A method and system for extracting and classifying and identifying multiple dimensions of white blood cells

Abstract

The present application relates to the technical field of cell feature extraction, and specifically discloses a white blood cell multi-dimensional feature extraction and classification identification method and system. The method comprises: analyzing the shape and extension state of red blood cells to obtain an eccentricity vector of a global push piece; dividing white blood cells into non-polluted targets and potential polluted targets based on the eccentricity vector; dividing into upwind areas and downwind areas, and extracting a side pollution index through optical density comparison; constructing a dynamic physiological tolerance for calibration to obtain a net feature reconstruction sequence; integrating the original features of non-polluted targets and the net feature reconstruction sequence into a non-interference feature group, and outputting a classification identification result. The system comprises a contour collection module, a target separation module, a side analysis module, a feature reconstruction module, and a feature shaping module. The present application is conducive to reducing physical feature pollution caused by one-way diffusion of nuclear substances during the push piece process, and improving the accuracy of white blood cell subtype classification identification by the system.

Description

Technical Field

[0001] This invention relates to the field of cell feature extraction technology, specifically to a method and system for extracting and classifying multidimensional features of white blood cells. Background Technology

[0002] In the field of fully automated digital blood smear analysis, the classification of white blood cells relies on the accurate extraction of cell morphological parameters and optical characteristics. However, in the actual slide preparation process, mechanical shearing forces can cause free red blood cells in the droplets to undergo irreversible physical tailing in the direction of the force, resulting in eccentric deformation.

[0003] Mechanical shearing forces can deform red blood cells and cause partial cell boundary damage, resulting in the free and dispersed internal nuclear material, thus forming abnormally ruptured smear cells in the slide. The free nuclear material inside the ruptured cells will diffuse unidirectionally along the off-center vector direction of the global slide. If a normally intact white blood cell happens to be in the direction of this unidirectional nuclear material diffusion, it will become a potential contamination target. When the cytoplasm of a white blood cell is unidirectionally covered and contaminated by external free nuclear material, its optical boundary will irregularly expand outward, introducing spatial positioning errors.

[0004] Furthermore, the accumulation of contaminants directly alters the optical absorption intensity of contaminated areas, leading to density imbalances between different regions of the cytoplasm. This results in external interference noise in the cytoplasmic geometric area and optical density parameters acquired by the system, interfering with the qualitative analysis of lymphocytes, monocytes, and various granulocytes by the mononuclear cytoplasmic feature model or granule cytoplasmic feature model. This can cause misclassification of secondary subtypes, limiting the overall accuracy and reliability of the leukocyte classification and identification system.

[0005] Therefore, the present invention provides a method and system for extracting and classifying multidimensional features of white blood cells. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for extracting and classifying multidimensional features of white blood cells to solve the aforementioned background problems.

[0007] The objective of this invention can be achieved through the following technical solutions: A method for extracting and classifying multidimensional features of white blood cells includes the following steps: Collect red blood cell contour deformation data across the entire field of view, analyze the morphological extension state of red blood cells in the background based on the contour deformation data, and obtain the global eccentric vector of the slide. The distribution location of ruptured cells was collected and individual white blood cells were identified. A one-way diffusion mapping analysis of white blood cells was performed using eccentric vectors. Based on the mapping analysis results, structurally intact white blood cells in the field of view were divided into uncontaminated targets and potentially contaminated targets. The cytoplasmic topology of potential contaminated targets is deconstructed along the eccentric vector of the global slide, dividing the potential contaminated targets into source-facing regions and back-source regions; by performing optical density comparison analysis between the source-facing and back-source regions, the eccentric contamination index is extracted and generated. A dynamic physiological tolerance is constructed based on the spatial distance attenuation characteristics. The bulk structural parameters of potential contaminated targets are extracted and the physiological tolerance is used for dynamic tolerance calibration. Based on the calibration results, pollution source stripping and spatial reconstruction are performed to obtain the net feature reconstruction sequence.

[0008] Furthermore, the method for analyzing the morphological extension state of the red blood cells is as follows: Obtain the morphological center of the internal region of the contour deformation data, and extract the contour line that passes through the morphological center and has the largest span, which is defined as the deformation principal axis; Along the principal axis of deformation, measure the first end distance and the second end distance from the morphological center to the edges of the two contours, respectively; Calculate the percentage of deviation between the first end distance and the second end distance to obtain the eccentric stretching ratio; Summarize the average eccentricity of all individual red blood cells within the current field of view; Construct a logic for determining eccentric deformation, input the average eccentricity ratio into the logic for determining eccentric deformation, and obtain the determination result of whether mechanical force eccentric deformation exists. If mechanical force eccentric deformation is determined, the deformation direction and eccentric stretching ratio of the abnormal red blood cell are extracted and processed by two-dimensional projection synthesis to obtain the eccentric vector of the global slide.

[0009] Furthermore, the method for performing the two-dimensional projection synthesis process is as follows: The individual red blood cell feature vector is constructed by taking the direction of the individual cell offset and the corresponding eccentric stretching ratio as the length. Project all individual feature vectors along the horizontal and vertical axes of the two-dimensional image to obtain the horizontal and vertical components of each individual feature vector. Calculate the arithmetic mean of all horizontal components and all vertical components respectively, and construct a two-dimensional vector using the obtained horizontal and vertical component mean values ​​as the eccentricity vector of the global push slice.

[0010] Furthermore, the method for performing the unidirectional diffusion mapping analysis is as follows: Using the coordinates of the rupture center of the ruptured smeared cells as the starting point of the geometric projection, and taking the eccentric vector of the slide as the direction of the central axis of the spatial projection, a fan-shaped influence area that diffuses outward in a unidirectional direction is constructed in the two-dimensional image coordinate system. Obtain the geometric centroid coordinates of each individual white blood cell to be tested, and compare the spatial positional inclusion relationship between the geometric centroid coordinates of the individual white blood cells to be tested and the constructed sector-shaped influence area. If the comparison results show that the geometric centroid coordinates of the white blood cell to be tested fall within the fan-shaped influence area, the white blood cell to be tested is determined to be a potential contaminated target; if it does not fall within the area, it is marked as an uncontaminated target.

[0011] Furthermore, the optical density comparison analysis is performed as follows: Extract the mean optical density of the incoming source region and simultaneously extract the mean optical density of the back source region. Perform a difference operation between the mean optical density of the incoming source region and the mean optical density of the back source region, and perform a difference check on the difference operation result to obtain the absolute difference. The absolute difference is normalized by dividing it by the mean optical density of the back source region to obtain the lateral pollution index.

[0012] Furthermore, the method for performing the pollution source stripping and spatial reconstruction process is as follows: Extract the physical edge contour of the back-source region of potential contaminated targets; Using the cell nucleus centroid coordinates as a reference, morphological symmetry mapping is performed using the physical edge contour to construct a reconstructed physical contour that covers the irregular features of the cell edge; The intrinsic texture features of the back source region are extracted and mapped to the front source region space corresponding to the reconstructed physical contour to generate a reconstructed cytoplasmic map. The net feature reconstruction sequence is extracted based on the reconstructed cytoplasmic map. The net feature reconstruction sequence includes: morphological geometric parameters, spatial texture distribution parameters, and optical density distribution parameters.

[0013] Furthermore, the method for performing the reconstructive assembly is as follows: Construct a multidimensional feature array space in computer memory; The ratio of the area of ​​the reconstructed physical contour to the perimeter area is used as the input morphological geometric parameters; The gray-level co-occurrence matrix parameters of the reconstructed cytoplasmic map are used as inputs for the spatial texture distribution parameters; The average optical density and contrast of the reconstructed cytoplasmic map are used as input parameters for optical density distribution. The multidimensional feature vectors are encapsulated to output the net feature reconstruction sequence.

[0014] Furthermore, it also includes: extracting the original features of uncontaminated targets and integrating them with the reconstructed sequences of net features to generate interference-free feature groups; establishing primary nuclear lineage identifiers of interference-free cell feature groups and performing secondary subtype characterization to obtain the final classification and identification results of leukocytes.

[0015] Furthermore, the method for integrating and generating the interference-free feature set is as follows: Extract the sum of the actual total geometric area and the actual total optical density of the cytoplasm of the uncontaminated target as the original feature; The original and net feature reconstruction sequences of uncontaminated targets are encapsulated into standard-length feature vectors according to the same geometric and optical dimension channel order. All standard-length feature vectors in the current field of view are merged to establish a global non-interference feature group.

[0016] A system for extracting and classifying multidimensional features of white blood cells includes the following modules: Contour acquisition module: used to acquire red blood cell contour deformation data within the entire field of view, analyze the red blood cell morphological extension state based on the contour deformation data, and obtain the eccentric vector; Target separation module: used to collect the distribution location of ruptured cells and identify individual white blood cells; combined with eccentric vector to perform unidirectional diffusion mapping analysis on white blood cells, and based on the mapping analysis results, white blood cells are divided into uncontaminated targets and potentially contaminated targets; Lateralization analysis module: used to perform cytoplasmic topology deconstruction of potential contaminated targets along the eccentric vector of the global slide, dividing potential contaminated targets into source-facing regions and back-source-facing regions; by performing optical density comparison analysis between the source-facing and back-source-facing regions, the lateralization contamination index is extracted and generated. Feature reconstruction module: It is used to construct physiological tolerance based on spatial distance attenuation characteristics, extract the ontological structural parameters and physiological tolerance for dynamic tolerance calibration, and perform pollution source stripping and spatial reconstruction processing based on the calibration results to obtain the net feature reconstruction sequence. Feature characterization module: used to extract the original features of uncontaminated targets and integrate them with the reconstructed sequence of net features to generate interference-free feature groups; establish primary nuclear system identifiers and perform secondary subtype characterization to obtain the final classification and recognition results.

[0017] The beneficial effects of this invention are as follows: 1. This invention collects the contour features of free red blood cells in the full field of view, compares the differences in asymmetric extension span to calculate the deformation measurement parameters of the population, and constructs a global slide eccentricity vector through spatial vector synthesis. This is beneficial for using the red blood cell population as a physical medium to record the mechanical shear force of the slide, and for quantifying the hydrodynamic deformation trend during the blood smear preparation process. The system distinguishes potential contamination risk targets from uncontamination targets by comparing geometric positions. This method of defining the specific physical path of free nuclear material penetration using the principles of flow and wind direction is beneficial for targeted identification of contamination populations in the diffusion path and reduces the risk of secondary feature distortion caused by over-processing of originally intact cells.

[0018] 2. After locating the contaminated target, the cytoplasm is topologically divided along the global eccentric vector, with the morphological center of the cell nucleus as the origin, to delineate the source region of the physically attacking nuclear material and the back-source region away from the wind direction. The high density and strong anti-interference properties of the cell nucleus are utilized to establish the cutting benchmark, reducing contour expansion errors. A lateral contamination index is generated by normalizing and comparing the optical density parameters of the two topological halves, quantifying the degree of intracellular photometric imbalance caused by unidirectional nuclear material coverage, and reflecting the magnitude of the impact of spatial directional interference on the distribution of matter within a single cell.

[0019] 3. For contamination targets, by allocating differentiated allowable ranges for targets at different spatial locations according to the physical laws of material diffusion attenuation over distance, the system's ability to resist false positives in contamination detection is enhanced. After confirming substantial contamination, spatial reconstruction based on symmetrical compensation within a safe half-zone helps correct for physical feature deficiencies and core parameter distortions caused by unidirectional permeation. In the final classification stage, a hierarchical structure determination strategy utilizes the physical stability of the cell nucleus as the first-level classification anchor point, converging the feature divergence space for major category determination. Simultaneously, secondary classification is activated based on primary population identifiers, and attribute comparison is performed using reconstructed cytoplasmic interference-free parameters, which helps improve the reliability of classifying polymorphic granulocytes and monocytes. Attached Figure Description

[0020] The invention will now be further described with reference to the accompanying drawings.

[0021] Figure 1 This is a flowchart of a method for extracting and classifying multidimensional features of white blood cells according to the present invention; Figure 2 This is a flowchart of the logic for constructing the eccentric deformation determination in this invention; Figure 3 This is a functional module diagram of a white blood cell multidimensional feature extraction and classification system in this invention. Detailed Implementation

[0022] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments. Example 1:

[0023] like Figure 1 As shown, a method for extracting and classifying multidimensional features of white blood cells includes the following steps: S10. Collect red blood cell contour deformation data within the entire field of view, analyze the morphological extension state of red blood cells in the background based on the contour deformation data, and obtain the global eccentric vector of the slide. The method for collecting red blood cell contour deformation data across the entire field of view is as follows: In some embodiments, a full-field digital blood smear image obtained by scanning with a microscopic device is acquired, and background segmentation and connected component analysis are performed on the blood smear image to extract the outer edge closed curves of all independent suspended objects in the image. From all the outer edge closed curves, impurities and white blood cell curves that do not conform to the physical morphological characteristics of human free red blood cells in terms of area and roundness are screened out, and the remaining outer edge closed curves are used as the initial red blood cell contour deformation data. Preferably, the method for screening out impurities and white blood cell curves is as follows: a standard red blood cell area reference range and a roundness reference lower limit are set, and closed curves whose enclosing area exceeds the area reference range or whose edge roundness is lower than the roundness reference lower limit are removed; the removal operation is used to eliminate the morphological interference of overlapping red blood cell groups, giant platelets and various white blood cells, so that the extracted red blood cell contour deformation data comes from a single layer of free red blood cells; Among them, the method of analyzing the morphological extension state of red blood cells in the background field of view based on contour deformation data to obtain the global slide offset vector is as follows: For each retained profile deformation data, obtain the geometric centroid coordinates of the internal region of the profile deformation data; Then, on the closed curve corresponding to the outer edge of the red blood cell, find the line connecting two points that pass through the geometric centroid coordinates and maximize the line segment span, and define it as the deformation major axis of the red blood cell; Along the deformation major axis, measure the distances from the intersection points of the two ends of the outer edge closed curve to the geometric centroid coordinates to obtain the first end distance and the second end distance; It should be noted that the value of the first end distance is higher than that of the second end distance; Calculate the difference between the first end distance and the second end distance, and divide the difference by the distance value of the shorter end to obtain the eccentric stretching ratio that reflects the degree of tailing on one side. Simultaneously extract the ray direction of the longer end as the single offset direction of the red blood cell; Summarize the eccentric stretching ratios of all individual red blood cells within the current field of view, and calculate the statistically average eccentricity ratio. Construct a logic for determining eccentric deformation, input the average eccentricity ratio into the logic for determining eccentric deformation, and obtain the determination result of whether mechanical force eccentric deformation exists during the film production process; Preferred, such as Figure 2 As shown, the method for constructing the judgment logic for eccentric deformation is as follows: If the average eccentricity ratio is less than or equal to the static smear background threshold, it indicates that the deformation of the local edge belongs to the normal physiological biconcave disk state characteristics of the cell itself or the imaging background noise. It is determined that the red blood cells in the background of the field of view maintain a natural resting state as a whole, and the risk of mechanical eccentric deformation is low. The system terminates the subsequent vector extraction operation. It should be noted that the static smear background threshold is determined as follows: the eccentric stretching ratio of all normal red blood cells in a large number of uncontaminated resting blood smear samples is extracted in advance, the statistical mean and standard deviation of the eccentric stretching ratio of normal red blood cells are calculated, and the statistical mean is added to the standard deviation by a specified multiple (e.g., three times) and set as the static smear background threshold. Conversely, if the average eccentricity ratio is greater than the static smear background threshold, it indicates that during the slide preparation operation of blood smear, mechanical shear force causes the red blood cells in the droplets to undergo irreversible physical tailing in the direction of force, confirming the existence of mechanical force eccentric deformation. If mechanical force eccentric deformation is confirmed, all abnormal red blood cells with an eccentric stretching ratio greater than the background threshold of the static smear are retrieved in reverse. Extract the monomer offset direction and eccentric stretching ratio corresponding to the individual abnormal red blood cells; The extracted individual unit offset directions and eccentric stretching ratios are subjected to two-dimensional projection synthesis to obtain the global push-piece eccentric vector. For example, the method for performing two-dimensional projection compositing is as follows: The individual red blood cell feature vector is constructed by taking the direction of the individual cell offset and the corresponding eccentric stretching ratio as the length. Project all individual feature vectors along the horizontal and vertical axes of the two-dimensional image to obtain the horizontal and vertical components of each individual feature vector. Each horizontal component and each vertical component is summed independently. The sum of the total horizontal and vertical components is divided by the total number of abnormal red blood cells involved in the calculation to obtain the mean of the horizontal and vertical components. A two-dimensional vector is constructed by combining the mean of the horizontal and vertical components, and the two-dimensional vector is marked as the eccentric vector of the global slide.

[0024] S20. Collect the distribution location of ruptured cells and identify individual white blood cells; combine the eccentric vector to perform unidirectional diffusion mapping analysis on white blood cells, and based on the mapping analysis results, classify the structurally intact white blood cells in the field of view into uncontaminated targets and potentially contaminated targets; The process of collecting the distribution location of ruptured cells and identifying individual white blood cells is as follows: In some embodiments, the outline integrity of nucleated cells in full-view digital blood smear images is screened to identify abnormal cell individuals with damaged cell boundaries and internal nuclear material in a free and dispersed state. Abnormal cells were marked as ruptured smeared cells, and the coordinates of the rupture center of the ruptured smeared cells were extracted; It should be noted that connected region grayscale detection is performed on the internal region of the ruptured smeared cell, and the residual main nucleus region with the lowest grayscale value, which is the region with the highest nuclear material density, is extracted. The geometric centroid of the residual main nucleus region is calculated and defined as the rupture center coordinate of the ruptured smeared cell. Simultaneously identify white blood cells with closed cell boundaries and normal morphology in full-view digital blood smear images, and mark the white blood cells with normal morphology as white blood cells to be tested; Preferably, the method for identifying ruptured smeared cells and individual white blood cells to be tested is: to physically distinguish them by detecting whether there are continuous closed cell membrane boundary lines around the nucleated cells; Nucleated cells with continuous closed boundary lines are classified as individual white blood cells to be tested, while nucleated cells with broken or defective boundary lines are classified as ruptured smear cells. Among them, the method of combining eccentric vector to perform unidirectional diffusion mapping analysis on leukocytes, and classifying structurally intact leukocytes in the field of view into uncontaminated targets and potentially contaminated targets based on the mapping analysis results is as follows: Using the rupture center coordinates of the ruptured smeared cells as the starting point of the geometric projection, and taking the global eccentric vector of the smear extracted by S10 as the direction of the central axis of the spatial projection, a fan-shaped influence area that diffuses outward in a unidirectional direction is constructed in the two-dimensional image coordinate system. Preferably, the method for constructing the fan-shaped influence region is as follows: With the fracture center coordinates as the center of the sector, the preset maximum dispersion distance of the nuclear material as the radius of the sector, and the line where the global pusher eccentric vector is located as the central axis of the sector, a preset diffusion angle is expanded on both sides along the central axis of the sector. The two-dimensional geometric space formed by the sector radius and the preset diffusion angle is the sector influence area. It should be noted that the method for establishing the sector radius and the preset diffusion angle is as follows: the preset maximum diffusion distance of nuclear material is determined based on the physical empirical values ​​of the standard blood drop volume and the slide pushing rate during smear preparation. The spatial resolution is calculated by combining the objective magnification of the current microscope equipment and the pixel size of the camera sensor. The theoretical physical diffusion length is then converted into the corresponding pixel length value using the spatial resolution as the sector radius. The preset diffusion angle is set between 30 and 45 degrees. The purpose of setting the preset diffusion angle is to cover the range of lateral dispersion of nuclear material caused by friction at the edge of the pusher fluid, thereby reducing the omission of edge contamination. Obtain the geometric centroid coordinates of each individual white blood cell to be tested, compare the spatial positional inclusion relationship between the geometric centroid coordinates of the individual white blood cells to be tested and the constructed sector-shaped influence area, and perform unidirectional diffusion mapping analysis. If the geometric centroid coordinates of the white blood cell to be tested fall within the fan-shaped influence area, it is determined that the white blood cell to be tested is in the physical wind direction of unidirectional nuclear material infiltration, and the white blood cell to be tested falling within the fan-shaped influence area is marked as a potential contaminated target. Conversely, if the geometric centroid coordinates of the individual white blood cell to be tested do not fall within the fan-shaped influence area, it is determined that the individual white blood cell to be tested is not contaminated by nuclear material covering the physical path, and the individual white blood cell to be tested that does not fall within the fan-shaped influence area is marked as an uncontaminated target. Example 2:

[0025] Please see Figure 1 As shown, a method for extracting and classifying multidimensional features of white blood cells includes the following steps: S30. Perform cytoplasmic topological decontamination of potential contaminated targets along the eccentric vector of the global slide, and divide potential contaminated targets into source-facing regions and back-source regions; extract and generate lateral contamination index by performing optical density comparison analysis between source-facing regions and back-source regions. The process of performing cytoplasmic topological decontamination of potential contaminated targets along the global slide offset vector, dividing the potential contaminated targets into source-facing and back-source regions, is as follows: In some embodiments, the nuclear outline boundary inside the potentially contaminated target is extracted, and the coordinates of the centroid of the cell formed by the nuclear outline boundary are calculated. Using the cell nucleus centroid coordinates as the spatial origin, a two-dimensional straight line is constructed that is perpendicular to the line containing the global slide eccentricity vector, and the two-dimensional straight line that is perpendicular to each other is defined as the spatial segmentation baseline. The cytoplasmic region of the potential contaminated target is divided into two topological halves located on either side of the spatial segmentation baseline using a spatial segmentation baseline. It should be noted that the reason for choosing the cell nucleus centroid coordinates as the cutting origin rather than the cell-wide centroid coordinates is that when the cytoplasm is unidirectionally covered and contaminated by nuclear material, the optical boundary of the cytoplasm will expand irregularly. If the cell-wide centroid coordinates are used, it will introduce spatial positioning errors. However, the cell nucleus has high density and is not easily affected by external diffused substances. Using the cell nucleus centroid coordinates as the origin can ensure that the spatial segmentation baseline accurately passes through the cell's most real physical center of symmetry. By performing directional mapping analysis between the two topological half-regions obtained from the cutting and the global push-piece eccentricity vector, the oncoming source region and the back source region are obtained; Preferably, the method for performing direction mapping analysis is as follows: The topological half-region located on the opposite side of the global pusher eccentricity vector is identified as the disaster surface that directly encounters the nuclear material dispersion path, and the topological half-region located on the opposite side of the global pusher eccentricity vector is marked as the source-facing region. Simultaneously, the topological half-region located on the positive side of the global push-film eccentricity vector is identified as a physical shielding surface that deviates from the dispersion wind direction, and the topological half-region located on the positive side of the global push-film eccentricity vector is marked as the back source region. The method for extracting and generating the lateral pollution index by comparing the optical density of the incoming and outgoing source regions is as follows: Extract the digital grayscale values ​​of all pixels within the source region, and convert the digital grayscale values ​​into equivalent light absorption intensity values ​​by querying the photometric response lookup table dictionary stored in the system; calculate the arithmetic mean of all light absorption intensity values ​​within the source region to obtain the average optical density of the source region. Among them, the photometric response lookup dictionary is a one-dimensional mapping array that is pre-formed by calculating the optical density value corresponding to the digital grayscale step based on the principle of optical transmission and then solidifying it. Simultaneously extract the light absorption intensity values ​​of all pixels inside the back source region, calculate the arithmetic mean of the light absorption intensity values ​​of all pixels inside the back source region, and calculate the average optical density of the back source region. The mean optical density of the incoming source area is subtracted from the mean optical density of the outgoing source area. If the result is positive, the positive number is extracted as the effective difference. If the result is zero or negative, it is determined that no one-way pollution has occurred, and the effective difference is forcibly set to zero. The obtained absolute difference was normalized by dividing it by the mean optical density of the back source region, and a dimensionless characteristic parameter reflecting the density imbalance ratio on both sides of the cytoplasm was calculated. The calculated dimensionless characteristic parameters are extracted and generated as a biased pollution index.

[0026] S40. Based on the spatial distance attenuation characteristics, a dynamic physiological tolerance is constructed, and the bulk structural parameters and physiological tolerance of the potential contaminated targets are extracted for dynamic tolerance calibration. Based on the calibration results, pollution source stripping and spatial reconstruction are performed to obtain the net feature reconstruction sequence. The process of constructing a dynamic physiological tolerance based on spatial distance attenuation characteristics and extracting the bulk structural parameters of potential contaminated targets is as follows: Preferably, the physical pixel distance between the geometric centroid coordinates of the potential contaminated target and the rupture center coordinates of the ruptured smeared cells is extracted and marked as the diffusion interval distance; Simultaneously extract the nuclear area features of potential contaminated targets determined in step S30 and label them as ontological structural parameters; Based on the spatial distance attenuation characteristics, a dynamic physiological tolerance is constructed to obtain a preset tolerance logic matrix; The dispersion interval distance and the body structure parameters are input into the tolerance logic matrix, and the real-time physiological steady-state tolerance upper limit corresponding to the potential contaminated target is obtained by retrieval and matching. The method for dynamically constructing physiological tolerance is as follows: S401. Based on prior statistical analysis of a large database of uncontaminated resting blood smear samples, the distribution characteristics of the optical density imbalance ratio of normal white blood cells are obtained as a statistical baseline. Multiple distance intervals from near to far are predefined, and a basic tolerance benchmark value is assigned to each distance interval based on the statistical baseline. The farther the physical distance represented by the distance interval, the lower the nuclear material diffusion concentration, and the larger the assigned basic tolerance benchmark value. A spatial lookup table dictionary with one-to-one correspondence between distance intervals and basic tolerance benchmark values ​​is generated to complete the establishment of positive correlation mapping relationship. S402. Predefine multiple core area intervals from small to large, assign a structural compensation weight to each core area interval. The larger the area value represented by the core area interval, the smaller the assigned structural compensation weight. Generate a structural lookup table dictionary that corresponds one-to-one with the core area intervals and structural compensation weights, and complete the establishment of the negative correlation mapping relationship. S403. After obtaining the dispersion interval distance and the body structure parameters, perform interval addressing in the spatial lookup dictionary and the structural lookup dictionary respectively, and extract the basic tolerance benchmark value and structural compensation weight of the address hit. S404. Perform a scalar product operation on the extracted basic tolerance benchmark value and the structural compensation weight. Use the value obtained from the product operation to form the output result of the tolerance logic matrix, and use it as the upper limit of the real-time physiological steady-state tolerance for the dynamic drift of the specific white blood cell individual, thus completing the adaptive calibration of the dynamic tolerance threshold. The lateral contamination index calculated in step S30 for the potential contaminated target is used to determine the out-of-bounds state by matching the upper limit of the real-time physiological steady-state tolerance. If the lateral contamination index is within the coverage range of the real-time physiological steady-state tolerance limit, the potential contaminated target is determined to be in a safe redundancy state. The original global geometric parameters and global optical parameters of the potential contaminated target are directly extracted as the uncontaminated net feature reconstruction sequence, and the subsequent stripping action is terminated. Conversely, if the unilateral contamination index exceeds the upper limit of the real-time physiological homeostasis tolerance, it confirms that the potential contaminated target has undergone substantial one-way infiltration of nuclear material, triggering a contamination source stripping signal. The method for obtaining the net feature reconstruction sequence by dynamically calibrating the ontological structural parameters and physiological tolerances, and then performing pollution source stripping and spatial reconstruction based on the calibration results is as follows: After triggering the pollution source stripping signal, the system extracts the physical edge contour of the back source area of ​​the potential contaminated target. It should be noted that the outer physical edge contour includes the true physiological morphology of the side of the leukocyte not covered by nuclear material during the slide preparation process (such as pseudopodia and the slight undulations of the cytoplasmic edge). Using the spatial segmentation baseline determined in S30 as the axis of symmetry, the physical edge contour of the back source region is projected onto the front source region side; The projected edges are smoothly connected by a nonlinear interpolation algorithm to construct a reconstructed physical contour that conforms to the physical laws of symmetry and the irregular edge features unique to white blood cells, thus achieving morphological symmetry mapping processing. The system extracts the gray-level co-occurrence matrix, local contrast, and grain distribution density of pixels inside the back source region, which are defined as intrinsic texture features. Using existing texture synthesis algorithms, the texture of the back source region is used as a sample to fill the space in the front source region that has been stripped (set to zero) to obtain a reconstructed cytoplasmic map. Multidimensional feature extraction was performed based on the reconstructed cytoplasmic map to obtain the net feature reconstruction sequence; Preferably, the method for obtaining the net feature reconstruction sequence by performing multidimensional feature extraction based on the reconstructed cytoplasmic map is as follows: S411. Construct a multidimensional blank feature array space with geometric dimension, photometric dimension and texture dimension in computer memory; S412. Extract the actual total geometric area of ​​the reconstructed physical contour, and calculate the numerical ratio of perimeter to area to obtain the perimeter-area ratio, which is then input into the geometric dimension channel as a morphological feature. S413. Extract the gray-level co-occurrence matrix and average optical density of the reconstructed cytoplasm map, and input them into the texture dimension channel and optical dimension channel, respectively. The gray-level co-occurrence matrix includes the second angular moment, entropy, and contrast. S414. Serialize and encapsulate the above channel data, output a feature array of a one-dimensional data structure, and define it as the final net feature reconstruction sequence.

[0027] S50. Extract the original features of uncontaminated targets and integrate them with the reconstructed sequences of net features to generate interference-free feature groups; establish the primary nuclear lineage identifiers of the interference-free cell feature groups and perform secondary subtype characterization to obtain the final classification and identification results of leukocytes; The process of extracting the original features of uncontaminated targets and integrating them with the net feature reconstruction sequence to generate an interference-free feature set is as follows: Preferably, the sum of the actual total geometric area and the actual total optical density of the cytoplasm of the uncontaminated target is extracted and defined as the original features of the uncontaminated target. The original and net feature reconstruction sequences of uncontaminated targets are uniformly encapsulated into standard-length feature vectors according to the same geometric and optical dimension channel order. All standard-length feature vectors in the current field of view are merged to establish a global non-interference feature group. The method for establishing primary nuclear lineage identifiers for interference-free cell characteristic groups and performing secondary subtype characterization to obtain the final classification and identification results of leukocytes is as follows: Obtain the physical outline of the nucleus for each individual white blood cell, calculate the area enclosed by the physical outline of the nucleus and the roundness of its edges, and define them as the morphological parameters of the nucleus. Extract the grayscale standard deviation and contrast of all pixels inside the cell nucleus, and define them as the texture parameters inside the cell nucleus. Independent nuclear morphological parameters and internal nuclear texture parameters are extracted independently to construct an independent nuclear feature subset. This independent nuclear feature subset is then input into a pre-trained primary nuclear morphology classification network for topology determination. It should be noted that the primary nuclear morphology classification network is an artificial neural network model or support vector machine model generated by supervised training using a set of white blood cell nuclear feature samples with known single-nucleus or multi-nucleus labels. Based on the output of the topology determination, white blood cells are divided into mononuclear cell populations or multinucleated cell populations, and assigned the corresponding primary nuclear lineage population identifiers. After generating the primary nuclear lineage identifier, geometric and optical features representing cytoplasmic properties are extracted from the non-interference feature set to construct an independent subset of cytoplasmic features. Read the population attributes of the primary nuclear lineage population identifier, and activate the matching specific cytoplasmic sub-model from the preset classification model library based on the population attributes; The classification model library is a set of multiple pre-trained classifiers pre-configured and stored in a computer-readable storage medium; the specific cytoplasmic classification sub-model is a random forest model or artificial neural network model pre-trained using a sample set of cytoplasmic features of a specific population with known cell subtype labels. If the primary nuclear lineage population identifier indicates a mononuclear cell population, then the mononuclear cytoplasmic sub-model is activated; if it indicates a multinucleated cell population, then the granular cytoplasmic sub-model is activated. A subset of independent cytoplasmic features is input into a specific cytoplasmic classification sub-model that is activated for pattern comparison. The specific cytoplasmic classification sub-model outputs a specific white blood cell subtype category label, which serves as the final classification and identification result of the white blood cells. Example 3:

[0028] Please see Figure 3 As shown, a system for extracting and classifying leukocyte multidimensional features includes the following modules: Contour acquisition module: used to acquire red blood cell contour deformation data in the entire field of view, analyze the morphological extension state of red blood cells in the background of the field of view based on the contour deformation data, and obtain the global eccentric vector of the slide. Target separation module: used to collect the distribution location of ruptured cells and identify individual white blood cells; combined with eccentric vector to perform unidirectional diffusion mapping analysis on white blood cells, and based on the mapping analysis results, divide the structurally intact white blood cells in the field of view into uncontaminated targets and potentially contaminated targets; Lateralization analysis module: used to perform cytoplasmic topology deconstruction of potential contaminated targets along the eccentric vector of the global slide, dividing potential contaminated targets into source-facing regions and back-source-facing regions; by performing optical density comparison analysis between the source-facing and back-source-facing regions, the lateralization contamination index is extracted and generated. Feature reconstruction module: It is used to construct dynamic physiological tolerance based on spatial distance attenuation characteristics, extract the bulk structural parameters of potential contaminated targets and perform dynamic tolerance calibration with physiological tolerance, and perform pollution source stripping and spatial reconstruction processing based on the calibration results to obtain net feature reconstruction sequence. Feature characterization module: used to extract the original features of uncontaminated targets and integrate them with the net feature reconstruction sequence to generate interference-free feature groups; establish the primary nuclear lineage identifier of the interference-free cell feature group and perform secondary subtype characterization to obtain the final classification and identification results of leukocytes.

[0029] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A method for extracting and classifying multi-dimensional features of white blood cells, characterized in that, Includes the following steps: Collect red blood cell contour deformation data across the entire field of view, analyze the morphological extension state of red blood cells in the background based on the contour deformation data, and obtain the global eccentric vector of the slide. Collect data on the distribution location of ruptured cells and identify individual white blood cells; By combining eccentric vectors to perform unidirectional diffusion mapping analysis on leukocytes, and based on the mapping analysis results, leukocytes with intact structures in the field of view are divided into uncontaminated targets and potentially contaminated targets. The cytoplasmic topology of potential contaminated targets is deconstructed along the eccentric vector of the global slide, dividing the potential contaminated targets into source-facing regions and back-source regions; by performing optical density comparison analysis between the source-facing and back-source regions, the eccentric contamination index is extracted and generated. A dynamic physiological tolerance is constructed based on the spatial distance attenuation characteristics. The bulk structural parameters of potential contaminated targets are extracted and the physiological tolerance is used for dynamic tolerance calibration. Based on the calibration results, pollution source stripping and spatial reconstruction are performed to obtain the net feature reconstruction sequence.

2. The method for extracting and classifying multi-dimensional features of white blood cells according to claim 1, characterized in that: The method for analyzing the morphological extension state of the red blood cells is as follows: Obtain the morphological center of the internal region of the contour deformation data, and extract the contour line that passes through the morphological center and has the largest span, which is defined as the deformation principal axis; Along the principal axis of deformation, measure the first end distance and the second end distance from the morphological center to the edges of the two contours, respectively; Calculate the percentage of deviation between the first end distance and the second end distance to obtain the eccentric stretching ratio; Summarize the average eccentricity of all individual red blood cells within the current field of view; Construct a logic for determining eccentric deformation, input the average eccentricity ratio into the logic for determining eccentric deformation, and obtain the determination result of whether mechanical force eccentric deformation exists. If mechanical force eccentric deformation is determined, the deformation direction and eccentric stretching ratio of the abnormal red blood cell are extracted and processed by two-dimensional projection synthesis to obtain the eccentric vector of the global slide.

3. The method for extracting and classifying multidimensional features of white blood cells according to claim 2, characterized in that: The method for performing the two-dimensional projection synthesis process is as follows: The individual red blood cell feature vector is constructed by taking the direction of the individual cell offset and the corresponding eccentric stretching ratio as the length. Project all individual feature vectors along the horizontal and vertical axes of the two-dimensional image to obtain the horizontal and vertical components of each individual feature vector. Calculate the arithmetic mean of all horizontal components and all vertical components respectively, and construct a two-dimensional vector using the obtained horizontal and vertical component mean values ​​as the eccentricity vector of the global push slice.

4. The method for extracting and classifying multi-dimensional features of white blood cells according to claim 1, characterized in that: The method for performing the unidirectional diffusion mapping analysis is as follows: Using the coordinates of the rupture center of the ruptured smeared cells as the starting point of the geometric projection, and taking the eccentric vector of the slide as the direction of the central axis of the spatial projection, a fan-shaped influence area that diffuses outward in a unidirectional direction is constructed in the two-dimensional image coordinate system. Obtain the geometric centroid coordinates of each individual white blood cell to be tested, and compare the spatial positional inclusion relationship between the geometric centroid coordinates of the individual white blood cells to be tested and the constructed sector-shaped influence area. If the comparison results show that the geometric centroid coordinates of the white blood cell to be tested fall within the fan-shaped influence area, the white blood cell to be tested is determined to be a potential contaminated target; if it does not fall within the area, it is marked as an uncontaminated target.

5. The method for extracting and classifying multidimensional features of leukocytes according to claim 1, characterized in that: The optical density comparison analysis is performed as follows: Extract the mean optical density of the incoming source region and simultaneously extract the mean optical density of the back source region. Perform a difference operation between the mean optical density of the incoming source region and the mean optical density of the back source region, and perform a difference check on the difference operation result to obtain the absolute difference. The absolute difference is normalized by dividing it by the mean optical density of the back source region to obtain the lateral pollution index.

6. The method for extracting and classifying multidimensional features of leukocytes according to claim 1, characterized in that: The method for performing the pollution source stripping and spatial reconstruction process is as follows: Extract the physical edge contour of the back-source region of potential contaminated targets; Using the cell nucleus centroid coordinates as a reference, morphological symmetry mapping is performed using the physical edge contour to construct a reconstructed physical contour that covers the irregular features of the cell edge; The intrinsic texture features of the back source region are extracted and mapped to the front source region space corresponding to the reconstructed physical contour to generate a reconstructed cytoplasmic map. The net feature reconstruction sequence is extracted based on the reconstructed cytoplasmic map. The net feature reconstruction sequence includes: morphological geometric parameters, spatial texture distribution parameters, and optical density distribution parameters.

7. The method for extracting and classifying multi-dimensional features of leukocytes according to claim 6, characterized in that, The method for performing the reconstructive assembly is as follows: Construct a multidimensional feature array space in computer memory; The ratio of the area of ​​the reconstructed physical contour to the perimeter area is used as the input morphological geometric parameters; The gray-level co-occurrence matrix parameters of the reconstructed cytoplasmic map are used as inputs for the spatial texture distribution parameters; The average optical density and contrast of the reconstructed cytoplasmic map are used as input parameters for optical density distribution. The multidimensional feature vectors are encapsulated to output the net feature reconstruction sequence.

8. The method for extracting and classifying multidimensional features of leukocytes according to claim 1, characterized in that: Also includes: Extract the original features of uncontaminated targets and integrate them with the net feature reconstruction sequence to generate an interference-free feature set; Primary nuclear lineage identifiers of interference-free cell characteristic groups were established and secondary subtypes were identified to obtain the final classification and identification results of leukocytes.

9. The method for extracting and classifying multidimensional features of leukocytes according to claim 8, characterized in that: The method for integrating and generating the interference-free feature set is as follows: Extract the sum of the actual total geometric area and the actual total optical density of the cytoplasm of the uncontaminated target as the original feature; The original and net feature reconstruction sequences of uncontaminated targets are encapsulated into standard-length feature vectors according to the same geometric and optical dimension channel order. All standard-length feature vectors in the current field of view are merged to establish a global non-interference feature group.

10. A system for extracting and classifying multi-dimensional features of white blood cells, used to implement the method for extracting and classifying multi-dimensional features of white blood cells as described in any one of claims 1-9, characterized in that, Includes the following modules: Contour acquisition module: used to acquire red blood cell contour deformation data within the entire field of view, analyze the red blood cell morphological extension state based on the contour deformation data, and obtain the eccentric vector; Target separation module: used to collect the distribution location of ruptured cells and identify individual white blood cells; By combining eccentric vector analysis to perform unidirectional diffusion mapping analysis on leukocytes, leukocytes are divided into uncontaminated targets and potentially contaminated targets based on the mapping analysis results. Lateralization analysis module: used to perform cytoplasmic topology deconstruction of potential contaminated targets along the eccentric vector of the global slide, dividing potential contaminated targets into source-facing regions and back-source-facing regions; by performing optical density comparison analysis between the source-facing and back-source-facing regions, the lateralization contamination index is extracted and generated. Feature reconstruction module: It is used to construct physiological tolerance based on spatial distance attenuation characteristics, extract the ontological structural parameters and physiological tolerance for dynamic tolerance calibration, and perform pollution source stripping and spatial reconstruction processing based on the calibration results to obtain the net feature reconstruction sequence. Feature characterization module: used to extract the original features of uncontaminated targets and integrate them with the reconstructed sequence of net features to generate interference-free feature groups; establish primary nuclear system identifiers and perform secondary subtype characterization to obtain the final classification and recognition results.

Images