Method for processing pathological medical images based on staining medical formulations

Abstract

This invention relates to the field of healthcare informatics technology and discloses a pathological medical image processing method based on stained medical preparations. The method includes: acquiring raw digital images through a data acquisition channel and extracting a set of pixels whose absorption spectral response deviates from the standard distribution range as geometric control anchor points; mapping the anchor points to curve control points and establishing an interference penalty term based on spatial steric hindrance characteristics; constructing an energy functional containing the interference penalty term and surface tension constraints, and determining closed vector spline curves of the cell nucleus and cytoplasmic entity boundaries through minimum convergence calculation; and assigning pixel component values ​​within the curve-defined region according to rendering parameters. This invention shifts the processing focus from pixel-level interpolation to tissue entity geometric modeling, compensates for data collapse caused by color rendering failure through spatial exclusive constraints, eliminates noise caused by carrier performance fluctuations, and improves the quality of standardized images.

Description

Technical Field

[0001] This invention belongs to the field of healthcare informatics technology, and in particular relates to a method for processing pathological medical images based on stained medical preparations. Background Technology

[0002] In current digital pathology analysis systems, acquiring images of biological tissue sections using scanning equipment is a fundamental step in building medical decision support systems. These images record cell morphology and tissue structure evolution at the microscopic scale, providing the original digital evidence for subsequent pathological analysis. Conventional methods use medical preparations such as hematoxylin and eosin to stain biological samples, utilizing the difference in absorption of specific wavelengths of light by the cell nucleus and cytoplasm to generate optical contrast. However, the physicochemical properties of medical preparations are constrained by objective physical factors such as concentration, storage time, environmental pH, and staining reaction temperature. When the performance of the preparations fluctuates, the color depth of the sample decreases, leading to irreversible decay of the color gradient at the pixel level. Due to the instability of the chemical reaction, underlying data noise is caused, essentially resulting in the loss of acquisition of physical boundary information at the cell edge.

[0003] Analysis of existing technologies reveals that most processing methods treat pathological images as discrete pixel arrays, focusing on color remapping at the image level. Specifically, existing technologies suffer from the following shortcomings: 1. The processing is limited to mathematical transformations of pixel grayscale values. When insufficient color rendering causes optical signals to fall below the detection limit of the imaging sensor, it cannot restore lost tissue edge information through simple contrast enhancement. 2. Compensation algorithms lack consideration for the inherent geometric and topological properties of biological tissues, making it difficult to handle spatial overlap of cell entities caused by dye diffusion. This easily leads to deviations between the reconstructed tissue morphology and the actual physical boundaries. Optimizing staining processes and imaging hardware can alleviate physical layer deviations. The stability of digital pathology analysis depends on backend image processing algorithms to complete missing information; existing software processing solutions and staining normalization control methods are insufficient to address this. Extreme color development failure conditions still have limitations. Chinese invention patent CN119399208B discloses a self-supervised staining normalization method and device for pathological images. It constructs a common feature domain through self-supervised learning and uses content and scene perceptrons to perform style transfer and color normalization on images with different centers. This type of algorithm based on deep neural networks belongs to pixel feature space mapping and statistical distribution alignment. The premise is that the original image retains an extractable content feature structure. In actual clinical scenarios, when the performance of medical preparations fluctuates drastically, causing the underlying spectral signal to collapse or fall below the detection limit of the imaging sensor, this solution lacks the modeling of the inherent physical geometric topology constraints of biological tissues. It cannot logically deduce the real tissue boundary from the signal missing area, which may produce morphological artifacts in low contrast areas and make it difficult to ensure the authenticity of the reconstructed image at the anatomical level.

[0004] Therefore, how to utilize the geometric topological constraints and physical space exclusivity logic of cell tissues to reconstruct the failed pathological image information caused by fluctuations in stained medical preparations has become the technical problem to be solved by this invention. Summary of the Invention

[0005] This invention provides a method for processing pathological medical images based on stained medical preparations, comprising the following steps:

[0006] Step S1: Acquire the original digital image representing the tissue morphology of the sample to be processed through the data acquisition channel, and extract the set of pixels whose absorption spectral response values ​​deviate from the preset standard distribution range in the original digital image as geometric control anchor points.

[0007] Step S2: Map the geometric control anchor points to the non-uniform rational B-spline curve control points in the computer-aided design logic, and establish an interference penalty term to characterize the geometric penetration risk of adjacent entity boundaries based on the spatial steric hindrance characteristics of biological entities in the sample to be processed.

[0008] Step S3: Construct an energy functional that includes an interference penalty term, surface tension constraint, and curvature smoothing factor, and perform a gradient descent-based minimum convergence calculation on the energy functional to determine the closed vector spline curves that characterize the boundaries of the cell nucleus entity and the cytoplasm entity in the sample to be processed.

[0009] Step S4: Obtain the pixel component response vector representing the standard colorimetric features as the material rendering parameter, and perform linear weighted mapping on the pixel component values ​​within the closed geometric region defined by the closed vector spline curve, so as to generate a standardized pathological digital image for medical auxiliary diagnosis and eliminate data noise caused by the concentration fluctuation of stained medical preparations.

[0010] Preferably, the logic for establishing the interference penalty term in step S2 includes: setting a nucleo-cytoplasmic ratio threshold that characterizes the limit of the ratio between the volume of the cell nucleus and the volume of the cytoplasm; monitoring the geometric trajectory formed by two adjacent nodes in real time during the iterative evolution of the control points of the non-uniform rational B-spline curve; when the geometric trajectory overlaps spatially or the nucleo-cytoplasmic ratio threshold exceeds the preset range of 0.1 to 0.8, calculating the energy gradient characterizing the spatial exclusive interference, and introducing the energy gradient as an interference penalty term into the energy functional to constrain the topological evolution path of the closed vector spline curve.

[0011] Preferably, the method for extracting geometric control anchor points in step S1 includes: performing spectral feature recognition on the original digital image to locate the coordinates of local pixels whose spectral response intensity is lower than a preset brightness threshold; calculating the gradient change direction of the local pixel coordinates in the local pixel matrix; and marking the local pixel coordinates as control vectors with determined topological weight attributes according to the gradient change direction, thereby forming geometric control anchor points.

[0012] Preferably, the closed vector spline curve has geometric continuity and is used to fill in the physically missing morphological boundaries in the original digital image by minimizing the tension energy constraint of the spline curve in areas where boundary information is missing due to spectral response failure.

[0013] Preferably, the material rendering parameters include preset standard color component weights for the color rendering characteristics of hematoxylin and eosin.

[0014] Preferably, the closed vector spline curve defines the geometric topological skeleton in the standardized digital pathology image, making the morphological feature data of the standardized digital pathology image independent of the pH and storage time of the stained medical preparation.

[0015] Preferably, during the minimum iterative convergence calculation of the energy functional, the penalty weight generated by the interference penalty term... Overlapping volume with adjacent entities The following mapping relationship is satisfied: ,in, The penalty weight is λ, which is a preset exclusivity adjustment coefficient. This represents the value of the overlapping volume between adjacent entities, calculated through geometric motion trajectories.

[0016] Preferably, in step S4, when performing linear weighted mapping assignment on pixel component values, global color distribution vector assignment is performed on restricted pixels within the closed geometric region defined by the closed vector spline curve to correct color deviation caused by batch differences in stained medical preparations.

[0017] Preferably, the method is applied to the automated analysis of pathological images in a medical information system, with standardized digital pathological images serving as the data source for the extraction of tissue morphological features by the medical information system.

[0018] Preferably, closed vector spline curves shift the logical focus of image processing from pixel-level numerical interpolation to organizational-level solid geometric modeling, thereby compensating for data collapse caused by the failure of spectral colorimetric carriers through spatial occupancy exclusive constraints.

[0019] Compared with existing technologies, the pathological medical image processing method based on stained medical preparations of this invention has the following advantages:

[0020] 1. In pathological medical image processing, by mapping the cell morphology edge pixel coordinates within the tissue matrix domain to geometric control points and constructing closed non-uniform rational B-spline curves, the dimensional transformation from a two-dimensional pixel array to a parameterized geometric topology model is achieved. This processing mechanism changes the absolute dependence on optical contrast gradients in traditional paths. In scenarios where the physical information of boundary pixels is lost due to storage time or environmental changes in stained medical preparations, the system calculates closed geometric boundaries that conform to physical morphology based on discrete effective anchor points and the continuity characteristics of spline curves, overcoming the information collapse problem caused by chemical color development failure.

[0021] 2. The spatial exclusivity equation and the surface tension energy minimization model integrated in the geometric constraint solver generate synergistic gains. The spatial exclusivity equation, by introducing a spatial interference penalty term, restricts the geometric penetration of adjacent cell models, simulating the inherent spatial occupancy rules of biological tissues. The surface tension energy minimization model, by adjusting the parametric equation of the spline curve, guides the geometric shape of the topological fracture zone to converge toward the energy functional minimum. This inherent coupling of multiple mechanisms enables the reconstructed digital image to have biologically meaningful morphological realism, correcting the geometric distortion produced by traditional color calibration algorithms under harsh staining conditions.

[0022] 3. By using the preset standard staining color distribution vector as the material rendering parameter, parametric color filling is performed inside the closed reconstructed spline curve. This transforms the performance fluctuations of stained medical preparations into a controllable geometric space material rendering process, converts the random differences in chemical dye concentration into physical geometric spectrum outputs with deterministic solutions, and eliminates data noise caused by differences in preparation concentration, pH, or manufacturer batches. The standardized pathological dataset generated by the system can accurately reflect the real tissue structure changes, improving the stability and automation level of the medical information system in pathological diagnosis and analysis. Attached Figure Description

[0023] Figure 1 This is a flowchart of the pathological and medical image processing method for eliminating color noise according to the present invention.

[0024] Figure 2 This is a modular processing diagram of the present invention, which transforms a two-dimensional pixel array into a geometric topology model. Detailed Implementation

[0025] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0026] It should be noted that all directional and positional terms used in this invention, such as: up, down, left, right, front, back, vertical, horizontal, inner, outer, top, low, lateral, longitudinal, center, etc., are only used to explain the relative positional relationship and connection between components in a specific state (as shown in the accompanying drawings). They are only for the convenience of describing this invention and do not require that this invention be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention. In addition, the descriptions of "first," "second," etc., in this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly indicating the number of technical features indicated.

[0027] In the description of this invention, unless otherwise explicitly specified and limited, the terms installation, connection, and linking should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections; they can refer to direct connections or indirect connections through an intermediate medium; they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0028] In the description of this specification, references to the terms "an embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example, and the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0029] A method for processing pathological medical images based on stained medical preparations includes the following steps:

[0030] Step S1: Acquire the original digital image representing the tissue morphology of the sample to be processed through the data acquisition channel, and extract the set of pixels whose absorption spectral response values ​​deviate from the preset standard distribution range in the original digital image as geometric control anchor points.

[0031] Step S2: Map the geometric control anchor points to the non-uniform rational B-spline curve control points in the computer-aided design logic, and establish an interference penalty term to characterize the geometric penetration risk of adjacent entity boundaries based on the spatial steric hindrance characteristics of biological entities in the sample to be processed.

[0032] Step S3: Construct an energy functional that includes an interference penalty term, surface tension constraint, and curvature smoothing factor, and perform a gradient descent-based minimum convergence calculation on the energy functional to determine the closed vector spline curves that characterize the boundaries of the cell nucleus entity and the cytoplasm entity in the sample to be processed.

[0033] Step S4: Obtain the pixel component response vector representing the standard colorimetric features as the material rendering parameter, and perform linear weighted mapping on the pixel component values ​​within the closed geometric region defined by the closed vector spline curve, so as to generate a standardized pathological digital image for medical auxiliary diagnosis and eliminate data noise caused by the concentration fluctuation of stained medical preparations.

[0034] Preferably, the logic for establishing the interference penalty term in step S2 includes: setting a nucleo-cytoplasmic ratio threshold that characterizes the limit of the ratio between the volume of the cell nucleus and the volume of the cytoplasm; monitoring the geometric trajectory formed by two adjacent nodes in real time during the iterative evolution of the control points of the non-uniform rational B-spline curve; when the geometric trajectory overlaps spatially or the nucleo-cytoplasmic ratio threshold exceeds the preset range of 0.1 to 0.8, calculating the energy gradient characterizing the spatial exclusive interference, and introducing the energy gradient as an interference penalty term into the energy functional to constrain the topological evolution path of the closed vector spline curve.

[0035] Preferably, the method for extracting geometric control anchor points in step S1 includes: performing spectral feature recognition on the original digital image to locate the coordinates of local pixels whose spectral response intensity is lower than a preset brightness threshold; calculating the gradient change direction of the local pixel coordinates in the local pixel matrix; and marking the local pixel coordinates as control vectors with determined topological weight attributes according to the gradient change direction, thereby forming geometric control anchor points.

[0036] Preferably, the closed vector spline curve has geometric continuity and is used to fill in the physically missing morphological boundaries in the original digital image by minimizing the tension energy constraint of the spline curve in areas where boundary information is missing due to spectral response failure.

[0037] Preferably, the material rendering parameters include preset standard color component weights for the color rendering characteristics of hematoxylin and eosin.

[0038] Preferably, the closed vector spline curve defines the geometric topological skeleton in the standardized digital pathology image, making the morphological feature data of the standardized digital pathology image independent of the pH and storage time of the stained medical preparation.

[0039] Preferably, during the minimum iterative convergence calculation of the energy functional, the penalty weight generated by the interference penalty term... Overlapping volume with adjacent entities The following mapping relationship is satisfied: ,in, The penalty weight is λ, which is a preset exclusivity adjustment coefficient. This represents the value of the overlapping volume between adjacent entities, calculated through geometric motion trajectories.

[0040] Preferably, in step S4, when performing linear weighted mapping assignment on pixel component values, global color distribution vector assignment is performed on restricted pixels within the closed geometric region defined by the closed vector spline curve to correct color deviation caused by batch differences in stained medical preparations.

[0041] Preferably, the method is applied to the automated analysis of pathological images in a medical information system, with standardized digital pathological images serving as the data source for the extraction of tissue morphological features by the medical information system.

[0042] Preferably, closed vector spline curves shift the logical focus of image processing from pixel-level numerical interpolation to organizational-level solid geometric modeling, thereby compensating for data collapse caused by the failure of spectral colorimetric carriers through spatial occupancy exclusive constraints.

[0043] Example 1: When the system faces high-throughput digital scanning conditions in a clinical pathology center, the biological tissue sections in the samples to be processed are affected by the prolonged storage time of hematoxylin and eosin stained medical preparations and fluctuations in environmental pH, resulting in a decrease in physical color development depth. This causes the edge pixel color gradient of cell nuclei and cytoplasm to drop below the detection limit of the imaging sensor, leading to the loss of tissue morphological boundary information and spatial adhesion between adjacent biological entities. To address the aforementioned optical boundary breakage, the system acquires raw digital images representing the morphological characteristics of the tissues in the samples to be processed through the data acquisition channel. It performs spectral feature recognition on the raw digital images, locates the coordinates of local pixels with spectral response intensity below a preset brightness threshold, calculates the gradient change direction of the local pixel coordinates in the local pixel matrix, and marks the local pixel coordinates as control vectors with defined topological weight attributes based on the gradient change direction, forming geometric control anchor points. To eliminate the impact of missing pixel signals, the system migrates the image processing logic from pixel-level numerical interpolation to tissue-level solid geometric modeling, and maps the extracted geometric control anchor points as non-uniform rational B-spline curve control points in the computer-aided design logic.

[0044] In its implementation, the system uses the arithmetic mean center of the geometric control anchor points as the pole reference. Following an increasing order of azimuth angles from 0 to 360 degrees in the polar coordinate system, all control anchor points are sorted and reorganized clockwise to establish a closed control point topological sequence. The curve order is uniformly set to 3. For the construction of node vectors, the system uses a quasi-uniform vector sequence with a first-to-last repetition of 4. Nodes within this sequence are evenly spaced with a step size of 0.01. The weight coefficients of all control points are uniformly preset to 1.0 to ensure that the curve possesses 2nd-order geometric continuity at the physical boundary. During the iterative evolution of the control points of the non-uniform rational B-spline curve, the system monitors the geometric trajectory formed by two adjacent nodes in real time. When the geometric trajectory spatially overlaps or the preset nucleus-to-mass ratio threshold exceeds the range of 0.1 to 0.8, the system adjusts the parameters based on the sample to be processed. The spatial steric hindrance characteristics of biological entities are calculated to characterize the energy gradient of spatial exclusive interference. An interference penalty term is established to characterize the geometric penetration risk of adjacent entity boundaries to constrain the topological evolution path. Geometric control anchor points are obtained, and the discrete pixel coordinate sequence is converted into control points of a third-order non-uniform rational B-spline curve through least-squares fitting. Based on the spatially sparse sampling of the anchor point density distribution, points that meet the minimum spacing requirement are retained as effective control points. The geometric center coordinates of the effective control points are calculated. Using these as the center, the radius of the initial evolution contour is selected as 0.8 times the average distribution radius of the control points, and a closed initial guiding curve is constructed. The node vector of the non-uniform rational B-spline curve is determined according to the cumulative chord length method. The curve has second-order geometric continuity at surface morphological feature points. The penalty weight generated by the interference penalty term and the overlapping volume between adjacent entities satisfy the following mapping relationship: ,in, The penalty weight is λ, which is a preset exclusivity adjustment coefficient. This represents the value of the overlapping volume between adjacent entities, calculated through geometric motion trajectories.

[0045] Based on this, the system constructs an energy functional that includes an interference penalty term, surface tension constraint, and curvature smoothing factor. This energy functional follows the expression below: ,in, Let be the overall energy functional, t be the parameter integral variable of the spline curve, C(t) be the parametric equation of the spline curve, C'(t) be the first derivative of the parametric equation of the spline curve, C″(t) be the second derivative of the property of the parametric equation of the spline curve, and α and β be preset mapping coefficients characterizing the elastic modulus of the tissue. As for the aforementioned spatial interference penalty term, at the physical implementation level of the underlying algorithm, the so-called overlapping volume actually refers to the cumulative area of ​​pixels covered by two closed vector spline curves within a 2D bitmap grid. The system establishes a logical buffer with a resolution of 1024x1024 and performs rasterization filling on the two evolving boundary curves. For each grid pixel unit within the buffer, the system performs a logical AND operation. If the result is true, the counter is incremented by 1. Finally, the specific value of the overlapping volume is obtained by multiplying the total value of the accumulated counter by the physical unit area of ​​0.0625 square micrometers represented by a single pixel. When the counter value exceeds 20 pixel units, the system generates a repulsive field gradient in the opposite direction. In the iterative calculation of the energy functional minimum value, the geometric region defined by the closed vector spline curve is mapped to the gridded two-dimensional bitmap space. The overlapping volume between adjacent entities is determined by counting the total number of pixels in the overlapping region. ;in, Let be the algebraic area value of the overlapping spatial region; when When the gradient is non-zero, the spatial exclusive interference energy gradient is calculated in real time and introduced into the energy functional as an interference penalty term, generating a repulsive force field directed towards the outside of the entity. The exclusive adjustment coefficient λ is preset according to the tissue type of the sample to be processed to determine the penalty weight. This avoids spatial adhesion regions during the evolution of spline curves.

[0046] The system calls the processing unit to perform gradient descent-based minimum iterative convergence calculation on the energy functional. Under the joint constraints of surface tension energy minimization constraint to complete the morphological boundary and spatial interference penalty term to restrict geometric penetration, the system determines the closed vector spline curves representing the boundaries of cell nuclei and cytoplasm in the sample to be processed. After determining the closed vector spline curves, they are defined as the geometric topological skeleton in the standardized pathological digital image. The system obtains the preset standard color component weights for hematoxylin staining features and eosin staining features as material rendering parameters. Within the closed geometric region defined by the closed vector spline curves, the system performs global color distribution vector assignment and linear weighted mapping on the restricted pixels according to the material rendering parameters. The morphological feature data of the output standardized pathological digital image is independent of the pH and storage time of the stained medical preparations and is directly used as the data source for the medical information system to perform automated pathological analysis and tissue morphological feature extraction.

[0047] Example 2: This example uses a bright-field digital pathology scanner with a spatial resolution of 0.25 μm as the acquisition channel to collect hematoxylin and eosin stained biological tissue sections stored for more than 60 months. The acquired raw digital images are superimposed with sensor Gaussian white noise with a signal-to-noise ratio of 15 dB and light source flicker perturbation signal with a frequency of 50 Hz. Affected by the fluctuation of environmental pH, the initial color contrast between the cell nucleus and cytoplasm decreased to 1.15, and the edge spectral response intensity deviated from the standard distribution range, showing an optical boundary breakage state. In order to establish an interference penalty term to constrain the non-uniform rational B-spline curve control The topological evolution path of the control point is determined by setting a kernel-mass ratio threshold. The kernel-mass ratio threshold is set based on achieving an engineering balance between the tolerance of natural morphological variations of biological entities and the spatial exclusivity constraint to prevent geometric penetration of adjacent entities. According to the spatial steric hindrance characteristics of biological entities in the sample to be processed, when the gradient change direction in the local pixel matrix tends to diverge, the system controls the kernel-mass ratio threshold to tend to the tightening range to ensure that the energy functional can smoothly perform the minimum value iteration convergence calculation based on gradient descent under the surface tension constraint. Based on this dual constraint of physics and algorithm, the system sets 0.1 to 0.8 as the effective optimization range of the kernel-mass ratio threshold.

[0048] To verify the convergence synergy effect of the energy functional and the physical effectiveness of the nucleus-to-mass ratio threshold boundary condition, the system established multiple sets of comparative and inventive sample groups. The system set the nucleus-to-mass ratio threshold to 0.45 to construct the inventive sample group, set the threshold to 0.05 to construct an out-of-range control group (below the lower limit), and set the threshold to 0.95 to construct an out-of-range control group (above the upper limit). A partially missing control group was constructed by removing the interference penalty term. The processing unit read the same original digital image and triggered iterative evolution calculations of the spline curve. Measurement data confirmed that the partially missing control group, due to containing only curvature smoothing factors and surface tension constraints, resulted in physical overlap between adjacent entity boundaries, with the measured overlap volume between adjacent entities reaching 28.6 μm³, leading to disordered tissue morphology boundary information. The out-of-range control group (below the lower limit) triggered the interference penalty term in the early stages of evolution, causing abnormal contraction of the closed vector spline curve, producing 145.2 isolated fragmented entities per square millimeter of field of view. The edges cause physical tearing of the continuous cytoplasmic entity boundaries. The control group above the upper limit has too low sensitivity to the exclusive regulation of geometric motion trajectory. When the geometric motion trajectory overlaps spatially, the energy gradient is insufficient to provide the reverse repulsive force, and the overlap volume between adjacent entities reaches 17.4 μm³. In the sample group of this invention, under the set nucleo-mass ratio threshold of 0.45, the interference penalty term and the surface tension constraint produce dynamic synergy. The gradient descent algorithm of the processing unit outputs the minimum convergence signal at the 18th iteration. The interference penalty term constrains the overlap volume between adjacent entities in the sample group of this invention to below 0.2 μm³. At the same time, the curvature continuity of the closed vector spline curve is maintained at an objective level of 98.5%. The data trajectory shows that nucleo-mass ratio thresholds below 0.1 or above 0.8 will cause nonlinear topological distortion or exponential increase in geometric penetration. The numerical range of 0.1 to 0.8 set by the system constitutes the optimal working window to ensure stable solution of the energy functional.

[0049] After determining the closed vector spline curves representing the boundaries of the cell nucleus and cytoplasm, the system obtains the preset standard color component weights as material rendering parameters. Within the closed geometric region defined by the closed vector spline curves, it performs global color distribution vector assignment and linear weighted mapping on the restricted pixels. The output standardized pathological digital image filters out the original 15dB white noise and power frequency disturbances. The generated geometric topological skeleton is independent of the fluctuations of the external chemical staining environment. Experimental data shows that through the parameterized co-evolution of the spatial interference penalty term and the geometric control anchor point, the system compensates for the optical gradient loss defect induced by the attenuation of the staining preparation and outputs a pathological digital image data source with morphological features.

[0050] Example 3: When the system processes biological tissue sections from different organ sources concurrently, there are physiological differences in the elastic modulus of the original tissue between the cell nucleus and cytoplasm in the samples to be processed. Furthermore, uneven section thickness causes shifts in the absorption spectral response of local areas, making it impossible to accurately constrain the energy functional simply by relying on constant parameters. To address this non-uniform optical distribution, the system acquires raw digital images representing the morphological characteristics of the tissue in the samples through the data acquisition channel, extracts the multi-channel color intensity vector for each pixel, and the processing unit converts this multi-channel color intensity vector into an optical density matrix according to Beer-Lambert's law. The matrix elements represent the surface absorbance of the corresponding local pixel coordinates, establishing the relationship between hematoxylin and iodine. The standard spectral center vector for red color is used to calculate the Euclidean distance between the spectral vector corresponding to each local pixel coordinate in the optical density matrix and the standard spectral center vector. The system sets the reference distance threshold to 0.45. Local pixel coordinates with Euclidean distances greater than this threshold are extracted as a set of pixels whose absorption spectral response values ​​deviate from the preset standard distribution range. The structure tensor of each local pixel coordinate in this set of pixels is calculated in the local pixel matrix. Eigenvalue decomposition is performed on the structure tensor to obtain the maximum eigenvector representing the gradient change direction. Based on the maximum eigenvector, this set of pixels is nominated as a geometric control anchor point with topological weight attributes and mapped to the non-uniform rational B-spline curve control point in the computer-aided design logic.

[0051] To construct the energy functional, the system executes a dynamic calibration procedure for the mapping coefficients characterizing the elastic modulus of tissue. The processing unit establishes a computational window containing 64 pixels centered on the control points of the non-uniform rational B-spline curve, generating a gray-level co-occurrence matrix of the optical density matrix elements within this window. The system calculates the local texture entropy based on the gray-level co-occurrence matrix, which is used to quantitatively characterize the surface structure density of the biological entity corresponding to the local pixel matrix. The system establishes the correlation between the mapping coefficients characterizing the elastic modulus of tissue and the local texture entropy, and sets the first mapping coefficient α used to control surface tension constraints according to the following expression: ,in, To calculate the obtained local texture entropy, The preset stiffness gain coefficient, As the basic tension constant, the system synchronously sets the second mapping coefficient β, used to control the curvature smoothing factor, as the ratio of the baseline curvature constant to the local texture entropy. After determining the first mapping coefficient α and the second mapping coefficient β, the system establishes an interference penalty term to characterize the geometric penetration risk of adjacent entity boundaries based on the spatial steric hindrance characteristics of biological entities in the sample to be processed. The system integrates the first mapping coefficient, the second mapping coefficient, and the interference penalty term to construct an energy functional that includes the interference penalty term, surface tension constraints, and curvature smoothing factor. The processing unit performs a minimum iterative convergence calculation based on gradient descent on this energy functional to determine the closed vector spline curves characterizing the boundaries of cell nuclei and cytoplasm entities in the sample to be processed. The system obtains the pixel component response vectors characterizing the standard colorimetric features as material rendering parameters. Within the closed geometric region defined by the closed vector spline curves, the pixel component values ​​are linearly weighted and mapped according to the material rendering parameters. The output standardized pathological digital image has morphological boundaries that have been density-calibrated. The medical information system receives this standardized pathological digital image and performs subsequent analysis, eliminating optical data biases caused by slice thickness and tissue heterogeneity.

[0052] Example 4: When the system is deployed on a new digital pathology platform in a clinical institution, there are physical response deviations in the optical systems of different batches of stained medical preparations and the hardware scanners to be connected. Before processing the actual samples, the system triggers a standard pre-baseline calibration procedure. It reads the reference digital image of the standard color calibration slide through the data acquisition channel and extracts the multi-channel color intensity vectors of each reference color block. The processing unit converts the multi-channel color intensity vectors into a reference optical density matrix according to Beer-Lambert's law, calculates the absorption characteristic extreme value of the matrix under the current optical sensor, and establishes a standard spectral center vector matching the current hardware physical environment based on the absorption characteristic extreme value. It locks this vector as the origin of the coordinates for calculating the Euclidean distance. The system simultaneously measures the edge gradient characteristics of the reference surface geometry with known density within the standard color calibration slide, extracts the reference curvature attenuation rate that characterizes the geometrical motion trajectory of the reference surface geometry's edge, and calculates the exclusive adjustment coefficient λ and the basic tension constant based on the reference curvature attenuation rate. The compensation offset is calculated, and the value after fusing the compensation offset is written into the physical register as the initial reference parameter for energy functional iteration.

[0053] After completing the physical alignment of the optical feature extraction benchmark, the system reads an offline cell morphology topology sample library containing multi-gradient tissue density, performs adaptive calibration of dynamic tension parameters, and the processing unit extracts the local texture entropy of each standard cell nucleus entity in the sample library under its benchmark state. This entropy is then input into the mapping coefficient logic set by the system for reverse deduction to calculate the stiffness gain coefficient required to maintain the non-uniform rational B-spline curve of the standard cell nucleus entity boundary in a minimum energy steady state. The system calibrates the convergence value. When the extracted local texture entropy exhibits isotropic characteristics in spatial distribution, the system controls the reference curvature constant for the batch of tissues to maintain a linear response state. When the spatial steric hindrance characteristics represented by the local texture entropy exceed the set upper limit of the nucleus-mass ratio threshold, the interference penalty term is activated and the corresponding penalty weight trigger limit is recorded. Based on this, a device-specific parameter package consisting of the standard spectral center vector and the calibrated gain coefficients is constructed. This device-specific parameter package is directly applied to the geometric modeling calculation of the raw digital images acquired in subsequent batches. The unique solution of the output value of the energy functional minimum iterative convergence solution process under different physical environments is established. Through this standardized engineering procedure, the standardized pathological digital images received by the medical information system have morphological boundary restoration accuracy that can resist hardware system error fluctuations.

[0054] Example 5: When the system faces the challenge of federated pathological image processing across medical institutions and scanning equipment models, to address the baseline absorbance shift caused by batch differences or environmental temperature and humidity fluctuations in stained medical preparations from different sources, the system executes a standard calibration procedure for non-uniform rational B-spline curve control points to determine the baseline distance threshold for extracting geometric control anchors. It extracts a standard hematoxylin and eosin tissue slide library covering high, medium, and low concentration staining intensities from the federated database and obtains the corresponding optical density matrix. The system iterates through the optical density matrix to calculate the number of geometric control anchors extracted at candidate distance thresholds increasing in 0.05 steps within the range of 0.1 to 0.9, constructing a control point density function. The control point density function is used to characterize the decay relationship between the number of geometric control anchor points per unit area and the candidate distance threshold. When the candidate distance threshold is lower than 0.3, the control point density function enters the oversaturation region, causing the topological weights within the calculation window to cancel each other out and making it impossible to establish an energy functional. When the candidate distance threshold is higher than 0.6, the control point density function enters the signal-to-noise ratio decay region, causing the closed vector spline curve to undergo geometric collapse due to the lack of control points. The system calculates the second derivative of the control point density function and extracts the extreme coordinates corresponding to its zero-crossing points as the geometric separation points of the current batch of tissue samples. The system confirms the extreme coordinates as a general reference distance threshold of 0.45 applicable to cross-device environments and writes this value into the decision logic.

[0055] To determine the computational logic between the local texture entropy and the first mapping coefficient α in the energy functional, the system executes a closed-loop calibration procedure for the tissue elastic modulus mapping coefficients to determine the computational expression. stiffness gain coefficient and fundamental tension constant ,in To calculate the obtained local texture entropy, the system simulates the surface deformation of pathological slides under pressure on a controlled test bench. The measured Young's modulus of standard tissue samples under gradient pressure is obtained using an atomic force microscope. Simultaneously, the system acquires the gray-level co-occurrence matrix under this deformation state to calculate the corresponding local texture entropy sequence. The processing unit uses the least squares method to perform linear regression fitting on the local texture entropy sequence and the measured Young's modulus values ​​to obtain the slope and intercept of the fitted line. The system converts the slope of the fitted line into a stiffness gain coefficient used to compensate for the geometric deformation of the digital image. The intercept of the fitted line is transformed into a fundamental tension constant that maintains the background stiffness of the geometric topological skeleton. When the medical information system receives a new batch of raw digital images, the processing unit uses the calibrated stiffness gain coefficient. With the fundamental tension constant The calculated local texture entropy The first mapping coefficient α is converted into the corresponding local pixel matrix. This calibration process matches the strength of the surface tension constraint with the physical parameters of the biological entity in the sample to be processed. Based on this, the system outputs a pathological digital image with definite morphological features, eliminating dimensional distortion deviations across hardware environments.

[0056] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit of this application and the scope of protection of this invention, and all of these forms are within the protection scope of this application.

Claims

1. A method for processing pathological medical images based on stained medical preparations, characterized in that, Includes the following steps: Step S1: Acquire the original digital image representing the tissue morphology of the sample to be processed through the data acquisition channel, and extract the set of pixels whose absorption spectral response values ​​deviate from the preset standard distribution range in the original digital image as geometric control anchor points. Step S2: Map the geometric control anchor points to the non-uniform rational B-spline curve control points in the computer-aided design logic, and establish an interference penalty term to characterize the geometric penetration risk of adjacent entity boundaries based on the spatial steric hindrance characteristics of biological entities in the sample to be processed. Step S3: Construct an energy functional that includes an interference penalty term, surface tension constraint, and curvature smoothing factor, and perform a gradient descent-based minimum convergence calculation on the energy functional to determine the closed vector spline curves that characterize the boundaries of the cell nucleus entity and the cytoplasm entity in the sample to be processed. Step S4: Obtain the pixel component response vector representing the standard colorimetric features as the material rendering parameter, and perform linear weighted mapping on the pixel component values ​​within the closed geometric region defined by the closed vector spline curve, so as to generate a standardized pathological digital image for medical auxiliary diagnosis and eliminate data noise caused by the concentration fluctuation of stained medical preparations.

2. The method of processing pathological medical images based on a stained medical preparation according to claim 1, characterized in that, The logic for establishing the interference penalty term in step S2 includes: setting a nucleo-cytoplasmic ratio threshold that represents the limit of the ratio between the volume of the cell nucleus and the volume of the cytoplasm; monitoring the geometric trajectory formed by two adjacent nodes in real time during the iterative evolution of the control points of the non-uniform rational B-spline curve; when the geometric trajectory overlaps spatially or the nucleo-cytoplasmic ratio threshold exceeds the preset range of 0.1 to 0.8, calculating the energy gradient representing the spatial exclusive interference, and introducing the energy gradient as an interference penalty term into the energy functional to constrain the topological evolution path of the closed vector spline curve.

3. The method of processing pathological medical images based on a stained medical preparation according to claim 1, characterized in that, The method for extracting geometric control anchor points in step S1 includes: performing spectral feature recognition on the original digital image to locate the coordinates of local pixels whose spectral response intensity is lower than a preset brightness threshold; calculating the gradient change direction of the local pixel coordinates in the local pixel matrix; and marking the local pixel coordinates as control vectors with defined topological weight attributes according to the gradient change direction, thereby forming geometric control anchor points.

4. The method of processing pathological medical images based on a stained medical preparation according to claim 1, characterized in that, Closed vector spline curves possess geometric continuity and are used to fill in the physically missing morphological boundaries in regions of the original digital image where boundary information is missing due to spectral response failure, based on the tension energy minimization constraint of the spline curve.

5. The method of processing pathological medical images based on staining medical formulations according to claim 1, characterized in that, The material rendering parameters include preset standard color component weights for the color rendering characteristics of hematoxylin and eosin.

6. The method of processing pathological medical images based on a stained medical preparation according to claim 1, characterized in that, Closed vector spline curves define the geometric topological skeleton in standardized digital pathological images, making the morphological features of standardized digital pathological images independent of the pH and storage time of stained medical preparations.

7. The method for processing pathological medical images based on stained medical preparations according to claim 2, characterized in that, The penalty weight generated by the interference penalty term during the minimum iterative convergence calculation of the energy functional. Overlapping volume with adjacent entities The following mapping relationship is satisfied: ,in, The penalty weight is λ, which is a preset exclusivity adjustment coefficient. This represents the value of the overlapping volume between adjacent entities, calculated through geometric motion trajectories.

8. The method for processing pathological medical images based on stained medical preparations according to claim 1, characterized in that, In step S4, when performing linear weighted mapping assignment on pixel component values, global color distribution vector assignment is performed on restricted pixels within the closed geometric region defined by the closed vector spline curve to correct color deviation caused by batch differences in stained medical preparations.

9. The method of processing pathological medical images based on staining medical formulations according to claim 1, characterized in that, The method is applied to the automated analysis of pathological images in a medical information system, with standardized digital pathological images serving as the data source for the extraction of tissue morphological features in the medical information system.

10. The method for processing pathological medical images based on stained medical preparations according to claim 1, characterized in that, Closed vector spline curves shift the logical focus of image processing from pixel-level numerical interpolation to organizational-level solid geometry modeling, thereby compensating for data collapse caused by the failure of spectral colorimetric carriers through spatial occupancy exclusive constraints.

Images