Abnormal cell screening method and system based on image analysis

Abstract

The application relates to the technical field of image processing, in particular to an abnormal cell screening method and system based on image analysis, which comprises the following steps: acquiring continuous image frames of a staining process, extracting average optical density, integral optical density and chromatin gray standard deviation of sample cells as staining parameters, and acquiring chromatin texture; forming a staining parameter space curve with the change of the staining parameters over time, fitting to obtain a continuous staining curve; aligning and fitting the continuous staining curve with a standard parameter curve calibrated in advance to obtain a coordinate transformation vector; acquiring an image frame after the completion of the staining, performing standardization correction on the staining parameters of the to-be-tested cells by using the coordinate transformation vector to obtain chromatin texture features, and then performing feature comparison and identification on abnormal cells. The application reduces the staining sensitivity difference caused by biological factors such as individual gene expression difference, different cell membrane protein compositions and pH value changes, and eliminates the influence of staining condition difference and equipment difference among different laboratories on the screening result.

Description

Image Analysis-Based Abnormal Cell Screening Method and System Technical Field

[0001] This invention relates to the field of image processing technology, specifically to an abnormal cell screening method and system based on image analysis. Background Technology

[0002] Cell image screening is an effective method of observing and analyzing cell morphology under a microscope. Sometimes, to facilitate screening, the differences in the properties of various cell types are used to differentiate cells through staining, thereby helping to identify target cells. Traditional Papanicolaou smears have a false negative rate as high as 5%-15%. The issue of staining standardization is a factor affecting the accuracy of cell screening. Even under the same staining conditions, staining differences still occur between cells from different individuals. This is mainly due to differences in gene expression among individuals, leading to variations in the types and quantities of cell membrane proteins. For example, major histocompatibility complex (MHC) proteins show significant differences between individuals, affecting the binding of staining agents. Differences in phospholipid composition and protein distribution in the cell membranes of different individuals result in varying abilities to absorb staining agents. Intracellular pH affects the charge of staining molecules, thus affecting their binding ability to cellular components. Therefore, it is necessary to process stained cell slide images to reduce individual staining errors and improve the accuracy of cell screening. Summary of the Invention

[0003] (1) Technical problems to be solved

[0004] The purpose of this invention is to provide an abnormal cell screening method and system based on image analysis, so as to solve the problem that the cell screening process is inaccurate due to the difference in staining characteristics caused by individual cell differences.

[0005] (2) Technical solution

[0006] To achieve the above objectives, in one aspect, the present invention provides an abnormal cell screening method based on image analysis, the method comprising:

[0007] Acquire continuous image frames of the cell slice to be tested during the staining process, identify and extract a preset number of sample cells from the image frames, record the average optical density, integrated optical density and chromatin gray standard deviation of the sample cells at different time points and take the corresponding mean as staining parameters, and obtain the corresponding chromatin texture based on the gray histogram of the sample cells.

[0008] The values ​​of the staining parameters of the sample cells as a function of time are projected onto a spatial coordinate system, and the projection points at different time points are connected to form a staining parameter space curve. The staining parameter space curve is fitted to obtain a continuous staining curve. The continuous staining curve is aligned and fitted with a pre-calibrated standard parameter curve, and the coordinate transformation vector of the alignment and fitting is obtained.

[0009] The image frames of the completed staining process are acquired, the staining parameters of all cells to be tested in the field of view are extracted, the extracted staining parameters are standardized and corrected using the coordinate transformation vector to obtain the chromatin texture features at the corresponding time, the corrected chromatin texture features are compared with the preset abnormal cell feature threshold, and when the number of feature dimensions exceeding the threshold reaches a predetermined value, they are marked as abnormal cells.

[0010] Furthermore, the method for recording the average optical density, integrated optical density, and chromatin gray standard deviation of the sample cells at different time points and taking the corresponding mean as the staining parameter includes:

[0011] In each sample cell At the point of time Average optical density Calculated as , For sample cells The number of pixels, For the first The grayscale value of each pixel; calculate the integrated optical density. chromatin gray standard deviation

[0012] For each time point Calculate the mean of the average optical density of all sample cells at this time point. Mean of integrated optical density Mean of standard deviation of chromatin gray level , This represents the number of cells in the sample.

[0013] Furthermore, the method for obtaining the corresponding chromatin texture based on the grayscale histogram of the sample cells includes:

[0014] For each sample cell, calculate its gray-level histogram. ,in Gray levels; from the histogram Extracting chromatin texture features, including histogram entropy. Histogram energy Histogram contrast ; denote chromatin texture as a vector

[0015] Further, the method of fitting the staining parameter space curve to obtain a continuous staining curve; aligning and fitting the continuous staining curve with a pre-calibrated standard parameter curve, and obtaining the coordinate transformation vector of the alignment fit includes:

[0016] For each coloring parameter type, the point set of the coloring parameter space curve To perform a fit, use a fitting polynomial function. By minimizing the sum of squared errors Coefficient of determination , thus obtaining the continuous staining curve ;

[0017] Continuous coloring curve Compared with standard parameter curve Perform alignment fitting, from Medium distance selection A set of points ,from Medium distance selection A set of points ;

[0018] By translation vector and scaling factor Point set Transform into ;

[0019] Compute point set and Sum of squared Euclidean distances between ;

[0020] By solving Seek The coordinate transformation vector is denoted as including... .

[0021] Furthermore, the method for obtaining the pre-calibrated standard parameter curve includes:

[0022] Acquire continuous image frames of known normal cells during the staining process, identify and extract a preset number of sample cells based on the image frames, record the average optical density, integrated optical density and chromatin gray standard deviation of the sample cells at different time points, and take the corresponding mean as the staining parameters, project the staining parameters into a spatial coordinate system to form a staining parameter space curve, and fit the staining parameter space curve to obtain the standard parameter curve.

[0023] Furthermore, the method for standardizing and correcting the extracted staining parameters using the coordinate transformation vector to obtain the chromatin texture features at the corresponding time step includes:

[0024] For each cell tested, its average optical density curve over time was extracted. Integral optical density curve chromatin gray standard deviation curve ;

[0025] Scaling factor in the coordinate transformation vector obtained from sample cells Horizontal displacement Calculate the corrected average optical density of the cells to be tested. Integrated optical density after calibration of the target cells Standard deviation of chromatin grayscale after correction of the test cells The chromatin texture features of the cells under test are represented as vectors. .

[0026] Further, the method of comparing the corrected chromatin texture features with a preset abnormal cell feature threshold, and marking a cell as an abnormal cell when the number of feature dimensions exceeding the threshold reaches a predetermined value, includes:

[0027] The corrected chromatin texture features for each test cell are vectors. ; with the preset abnormal cell feature threshold as a vector At a preset designated time of and Feature comparison is performed, and if the cell meets the characteristics of an abnormal cell, it is marked as an abnormal cell.

[0028] Based on the same inventive concept, in another aspect, the present invention also provides an image analysis-based method for screening abnormal cells, for performing any of the image analysis-based methods for screening abnormal cells.

[0029] (3) Beneficial effects

[0030] Compared with existing technologies, the beneficial effects of this invention are that it reduces the differences in staining sensitivity caused by biological factors such as individual gene expression differences, different cell membrane protein compositions, and pH changes, and eliminates the impact of differences in staining conditions and equipment between different laboratories on screening results. Attached Figure Description

[0031] Figure 1 is a flowchart of the abnormal cell screening method based on image analysis according to Embodiment 1 of the present invention;

[0032] Figure 2 is a schematic diagram of the steps of the abnormal cell screening method based on image analysis in Embodiment 1 of the present invention. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] Before providing examples, it's necessary to explain the application scenario of this invention. In the practice of screening for abnormal breast cells, pathologists need to assess HER2 expression levels through immunohistochemical staining. According to the ASCO-CAP guidelines, the HER2 IHC score is divided into four levels: 0, 1+, 2+, and 3+. Patients with low HER2 expression (1+ or 2+ and FISH negative) can now benefit from novel ADC drug therapy. However, staining differences between different laboratories lead to imperfections in diagnostic consistency. Even with the same staining procedure and method, individual cell variations can cause errors in diagnosis based on a feature database trained from historical images. Under staining conditions, the mean optical density (AOD), integrated optical density (IOD), chromatin gray standard deviation, and chromatin texture characteristics of cells exhibit corresponding feature changes over time. To eliminate the sensitivity of similar cells from different individuals to staining, the staining changes of different individual cells over time are matched with features to prevent staining differences from affecting image interpretation and improve the accuracy of abnormal cell screening. For example, in HER2 testing for breast cancer, the liquid-based cytology test (TCT) combined with the image processing of cell sections according to this invention can improve diagnostic concordance from 89% to over 95%. Projecting features such as average optical density (AOD) and integrated optical density (IOD) onto spatial coordinates and fitting a time curve essentially extracts core signals based on the kinetic laws of the staining reaction (such as the first-order binding reaction model). In clinical practice, the staining dynamic curve of normal breast cells has been proven to have stable morphological characteristics (such as slope and peak inflection point), which can be used as a standardized template. Curve fitting can effectively eliminate random noise in the staining process (such as uneven distribution of staining solution and image acquisition interference), and spatial mapping calibration based on the standard template can accurately offset individual feature shifts caused by individual heterogeneity. As shown in Figure 2, a specific embodiment of this invention elaborates on this process as follows.

[0035] Example 1: As shown in Figure 1, this example provides an abnormal cell screening method based on image analysis, the method including:

[0036] S1. Acquire continuous image frames of the cell slice to be detected during the staining process. Identify and extract a preset number of sample cells from the image frames. Record the average optical density, integrated optical density and chromatin gray standard deviation of the sample cells at different time points and take the corresponding mean as staining parameters. At the same time, obtain the corresponding chromatin texture based on the gray histogram of the sample cells.

[0037] S2. Project the values ​​of the sample cell staining parameters as a function of time onto a spatial coordinate system, connect the projection points at different time points to form a staining parameter space curve, fit the staining parameter space curve to obtain a continuous staining curve; align and fit the continuous staining curve with a pre-calibrated standard parameter curve, and obtain the coordinate transformation vector of the alignment fit.

[0038] S3. Acquire image frames of the completed staining process, extract staining parameters of all cells to be tested within the field of view, use the coordinate transformation vector to standardize and correct the extracted staining parameters to obtain chromatin texture features at the corresponding time, compare the corrected chromatin texture features with a preset abnormal cell feature threshold, and mark the cells as abnormal when the number of feature dimensions exceeding the threshold reaches a predetermined value.

[0039] For example, a section of breast ductal epithelial tissue was used for HER2 immunohistochemical staining. The final staining results were evaluated according to the ASCO-CAP guidelines. The staining reaction lasted 20 minutes. The binding of the antibody to the HER2 protein followed a first-order kinetic model, with the first 10 minutes being the rapid binding phase and the following 10 minutes being the slow equilibrium phase. A microscope equipped with a CCD camera was used to acquire one image every 30 seconds during the staining process. The 30-second interval was chosen because changes in cell staining parameters at this timescale could capture the dynamic process while avoiding oversampling. A total of 40 images were acquired to fully record the staining kinetics. Twenty morphologically normal and similarly sized epithelial cells were identified and extracted from the first image as sample cells. These sample cells were used to establish an individualized staining baseline for the current section, avoiding the influence of differences in gene expression and cell membrane protein composition between different patients on staining sensitivity.

[0040] The average optical density, integrated optical density, and chromatin grayscale standard deviation of these 20 sample cells were recorded at each time point, and the corresponding mean values ​​were calculated as staining parameters. For example, at the 2-minute mark of staining, the average optical density of the 20 sample cells was 42.3, the average integrated optical density was 1180, and the average chromatin grayscale standard deviation was 7.9; at the 10-minute mark, the corresponding values ​​changed to 55.7, 1620, and 13.4; at the 18-minute mark, they reached 61.2, 1780, and 16.8; and at the 20-minute mark, the staining endpoint, they stabilized at 62.1, 1810, and 17.2. Over time, the staining molecules gradually penetrated the cell membrane and bound to the target protein. A 256-level grayscale histogram was extracted for each sample cell, and the corresponding chromatin texture features, including histogram entropy, energy, and contrast, were calculated. These texture features quantify the uniformity and complexity of the intracellular staining distribution.

[0041] Forty data points showing the changes of the three staining parameters over time were projected onto a two-dimensional coordinate system with time as the abscissa and parameter values ​​as the ordinate. Connecting the projection points at each time point formed three spatial curves for the staining parameters. Due to random interference such as microscope vibration and light source fluctuations during actual data acquisition, the original data points exhibited sawtooth-like fluctuations. Therefore, a fourth-order polynomial fitting method was used for each spatial curve to obtain a smooth, continuous staining curve, effectively eliminating measurement noise and preserving the essential characteristics of staining dynamics. The fitted continuous staining curves were aligned and fitted with standard parameter curves established using 100 normal breast tissue samples. The sum of squared Euclidean distances between the two sets of curves was minimized using an iterative optimization algorithm to obtain the optimal coordinate transformation vector. This vector includes a time dimension translation of -1.2 minutes, a parameter value translation of +5.8 units, and an overall scaling factor of 0.92. The transformation parameters reflect the individualized staining characteristic deviation of the current patient relative to the standard population; for example, some sample cells stain quickly on the time axis, while others stain slowly.

[0042] The final image frame was acquired 20 minutes after staining. 156 intact epithelial cells were identified within a 400x microscope field of view, and their final staining parameter values ​​were extracted one by one. The staining parameters of these cells were standardized using the aforementioned coordinate transformation vector, uniformly calibrating individual-specific differences in staining sensitivity to a standard baseline level, resulting in a standardized chromatin texture feature vector. The standardized texture features were compared item by item with the HER2 overexpression abnormal cell characteristic thresholds established based on large-sample clinical data. Cells were marked as abnormal cells according to the HER2-positive criteria if any two or more of the following three thresholds were met: corrected mean optical density exceeding 68, corrected integrated optical density exceeding 2100, and corrected chromatin grayscale standard deviation exceeding 20. Using this standardized correction method, 31 HER2 overexpression abnormal cells were accurately identified from the 156 cells tested.

[0043] It's important to note that cell staining is essentially a chemical reaction between dye molecules and cellular components (proteins, nucleic acids, etc.). The three parameters—AOD, IOD, and standard deviation—describe the characteristics of the staining state. Staining curves for the same type of cells from different individuals may be generally higher or lower (translation), or the rate of change may differ (scaling), but the temporal evolution pattern of the curves is similar. Spatial corrections are already covered by the scaling factor α and the translation vector β, so complex spatial rotations or nonlinear transformations of the curves are unnecessary. Curve fitting is essentially similar to fitting along the time axis. AOD reflects the staining intensity per unit area and is directly related to antigen expression density (e.g., AOD directly corresponds to the expression level of HER2 protein). IOD reflects the total amount of staining in the cells, combining expression density and cell size. Some abnormal cells exhibit increased volume and enhanced staining; IOD can better capture this change. For example, in breast cancer cell analysis, cancer cells are typically larger and have high HER2 expression, making IOD an important indicator.

[0044] Further feasible optimizations in this embodiment include introducing image stitching technology in the image frame acquisition and processing stage to expand the field of view and ensure the integrity of cell images, reducing the omission of some cells. This embodiment uses a pre-trained recognition model. Optionally, in the feature extraction and comparison stage, a deep learning network model (such as a convolutional neural network) is combined to perform deep feature learning on staining parameters (including average optical density, integral optical density, and chromatin gray standard deviation) and chromatin texture. At the same time, a generative model (such as a generative adversarial network) is used to enhance the diversity of training data, thereby improving the screening accuracy on public datasets (such as the HERLEV dataset) and improving the recognition speed (feature extraction is more accurate, the computational dimension is reduced to reduce the amount of computation, and screening is achieved with less comparison calculation). In addition, for other cell situations besides those listed in this embodiment (such as adherent cells or rare cells), by fusing morphological feature analysis (for example, cell roundness can also be calculated) or correlation imaging technology, noise interference can be effectively reduced and the targeting of target cells can be improved. In the process of training the recognition model, in order to balance performance and computing resources, a lightweight model design (such as knowledge distillation technology, which further compresses and refines the knowledge model during the model training stage) is adopted. This ensures that the accuracy does not decrease significantly, but the amount of computation is greatly reduced.

[0045] Furthermore, the method for recording the average optical density, integrated optical density, and chromatin gray standard deviation of the sample cells at different time points and taking the corresponding mean as the staining parameter includes:

[0046] In each sample cell At the point of time Average optical density Calculated as , For sample cells The number of pixels, For the first The grayscale value of each pixel; calculate the integrated optical density. chromatin gray standard deviation ;

[0047] For each time point Calculate the mean of the average optical density of all sample cells at this time point. Mean of integrated optical density Mean of standard deviation of chromatin gray level , This represents the number of cells in the sample.

[0048] For example, taking the image at the 10-minute time point as an example, we will calculate the staining parameter values ​​of 20 sample cells in detail. The first sample cell is selected for demonstration. This cell contains 320 pixels in the image. The grayscale value of each pixel is read one by one and statistically analyzed. To calculate the average optical density of this cell, the grayscale values ​​of the 320 pixels are summed to obtain a total of 17856. This sum is then divided by the total number of pixels (320) to obtain an average optical density of 55.8. To calculate the integrated optical density, the sum of the pixel grayscale values ​​(17856) is directly used as the integrated optical density value of this cell. To calculate the standard deviation of chromatin grayscale, the square of the difference between each pixel's grayscale value and the average value (55.8) is calculated first. The sum of the squares of the 320 differences is obtained as 1705.6. This square root is then taken after dividing by the total number of pixels to obtain the standard deviation of chromatin grayscale for this cell, which is 13.1.

[0049] Using the same method, the three staining parameters of the remaining 19 cell samples at the 10-minute time point were calculated one by one, yielding 20 average optical density values: 55.8, 54.2, 57.1, 56.9, 54.8, 55.3, 56.7, 55.1, 57.4, 56.2, 54.9, 55.6, 57.8, 55.4, 56.1, 54.7, 57.2, 55.9, 56.4, and 55.7. Summing these 20 values ​​yielded 1115.2, which, divided by the sample size of 20, gave the mean optical density at the 10-minute time point of 55.76. Rounding this down to 55.8, it matched the previous data. Similarly, the sum of the integrated optical density values ​​of the 20 cell samples was calculated to be 32400, which, divided by 20, yielded a mean of 1620. The sum of the standard deviations of chromatin gray values ​​from 20 sample cells was calculated to be 268, which, when divided by 20, yielded a mean of 13.4. This time-point parameter calculation method established a complete sequence of staining kinetic data, facilitating subsequent curve fitting.

[0050] Furthermore, the method for obtaining the corresponding chromatin texture based on the grayscale histogram of the sample cells includes:

[0051] For each sample cell, calculate its gray-level histogram. ,in Gray levels; from the histogram Extracting chromatin texture features, including histogram entropy. Histogram energy Histogram contrast ; denote chromatin texture as a vector .

[0052] For example, continuing with the first sample cell at the 10-minute time point, we extract the chromatin texture feature parameters of this cell. This cell contains 320 pixels, with gray values ​​distributed in the range of 0 to 255. We count the number of pixels at each gray level to construct a gray-level histogram. For example, gray value 45 appears 12 times, gray value 46 appears 15 times, gray value 47 appears 18 times, and so on, completing the frequency statistics for 256 gray levels. Dividing the number of pixels at each gray level by the total number of pixels, 320, yields the corresponding probability distribution. For example, the probability of gray value 45 is 12 divided by 320, which equals 0.0375, and the probability of gray value 46 is 15 divided by 320, which equals 0.0469.

[0053] When calculating the histogram entropy, the probability value of each gray level is multiplied by the negative of its logarithm, and then the results of 256 gray levels are summed, resulting in a histogram entropy of 7.32 for this cell. A larger value indicates a more uniform and complex gray-level distribution. When calculating the histogram energy, the probability value of each gray level is squared, and then all squared values ​​are summed, resulting in a histogram energy of 0.0089 for this cell. A larger value indicates that some gray values ​​are dominant. When calculating the histogram contrast, the square of each gray level value is multiplied by its corresponding probability, and then summed, resulting in a histogram contrast of 3254.7 for this cell, reflecting the degree of dispersion of gray values. Combining these three texture features into a feature vector represents histogram entropy 7.32, histogram energy 0.0089, and histogram contrast 3254.7. This vector quantitatively describes the spatial heterogeneity of the intracellular staining agent distribution, providing texture criteria for abnormal cell identification.

[0054] Further, the method of fitting the staining parameter space curve to obtain a continuous staining curve; aligning and fitting the continuous staining curve with a pre-calibrated standard parameter curve, and obtaining the coordinate transformation vector of the alignment fit includes:

[0055] For each coloring parameter type, the point set of the coloring parameter space curve To perform a fit, use a fitting polynomial function. By minimizing the sum of squared errors Coefficient of determination , thus obtaining the continuous staining curve ;

[0056] Continuous coloring curve Compared with standard parameter curve Perform alignment fitting, from Medium distance selection A set of points ,from Medium distance selection A set of points ;

[0057] By translation vector and scaling factor Point set Transform into ;

[0058] Compute point set and Sum of squared Euclidean distances between ;

[0059] By solving Seek The coordinate transformation vector is denoted as including... .

[0060] For example, curve fitting was performed using 40 data points showing the average optical density variation over time obtained earlier. These 40 data points were distributed from 0 minutes to 20 minutes on the time axis, with the corresponding average optical density values ​​gradually increasing from an initial 38.2 to a final 62.1. A fourth-order polynomial fitting method was used to process these discrete data points because the kinetics of the HER2 antibody binding reaction exhibit a typical S-shaped growth curve, and the fourth-order polynomial can accurately describe the three stages of the reaction: the initial slow phase, the intermediate rapid phase, and the final equilibrium phase. The fitting coefficients were calculated using the least squares method to minimize the sum of squared errors between the original data points and the fitted curve, resulting in a smooth, continuous staining curve that eliminates random fluctuations during the measurement process. Similarly, the same fourth-order polynomial fitting was performed on the time series data of integrated optical density and chromatin grayscale standard deviation to obtain three complete continuous staining curves.

[0061] The fitted average optical density continuous curve is aligned with a pre-established standard parameter curve. Specifically, 20 feature points are selected at equal intervals from the continuous curve of the current sample, with time coordinates from 1 minute to 20 minutes, corresponding to average optical density values ​​of 40.1, 42.8, 45.6, 48.2, 50.9, 53.4, 55.8, 57.9, 59.7, 61.2, 62.1, 62.3, 62.4, 62.5, 62.5, 62.6, 62.6, 62.6, 62.6, and 62.6. The same number of corresponding points are selected from the standard parameter curve, with time coordinates and parameter values ​​of 46.3 for 1 minute, 49.1 for 2 minutes, 51.8 for 3 minutes, and so on. The optimal parameters for transforming the standard curve to the current sample curve are calculated. Through iterative optimization, the time dimension shift of -1.2 minutes indicates that the staining reaction of the current sample is 1.2 minutes earlier than the standard case. The parameter dimension shift of +5.8 indicates that the baseline shift is 5.8 optical density units. The scaling factor of 0.92 indicates that the staining sensitivity of the current sample is 8% lower than the standard case.

[0062] Furthermore, the method for obtaining the pre-calibrated standard parameter curve includes:

[0063] Acquire continuous image frames of known normal cells during the staining process, identify and extract a preset number of sample cells based on the image frames, record the average optical density, integrated optical density and chromatin gray standard deviation of the sample cells at different time points, and take the corresponding mean as the staining parameters, project the staining parameters into a spatial coordinate system to form a staining parameter space curve, and fit the staining parameter space curve to obtain the standard parameter curve.

[0064] For example, the standard parameter curve was established using 100 pathologically confirmed normal mammary ductal epithelial tissue sections. These samples came from healthy women of different ages to ensure coverage of individual differences under normal physiological conditions. Each section was processed using the same HER2 immunohistochemical staining procedure, with strict control over key parameters such as antibody concentration, incubation temperature, and washing time to eliminate the influence of experimental condition differences on the results. Using the same microscope equipment and image acquisition parameters, the staining process of each section was continuously monitored for 20 minutes, with one image acquired every 30 seconds for a total of 40 frames, establishing a standardized image database.

[0065] From the initial image of each tissue section, 15 to 25 morphologically normal epithelial cells were identified and extracted as sample cells. Typical representative cells were selected based on cell size and chromatin distribution characteristics, excluding atypical cells such as apoptotic cells and cells in the mitotic phase. The mean optical density, integrated optical density, and standard deviation of chromatin grayscale values ​​for each sample cell were recorded at each time point, and the mean of the corresponding parameters for all sample cells at the same time point was calculated. For example, the mean optical density of 100 normal tissues at the 10-minute time point were 49.7, 51.2, 52.8, 50.4, and 53.1, respectively. The mean of these 100 values ​​was calculated again to obtain 51.24, which was used as the baseline parameter for the standard population at the 10-minute time point.

[0066] The staining parameters of each slide were projected over time onto a time-parameter coordinate system, resulting in 100 individualized staining parameter spatial curves. A fourth-order polynomial fitting method was used to obtain a continuous curve for each curve. Then, the parameter values ​​of the 100 continuous curves at each time point were statistically analyzed, and the mean and standard deviation were calculated to determine the normal range. Finally, a standard parameter curve representing the characteristics of a normal breast cell population was obtained by comprehensively fitting the 100 individual curves. This curve reflects the binding kinetics of HER2 staining agent in normal cells.

[0067] Furthermore, the method for standardizing and correcting the extracted staining parameters using the coordinate transformation vector to obtain the chromatin texture features at the corresponding time step includes:

[0068] For each cell tested, its average optical density curve over time was extracted. Integral optical density curve chromatin gray standard deviation curve ;

[0069] Scaling factor in the coordinate transformation vector obtained from sample cells Horizontal displacement Calculate the corrected average optical density of the cells to be tested. Integrated optical density after calibration of the target cells Standard deviation of chromatin grayscale after correction of the test cells The chromatin texture features of the cells under test are represented as vectors. .

[0070] For example, the first cell to be tested in the staining endpoint image at the 20-minute mark is used for standardized correction calculation. This cell contains 298 pixels, and the calculated original average optical density is 67.4, the integrated optical density is 2009, and the standard deviation of chromatin grayscale is 18.6. Since this cell to be tested comes from the same patient slice as the sample cells, its staining sensitivity exhibits the same individual-specific bias, requiring correction using the aforementioned coordinate transformation vector. Based on the determined transformation parameters, with a scaling factor of 0.92 and a parameter dimension translation of +5.8 units, the three staining parameters of this cell to be tested are standardized and corrected one by one.

[0071] The original average optical density of the test cells, 67.4, was multiplied by a scaling factor of 0.92 to obtain 62.0. Adding a translation factor of 5.8 yielded the corrected average optical density of 67.8. The corrected integrated optical density was calculated using the same method: the original value 2009 multiplied by 0.92 plus 5.8 equals 1854.3. The corrected standard deviation of chromatin grayscale was the original value 18.6 multiplied by 0.92 plus 5.8 equals 22.9. This correction process transforms the staining parameters of the test cells from an individualized state to a standardized state, eliminating patient-specific differences in staining sensitivity and allowing direct comparison with the abnormal cell detection threshold established based on a standard population. The three corrected parameters are combined into a chromatin texture feature vector, represented as average optical density 67.8, integrated optical density 1854.3, and chromatin grayscale standard deviation 22.9. This vector accurately reflects the true staining characteristics of the cells under standardized conditions.

[0072] Further, the method of comparing the corrected chromatin texture features with a preset abnormal cell feature threshold, and marking a cell as an abnormal cell when the number of feature dimensions exceeding the threshold reaches a predetermined value, includes:

[0073] The corrected chromatin texture features for each test cell are vectors. ; with the preset abnormal cell feature threshold as a vector At a preset designated time of and Feature comparison is performed, and if the cell meets the characteristics of an abnormal cell, it is marked as an abnormal cell.

[0074] For example, a preliminary analysis of cell staining characteristics from 500 HER2-positive breast cancer patients was conducted. Based on the criteria for strong HER2 protein 3+ positive expression in the ASCO-CAP guidelines, three key thresholds for abnormal cells were determined: corrected mean optical density exceeding 68, corrected integrated optical density exceeding 2100, and corrected chromatin gray standard deviation exceeding 20. When the corrected chromatin texture characteristics of the tested cells exceeded the corresponding thresholds in any two or more of the three dimensions, the cells were identified as HER2 overexpressing abnormal cells.

[0075] The corrected chromatin texture feature vector of the test cell was compared with the abnormal cell determination threshold item by item. The corrected average optical density of this test cell was 67.8, which was less than the threshold of 68, and did not meet the first determination criterion. The corrected integrated optical density was 1854.3, which was less than the threshold of 2100, and did not meet the second determination criterion. The corrected chromatin gray standard deviation was 22.9, which was greater than the threshold of 20, and met the third determination criterion. Since this test cell exceeded the threshold in only one of the three dimensions, it did not meet the joint determination condition of exceeding the threshold in two or more dimensions. Therefore, this cell was determined to be a normal cell. The same comparison process was used to compare the features of the remaining 155 test cells in the field of view one by one. Finally, 31 cells were identified as exceeding the determination threshold in two or more dimensions. These 31 cells were marked as HER2 positive abnormal cells. For example, the corrected chromatin texture features of one abnormal cell were: average optical density 72.1, integrated optical density 2247, and chromatin gray standard deviation 21.8. It exceeded the threshold in both the average optical density and integrated optical density dimensions, and met the abnormal cell determination condition. By combining standardized correction with multi-dimensional threshold comparison, this method transforms traditional subjective visual assessment into objective quantitative detection, significantly improving the accuracy and reproducibility of HER2 detection.

[0076] It should be noted that this embodiment focuses on how to eliminate the disturbance of the judgment result caused by the staining characteristics caused by the individual differences in staining on the time axis. Therefore, this embodiment only uses the simplest numerical comparison method. However, how to make a judgment based on the characteristics in the end, such as other methods such as clustering, principal component analysis, Euclidean distance, etc., is not the focus of this invention and will not be elaborated here.

[0077] Example 2: Based on the same inventive concept, this example also provides an image analysis-based abnormal cell screening method for performing any of the image analysis-based abnormal cell screening methods described above.

[0078] It should be noted that the specific methods of performing operations in the systems described in the above embodiments have been described in detail in the embodiments related to this method, and will not be elaborated here.

[0079] Finally, it should be noted that although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An abnormal cell screening method based on image analysis, characterized in that, The method includes: acquiring continuous image frames of the cell slice to be tested during the staining process; identifying and extracting a preset number of sample cells from the image frames; recording the average optical density, integrated optical density, and chromatin grayscale standard deviation of the sample cells at different time points and taking the corresponding mean as staining parameters; simultaneously obtaining the corresponding chromatin texture based on the grayscale histogram of the sample cells; projecting the values ​​of the sample cell staining parameters changing with the time axis onto a spatial coordinate system; connecting the projection points at different time points to form a staining parameter space curve; fitting the staining parameter space curve to obtain a continuous staining curve; aligning and fitting the continuous staining curve with a pre-calibrated standard parameter curve and obtaining the coordinate transformation vector of the alignment fit; acquiring image frames of the completed staining process; extracting the staining parameters of all cells to be tested within the field of view; using the coordinate transformation vector to standardize and correct the extracted staining parameters to obtain the chromatin texture features at the corresponding time point; comparing the corrected chromatin texture features with a preset abnormal cell feature threshold; and marking cells as abnormal when the number of feature dimensions exceeding the threshold reaches a predetermined value.

2. The abnormal cell screening method based on image analysis according to claim 1, characterized in that, The method for recording the average optical density, integrated optical density, and chromatin gray standard deviation of the sample cells at different time points and taking the corresponding mean as the staining parameters includes: for each sample cell At the point of time Average optical density Calculated as ， For sample cells The number of pixels, For the first The grayscale value of each pixel; calculate the integrated optical density. chromatin gray standard deviation For each time point Calculate all sample cells at time points The mean of the average optical density Mean of integrated optical density Mean of standard deviation of chromatin gray level ， This represents the number of cells in the sample.

3. The abnormal cell screening method based on image analysis according to claim 2, characterized in that, The method for obtaining the corresponding chromatin texture based on the gray-level histogram of sample cells includes: calculating the gray-level histogram for each sample cell. ,in Gray levels; from the histogram Extracting chromatin texture features, including histogram entropy. Histogram energy Histogram contrast ; denote chromatin texture as a vector 。 4. The abnormal cell screening method based on image analysis according to claim 1, characterized in that, The continuous staining curve is obtained by fitting the space curve of the staining parameters. The method for aligning and fitting continuous coloring curves with pre-calibrated standard parameter curves and obtaining the coordinate transformation vector of the alignment fit includes: for each coloring parameter type, the point set of the coloring parameter space curve is... Perform fitting, where For the first At a certain point in time, These are the staining parameter values ​​at the corresponding time points; Using a fitted polynomial function By minimizing the sum of squared errors Coefficient of determination , thus obtaining the continuous staining curve ; Continuous coloring curve Compared with standard parameter curve Perform alignment fitting, from Medium distance selection A set of points ,from Medium distance selection A set of points ; by translating vectors and scaling factor Point set Transform into ; Compute point set and Sum of squared Euclidean distances between By solving , , Seek , , The coordinate transformation vector is denoted as 。 5. The abnormal cell screening method based on image analysis according to claim 4, characterized in that, The method for obtaining the pre-calibrated standard parameter curve includes: acquiring continuous image frames of known normal cells during the staining process; identifying and extracting a preset number of sample cells based on the image frames; recording the average optical density, integrated optical density, and standard deviation of chromatin grayscale of the sample cells at different time points and taking the corresponding mean as staining parameters; projecting the staining parameters into a spatial coordinate system to form a staining parameter space curve; and fitting the staining parameter space curve to obtain the standard parameter curve.

6. The abnormal cell screening method based on image analysis according to claim 4, characterized in that, The method for standardizing and correcting the extracted staining parameters using the coordinate transformation vector to obtain the chromatin texture features at the corresponding time step includes: for each cell to be tested, extracting its average optical density curve over time. Integral optical density curve chromatin gray standard deviation curve ; Scaling factor in the coordinate transformation vector obtained from sample cells Translational displacement Calculate the corrected average optical density of the cells to be tested. Integrated optical density after calibration of the target cells Standard deviation of chromatin grayscale after correction of the test cells The chromatin texture features of the cells under test are represented as vectors. 。 7. The abnormal cell screening method based on image analysis according to claim 6, characterized in that, The method of comparing the corrected chromatin texture features with a preset abnormal cell feature threshold, and marking cells as abnormal when the number of feature dimensions exceeding the threshold reaches a predetermined value, includes: for each cell to be tested, the corrected chromatin texture features are vectors. ; with the preset abnormal cell feature threshold as a vector At a preset designated time of and Feature comparison is performed, and if the cell meets the characteristics of an abnormal cell, it is marked as an abnormal cell.

8. An abnormal cell screening system based on image analysis, characterized in that, The system is used to perform any one of the image analysis-based abnormal cell screening methods according to claims 1-7.

Images