A method and system for analyzing the state of immune cells

Abstract

The present invention discloses a state analysis method and system for immune cells, and relates to the field of image processing technology. The method includes: using a microscopic imaging device to collect a motion image sequence of immune cells; performing edge detection on the motion image sequence, marking the contour pixel data label of the immune cells; tracking the trajectory of the immune cells in a time series, and outputting motion trajectory pixel data; inputting the motion trajectory pixel data into a dynamic change rate recognition model, obtaining the first-order change rate, the second-order change rate, and the change rate mean, and generating a dynamic change feature embedding vector; inputting the state prediction model of the immune cell to perform state prediction, obtaining an activity state prediction index, and marking the activity of the immune cells according to the activity state prediction index. The technical problem of inaccurate analysis and incomplete prediction of the activity state of immune cells in the prior art is solved, and the technical effect of improving the accuracy and comprehensiveness of the prediction of the activity state of immune cells is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a method and system for analyzing the state of immune cells. Background Art

[0002] The activity state of immune cells is an important indicator for evaluating the body's immune function, screening the efficacy of potential drugs, and formulating personalized treatment plans. Immune cells play a vital role in the body's defense mechanism, and their activity state directly affects the occurrence and development of diseases and the evaluation of treatment effects. Therefore, accurate analysis of the activity state of immune cells can not only reveal their functional characteristics, but also provide scientific support for the diagnosis and treatment of related diseases. However, the current traditional analysis methods of immune cell activity state mainly rely on static observation or manual detection, which is limited by detection conditions and experimental environment, and has certain subjectivity and errors. At the same time, because immune cells themselves have significant dynamics, their activity state is often accompanied by movement behavior and morphological changes. Therefore, it is difficult to fully reflect their true state only through static data, especially the small changes in dynamic behavior are often ignored. This technical limitation restricts the accuracy and comprehensiveness of the analysis results, seriously affecting the in-depth development of immune cell research and its potential for clinical application. Summary of the invention

[0003] The present application provides a method and system for analyzing the status of immune cells, which solves the technical problems of inaccurate analysis and incomplete prediction of the activity status of immune cells in the prior art.

[0004] In view of the above problems, the present application provides a method and system for analyzing the status of immune cells.

[0005] In a first aspect of the present application, a method for analyzing the state of immune cells is provided, the method comprising:

[0006] Use a microscopic imaging device to collect a motion image sequence of immune cells; perform edge detection on the motion image sequence and mark the contour pixel data labels of the immune cells; track the trajectory of the immune cells in a time series according to the contour pixel data labels, and output motion trajectory pixel data; input the motion trajectory pixel data into a dynamic change rate recognition model, obtain the first-order change rate, the second-order change rate and the mean of the change rate according to the dynamic change rate recognition model, and generate a dynamic change feature embedding vector from the first-order change rate, the second-order change rate and the mean of the change rate; input the dynamic change feature embedding vector into the immune cell state prediction large model for state prediction, obtain the activity state prediction index, and mark the activity of the immune cells according to the activity state prediction index.

[0007] The second aspect of the present application provides a system for analyzing the state of immune cells, the system comprising:

[0008] Image acquisition module: uses a microscopic imaging device to collect a motion image sequence of immune cells; edge detection module: performs edge detection on the motion image sequence and marks the contour pixel data labels of the immune cells; trajectory tracking module: tracks the trajectory of the immune cells in a time series according to the contour pixel data labels, and outputs motion trajectory pixel data; recognition module: inputs the motion trajectory pixel data into a dynamic change rate recognition model, obtains the first-order change rate, the second-order change rate and the mean change rate according to the dynamic change rate recognition model, and generates a dynamic change feature embedding vector from the first-order change rate, the second-order change rate and the mean change rate; state prediction module: inputs the dynamic change feature embedding vector into the immune cell state prediction large model for state prediction, obtains an activity state prediction index, and performs activity marking on the immune cells according to the activity state prediction index.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0010] First, a motion image sequence of immune cells is collected using a microscopic imaging device. Next, edge detection is performed on the motion image sequence, and the contour pixel data labels of the immune cells are marked. Further, the trajectory of the immune cells is tracked in a time series according to the contour pixel data labels, and the motion trajectory pixel data is output. Then, the motion trajectory pixel data is input into the dynamic change rate recognition model, and the first-order change rate, the second-order change rate and the mean of the change rate are obtained according to the dynamic change rate recognition model. The first-order change rate, the second-order change rate and the mean of the change rate are used to generate a dynamic change feature embedding vector. Finally, the dynamic change feature embedding vector is input into the immune cell state prediction model for state prediction, and the activity state prediction index is obtained. The immune cells are marked for activity according to the activity state prediction index. The technical problems of inaccurate analysis and incomplete prediction of the activity state of immune cells in the prior art are solved, and the technical effect of improving the accuracy and comprehensiveness of the prediction of the activity state of immune cells is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0012] Figure 1 A schematic diagram of a method for analyzing the state of immune cells provided in an embodiment of the present application;

[0013] Figure 2 A schematic diagram of the structure of an immune cell status analysis system provided in an embodiment of the present application.

[0014] Explanation of reference numerals: image acquisition module 11 , edge detection module 12 , trajectory tracking module 13 , recognition module 14 , state prediction module 15 . DETAILED DESCRIPTION

[0015] The present application solves the technical problems of inaccurate analysis and incomplete prediction of the activity status of immune cells in the prior art by providing a method and system for analyzing the status of immune cells.

[0016] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0017] It should be noted that the terms "including" and "having" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or are inherent to these processes, methods, products or devices.

[0018] Embodiment 1, as Figure 1 As shown, the present application provides a method for analyzing the state of immune cells, wherein the method comprises:

[0019] Use a microscopic imaging device to capture moving image sequences of immune cells.

[0020] Microscopic imaging equipment is a high-resolution imaging system that usually includes a microscope, an imaging sensor (such as a CCD or CMOS camera) and a light source, which is used to capture the subtle movement characteristics of immune cells at high magnification. Through microscopic imaging equipment, immune cells can be continuously monitored in liquid culture medium or a specific environment, and their dynamic behavior within a certain time range can be recorded to obtain a sequence of immune cell motion images. These motion image sequences not only contain the position changes of cells, but also carry the time series information of cell movement characteristics, which lays an important foundation for analyzing their activity status.

[0021] Edge detection is performed on the motion image sequence, and contour pixel data labels of the immune cells are marked.

[0022] The motion image sequence is processed by edge detection algorithms (such as Canny or Sobel edge detection) to identify the outlines of immune cells in the image, mark the detected cell edges, and convert the cell outlines into pixel data labels. Specifically, the pixel points belonging to the cell edge in the image are identified to form a closed or semi-closed outline; a unique label (such as a number or a specific color) is assigned to each cell outline pixel to distinguish different cells in the subsequent trajectory tracking process.

[0023] Furthermore, edge detection is performed on the motion image sequence to mark the contour pixel data labels of the immune cells, and the method includes:

[0024] The motion image sequence is preprocessed to obtain a preprocessed motion image sequence; the gradient of the preprocessed motion image sequence is calculated by a Sobel operator to output a gradient amplitude; a double edge threshold is set, and the gradient amplitude is assigned according to the double edge threshold to obtain an edge image of the immune cell; and according to the edge image of the immune cell, the contour pixel data of the immune cell is marked.

[0025] Preprocess the motion image sequence to improve the effect of edge detection; preprocessing includes denoising, grayscale and brightness standardization of the image; denoising is usually performed through Gaussian filtering to eliminate random noise in the image while retaining the edge features of the cells; grayscale processing converts color images into grayscale images to reduce computational complexity; brightness standardization ensures that the image has consistent contrast and brightness between different frames through histogram equalization or brightness normalization.

[0026] After the image preprocessing is completed, the Sobel operator is used to calculate the gradient of the preprocessed motion image sequence. Specifically, the Sobel operator is used to calculate the gradient values ​​(i.e., gray value change rate) of the image in the horizontal and vertical directions respectively, and the horizontal gradient G is obtained. x and the vertical gradient G y ; Based on these gradient values, the gradient amplitude G is calculated, which is a comprehensive indicator of the change in pixel grayscale value. The formula is: The higher the gradient magnitude, the more likely the pixel is to be an edge in the image.

[0027] The double edge threshold is set to process the gradient amplitude to obtain the edge image of the immune cell; the double edge threshold includes a high threshold and a low threshold. For the gradient amplitude higher than the high threshold, it is marked as a strong edge; the gradient amplitude between the high threshold and the low threshold is marked as a weak edge; the gradient amplitude lower than the low threshold is ignored. Through connectivity analysis, the weak edges connected to the strong edges are also marked as valid edges, thereby generating a complete edge image. The edge of each immune cell is traced by the edge-detected image to extract its closed or semi-closed boundaries, that is, the outline of each cell in the edge image. These outlines are usually represented by a set of closed pixel coordinates, which constitute the outline pixel data label of each cell.

[0028] The immune cell is tracked in time series according to the contour pixel data label, and the motion trajectory pixel data is output.

[0029] According to the marked contour pixel data labels, the trajectory of the immune cells is tracked in time series, and the motion trajectory pixel data of the immune cells are output, which describes the position changes of the immune cells in continuous image frames.

[0030] Furthermore, the immune cell is tracked in a time series according to the contour pixel data label, and the motion trajectory pixel data is output, and the method includes:

[0031] Establish an immune cell association relationship between each frame image in the motion image sequence; locate the continuous contour pixel data matching the contour pixel data label in each frame image according to the association relationship; connect the continuous contour pixel data according to the time series of the motion image sequence, and output motion trajectory pixel data.

[0032] By analyzing the centroid position, contour area and shape of each immune cell in adjacent frames, the continuity of these features in time is determined, so as to establish the immune cell association relationship between each frame image. Specifically, the Euclidean distance of the centroid of each immune cell contour in adjacent frames is calculated, and the two centroids with the smallest distance are matched. At the same time, the accuracy of the match is further confirmed by combining the similarity of contour area and shape. Optionally, the area of ​​the immune cell contour in the current frame and the previous frame is calculated, which is usually obtained by counting the number of contour pixels; the area difference of adjacent contours in the two frames is compared. If the area difference exceeds the set threshold (for example, the area change exceeds 30%), it is judged that the two may not be the same cell, thereby avoiding mismatching; secondly, the geometric consistency of the adjacent contour shapes in the two frames is analyzed by using shape similarity measurement, such as calculating the Hu moment feature or perimeter ratio of the contour. If the characteristic value of the shape changes too much, it can also be used as a basis for invalid matching. By comprehensively considering the three key parameters of centroid distance, contour area and shape similarity, the association relationship of immune cells in adjacent frames can be more accurately determined, thereby effectively avoiding mismatching and trajectory confusion, and providing a reliable data basis for the final generation of accurate motion trajectories.

[0033] According to the established association relationship, the continuous contour data of each immune cell matching its contour pixel data label in the time series is located frame by frame. In each frame, the contour pixel data associated with the current immune cell is extracted, and its centroid coordinates and corresponding pixel point set are recorded. Through this frame-by-frame matching and positioning, it is ensured that the contour data of each immune cell can be fully tracked in the time series. According to the time sequence of the motion image sequence, the matching data of each immune cell in each frame are connected in series to generate the motion trajectory pixel data of the immune cell.

[0034] The motion trajectory pixel data is input into a dynamic change rate recognition model, and the first-order change rate, the second-order change rate and the mean change rate are obtained according to the dynamic change rate recognition model. The first-order change rate, the second-order change rate and the mean change rate are used to generate a dynamic change feature embedding vector.

[0035] The pixel data of the immune cell's motion trajectory contains the position information of the immune cells in the time series, that is, the pixel coordinates of each time point; the pixel data of the immune cell's motion trajectory is input into the dynamic change rate recognition model, which can calculate the first-order change rate (i.e., speed), second-order change rate (i.e., acceleration) and the mean of the change rate of the immune cell's motion trajectory. These dynamic change rates reflect the speed, direction change and other characteristics of the immune cell's movement; the calculated first-order change rate, second-order change rate and mean of the change rate are combined into a dynamic change feature embedding vector, which is a high-dimensional feature representation that contains the dynamic feature information of the immune cell's movement.

[0036] Furthermore, the motion trajectory pixel data is input into a dynamic change rate recognition model, and the first-order change rate, the second-order change rate and the mean change rate are obtained according to the dynamic change rate recognition model; the motion trajectory pixel data is subjected to pixel coordinate change analysis according to the dynamic change rate recognition model, and the pixel displacement change speed and the pixel displacement change acceleration in each sequence, as well as the mean pixel displacement change speed and the mean pixel displacement change acceleration in the motion image sequence are calculated; the pixel displacement change speed in each sequence is output as a first-order change rate; the pixel displacement change acceleration in each sequence is output as a second-order change rate; the mean pixel displacement change speed and the mean pixel displacement change acceleration in the motion image sequence are output as the mean change rate.

[0037] After the pixel data of the motion trajectory is input into the dynamic change rate recognition model, the pixel coordinates of the motion trajectory are first analyzed for changes, and the pixel displacement change speed and acceleration of the immune cells in the time series are calculated frame by frame. By analyzing the pixel coordinate changes of the center of mass in every two consecutive frames, the pixel displacement change speed is calculated to represent the first-order change rate of the cell, reflecting its movement intensity. At the same time, the pixel displacement change acceleration is further calculated based on the change in speed as the second-order change rate, which is used to describe the fluctuation amplitude and dynamic change characteristics of the cell movement state. In addition, the speed and acceleration in the entire motion image sequence are averaged respectively, and the mean pixel displacement change speed and the mean pixel displacement change acceleration are calculated as the global features of the cell motion trajectory, reflecting its overall dynamic behavior. Finally, the dynamic change rate recognition model outputs the frame-by-frame pixel displacement change speed as the first-order change rate, the frame-by-frame pixel displacement change acceleration as the second-order change rate, and the mean speed and mean acceleration in the motion image sequence as the mean change rate.

[0038] The dynamic change feature embedding vector is input into the immune cell state prediction large model to perform state prediction, obtain the activity state prediction index, and the immune cell is marked for activity according to the activity state prediction index.

[0039] The dynamically changing feature embedding vector is input into the immune cell state prediction model. The immune cell state prediction model predicts the state based on the input feature vector and outputs the activity state prediction index of the immune cell, such as whether the cell is in a high-activity state, a low-activity state or an inactive state. The immune cells are marked for activity based on the activity state prediction index. For example, the high-activity cells in the prediction results are marked as "active", the low-activity cells are marked as "static", and the inactive cells are marked as "dormant". This marking can be presented visually (for example, different colors are used to represent different states in the cell image).

[0040] Furthermore, the dynamic change feature embedding vector is input into the immune cell state prediction model for state prediction. The expression of the immune cell state prediction model is as follows: ; Among them, y i is the cell state prediction result after analysis based on the dynamic change feature embedding vector. Softmax is used to map the input dynamic change feature embedding vector to a probability distribution function for classification. j is the weight of each feature in the dynamically changing feature embedding vector, is the vector value of the jth dynamically changing feature in the normalized input data, b j is the bias term to be corrected, j is the index of the feature, and n is the number of features in the dynamically changing feature embedding vector.

[0041] y i It is the cell state prediction result obtained after analyzing the dynamic change feature embedding vector. It is a multi-classification probability distribution that indicates the possibility of immune cells being in different states (such as active, static, dormant, etc.).

[0042] Normalize each eigenvalue F[j] in the dynamically changing feature embedding vector to obtain the normalized eigenvalue Normalization can eliminate the dimensional differences of different eigenvalues ​​and ensure that the eigenvalues ​​of the input data are within a unified range, which is beneficial to improving the calculation stability and accuracy of the model. j and the bias term b j , calculate the contribution of each feature to the cell state prediction result, weight w j Indicates the importance of the jth feature to the prediction result, bias term b j Used to correct the deviation of feature input. After combining the weighted sum of all features, the original score of cell state prediction is obtained. The result of the above weighted sum is input into the Softmax function to generate a probability distribution for classification. The function of Softmax is to map any real value to between [0,1], and the sum of the probabilities of all categories is 1. Where C is the number of possible cell states, y k For each category, the score corresponding to the Softmax output y i Indicates the probability that the immune cell belongs to the i-th category state. According to the probability distribution of the Softmax output, the state with the highest probability is selected as the predicted state of the immune cell. For example, if the predicted probability of a cell is the largest in the "active" state, the cell is marked as "active".

[0043] Furthermore, the method of training a large model for predicting immune cell status also includes:

[0044] Perform visual image recognition on the motion image sequence to obtain visual morphological features of the immune cells, wherein the visual morphological features include cell area, cell shape and cell texture; perform cluster analysis based on the visual morphological features to determine cell subtypes; obtain immune cell samples based on the cell subtypes and obtain morphological feature embedding vector samples; perform modal fusion training on the immune cell state prediction large model based on the morphological feature embedding vector samples, and output the immune cell state prediction large model.

[0045] Optionally, visual image recognition is performed on the motion image sequence to extract the visual morphological features of the immune cells. These visual morphological features include the area, shape and texture information of the cells. For example, the cell area can be obtained by calculating the number of pixels of the cell contour, the shape feature can be extracted by analyzing the geometric parameters of the cell contour (such as roundness, aspect ratio, etc.), and the texture feature is quantified by image processing methods (such as gray-level co-occurrence matrix or local binary pattern) to quantify the fine structural features inside the cell; based on the extracted visual morphological features, the immune cells are clustered and analyzed, and the immune cells are divided into different subtypes according to the similarity of the cell's area, shape, texture and other features through clustering algorithms (such as K-means clustering, hierarchical clustering or DBSCAN algorithm). Cluster analysis can identify cell populations with similar characteristics, such as active immune cells or static immune cells in a specific state, thereby providing more refined data classification for the training of the state prediction model; based on the cell subtypes obtained by clustering analysis, an immune cell sample data set is generated, and on this basis, the visual morphological features are further converted into The method converts the multidimensional feature data into high-dimensional and structured embedding vectors through feature mapping methods (such as deep learning embedding layer or principal component analysis). These embedding vectors not only contain the appearance characteristics of the cells, but also capture the intrinsic correlation between different features through embedding, so as to provide a unified data expression for model training. The generated morphological feature embedding vector samples are used to perform modal fusion training on the large model for predicting the state of immune cells, so that the dynamically changing feature embedding vectors and the morphological feature embedding vectors can be jointly input into the model, and the complementarity of the two modal data can be used to improve the prediction accuracy. During the training process, the model learns the deep correlation between dynamic features and visual morphological features by optimizing the loss function (such as cross entropy loss), and outputs the final immune cell state prediction results by combining the two modal data. After modal fusion training, the large model for predicting the state of immune cells is output.

[0046] Furthermore, the immune cell samples are stained and marked to obtain activity detection data based on the staining and marking; the immune cell state prediction model is supervised and trained using the activity detection data as label data for model training, and an immune cell state prediction model with a prediction accuracy reaching a preset threshold is output.

[0047] Preferably, the extracted immune cell samples are stained and labeled to reflect their biological state or functional activity by staining the immune cells. For example, flow cytometry (FACS) or immunofluorescence staining techniques can be used to label the active molecules of the cells, such as the enzyme activity, metabolic level or expression of surface markers in the cells using specific active dyes or antibodies. Activity detection data is obtained through staining and labeling experiments. The activity detection data includes the category information of immune cells in different states after activity staining. As label data for model training, the label data can be concretized as a numerical expression of the activity level (such as high activity is 1, low activity is 0.5, and inactivity is 0), or classification data (such as high activity, low activity, inactivity, etc.), which correspond to the dynamic change feature embedding vector and visual morphology feature embedding vector of immune cells, forming a complete training data set. The training data set is used to conduct supervised training on the immune cell state prediction model. The core of supervised training is to minimize the error between the model prediction output and the actual label data by adjusting the model parameters. The cross entropy loss function is often used as the optimization target. The model embeds the vector according to the dynamic change characteristics and visual morphological characteristics of the input, and gradually learns the mapping relationship between these features and the active state label. During the training process, the weight parameters and bias items of the model are updated through the back propagation algorithm to improve the prediction ability of the model. The training effect of the model is evaluated by setting a preset prediction accuracy threshold. When the prediction accuracy of the model on the validation set reaches or exceeds the set threshold, the final immune cell state prediction model is output.

[0048] In summary, the embodiments of the present application have at least the following technical effects:

[0049] First, a motion image sequence of immune cells is collected using a microscopic imaging device. Next, edge detection is performed on the motion image sequence, and the contour pixel data labels of the immune cells are marked. Further, the trajectory of the immune cells is tracked in a time series according to the contour pixel data labels, and the motion trajectory pixel data is output. Then, the motion trajectory pixel data is input into the dynamic change rate recognition model, and the first-order change rate, the second-order change rate and the mean of the change rate are obtained according to the dynamic change rate recognition model. The first-order change rate, the second-order change rate and the mean of the change rate are used to generate a dynamic change feature embedding vector. Finally, the dynamic change feature embedding vector is input into the immune cell state prediction model for state prediction, and the activity state prediction index is obtained. The immune cells are marked for activity according to the activity state prediction index. The technical problems of inaccurate analysis and incomplete prediction of the activity state of immune cells in the prior art are solved, and the technical effect of improving the accuracy and comprehensiveness of the prediction of the activity state of immune cells is achieved.

[0050] Embodiment 2, based on the same inventive concept as the method for analyzing the state of immune cells in the above embodiment, Figure 2 As shown, the present application provides a system for analyzing the state of immune cells, wherein the system comprises:

[0051] Image acquisition module 11: uses a microscopic imaging device to collect a motion image sequence of immune cells; edge detection module 12: performs edge detection on the motion image sequence and marks the contour pixel data labels of the immune cells; trajectory tracking module 13: tracks the trajectory of the immune cells in a time series according to the contour pixel data labels, and outputs motion trajectory pixel data; recognition module 14: inputs the motion trajectory pixel data into a dynamic change rate recognition model, obtains the first-order change rate, the second-order change rate and the mean change rate according to the dynamic change rate recognition model, and generates a dynamic change feature embedding vector from the first-order change rate, the second-order change rate and the mean change rate; state prediction module 15: inputs the dynamic change feature embedding vector into the immune cell state prediction large model for state prediction, obtains an activity state prediction index, and performs activity marking on the immune cells according to the activity state prediction index.

[0052] Furthermore, the edge detection module 12 is used to perform the following method:

[0053] The motion image sequence is preprocessed to obtain a preprocessed motion image sequence; the gradient of the preprocessed motion image sequence is calculated by a Sobel operator to output a gradient amplitude; a double edge threshold is set, and the gradient amplitude is assigned according to the double edge threshold to obtain an edge image of the immune cell; and according to the edge image of the immune cell, the contour pixel data of the immune cell is marked.

[0054] Furthermore, the trajectory tracking module 13 is used to execute the following method:

[0055] Establish an immune cell association relationship between each frame image in the motion image sequence; locate the continuous contour pixel data matching the contour pixel data label in each frame image according to the association relationship; connect the continuous contour pixel data according to the time series of the motion image sequence, and output motion trajectory pixel data.

[0056] Furthermore, the identification module 14 is used to execute the following method:

[0057] The motion trajectory pixel data is input into a dynamic change rate recognition model, and the first-order change rate, the second-order change rate and the mean change rate are obtained according to the dynamic change rate recognition model; the motion trajectory pixel data is subjected to pixel coordinate change analysis according to the dynamic change rate recognition model, and the pixel displacement change speed and the pixel displacement change acceleration under each sequence, as well as the mean pixel displacement change speed and the mean pixel displacement change acceleration under the motion image sequence are calculated; the pixel displacement change speed under each sequence is output as a first-order change rate; the pixel displacement change acceleration under each sequence is output as a second-order change rate; the mean pixel displacement change speed and the mean pixel displacement change acceleration under the motion image sequence are output as the mean change rate.

[0058] Furthermore, the state prediction module 15 is used to perform the following method:

[0059] The expression of the immune cell status prediction model is as follows:

[0060] Among them, y i is the cell state prediction result after analysis based on the dynamic change feature embedding vector. Softmax is used to map the input dynamic change feature embedding vector to a probability distribution function for classification. j is the weight of each feature in the dynamically changing feature embedding vector, is the vector value of the jth dynamically changing feature in the normalized input data, b j is the bias term to be corrected, j is the index of the feature, and n is the number of features in the dynamically changing feature embedding vector.

[0061] Furthermore, the state prediction module 15 is used to perform the following method:

[0062] Perform visual image recognition on the motion image sequence to obtain visual morphological features of the immune cells, wherein the visual morphological features include cell area, cell shape and cell texture; perform cluster analysis based on the visual morphological features to determine cell subtypes; obtain immune cell samples based on the cell subtypes and obtain morphological feature embedding vector samples; perform modal fusion training on the immune cell state prediction large model based on the morphological feature embedding vector samples, and output the immune cell state prediction large model.

[0063] Furthermore, the state prediction module 15 is used to perform the following method:

[0064] The immune cell sample is stained and marked to obtain activity detection data based on the staining mark; the immune cell state prediction model is supervised and trained using the activity detection data as label data for model training, and an immune cell state prediction model with a prediction accuracy reaching a preset threshold is output.

[0065] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0066] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0067] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A method for analyzing the state of immune cells, characterized in that: The method comprises: Use microscopic imaging equipment to collect moving image sequences of immune cells; Performing edge detection on the motion image sequence and marking contour pixel data labels of the immune cells; Tracking the trajectory of the immune cell in a time series according to the contour pixel data label, and outputting motion trajectory pixel data; Inputting the motion trajectory pixel data into a dynamic change rate recognition model, obtaining a first-order change rate, a second-order change rate and a change rate mean according to the dynamic change rate recognition model, and generating a dynamic change feature embedding vector by using the first-order change rate, the second-order change rate and the change rate mean; The dynamic change feature embedding vector is input into the immune cell state prediction large model to perform state prediction, obtain an activity state prediction index, and perform activity marking on the immune cell according to the activity state prediction index; Training a large model for predicting immune cell status, the method also includes: Performing visual image recognition on the motion image sequence to obtain visual morphological features of the immune cells, wherein the visual morphological features include cell area, cell shape, and cell texture; Performing cluster analysis based on the visual morphological features to determine cell subtypes; Obtain an immune cell sample according to the cell subtype, and obtain a morphological feature embedding vector sample; Performing modal fusion training on the immune cell state prediction large model according to the morphological feature embedding vector samples, and outputting the immune cell state prediction large model; Staining and marking the immune cell sample to obtain activity detection data based on the staining and marking; The immune cell state prediction large model is supervised and trained using the activity detection data as label data for model training, and an immune cell state prediction large model whose prediction accuracy reaches a preset threshold is output.

2. The method for analyzing the state of immune cells according to claim 1, characterized in that: Performing edge detection on the motion image sequence and marking contour pixel data labels of the immune cells, the method comprising: Preprocessing the motion image sequence to obtain a preprocessed motion image sequence; The gradient of the preprocessed motion image sequence is calculated by Sobel operator, and the gradient amplitude is output; Setting a double edge threshold, assigning a value to the gradient amplitude according to the double edge threshold, and obtaining an edge image of the immune cell; According to the edge image of the immune cell, the outline pixel data of the immune cell is marked.

3. The method for analyzing the state of immune cells according to claim 1, characterized in that: Tracking the immune cell's trajectory in a time series according to the contour pixel data label and outputting the motion trajectory pixel data, the method comprising: Establishing an immune cell association relationship between each frame of the motion image sequence; According to the association relationship, locate the continuous contour pixel data matching the contour pixel data label in each frame of the image; The continuous contour pixel data are connected according to the time series of the motion image sequence, and motion trajectory pixel data are output.

4. The method for analyzing the state of immune cells according to claim 1, characterized in that: Inputting the motion trajectory pixel data into a dynamic change rate recognition model, and obtaining a first-order change rate, a second-order change rate, and a change rate mean according to the dynamic change rate recognition model; Performing pixel coordinate change analysis on the motion trajectory pixel data according to the dynamic change rate recognition model, calculating the pixel displacement change speed and pixel displacement change acceleration in each sequence, and the average value of the pixel displacement change speed and the average value of the pixel displacement change acceleration in the motion image sequence; Outputting the pixel displacement change speed in each sequence as a first-order change rate; Outputting the pixel displacement change acceleration in each sequence as a second-order change rate; The average value of the pixel displacement change speed and the average value of the pixel displacement change acceleration in the motion image sequence are output as the change rate average value.

5. The method for analyzing the state of immune cells according to claim 1, characterized in that: The dynamic change feature embedding vector is input into the immune cell state prediction model for state prediction. The expression of the immune cell state prediction model is as follows: ； Among them, y i is the cell state prediction result after analysis based on the dynamic change feature embedding vector. Softmax is used to map the input dynamic change feature embedding vector to a probability distribution function for classification. j is the weight of each feature in the dynamically changing feature embedding vector, is the vector value of the jth dynamically changing feature in the normalized input data, b j is the bias term to be corrected, j is the index of the feature, and n is the number of features in the dynamically changing feature embedding vector.

6. A system for analyzing the state of immune cells, characterized in that: A method for analyzing the state of an immune cell according to any one of claims 1 to 5, wherein the system comprises: Image acquisition module: uses microscopic imaging equipment to collect motion image sequences of immune cells; Edge detection module: performs edge detection on the motion image sequence and marks contour pixel data labels of the immune cells; Trajectory tracking module: tracking the trajectory of the immune cell in time series according to the contour pixel data label, and outputting motion trajectory pixel data; Recognition module: input the motion trajectory pixel data into a dynamic change rate recognition model, obtain the first-order change rate, the second-order change rate and the change rate mean according to the dynamic change rate recognition model, and generate a dynamic change feature embedding vector by taking the first-order change rate, the second-order change rate and the change rate mean; State prediction module: embedding the dynamic change feature into a vector and inputting it into a large immune cell state prediction model to perform state prediction, obtain an activity state prediction index, and perform activity marking on the immune cells according to the activity state prediction index.

Images