, and the morphological distortion value is compared with a preset threshold to output a conclusion.

5. The method for identifying breast lesion areas based on big data according to claim 4, characterized in that, Calculate the difference between the feature descriptors of the tested morphological field and the normal morphological field, using Euclidean distance. As a measure of difference: in, and These are the values ​​of the test sample and a normal sample on the i-th feature dimension, respectively, where i is the index of the feature dimension. This represents the total dimension of the feature descriptor.

6. A breast lesion area identification system based on big data, characterized in that, The aforementioned big data-based breast lesion area identification system includes: The processing module processes the breast magnetic resonance imaging data, eliminates positional deviations, and generates time-series image data; it extracts the enhancement dynamics curves of pixels in the time-series image data, calculates the maximum enhancement intensity and minimum baseline intensity of the curves, and performs normalization calculations based on the difference between the maximum and minimum to generate normalized intensity data. The extraction module selects image data from the normalized intensity data as the steady-state driving source data for the reaction-diffusion system. The computational module, based on the dynamic parameters of the reaction-diffusion system, including the activator diffusion coefficient, inhibitor diffusion coefficient, and reaction rate constant, establishes a computational grid matching the steady-state driving source. It initializes the activator field and inhibitor field on this grid, setting the activator field as the intensity distribution of the steady-state driving source and the inhibitor field as a uniform distribution. The reaction-diffusion equations are discretized, and the diffusion and reaction terms are solved within discrete time steps to update the activator and inhibitor fields. The numerical residual of the activator field is calculated after each time step; when the numerical residual is less than a preset tolerance, a steady state is determined. The spatial distribution of the activator field in the steady state is extracted to generate a morphological field. The construction module sequentially processes the magnetic resonance imaging data of normal breast tissue through the driving source data, steady-state driving source, and morphological field to form a standard morphological field cloud library. The identification module obtains the morphological field to be tested based on the data of the breast region to be tested, calculates the difference between the morphological field to be tested and the standard morphological field cloud library to obtain the morphological distortion value, compares the morphological distortion value with the preset value, and if it exceeds the preset value, it is a lesion area.

Images