CAR-T cell culture monitoring system based on image recognition

Abstract

The invention relates to the technical field of medical image analysis, in particular to a CAR-T cell culture monitoring system based on image recognition, which comprises a cell region division module, a cell morphology analysis module, a cell behavior dynamic analysis module and a cell population collaborative monitoring module. According to the method, through cell image gradient intensity analysis, refined cell edge detection and dynamic adjustment of a segmentation threshold value, high-precision extraction of cell boundaries is ensured, noise interference is avoided, cell shapes, textures and geometric features are refined, the identification degree of morphological features is improved, and tiny changes of cell morphologies are accurately captured; dynamic tracking and trend evaluation are carried out on cell behaviors based on adjacent frame images, tiny dynamic changes such as cell division, aggregation and migration are accurately reflected, excessive smoothness in the dynamic process is avoided, cell population behavior monitoring is combined with interaction and corresponding speed calculation, the population synergistic effect change trend is comprehensively evaluated, and the cell population behavior monitoring effect is improved. And finer cell culture environment optimization and clinical research support are provided.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image analysis, and particularly to a CAR-T cell culture monitoring system based on image recognition. Background Art

[0002] The technical field of medical image analysis includes technical methods based on image processing and analysis and their applications in medical scenarios. The core content of this technical field is to assist clinical diagnosis, treatment, and research work through the acquisition, processing, analysis, and interpretation of medical images. Specific technologies include image segmentation, feature extraction, target recognition, and classification, etc., which are widely used in fields such as pathological analysis, surgical navigation, and imaging assessment. The overall technical field of medical image analysis combines computer vision, biomedical engineering, and artificial intelligence technologies to process various types of medical image data, such as microscope images, CT images, MRI images, and ultrasound images, in an accurate and efficient manner, providing important technical support for medical scenarios.

[0003] Among them, the CAR-T cell culture monitoring system refers to a system used to monitor and evaluate the culture status in real time through image analysis technology during the CAR-T cell culture process. This system aims at technical matters such as cell proliferation, differentiation, and state changes in CAR-T cell culture, uses a high-resolution image acquisition device to obtain image data of cells during the culture process, and completes cell contour detection, cell number counting, and morphological analysis through image processing means. The system uses an image recognition-based method to analyze and record specific features of cells during the culture process in order to achieve the determination of the cell growth state in the culture environment and the feature identification of key stages.

[0004] In the existing technology during the image acquisition and analysis process, it relies on traditional cell segmentation methods, which have great limitations in the detailed analysis of cell morphology and behavior. Due to the noise interference of image data and the challenges of cell morphology complexity, it is difficult for the existing technology to accurately divide cell boundaries, resulting in insufficiently detailed extraction of cell morphological features. Especially in the case of high-density culture or high similarity of cell morphology, recognition errors are likely to occur. The existing technology has relatively limited dynamic tracking ability for cell behavior and relies on image comparison at long time intervals, which makes it have large errors in capturing small cell behavior changes and cannot reflect the precise dynamics of cells in real time. In terms of population cooperation analysis, the existing technology simplifies the evaluation of the interrelationship between cells and cannot effectively measure the complex interactions within the cell population and the corresponding speed, resulting in relatively rough analysis of population behavior patterns and being unable to fully reflect the changes in population cooperation effects. The existing technology has certain deficiencies in the precise monitoring, dynamic analysis of cell behavior, and comprehensive evaluation of cell population cooperation, affecting the efficiency and accuracy of various operations during cell culture. Summary of the Invention

[0005] The object of the present invention is to solve the disadvantages existing in the prior art, and a CAR-T cell culture monitoring system based on image recognition is proposed.

[0006] To achieve the above object, the present invention adopts the following technical solution: The CAR-T cell culture monitoring system based on image recognition includes:

[0007] The cell region division module analyzes the gradient intensity of the cell image based on the cell culture image, divides the image into a high-feature region and a low-feature region, performs cell edge detection on the high-feature region, extracts the edge position and cell morphological features, refines the cell shape and texture, dynamically adjusts the segmentation threshold, and obtains the cell boundary segmentation result;

[0008] The cell morphology analysis module analyzes the morphology of the cells in the high-feature region based on the cell boundary segmentation result, extracts the local texture information of the cells, analyzes the cell morphological differences through local feature comparison, refines the geometric and structural features of the cells, and obtains the cell morphology feature dataset;

[0009] The cell behavior dynamic analysis module tracks and analyzes the cell behavior based on the cell morphology feature dataset, monitors the cell division, aggregation, and migration behaviors, evaluates the distance and behavior trend through adjacent frame images, records the dynamic behavior of the cells, analyzes the dynamic changes in cell culture, and obtains the cell behavior trend analysis result;

[0010] The cell population cooperation monitoring module analyzes the cooperation between cell populations based on the cell behavior trend analysis result, monitors the trend of cell aggregation and dispersion, combines the mutual relationship between cells, calculates the distance and corresponding speed between cells, evaluates the overall behavior pattern of the cell population, dynamically monitors the cooperation effect of the cell population, and obtains the cell population cooperation behavior monitoring result.

[0011] As a further solution of the present invention, the specific steps for obtaining the edge position and cell morphological features are as follows:

[0012] Based on the cell culture image, the image is converted into a grayscale image, the gray difference between the brightness value of each pixel point and the surrounding pixels is analyzed, the gradient value of each pixel is calculated, the image region is divided using the gradient value, and the region features are judged according to the gradient value to obtain the high-low feature region division result;

[0013] Based on the high-low feature region division result, the pixels in the high-feature region are subjected to regional enhancement processing, the local contrast of each pixel in the high-feature region is analyzed, and weight distribution is performed according to the difference in surrounding pixel values to obtain the enhanced data of the corresponding region;

[0014] Based on the enhanced data of the corresponding region, locate the edge position of the cell through contour, analyze the gradient change of each pixel in the high-feature region, determine the edge pixel points, and analyze the cell morphology based on the edge information to obtain the edge position and cell morphology features.

[0015] As a further solution of the present invention, the steps for obtaining the cell boundary segmentation result are specifically as follows:

[0016] Based on the edge position and cell morphology features, extract the edge information in the cell image, identify the position information of each cell boundary, and generate a cell boundary position feature dataset by analyzing cell morphology features including curvature and edge smoothness;

[0017] Combine the cell boundary position feature dataset with the cell texture information, extract the texture features in the image, analyze the change law of the texture features in the differential edge region, dynamically adjust the segmentation threshold, and obtain the segmentation threshold adjustment coefficient;

[0018] According to the segmentation threshold adjustment coefficient, using the optimal matching of cell morphology and texture features, adopt the formula:

[0019]

[0020] Generate the cell boundary segmentation result;

[0021] where, T final represents the cell boundary segmentation threshold, T base represents the initial segmentation threshold, ω i represents the weight coefficient, X i represents the adjustment coefficient related to the association between cell morphology and texture features, and n is the number of features.

[0022] As a further solution of the present invention, the steps for obtaining the cell local texture information are specifically as follows:

[0023] Based on the cell boundary segmentation result, perform boundary refinement and correction, remove irregular edges through local smoothing, repair the boundary in combination with cell contour features, avoid breaks and discontinuities, and perform morphological optimization according to pixel distribution to obtain the corrected cell boundary;

[0024] Based on the corrected cell boundary, perform geometric feature extraction, identify the area and perimeter of each cell, calculate the aspect ratio and shape factor using boundary points, analyze the symmetry and circularity of the cell, and screen the cell shape in combination with the area-to-perimeter ratio to obtain the cell size and shape characteristics;

[0025] Based on the cell size and shape characteristics, a local area is selected for texture analysis. The pixel value changes in the local area of the cell are statistically analyzed to quantify the texture variation degree and consistency. Combining the boundary information, the texture distribution pattern is analyzed, and the texture differences and consistencies between regions are evaluated to obtain the local texture information of the cell.

[0026] As a further solution of the present invention, the obtaining steps of the cell morphological feature dataset are specifically as follows:

[0027] According to the local texture information of the cell, the cell edge, texture pattern, and contrast between regions are extracted. Combining with the region division, the gray-scale change, local contrast, and texture direction parameters of the image are identified, and the geometric differences between local regions are analyzed to obtain the cell texture features.

[0028] Based on the cell texture features, the cell morphological differences and geometric features between different regions are quantified and compared with the texture similarity metric between regions. Using the formula:

[0029]

[0030] The cell morphological difference data is obtained;

[0031] wherein, δ represents the local texture difference degree, T r,j is the texture feature of the right region, T l,j is the texture feature of the left region, and N is the number of texture points;

[0032] Based on the cell morphological difference data, through clustering and segmentation methods, the geometric and structural features of the cell are refined. Using distance measurement judgment, the regions with the same geometric features are grouped to obtain the cell morphological feature dataset.

[0033] As a further solution of the present invention, the obtaining steps of the cell behavior trend analysis result are specifically as follows:

[0034] Each frame of image in the cell morphological feature dataset is analyzed to extract the cell position and identify the distance between adjacent frames. Using the distance to judge the cell division, aggregation, and migration behaviors. Through time series, and combining with the pixel position of the image, the cell position is tracked to obtain the cell behavior tracking data;

[0035] Based on the cell behavior tracking data, according to the motion trajectory and change trend between adjacent frames, the cell dynamic behavior is evaluated, and the displacement distance value between adjacent frames of the cell is calculated. Using the formula:

[0036]

[0037] The cell motion trend data is obtained;

[0038] Among them, d represents the displacement distance value of the cell between adjacent frames, x t+1 and y t+1 are the coordinates of the cell in the (t + 1)-th frame, x t and y t are the coordinates of the cell in the t-th frame;

[0039] Based on the cell movement trend data, analyze cell division, aggregation and migration behaviors, evaluate the cell behavior trend, analyze whether there is a changing trend, and judge the key points of cell behavior changes by setting thresholds, so as to obtain the cell behavior trend analysis result.

[0040] As a further solution of the present invention, the steps for obtaining the distance between cells and the corresponding speed are specifically as follows:

[0041] Based on the cell behavior trend analysis result, analyze the positional relationship between cells, extract the coordinate information of each pair of cells, identify the straight-line distance between cells, extract cell coordinates and track the position changes of each pair of cells, so as to obtain the distance dataset between cells;

[0042] Based on the distance dataset between cells, calculate the speed between cells according to the displacement of each pair of cells in adjacent frames, perform derivative calculation through the cell coordinate difference, analyze the speed change between cells, and adopt the formula:

[0043]

[0044] to obtain the distance between cells and the corresponding speed;

[0045] Among them, v represents the speed value of each pair of cells, D t+1 and D t are the distances between cells between the (t + 1)-th frame and the t-th frame, and Δt is the time interval.

[0046] As a further solution of the present invention, the steps for obtaining the monitoring result of the cooperative behavior of the cell population are specifically as follows:

[0047] Based on the distance between cells and the corresponding speed, capture the spatial position coordinates and motion states of each cell, monitor the corresponding speed of the cells, record the distances and speeds of each cell from the surrounding cells, and generate the cell population position and speed dataset;

[0048] Based on the cell population position and speed dataset, analyze the corresponding positions and motion states between each cell and the surrounding cells, screen the cells with the interaction intensity greater than the set threshold, delimit the interaction range of each pair of cells according to the distance, speed and interaction time, and construct the cell interaction relationship dataset;

[0049] Using the dataset of cell - cell interaction relationships, analyze the cooperative behavior of cell populations. By the interactions between cells and population trends, judge the dynamic changes of population behavior, and obtain the monitoring results of cell population cooperative behavior.

[0050] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0051] In the present invention, through the gradient intensity analysis of cell images, the images are effectively divided into high - feature regions and low - feature regions, and then the edge detection of cells is refined. Dynamically adjusting the segmentation threshold can ensure the high - precision extraction of cell boundaries without being affected by interference such as illumination and noise. Conducting cell morphology analysis on the high - feature regions, refining cell shape, texture, and geometric features helps to accurately capture the minute morphological changes of cells, thereby providing more in - depth data on cell states. This targeted and refined analysis effectively improves the recognition rate and accuracy of cell morphological features. The dynamic tracking and trend evaluation of cell behavior provide higher spatio - temporal resolution for the real - time monitoring of cell division, aggregation, migration, etc., making the capture of cell behavior patterns more accurate. The evaluation of cell behavior based on adjacent frame images enables the analysis results to finely reflect the minute change trends of cells in a short period of time, effectively avoiding the situation of over - smoothing of the dynamic process or neglecting local details. The behavior monitoring of cell populations, combined with the calculation of interactions between cells and corresponding speeds, can comprehensively evaluate the change trends of cell population synergy effects, provide a panoramic analysis of cell population behavior patterns, and provide strong data support for more efficient optimization of cell culture environments and clinical research. Brief Description of the Drawings

[0052] Figure 1 It is the system flow chart of the present invention;

[0053] Figure 2 It is the flow chart of the edge position and cell morphological features in the present invention;

[0054] Figure 3 It is the flow chart of the cell boundary segmentation result in the present invention;

[0055] Figure 4 It is the flow chart of the local texture information of cells in the present invention;

[0056] Figure 5 It is the flow chart of the cell morphological feature dataset in the present invention;

[0057] Figure 6 It is the flow chart of the cell behavior trend analysis result in the present invention;

[0058] Figure 7 It is the flow chart of the distance between cells and corresponding speeds in the present invention;

[0059] Figure 8 This is a flowchart of the monitoring results of the cooperative behavior of cell populations in the present invention. Detailed implementation manners

[0060] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0061] In the description of the present invention, it should be understood that the orientation or positional relationships indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. are based on the orientation or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as limiting the present invention. In addition, in the description of the present invention, "a plurality of" means two or more unless otherwise specifically defined.

[0062] Please refer to Figure 1 , the CAR-T cell culture monitoring system based on image recognition includes:

[0063] The cell region division module analyzes the gradient intensity of the cell image based on the cell culture image, divides the image into a high-feature region and a low-feature region, performs cell edge detection on the high-feature region, extracts the edge positions and cell morphological features, refines the cell shape and texture, dynamically adjusts the segmentation threshold, and obtains the cell boundary segmentation result;

[0064] The cell morphology analysis module analyzes the morphology of the cells in the high-feature region based on the cell boundary segmentation result, extracts the local texture information of the cells, analyzes the cell morphological differences through local feature comparison, refines the geometric and structural features of the cells, and obtains the cell morphology feature dataset;

[0065] The cell behavior dynamic analysis module tracks and analyzes the cell behavior based on the cell morphology feature dataset, monitors the cell division, aggregation, and migration behaviors, evaluates the distance and behavior trend through adjacent frame images, records the dynamic behavior of the cells, analyzes the dynamic changes in cell culture, and obtains the cell behavior trend analysis result;

[0066] The cell population cooperation monitoring module analyzes the cooperation between cell populations based on the cell behavior trend analysis result, monitors the trend of cell aggregation and dispersion, combines the mutual relationships between cells, calculates the distance and corresponding speed between cells, evaluates the overall behavior pattern of the cell population, dynamically monitors the cooperation effect of the cell population, and obtains the monitoring result of the cooperative behavior of the cell population.

[0067] The cell boundary segmentation results include the cell edge position, cell morphological features, and cell segmentation threshold. The cell morphological feature dataset includes cell shape, cell size, cell edge, local texture, cell geometric features, and cell structural features. The cell behavior trend analysis results include cell division behavior, cell aggregation behavior, cell migration behavior, cell behavior trend, and cell dynamic changes. The monitoring results of cell population cooperative behavior include cell aggregation trend, cell dispersion trend, intercellular distance, corresponding cell speed, cell population behavior pattern, and cell cooperative effect.

[0068] Please refer to Figure 2 , and the steps for obtaining the edge position and cell morphological features are specifically as follows:

[0069] Based on the cell culture image, convert the image into a grayscale image, analyze the gray difference between the brightness value of each pixel and the gray levels of the surrounding pixels, calculate the gradient value of each pixel, use the gradient value for image region division, and judge the region features based on the gradient value to obtain the high and low feature region division results;

[0070] Convert the cell culture image into a grayscale image for subsequent analysis. The conversion process depends on extracting the brightness value of each pixel of the original image. By weighted averaging the pixel values of the red, green, and blue channels, the corresponding gray value is obtained. Analyze the gray difference between the brightness value of each pixel and the gray levels of the surrounding pixels, and calculate the gradient value accordingly. The gradient value represents the rate of change of pixel intensity in the image and can reflect the edges and changing areas of the image. When calculating the gradient value for each pixel, select a preset-sized window and derive the gradient value of the pixel by calculating the differences in the brightness values of the pixels within the window. The specific process can be obtained by calculating the differences of the surrounding pixels, which can quantify the changing areas and static areas in the cell image. Then, based on the calculated gradient value, perform region division. By setting a gradient threshold, the image is divided into a high feature region and a low feature region. The high feature region reflects stronger cell activities, while the low feature region is relatively stable. In this way, the high and low feature region division results of the cell image can be obtained. Through the gradient feature analysis of the regions, the cell culture image can be preliminarily classified, laying a foundation for subsequent cell morphological analysis.

[0071] Based on the high and low feature region division results, perform region enhancement processing on the pixels in the high feature region. Analyze the local contrast of each pixel in the high feature region and allocate weights according to the differences in the surrounding pixel values to obtain the enhanced data for the corresponding region;

[0072] Enhance the high-feature regions, especially in cell culture images, which are regions with high features reflecting cell activity or significant changes. To enhance the distinguishability of high-feature regions, first calculate the local contrast of each pixel. The local contrast can be achieved by analyzing the gray-scale difference between the pixel and its surrounding pixels. Use a window to analyze the difference between the pixels within the window and their neighborhood, and calculate the local contrast. After obtaining the local contrast, ensure that the enhancement effect conforms to the characteristics of the cell image, and weight distribution can be carried out according to the contrast magnitude. For regions with larger contrast, the weight value is higher. In this way, during the enhancement process, the parts with greater changes will be enhanced more, and the details of the high-feature regions can be highlighted without affecting the low-feature regions, making the cell morphology and features in the image more distinct, and obtaining the enhanced data for the corresponding regions.

[0073] Based on the enhanced data of the corresponding region, locate the edge position of the cell through contour, analyze the gradient change of each pixel within the high-feature region, determine the edge pixel points, and analyze the cell morphology based on the edge information to obtain the edge position and cell morphology characteristics.

[0074] Within the high-feature region, analyze the gradient change of each pixel. The change in gradient value can often accurately reflect the existence of the edge. By analyzing the gradient change, the position of the edge pixel points can be determined, and then the cell edge can be inferred. Through precise edge positioning, the contour of the cell can be recognized, and further analyze the cell morphology to judge the morphological characteristics of the cell, such as the size and shape of the cell. The analysis results can help better understand the growth and distribution status of the cells, providing a more accurate reference for the monitoring of CAR-T cell culture. By carefully analyzing the edge information, not only can the changes in cell morphology be understood, but also the interaction between the cell and the environment can be further explored, thereby promoting the real-time monitoring and optimization in the cell culture process, and obtaining the edge position and cell morphology characteristics.

[0075] Please refer to Figure 3 , and the specific steps for obtaining the cell boundary segmentation result are as follows:

[0076] Based on the edge position and cell morphology characteristics, extract the edge information in the cell image and identify the position information of each cell boundary. By analyzing the cell morphology characteristics, including curvature and edge smoothness, generate the cell boundary position feature dataset.

[0077] First, extract the edge information in the cell image, perform edge detection through image processing techniques, and use methods such as image gradient or Canny edge detection to extract the cell boundaries. Extract the edge positions of each cell and determine the coordinate points of each cell boundary. Combine the cell morphological features, such as morphological features, aspect ratio of the cell, contour smoothness, etc., to conduct further refined analysis of the cell boundaries. Use morphological filtering or contour analysis to calculate features such as the curvature and edge smoothness of the cells, forming a morphological feature vector. In this way, it is possible to ensure accurate capture of the shape changes and edge features of the cells, thereby preparing sufficient morphological data for subsequent cell segmentation, generating a cell boundary position feature dataset, which will serve as a key input in subsequent analysis to ensure that the subsequent steps can perform dynamic adjustment of the segmentation threshold based on this dataset.

[0078] By combining the cell boundary position feature dataset with the cell texture information, extract the texture features in the image, analyze the variation law of the texture features in the differential edge regions, and dynamically adjust the segmentation threshold to obtain the segmentation threshold adjustment coefficient.

[0079] Combine the cell texture feature information to further analyze the cell texture distribution in the image. Extract the texture features of each cell through texture analysis methods such as the gray-level co-occurrence matrix (GLCM). The features include contrast, homogeneity, energy, and entropy, etc., which can reflect the texture information in the cell edge regions. Combine the texture features with the edge position feature data to analyze the variation law of the texture features in different edge regions. Use image region segmentation methods, such as the K-means clustering algorithm, to refine the cell regions and identify different texture regions of the cells. This can not only accurately obtain the positions of the cell edges but also obtain the texture features of different regions inside and outside the cells, providing more accurate data support for threshold adjustment. According to the texture analysis results, adopt a dynamic adjustment method based on entropy value to update the segmentation threshold to obtain the segmentation threshold adjustment coefficient.

[0080] According to the segmentation threshold adjustment coefficient, utilize the optimal matching of cell morphology and texture features, and adopt the formula:

[0081]

[0082] Generate the cell boundary segmentation result;

[0083] Among them, T final represents the cell boundary segmentation threshold, T base represents the initial segmentation threshold, ω i represents the weight coefficient, X i represents the adjustment coefficient related to the association between cell morphology and texture features, and n is the number of features;

[0084] The advantage of the formula is that by introducing the weight coefficients of morphological features and texture features and combining with the adjustment coefficient of the threshold, the segmentation threshold can more accurately match the complex features in the cell image, thereby improving the accuracy and reliability of the segmentation result; the initial segmentation threshold T base is preset according to the basic features of the cell image. By comprehensively considering the cell morphology and texture features, the adjustment coefficient X of each feature is calculated i , and the weight coefficient ω is set according to its influence degree i . The coefficient is obtained through experimental data or previous model training, and the weight coefficient is set by comparing the influence degrees of different features on the final segmentation result. For example, if the morphological feature has a greater influence on the segmentation result, a higher weight is assigned;

[0085] The specific calculation process is as follows: T final = T base ·(1 + (ω1·X1 + ω2·X2 + … + ω n ·X n ));

[0086] Among them, T base is the initial threshold, ω i is the weight coefficient of the i-th feature, X i is the adjustment coefficient of the i-th feature, and n is the number of features;

[0087] This result shows that by dynamically adjusting the segmentation threshold, the cell boundary can be accurately identified in the image, avoiding over-segmentation or under-segmentation, and thus improving the segmentation accuracy. In this example, if the initial segmentation threshold T base is 100, the weight coefficients ω1 = 0.7, ω2 = 0.3, and the adjustment coefficients X1 = 5, X2 = 3;

[0088] Then the finally calculated segmentation threshold is: T final = 100·(1 + (0.7·5 + 0.3·3)) = 100·(1 + 3.9) = 490

[0089] Therefore, the final segmentation threshold is 490.

[0090] Please refer to Figure 4 , the specific steps for obtaining the local texture information of the cell are as follows:

[0091] Based on the cell boundary segmentation result, the boundary is refined and corrected. The irregular edges are removed by local smoothing, the boundary is repaired by combining the cell contour features to avoid breaks and discontinuities, and the morphology is optimized according to the pixel distribution to obtain the corrected cell boundary;

[0092] First, smooth the boundaries in the original segmentation results. Use local pixel averaging to remove small sharp protrusions or noise. By calculating the local gradient values of boundary pixels, mark the sharp changes at the edges, and apply a gradually decreasing smoothing window for adjustment. This process ensures the coherence of the cell boundaries by combining cell contour features. For parts that are broken or discontinuous, according to the spatial distribution of surrounding pixels, use interpolation techniques to repair the boundaries. Interpolation not only considers pixel adjacency relationships but also combines the morphological features of the cells themselves, such as circularity, ellipticity, etc., to ensure that the repaired boundaries conform to the natural morphology of the cells. Then, perform morphological optimization. Use the method of pixel clustering to re-evaluate the boundaries, remove the irregular edges caused by incorrect segmentation, and finally obtain the corrected cell boundaries. The boundaries are smoother and more continuous, reducing the morphological distortion caused by segmentation errors.

[0093] Based on the corrected cell boundaries, extract geometric features. By identifying the area and perimeter of each cell, calculate the aspect ratio and shape factor using boundary points, analyze the symmetry and circularity of the cells, and combine the area-to-perimeter ratio to screen cell shape features, so as to obtain cell size and shape characteristics.

[0094] First, extract the area and perimeter of each cell. These two basic geometric features can be obtained by pixel counting of boundary points and the length of the boundary path. The area is obtained by calculating the number of pixel points within the enclosed region, and the perimeter is obtained by calculating the distances between boundary points one by one and summing them up. Calculate the aspect ratio of the cell, which is obtained by identifying the bounding box of the cell and calculating its dimensions. To accurately reflect the shape characteristics of the cell, the shape factor is also calculated. This factor combines the ratio of area to perimeter and quantitatively describes the circularity and symmetry of the cell. Analyze the symmetry and circularity of the cell shape. If the cell shows high symmetry and a shape close to a circle, it indicates good growth. Combine the ratio of area to perimeter to screen out eligible cell shape features, obtain cell size and shape characteristics, and provide a reference for further biological analysis.

[0095] Based on cell size and shape characteristics, select local regions for texture analysis. Statistically analyze the pixel value changes in the local regions of the cells, quantify the texture variation degree and consistency, combine the boundary information to analyze the texture distribution pattern, and evaluate the texture differences and consistency between regions to obtain local cell texture information.

[0096] First, according to the size and shape characteristics of cells, a local area of the cell is selected. The area represents the active part or the area with special functions of the cell. By counting the pixel values within the local area, the range of change and distribution of the pixel values are calculated to quantify the degree of variation and consistency of the texture. To quantify the degree of variation of the texture, the roughness and uniformity of the texture can be obtained by calculating the gradient difference between adjacent pixel values within the local area, and the consistency is evaluated by counting the distribution balance of the pixel values. Next, combined with the boundary information, the distribution pattern of the local texture is analyzed, the texture characteristics in the area near the boundary are observed, and the texture differences between different areas are compared. Such texture differences not only help to understand the internal structural characteristics of the cell, but also can reveal the functional state or health status of the cell. By operating to obtain the local texture information of the cell, it is used to support further cell analysis and evaluation.

[0097] Please refer to Figure 5 , the steps for obtaining the cell morphological feature dataset are specifically as follows:

[0098] According to the local texture information of the cell, extract the contrast between the cell edge, texture pattern and regions. Combined with region division, identify the gray-scale change, local contrast, texture direction parameters of the image, analyze the geometric differences between local regions, and obtain the cell texture features;

[0099] First, perform image segmentation on each local area. Use image processing algorithms to segment the image into multiple regions of interest, and by calculating the texture features within each region, such as indicators like local contrast, gray-level co-occurrence matrix (GLCM), etc., extract the texture feature values of each region. Further, reflect the structural features such as the cell edge or cell membrane through the regional texture contrast. By calculating the comparison of regional gray values, extract the texture features of each region. By comparing the features with the original image, analyze the morphological differences of the cells, and finally obtain the cell texture features, and further quantify the texture differences of each region.

[0100] Based on the cell texture features, quantify the morphological differences and geometric features of the cells between the differentiated regions, compare them with the texture similarity measurement indicators between regions, and use the formula:

[0101]

[0102] Obtain the cell morphological difference data;

[0103] where, δ represents the local texture difference degree, T r,j is the texture feature of the right region, T l,j is the texture feature of the left region, and N is the number of texture points;

[0104] The advantage of the formula is that through texture contrast analysis, the morphological differences of cells can be quantified, and the accuracy of cell morphological feature extraction can be improved;

[0105] δ represents the local texture difference degree, which measures the morphological difference by calculating the texture difference between the right region and the left region of the cell, T r,j is the texture feature of the right region, T l,j is the texture feature of the left region, N is the number of texture points. When calculating, the sum of the squared differences of each pair of compared textures is calculated and averaged to obtain the difference degree of this region. Set the texture value of the right region as T r,j =[5, 8, 7] and the texture value of the left region as T l,j =[4, 9, 6], and the operation is carried out according to the formula:

[0106]

[0107] The result shows that the texture difference degree between the right region and the left region of the cell is 1, indicating that there are certain differences in their local texture features, which in turn affects the analysis of cell morphological difference data.

[0108] Based on the cell morphological difference data, through clustering and segmentation methods, the geometric and structural features of the cells are refined. Using distance measurement judgment, the regions with the same type of geometric features are grouped to obtain the cell morphological feature dataset;

[0109] Use the clustering algorithm to group the data. The regions with the same type of geometric and structural features are divided into the same group. Adopt the clustering method based on Euclidean distance. First, calculate the texture difference degree of each pair of regions. Use the previously calculated difference degree as the sample data. Through the hierarchical clustering algorithm, gradually classify the regions with the same type of texture together. Finally, generate the cell region grouping with obvious features, and further optimize and integrate to obtain the cell morphological feature dataset. By setting thresholds to identify the significant regions of cell morphological changes, mark the regions with larger changes, and further refine the geometric structure features of the regions.

[0110] Please refer to Figure 6 , the specific steps for obtaining the analysis result of cell behavior trend are as follows:

[0111] Analyze each frame of image in the cell morphological feature dataset, extract the cell position and identify the distance between adjacent frames. Use the distance to judge the cell division, aggregation and migration behaviors. Through the time series, and combine the image pixel positions to track the cell positions to obtain the cell behavior tracking data;

[0112] For each frame of the cell morphological feature dataset, the position of the cells needs to be extracted through image processing methods. The coordinates of the cells can be obtained from the pixel positions of the image. For each frame of the image, first define a reference area, compare the position changes of the cells within the area, calculate the displacement of the cells frame by frame, and use the pixel differences of the image to represent the movement of the cells. To identify cell division, aggregation, and migration behaviors, it is necessary to calculate the movement speed and direction of the cells per unit time based on the position differences between adjacent images of each frame. By calculating the displacement of the cells in each frame, the distance between adjacent frames is obtained, and further, the behavior trend of the cells is evaluated through displacement calculation to obtain the dynamic behavior of each cell. Taking a cell as an example, assume that in the first frame, the coordinates of the cell are (100, 200), and in the second frame, the coordinates of the cell are (110, 210). By calculating the distance between these two points, the movement distance of the cell is obtained, and the calculation result finally yields the cell behavior tracking data.

[0113] Based on the cell behavior tracking data, evaluate the dynamic behavior of the cells according to the movement trajectories and change trends between adjacent frames, calculate the displacement distance value of the cells between adjacent frames, and use the formula:

[0114]

[0115] Obtain the cell movement trend data;

[0116] where d represents the displacement distance value of the cells between adjacent frames, x t+1 and y t+1 are the coordinates of the cell in the (t + 1)-th frame, x t and y t are the coordinates of the cell in the t-th frame;

[0117] The advantage of the formula is that by calculating the displacement of the cells between adjacent frames, the dynamic behavior of the cells can be quantified, and the movement trajectory of the cells can be accurately tracked;

[0118] d represents the displacement of the cells between adjacent frames, x t+1 and y t+1 are the coordinates of the cell in the (t + 1)-th frame, x t and y t are the coordinates of the cell in the t-th frame. When calculating, first take the cell position coordinates in two adjacent frame images, then calculate using the square of the coordinate difference, and then take the square root to obtain the actual distance between the two frames. In practical applications, assume that the cell position in the first frame is (100, 200) and the cell position in the second frame is (110, 210), then calculate according to the formula:

[0119]

[0120] The results show that the moving distance value of the cells between two adjacent frames is about 14.14 pixels, which further reflects the dynamic change trend of the cells and helps to analyze the behaviors of cell division, aggregation, migration, etc.

[0121] Based on the cell movement trend data, analyze the behaviors of cell division, aggregation and migration, evaluate the cell behavior trend, analyze whether there is a change trend, judge the key points of cell behavior changes by setting thresholds, and obtain the cell behavior trend analysis results;

[0122] First, analyze the displacement data of all cells, screen the cells with significant behavior changes, use the threshold method to judge whether the behaviors of the cells show the characteristics of aggregation, division or migration, judge the significance of cell behaviors by setting thresholds. If the displacement speed of a cell exceeds the set threshold within a specific time period, it can be determined that the cell has a migration behavior. If the displacement amplitude is small, it means that the cell is in a stationary or division stage. Based on this, the cells can be divided into different groups, classified according to the movement trend, and then the laws of cell behavior changes can be analyzed. For the actual application scenario, a specific displacement threshold (such as 10 pixels) can be set to distinguish the migration behavior from the stationary behavior. If the cell displacement is greater than 10 pixels, it is considered that the cell shows a migration behavior, otherwise it is a division or stationary behavior. Finally, the cell behavior trend analysis results are formed.

[0123] Please refer to Figure 7 , and the steps for obtaining the distance and corresponding speed between cells are specifically as follows:

[0124] Based on the cell behavior trend analysis results, analyze the positional relationship between cells, extract the coordinate information of each pair of cells, identify the straight-line distance between cells, extract cell coordinates and track the position changes of each pair of cells to obtain the distance dataset between cells;

[0125] First, extract the position coordinates of the cells, obtain the coordinate points of the cells in each frame of the image. The coordinate points are automatically extracted by image processing technology, and the gray-scale changes of the image pixels are used to calibrate the precise positions of the cells. Use the cell coordinates in adjacent frames to calculate the straight-line distance between each pair of cells. The distance calculation method uses the Euclidean distance formula to obtain the distance between each pair of cells. Through the coordinate differences between consecutive time frames, the corresponding position changes between cells can be tracked, providing basic data for subsequent behavior trend analysis. Suppose in the first frame, the coordinates of cell A are (100, 150), the coordinates of cell B are (120, 180), in the second frame, the coordinates of cell A are (110, 160), and the coordinates of cell B are (125, 185). Calculate the distance changes of cell A and cell B between the first frame and the second frame, and obtain the relative distance data between the corresponding cells through the formula calculation. Finally, obtain the distance dataset between cells.

[0126] Based on the intercellular distance dataset, calculate the velocity between cells according to the displacement of each pair of cells in adjacent frames, perform differentiation through the difference in cell coordinates, analyze the change in the velocity between cells, and use the formula:

[0127]

[0128] Obtain the distance between cells and the corresponding velocity;

[0129] Among them, v represents the velocity value of each pair of cells, D t+1 and D t are the distances between cells between the (t + 1)-th frame and the t-th frame, and Δt is the time interval;

[0130] The benefit of the formula is that through the calculation of the corresponding velocity between cells, the movement rate between cells can be quantified, and the dynamic behavior pattern of cells can be further analyzed;

[0131] v represents the corresponding velocity between cells, D t+1 and D t are the distances between cells between the (t + 1)-th frame and the t-th frame respectively, and Δt is the time interval. When calculating, the corresponding velocity between cells is obtained by dividing the change in distance by the time interval, which helps to evaluate whether the cells are aggregating or dispersing. Assume that in the first frame and the second frame, the distances between cell A and cell B are 10 and 15 pixels respectively, and the time interval is 1 second;

[0132] Then the corresponding velocity between cells is:

[0133] This result shows that the corresponding velocity between cell A and cell B is 5 pixels per second, and the cells are rapidly dispersing. Further analysis of the behavior pattern will infer whether the cells will continue to disperse or tend to aggregate based on this velocity, and obtain the distance between cells and the corresponding velocity.

[0134] Please refer to Figure 8 , and the specific steps for obtaining the monitoring results of the cooperative behavior of cell populations are as follows:

[0135] Based on the distance between cells and the corresponding velocity, capture the spatial position coordinates and motion states of each cell, monitor the corresponding velocity of the cells, record the distances and velocities between each cell and its surrounding cells, and generate a cell population position and velocity dataset;

[0136] First, the spatial position coordinates of each cell in the microscope image are determined through image recognition technology. The coordinates are converted from the pixel positions of the cells in the image to the actual spatial positions. The motion state of each cell is obtained by comparing consecutive image frames, calculating the displacement of the cell and the change of displacement over time, and then deriving its corresponding velocity. To accurately monitor the corresponding velocities between cells, the distances between each cell and its surrounding neighboring cells are calculated in real time. By comparing their respective spatial position coordinates, the straight-line distance between each pair of cells is obtained. The velocities of each cell and its surrounding cells are calculated by tracking the change of their positions over time, thereby obtaining the motion direction and velocity of the cells. This involves precise matching and trajectory tracking of cells between consecutive image frames, generating a dataset of cell population positions and velocities, which provides a basis for subsequent cell behavior analysis.

[0137] Based on the dataset of cell population positions and velocities, analyze the corresponding positions and motion states between each cell and its surrounding cells, screen out the cells with interaction strength greater than the set threshold, delimit the interaction range of each pair of cells according to distance, velocity, and interaction time, and construct a dataset of cell - cell interaction relationships;

[0138] First, use the spatial position and velocity data of the cell population to calculate the corresponding positions and motion states of each cell and its surrounding cells. The core of the analysis is to evaluate the relative motion direction and velocity change between cells. By calculating the corresponding velocities of each pair of cells, it is evaluated whether they are approaching or moving away from each other. According to the preset interaction strength threshold, cell pairs with significant interactions are screened out. When the corresponding velocity between cells is less than a certain specific value, it is considered that there is a strong interaction between the cells. Based on the screened cell pairs, combined with the spatial distance, velocity difference, and interaction duration, the interaction range of each pair of cells is further delimited. The range is determined by calculating the change trend of the interaction time and distance between cells, and a dataset of cell - cell interaction relationships is constructed. Through the dataset, a more in - depth analysis of the behavior patterns and interactions between cells can be carried out.

[0139] Use the dataset of cell - cell interaction relationships to analyze the cooperative behavior of the cell population. By the interactions between cells and the population trend, judge the dynamic changes of the population behavior and obtain the monitoring results of the cell population cooperative behavior;

[0140] First, conduct an in-depth analysis of the dataset of cell-cell interaction relationships, evaluate the intensity and time span of cell-cell interactions, so as to reveal the collaborative behavior patterns among cell populations. Through the data of the distance and corresponding speed between cells, identify the cell pairs with close cooperation in the population, and then analyze how cells adjust their population behavior through interactions. The behavioral changes of cell populations are manifested as the enhancement or weakening of cell-cell interactions. Based on the interaction information in the dataset, the changing trend of synergistic effects within the population can be traced. The interaction between cells not only depends on physical distance, but is also affected by factors inside and outside the cell. Therefore, dynamically monitoring the changes in cell behavior is crucial for understanding the health status of cell populations. Through continuous monitoring and data update, the dynamic changes in the behavior of cell populations can be accurately judged, and the monitoring results of the collaborative behavior of cell populations can be generated.

[0141] The above are only the preferred embodiments of the present invention, and do not limit the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A CAR-T cell culture monitoring system based on image recognition, characterized in that The system includes: The cell region division module analyzes the gradient intensity of the cell culture image based on the cell culture image, divides the image into a high-feature region and a low-feature region, performs cell edge detection on the high-feature region, extracts the edge position and cell morphological features, refines the cell shape and texture, dynamically adjusts the segmentation threshold, and obtains the cell boundary segmentation result; The cell morphology analysis module analyzes the morphology of the cells in the high-feature region based on the cell boundary segmentation result, extracts the local texture information of the cells, analyzes the cell morphological differences through local feature comparison, refines the geometric and structural features of the cells, and obtains the cell morphology feature dataset; The cell behavior dynamic analysis module tracks and analyzes the cell behavior based on the cell morphology feature dataset, monitors the cell division, aggregation, and migration behaviors, evaluates the distance and behavior trend through adjacent frame images, records the dynamic behavior of the cells, analyzes the dynamic changes in cell culture, and obtains the cell behavior trend analysis result; The cell population cooperation monitoring module analyzes the cooperation between cell populations based on the cell behavior trend analysis result, monitors the trend of cell aggregation and dispersion, combines the mutual relationship between cells, calculates the distance and corresponding speed between cells, evaluates the overall behavior pattern of the cell population, dynamically monitors the cooperation effect of the cell population, and obtains the cell population cooperation behavior monitoring result.

2. The CAR-T cell culture monitoring system based on image recognition according to claim 1, characterized in that, The specific steps for obtaining the edge position and cell morphological features are as follows: Based on the cell culture image, convert the image into a grayscale image, analyze the gray difference between the brightness value of each pixel point and the surrounding pixels, calculate the gradient value of each pixel, use the gradient value for image region division, and judge the region features based on the gradient value to obtain the high-low feature region division result; Based on the high-low feature region division result, perform region enhancement processing on the pixels in the high-feature region, analyze the local contrast of each pixel in the high-feature region, and allocate weights according to the difference in surrounding pixel values to obtain the enhanced data for the corresponding region; Based on the enhanced data for the corresponding region, locate the edge position of the cell through the contour, analyze the gradient change of each pixel in the high-feature region, judge the edge pixel points, and analyze the cell morphology based on the edge information to obtain the edge position and cell morphological features.

3. The CAR-T cell culture monitoring system based on image recognition according to claim 2, characterized in that, The specific steps for obtaining the cell boundary segmentation result are as follows: Based on the edge position and cell morphological features, extract the edge information in the cell image and identify the position information of each cell boundary. By analyzing the cell morphological features, including curvature and edge smoothness, generate the cell boundary position feature dataset; Combine the cell boundary position feature dataset with the cell texture information to extract the texture features in the image, analyze the change law of the texture features in the differential edge region, and dynamically adjust the segmentation threshold to obtain the segmentation threshold adjustment coefficient; According to the segmentation threshold adjustment coefficient, use the optimal matching of cell morphology and texture features, and adopt the formula: Generate the cell boundary segmentation result; Among them, T final represents the cell boundary segmentation threshold, T base represents the initial segmentation threshold, ω i represents the weight coefficient, X i represents the adjustment coefficient related to the association between cell morphology and texture features, and n is the number of features.

4. The CAR-T cell culture monitoring system based on image recognition according to claim 3, characterized in that, The specific steps for obtaining the cell local texture information are as follows: Based on the cell boundary segmentation results, boundary refinement and correction are carried out. Irregular edges are removed through local smoothing, and the boundary is repaired by combining cell contour features to avoid breaks and discontinuities. Morphological optimization is performed according to pixel distribution to obtain the corrected cell boundary. Based on the corrected cell boundary, geometric feature extraction is carried out. By identifying the area and perimeter of each cell, the aspect ratio and shape factor are calculated using boundary points, the symmetry and circularity of the cells are analyzed, and the cell shape is screened by combining the area-to-perimeter ratio to obtain the cell size and shape characteristics. Based on the cell size and shape characteristics, a local area is selected for texture analysis. The pixel value changes in the local area of the cell are statistically analyzed, the texture variation degree and consistency are quantified, the texture distribution pattern is analyzed by combining boundary information, and the texture difference and consistency between regions are evaluated to obtain the local texture information of the cell.

5. The CAR-T cell culture monitoring system based on image recognition according to claim 4, wherein, The specific steps for obtaining the cell morphological feature dataset are as follows: According to the local texture information of the cell, the contrast between the cell edge, texture pattern, and regions is extracted. By combining region division, the gray-scale change, local contrast, and texture direction parameters of the image are identified, and the geometric differences between local regions are analyzed to obtain the cell texture features. Based on the cell texture features, the morphological differences and geometric features between cells in different regions are quantified and compared with the texture similarity metric index between regions. Using the formula: Cell morphological difference data is obtained. Among them, δ represents the local texture difference degree, T r,j is the texture feature of the right region, T l,j is the texture feature of the left region, and N is the number of texture points; Based on the cell morphological difference data, through clustering and segmentation methods, the geometric and structural features of the cells are refined. Using distance measurement judgment, regions with the same geometric features are grouped to obtain the cell morphological feature dataset.

6. The CAR-T cell culture monitoring system based on image recognition according to claim 5, wherein, The specific steps for obtaining the analysis result of cell behavior trend are as follows: Each frame of image in the cell morphological feature dataset is analyzed. Cell position extraction and adjacent frame distance recognition are carried out. Cell division, aggregation, and migration behaviors are judged using distance. Through time series and combined with the pixel position of the image, cell position tracking is performed to obtain cell behavior tracking data. Based on the cell behavior tracking data, according to the movement trajectory and change trend between adjacent frames, the dynamic behavior of the cells is evaluated. The displacement distance value between adjacent frames of the cells is calculated. Using the formula: Cell movement trend data is obtained. where d represents the displacement distance value of the cell between adjacent frames, x t+1 and y t+1 are the coordinates of the cell in the (t + 1)-th frame, and x t and y t are the coordinates of the cell in the t-th frame; Based on the cell movement trend data, cell division, aggregation, and migration behaviors are analyzed, the cell behavior trend is evaluated, and whether there is a change trend is analyzed. By setting a threshold, the key point of cell behavior change is judged to obtain the analysis result of cell behavior trend.

7. The CAR-T cell culture monitoring system based on image recognition according to claim 6, characterized in that, The specific steps for obtaining the distance between cells and the corresponding speed are as follows: Based on the analysis result of cell behavior trend, the positional relationship between cells is analyzed. The coordinate information of each pair of cells is extracted, the straight-line distance between cells is identified, the cell coordinates are extracted, and the position changes of each pair of cells are tracked to obtain the distance dataset between cells. Based on the distance dataset between cells, the speed between cells is calculated according to the displacement of each pair of cells in adjacent frames. By taking the derivative of the cell coordinate difference, the speed change between cells is analyzed. Using the formula: The distance between cells and the corresponding speed are obtained. where v represents the velocity value of each pair of cells, D t+1 and D t is the intercellular distance between the (t + 1)-th frame and the t-th frame, and Δt is the time interval.

8. The CAR-T cell culture monitoring system based on image recognition according to claim 7, wherein, The specific steps for obtaining the monitoring result of cell population collaborative behavior are as follows: Capture the spatial position coordinates and motion states of each cell based on the distance and corresponding speed between cells, monitor the corresponding speed of cells, record the distances and speeds of each cell from its surrounding cells, and generate a cell population position and speed data set; Based on the cell population position and speed data set, analyze the corresponding positions and motion states between each cell and its surrounding cells, screen cells with an interaction strength greater than a set threshold, delimit the interaction range of each pair of cells according to distance, speed, and interaction time, and construct a cell-cell interaction relationship data set; Utilize the cell-cell interaction relationship data set to analyze the cooperative behavior of the cell population, judge the dynamic changes of the population behavior through the interactions between cells and the population trend, and obtain the monitoring results of the cell population cooperative behavior.

Images