Antibody titer prediction method and system based on titer prediction model

Abstract

The invention discloses an antibody titer prediction method and system based on a titer prediction model, and the method comprises the steps: carrying out the image collection of an obtained antibody titer sample, and generating an image sample set; performing instance segmentation on each cell in the image sample set according to a cellpose algorithm to obtain a cell set image; inputting the cell set image into a preset titer prediction model to obtain a corresponding feature map, extracting regional features corresponding to the feature map, respectively extracting corresponding statistical features from a positive cell feature map and a negative cell feature map, and predicting a titer value of an antibody based on the regional features and the statistical features, the feature maps comprise a positive cell feature map, a negative cell feature map and a background feature map. Through automatic image processing and titer prediction model prediction, rapid, accurate and objective evaluation of the antibody titer is realized, and the sample preparation cost is reduced.

Description

Technical Field

[0001] This application relates to the field of bioinformatics processing, and in particular, to an antibody titer prediction method and system based on a titer prediction model. Background Art

[0002] Common methods for detecting antibodies include indirect immunofluorescence assay, ELISA, radioimmunoassay, etc. Among them, indirect immunofluorescence assay (IFA) is the most commonly used method for current detection. This method mainly determines positive or negative by manually observing fluorescence intensity under a microscope. However, on the one hand, the IFA technology highly depends on the experience and skills of professionals, and the interpretation process is highly subjective. On the other hand, the preparation and interpretation processes of IFA are cumbersome and time-consuming, and it is not suitable for quickly processing a large number of samples.

[0003] With the development of computer technology, many automatic antibody titer estimation algorithms have emerged. After preparing a series of antibody titer samples with different dilutions, the automatic estimation algorithm is used to judge negative / positive for each dilution sample, and the judgment results of samples with different dilutions are statistically analyzed to obtain the estimated value of the antibody titer. Although this algorithm solves the subjectivity of negative / positive discrimination, it also requires preparing multiple dilution samples of antibody titers and cannot reduce the cost of sample preparation.

[0004] Application content

[0005] This application provides an antibody titer prediction method and system based on a titer prediction model. Through automated image processing and titer prediction model prediction, it not only realizes rapid, accurate, and objective evaluation of antibody titers, but also reduces the cost of sample preparation.

[0006] In a first aspect, this application provides an antibody titer prediction method based on a titer prediction model, including:

[0007] Performing image acquisition on the obtained antibody titer samples to generate an image sample set;

[0008] Performing instance segmentation on each cell in the image sample set according to the cellpose algorithm to obtain a cell set image;

[0009] Inputting the cell set image into a preset titer prediction model to obtain a corresponding feature map, extracting the corresponding regional features of the feature map, and respectively extracting the corresponding statistical features from the positive cell feature map and the negative cell feature map. Based on the regional features and the statistical features, predicting the titer value of the antibody. Among them, the feature map includes a positive cell feature map, a negative cell feature map, and a background feature map, and the titer prediction model includes a UNet positive cell segmenter, a deep learning feature extractor, a statistical feature extractor, and an omics titer regression predictor.

[0010] In the embodiments of the present application, by collecting images of the obtained antibody titer samples, it is convenient to perform instance segmentation on the collected images subsequently to determine the cell boundaries in the image sample set, and further convenient to quickly, accurately and objectively predict the antibody titer value subsequently; by using the cellpose algorithm to perform instance segmentation on each cell in the image sample set, the cell boundaries of each cell in the image sample set can be determined, which is convenient to determine whether the cell is a positive cell or a negative cell subsequently, and further convenient to quickly, accurately and objectively predict the antibody titer value subsequently; by using the titer prediction model, the positive cell feature map and negative cell feature map in the image sample set can be quickly segmented, and the titer is predicted by combining regional features and statistical features, which not only realizes the rapid, accurate and objective evaluation of the antibody titer, but also reduces the cost of sample preparation.

[0011] Further, the collecting images of the obtained antibody titer samples to generate an image sample set specifically includes:

[0012] Pre-set the exposure parameters of the image acquisition device, where the exposure parameters include the exposure value and the exposure time;

[0013] Based on the exposure parameters, collect multiple images of the antibody titer sample, and stack the multiple images with different exposures to obtain an image sample set.

[0014] In this way, by collecting images of the obtained antibody titer samples, it is convenient to perform instance segmentation on the collected images subsequently to determine the cell boundaries in the image sample set, and further convenient to quickly, accurately and objectively predict the antibody titer value subsequently.

[0015] Further, the performing instance segmentation on each cell in the image sample set according to the cellpose algorithm to obtain a cell set image specifically includes:

[0016] Perform instance segmentation on each cell in the image sample set according to the cellpose algorithm to obtain an instance segmentation result;

[0017] Eliminate the cells in the instance segmentation result that do not meet the preset cell area threshold to obtain a first processing result;

[0018] Count the detection times of each cell in the first processing result, and retain the cells that meet the preset detection times to obtain a second processing result;

[0019] Eliminate the overlapping cells in the second processing result to obtain a cell set image.

[0020] In this way, by using the cellpose algorithm to perform instance segmentation on each cell in the image sample set, the cell boundaries of each cell in the image sample set can be determined, which is convenient for subsequent determination of positive cells or negative cells, and further convenient for subsequent rapid, accurate and objective prediction of the antibody titer value.

[0021] Furthermore, the statistical features include the average fluorescence intensity and the standard deviation of fluorescence intensity corresponding to the positive cell feature map, the average fluorescence intensity and the standard deviation of fluorescence intensity corresponding to the negative cell feature map, the overall average fluorescence intensity and the standard deviation of fluorescence intensity.

[0022] Furthermore, the calculation formula of the statistical features is specifically as follows:

[0023]

[0024] In the formula, mean p is the average fluorescence intensity of the positive cell feature map, V is the average brightness image, S p is the distribution probability of positive cells, std p is the standard deviation of fluorescence intensity of the positive cell feature map, mean n is the average fluorescence intensity of the negative cell feature map, S n is the distribution probability of negative cells, std n is the standard deviation of fluorescence intensity of the negative cell feature map, mean p+n is the overall average fluorescence intensity, std p+n is the overall standard deviation of fluorescence intensity.

[0025] In this way, the positive cell feature map and the negative cell feature map in the image sample set can be quickly segmented by the UNet positive cell segmenter, and the corresponding statistical features in the positive cell feature map and the negative cell feature map are calculated respectively, which is convenient for subsequent rapid, accurate and objective prediction of the antibody titer value based on the omics titer prediction model.

[0026] In a second aspect, the present application provides an antibody titer prediction system based on a titer prediction model, including: an acquisition module, a segmentation module and a prediction module;

[0027] The acquisition module is used to perform image acquisition on the acquired antibody titer samples to generate an image sample set;

[0028] The segmentation module is used to perform instance segmentation on each cell in the image sample set according to the cellpose algorithm to obtain a cell set image;

[0029] The prediction module is used to input the cell collection image into a preset titer prediction model to obtain a corresponding feature map, extract the regional features corresponding to the feature map, and respectively extract corresponding statistical features from the positive cell feature map and the negative cell feature map, and predict the titer value of the antibody based on the regional features and the statistical features, wherein the feature map includes a positive cell feature map, a negative cell feature map and a background feature map, and the titer prediction model includes a UNet positive cell segmentor, a deep learning feature extractor, a statistical feature extractor and an omics titer regression predictor.

[0030] The embodiment of the present application performs image acquisition on the acquired antibody titer samples, which can facilitate subsequent instance segmentation of the acquired images to determine the cell boundaries in the image sample set, thereby facilitating subsequent rapid, accurate and objective prediction of the antibody titer value; by performing instance segmentation on each cell in the image sample set through the cellpose algorithm, the cell boundary of each cell in the image sample set can be determined, which facilitates subsequent determination of whether it is a positive cell or a negative cell, thereby facilitating subsequent rapid, accurate and objective prediction of the antibody titer value; by using the titer prediction model, the positive cell feature map and the negative cell feature map in the image sample set can be quickly segmented, and the titer prediction is performed in combination with the regional features and the statistical features, which not only realizes the rapid, accurate and objective evaluation of the antibody titer, but also reduces the cost of sample preparation.

[0031] Furthermore, the acquisition module includes: a preset unit and a collection unit;

[0032] The preset unit is used to preset the exposure parameters of the image acquisition device, wherein the exposure parameters include the exposure value and the exposure time;

[0033] The acquisition unit is used to acquire multiple images of the antibody titer sample based on the exposure parameters, and stack multiple images with different exposures to obtain an image sample set.

[0034] Further, the segmentation module includes: a segmentation unit, a first processing unit, a second processing unit and a third processing unit;

[0035] The segmentation unit is used to perform instance segmentation on each cell in the image sample set according to the cellpose algorithm to obtain an instance segmentation result;

[0036] The first processing unit is used to remove cells that do not meet a preset cell area threshold in the instance segmentation result to obtain a first processing result;

[0037] The second processing unit is used to count the number of times each cell is detected in the first processing result, and retain cells that meet a preset number of detections to obtain a second processing result;

[0038] The third processing unit is configured to eliminate overlapping cells in the second processing result to obtain a cell set image.

[0039] Further, the statistical features include the average fluorescence intensity and the standard deviation of the fluorescence intensity corresponding to the positive cell feature map, the average fluorescence intensity and the standard deviation of the fluorescence intensity corresponding to the negative cell feature map, the overall average fluorescence intensity, and the standard deviation of the fluorescence intensity.

[0040] Further, the calculation formula of the statistical features is specifically as follows:

[0041]

[0042]

[0043] In the formula, mean p is the average fluorescence intensity of the positive cell feature map, V is the average brightness image, S p is the distribution probability of positive cells, std p is the standard deviation of the fluorescence intensity of the positive cell feature map, mean n is the average fluorescence intensity of the negative cell feature map, S n is the distribution probability of negative cells, std n is the standard deviation of the fluorescence intensity of the negative cell feature map, mean p+n is the overall average fluorescence intensity, std p+n is the standard deviation of the overall fluorescence intensity. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is a schematic flowchart of an embodiment of an antibody titer prediction method based on a titer prediction model provided by the present application;

[0045] Figure 2 is a schematic diagram of the prediction process of the titer prediction model provided by the present application;

[0046] Figure 3 is a schematic structural diagram of an embodiment of an antibody titer prediction system based on a titer prediction model provided by the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0048] It should be understood that the step numbers used in the text are only for convenience of description and do not limit the order of execution of the steps.

[0049] It should be understood that the terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include the plural forms.

[0050] The terms "comprising" and "including" indicate the presence of the described features, wholes, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or their combinations.

[0051] The term "and / or" refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0052] Common methods for detecting antibodies include indirect immunofluorescence assay, ELISA, radioimmunoassay, etc. Among them, IFA is the most commonly used. This method mainly determines the negativity or positivity by manually observing the fluorescence intensity under a microscope. However, IFA highly depends on the experience and skills of professionals, and the interpretation process is subjective, cumbersome and time-consuming. With the development of computer technology, an automatic antibody titer estimation algorithm has emerged. By using the algorithm to judge the negativity or positivity of samples at different dilutions and statistically analyzing the judgment results of samples at different dilutions, the estimated value of the antibody titer can be obtained, which solves the subjectivity problem. However, it still requires the preparation of antibody titer samples at multiple dilutions and cannot reduce the cost of sample preparation.

[0053] Next, the nouns involved in this application are analyzed:

[0054] Cellpose is a deep learning algorithm for cell image segmentation. It can accurately extract cells from complex images automatically. This algorithm is based on the architecture of convolutional neural network (CNN) and learns how to distinguish cells from the background through training. Its principle and implementation method make it have important application value in the fields of cell biology research and medical diagnosis, etc.

[0055] The inflection point detection algorithm is a class of techniques for identifying mutation points in time series or data, and these mutation points are usually called changepoints or breakpoints.

[0056] Ordinal Regression, also known as ordinal classification, is a statistical method, specifically a regression method for data with an ordinal scale dependent variable. It is applicable to scenarios where the dependent variable is data of an ordinal categorical scale type. This regression method can use predictor variables (which can be categorical variables and numerical variables) to model the ordinal categorical scale response variable.

[0057] Based on this, the embodiments of the present application provide an antibody titer prediction method and system based on a titer prediction model. Through automated image processing and titer prediction model prediction, it not only realizes the rapid, accurate and objective evaluation of antibody titers, but also reduces the cost of sample preparation.

[0058] The antibody titer prediction method and system based on a titer prediction model provided by the embodiments of the present application are specifically described through the following embodiments. First, the antibody titer prediction method based on a titer prediction model in the embodiments of the present application is described.

[0059] The antibody titer prediction method based on a titer prediction model provided by the embodiments of the present application relates to the field of bioinformatics processing. The antibody titer prediction method based on a titer prediction model provided by the embodiments of the present application can be applied to a terminal, or to a server side, or can also be software running on a terminal or a server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a notebook computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or can be configured as a server cluster or distributed system composed of multiple physical servers, or can also be configured as a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can implement an application of an antibody titer prediction method based on a titer prediction model, etc., but is not limited to the above forms.

[0060] This application can also be used in numerous general-purpose or special-purpose computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and so on. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This application can also be practiced in a distributed computing environment where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0061] Embodiment 1

[0062] Please refer to Figure 1 , Figure 1 which is a schematic flowchart of an embodiment of an antibody titer prediction method based on a titer prediction model provided by this application, including: steps S101 to S103;

[0063] Step S101: Perform image acquisition on the obtained antibody titer samples to generate an image sample set;

[0064] It is understandable that the antibody titer sample can be directly obtained or a sample prepared through experiments. That is, this application is not limited to the experimental stage. The preparation method of the antibody titer sample can refer to the IFA detection process. Induce and express the corresponding cells into the corresponding antigens, then fix the cells on a glass slide, and incubate the serum sample to be tested with the cells to bind the antibodies therein to the EBV antigen, obtaining the antibody titer sample. Among them, the antigen can be but is not limited to the EA antigen (EBV early antigen) and the VCA antigen (EBV viral capsid antigen). Exemplarily rather than restrictively, when preparing the antibody titer sample corresponding to the EA antigen, the specific operation method is as follows: (1) Cell acquisition and culture: Use the Raji cell line (CCL-86) purchased from the American Type Culture Collection (ATCC), culture the Raji cells (CCL-86) in RPMI-1640 medium containing 10% fetal bovine serum (FBS), and place the cells under the conditions of 37°C and 5% CO2 for suspension culture; (2) Cell induction: Under the conditions of 37°C and 5% CO2, use 20 ng / ml of Phorbol 12-myristate 13-acetate (TPA) and 3.3 mg / ml of sodium butyrate (SB) to induce the cells, where the induction time is 2 days; (3) Determination of cell viability and quantity: Use the trypan blue exclusion method to evaluate the cell viability, and use an automatic cell counter (Biorad TC 10 TM ) to determine the quantity of cells; (4) Cell treatment and fixation: After washing the cells in 1× phosphate buffer solution (PBS), drop them onto a teflon-coated glass slide at a density of approximately 4000 cells per well, and use acetone to fix the cells on the glass slide. The fixation time is 10 minutes. Finally, dry the fixed glass slide and store it at -20°C for subsequent use. (5) Binding of antibody and antigen: Take out the fixed and stored glass slide, incubate the serum sample to be tested with the cells on the glass slide to bind the antibodies in the serum to the EBV antigen (present on the Raji cells), obtaining the antibody titer sample. Exemplarily rather than restrictively, when preparing the antibody titer sample corresponding to the VCA antigen, the specific operation method is as follows: (1) Cell acquisition and culture: Purchase the Raji cell line (CCL-86) from the American Type Culture Collection (ATCC), culture the Raji cells in RPMI-1640 medium containing 10% fetal bovine serum (FBS), and place the cells under the conditions of 37°C and 5% CO2 for suspension culture. (2) Evaluation of cell viability and quantity: Use the trypan blue exclusion method to evaluate the cell viability, and use an automatic cell counter (Biorad TC 10 TM) Determine the number of cells. (3) Cell processing and slide preparation: Wash the cells in 1× phosphate buffered saline (PBS) to remove the culture medium and other impurities, and then drop the washed cells onto a teflon-coated slide at a density of approximately 4000 cells per well for subsequent experimental operations; (4) Cell fixation and storage: Fix the cells on the slide with acetone for 10 minutes to maintain the cell morphology and structure for subsequent staining and observation. Then dry the fixed slide and store it at -20°C to preserve the cell sample for a long time and ensure the stability and reliability of subsequent experimental results. (5) Antibody-antigen binding: Take out the fixed and stored slide, incubate the serum sample to be tested with the cells on the slide to allow the antibodies in the serum to bind to the VCA antigen (present on B958 cells), obtaining an antibody titer sample.

[0065] It can be understood that after obtaining the antibody titer sample, it is necessary to dilute the antibody titer sample and perform image acquisition on the diluted sample. Among them, the specific method of image acquisition is: preset the exposure parameters of the image acquisition device, where the exposure parameters include the exposure value and the exposure time; perform multiple image acquisitions on the antibody titer sample based on the exposure parameters, and stack multiple images with different exposures to obtain an image sample set.

[0066] It should be noted that the image acquisition device can be, but is not limited to, an EUROstar IIIPlus microscope and a Basler's Ace sensor camera, etc.

[0067] It should be noted that in single-image recognition, problems such as unstable cell boundaries may occur due to poor image quality or complex cell morphology. At the same time, problems such as false positives and false negatives of cells may occur due to noise in the image or cell overlap. Therefore, for the images of multiple antibody titer samples obtained, images of the titer samples can be obtained by using different exposure parameters, and multiple images with different exposures are stacked to obtain an image sample set.

[0068] In this way, by performing image acquisition on the obtained antibody titer sample, it is convenient to perform instance segmentation on the acquired image to determine the cell boundaries in the image sample set, and then it is convenient to quickly, accurately and objectively predict the antibody titer value subsequently.

[0069] Step S102: Perform instance segmentation on each cell in the image sample set according to the cellpose algorithm to obtain a cell set image;

[0070] It can be understood that after obtaining the image sample set, it is necessary to predict the boundaries of cells, which facilitates the subsequent determination of positive or negative cells. Specifically: First, perform instance segmentation on each cell in the image sample set according to the cellpose algorithm to obtain the instance segmentation result. Among them, the cellpose algorithm can, but is not limited to, select the cyto3 pre-trained model to perform cell instance segmentation on each image in the image sample set one by one. At the same time, when performing cell segmentation, parameters such as cell diameter and watershed value can be continuously adjusted to obtain a better instance segmentation result. Second, eliminate the cells in the instance segmentation result that do not meet the preset cell area threshold to obtain the first processing result. Among them, the cell area threshold can be preset, including the minimum cell area threshold and the maximum cell area threshold. By comparing the area of each cell in the instance segmentation result with the preset cell area threshold, cells outside the reasonable cell size range can be excluded. Third, count the number of detections of each cell in the first processing result and retain the cells that meet the preset detection times to obtain the second processing result. Among them, since ideally, the positions and regions of cells should be exactly the same, the number of detections of each cell in the first processing result is counted. When the number of detections is greater than the preset detection times, it can be considered that the detected cells in this region are valid, and thus the second processing result is obtained. Finally, eliminate the overlapping cells in the second processing result to obtain the cell set image. Among them, since there will be multiple detection results at the same position in the second processing result, it is necessary to exclude overlapping cells so that there is only one detected cell at each position, and finally obtain the cell set image.

[0071] It should be noted that the non-maximum suppression (NMS) method can be used, but is not limited to, to exclude overlapping cells. Specifically, the non-maximum suppression method sorts the cells from large to small, preferentially selects the cells with larger areas, and traverses the remaining cell set to calculate the intersection over union (IoU) with the currently selected cell. When the IoU is greater than the threshold, the cell is excluded from the remaining cell set, thus ensuring that there is only one detected cell at each position.

[0072] In this way, by performing instance segmentation on each cell in the image sample set through the cellpose algorithm, the cell boundaries of each cell in the image sample set can be determined, which facilitates the subsequent determination of whether the cell is a positive cell or a negative cell, and further facilitates the subsequent rapid, accurate and objective prediction of the antibody titer value.

[0073] Step S103: Input the cell set image into a preset titer prediction model to obtain a corresponding feature map, extract the regional features corresponding to the feature map, and respectively extract the corresponding statistical features from the positive cell feature map and the negative cell feature map. Predict the titer value of the antibody based on the regional features and the statistical features. Among them, the feature map includes a positive cell feature map, a negative cell feature map, and a background feature map, and the titer prediction model includes a UNet positive cell segmenter, a deep learning feature extractor, a statistical feature extractor, and an omics titer regression predictor.

[0074] It can be understood that after obtaining the cell set image, the positive cells and negative cells can be determined. Specifically, the cell set image can be input into a preset titer prediction model. Among them, the schematic diagram of the prediction process of the titer prediction model is as Figure 2 shown. Through the UNet positive cell segmenter in the titer prediction model, a positive cell feature map, a negative cell feature map, and a background feature map can be obtained. Among them, the feature map is also the semantic segmentation result.

[0075] It should be noted that the UNet positive cell segmenter can be trained but is not limited to using the UNet architecture, wherein the UNet network structure uses ResNet18 as an encoder. By way of example but not limitation, the training process of the UNet positive cell segmenter is as follows: first, a training image sample set is prepared in advance; second, since there is an obvious difference in brightness values ​​between the background image, negative cell image and positive cell image in the training image sample set, that is, the positive cell image brightness value > negative cell image brightness value > background area brightness value, based on this feature, the average brightness value in each cell area in the training image sample set is extracted as the brightness representative value of the cell, and the brightness representative values ​​of all cells are sorted to obtain a brightness representative value sorting curve with the horizontal axis being the sorted cell number and the vertical axis being the brightness value; third, the brightness sorting curve is analyzed by an inflection point detection algorithm, and then the curve is extracted. The inflection point represents the potential dividing point between positive cells and negative cells, wherein the inflection point detection algorithm may be, but is not limited to, KneeLocator in the knee library of python. When there is an obvious inflection point in the curve, the brightness value corresponding to the inflection point can be used as the judgment threshold of the positive cell, and according to the determined judgment threshold, the cell instance with a brightness value higher than the threshold is marked as a positive cell, otherwise it is a negative cell, so that the positive cell feature map and the negative cell feature map can be preliminarily determined; the preliminarily determined positive cell feature map and negative cell feature map are used as the target of initialization training of the UNet positive cell segmentor, and the training image sample set is input into the UNet positive cell segmentor, and the UNet positive cell segmentor uses the softmax function to correspond to the probability distribution S of the three categories of the background image, the negative cell feature map and the positive cell feature map for each cell in the training image sample set. b , S n and S p , and then based on the probability distribution S b , S n and S p Determine the corresponding feature map.

[0076] It should be noted that in order to keep the semantic segmentation result of the UNet positive cell segmenter differentiable, we do not use the argmax function to obtain the segmentation result of each pixel, but use softmax to obtain the classification probability map S of the three regions of background, negative cells and positive cells. b , S n and S p , and there is S b +S n +S p =1, the corresponding feature map can be determined based on the classification probability map.

[0077] It should be noted that UNet can be replaced by other semantic segmentation models.

[0078] It should be noted that the positive cell feature map and negative cell feature map obtained through inflection point detection are only roughly estimated images. Therefore, in order to ensure segmentation accuracy, it is necessary to further train the UNet positive cell segmenter.

[0079] It can be understood that after obtaining the feature maps, weighted average features will be extracted from each feature map (positive cell feature map, negative cell feature map, and background feature map) through the deep learning feature extractor in the titer prediction model. At the same time, global average pooling features will be extracted, and the weighted average features and global average pooling features will be concatenated to obtain regional features, and the regional features will be reduced to a fixed dimension (32 dimensions) through a multi-layer perceptron.

[0080] It can be understood that after obtaining the feature maps, corresponding statistical features (7 dimensions) also need to be extracted from the positive cell feature map and negative cell feature map through the statistical feature extractor in the titer prediction model. Among them, the statistical features include the average fluorescence intensity and standard deviation of fluorescence intensity corresponding to the positive cell feature map, the average fluorescence intensity and standard deviation of fluorescence intensity corresponding to the negative cell feature map, the overall average fluorescence intensity and standard deviation of fluorescence intensity; the statistical features are input into the omics titer prediction model to obtain the titer value of the antibody. Among them, the calculation formula of the statistical features is specifically:

[0081]

[0082]

[0083] In the formula, mean p is the average fluorescence intensity of the positive cell feature map, V is the average brightness image, S p is the distribution probability of positive cells, std p is the standard deviation of fluorescence intensity of the positive cell feature map, mean n is the average fluorescence intensity of the negative cell feature map, S n is the distribution probability of negative cells, std n is the standard deviation of fluorescence intensity of the negative cell feature map, mean p+n is the overall average fluorescence intensity, std p+n is the overall standard deviation of fluorescence intensity.

[0084] In this way, the positive cell feature map and the negative cell feature map in the image sample set can be quickly segmented by the UNet positive cell segmenter, and the corresponding statistical features in the positive cell feature map and the negative cell feature map are calculated respectively, which facilitates the subsequent rapid, accurate and objective prediction of the antibody titer value based on the omics titer prediction model.

[0085] It can be understood that after obtaining the regional features and statistical features, it is necessary to splice the regional features and statistical features and use the titer regression predictor in the titer prediction model for regression prediction to predict the antibody titer value. Among them, the titer regression predictor is a multi-layer perceptron with dropout.

[0086] It should be noted that the titer prediction model is weakly supervised trained through a multi-task learning framework. By simultaneously optimizing the two tasks of cell region segmentation and ordinal regression, the performance of the model is improved. During the training process, a variety of data augmentation methods are used to increase the diversity of data, the Adam optimizer and the cosine annealing learning rate scheduling strategy are used to optimize the model parameters, weighted random sampling is used to handle the data imbalance problem, and a memory bank is used to store features and prediction results to achieve the effect of global optimization.

[0087] It should be noted that the titer prediction model includes two main loss functions, namely the mask segmentation loss and the ordinal regression loss. Among them, the mask segmentation loss uses the cross-entropy loss function to optimize cell region segmentation, and the ordinal regression loss uses the global ordinal loss to ensure that the prediction results satisfy the ordinal relationship of the titer. The calculation formula of the loss function is: Total_Loss = w1 * Mask_Loss + w2 * Ordinal_Loss, where Total_Loss is the total loss value, Mask_Loss and Ordinal_Loss are the cross-entropy loss value and the ordinal regression loss value respectively, and w1 and w2 are weight coefficients used to balance the importance of the two tasks.

[0088] It should be noted that when training the titer prediction model, it is necessary to perform data augmentation on the training data set. The means of data augmentation include but are not limited to: random rotation (±30 degrees), random cropping, and scaling while maintaining the aspect ratio, etc.

[0089] It should be noted that the training process is as follows: First, in terms of the optimizer, the Adam optimizer with excellent performance was selected, and its initial learning rate was set to 0.0005 to ensure that the model can converge smoothly and quickly in the initial stage of training. To further improve the training effect, a cosine annealing learning rate scheduling strategy was introduced, which can dynamically adjust the learning rate according to the training progress, so as to give the model the most appropriate update step size at different stages of training. In addition, special attention was paid to the selection of the batch size during training, and the batch size was set to 16. This setting not only ensures the training efficiency but also ensures that the model can fully learn the internal characteristics of the data. At the same time, 100 training epochs were set to ensure that the titer prediction model has enough time to fully fit the data and achieve stable performance. To address the common problem of data imbalance, a weighted random sampling method was adopted. By giving different sampling weights to different categories of data, the data distribution was effectively balanced, and overfitting of the model to the majority-class data was avoided. Finally, to further improve the generalization ability and global optimization effect of the model, a MemoryBank mechanism was introduced. This mechanism can store the features and prediction results during the training process, providing more context information for the model and helping the model to perform more accurate optimization globally.

[0090] It should be noted that it is also necessary to evaluate the two trained models using the validation dataset, and measure the model performance through indicators such as the confusion matrix, accuracy, Kappa coefficient, Pearson correlation coefficient, and R2 score. The confusion matrix can evaluate the prediction performance of different titers, and the higher the accuracy, Kappa coefficient, Pearson correlation coefficient, and R2 score, the better. Adjust the training strategies of the two models according to the indicators, including parameters such as data balance and learning rate, until the two models with the best performance are obtained.

[0091] In the embodiment of the present application, by collecting images of the obtained antibody titer samples, it is convenient to perform instance segmentation on the collected images subsequently to determine the cell boundaries in the image sample set, and then it is convenient to quickly, accurately, and objectively predict the titer value of the antibody subsequently; by using the cellpose algorithm to perform instance segmentation on each cell in the image sample set, the cell boundaries of each cell in the image sample set can be determined, which is convenient to determine whether it is a positive cell or a negative cell subsequently, and then it is convenient to quickly, accurately, and objectively predict the titer value of the antibody subsequently; through the titer prediction model, the positive cell feature map and negative cell feature map in the image sample set can be quickly segmented, and the titer is predicted by combining regional features and statistical features, which not only realizes the rapid, accurate, and objective evaluation of the antibody titer but also reduces the cost of sample preparation.

[0092] Embodiment 2

[0093] Please refer to Figure 3, Figure 3 Figure 3 is a schematic structural diagram of an embodiment of an antibody titer prediction system based on a titer prediction model provided by the present application, including: an acquisition module 100, a segmentation module 200, and a prediction module 300;

[0094] The acquisition module 100 is configured to perform image acquisition on the acquired antibody titer samples to generate an image sample set;

[0095] It can be understood that the antibody titer samples can be directly obtained or samples prepared through experiments, that is, the present application is not limited to the experimental stage, and the preparation method of the antibody titer samples can refer to the IFA detection process. The corresponding cells are induced and expressed as the corresponding antigens, and then the cells are fixed on a glass slide, and the serum sample to be tested is incubated with the cells to bind the antibodies therein to the EBV antigen to obtain the antibody titer samples. Among them, the antigen can be but is not limited to the EA antigen (EBV early antigen) and the VCA antigen (EBV viral capsid antigen). Exemplarily but not restrictively, when preparing the antibody titer sample corresponding to the EA antigen, the specific operation method is: (1) Cell acquisition and culture: Use the Raji cell line (CCL-86) purchased from the American Type Culture Collection (ATCC), and culture the Raji cells (CCL-86) in RPMI-1640 medium containing 10% fetal bovine serum (FBS), and place the cells under the conditions of 37°C and 5% CO2 for suspension culture; (2) Cell induction: Under the conditions of 37°C and 5% CO2, use 20 ng / ml of Phorbol 12-myristate 13-acetate (TPA) and 3.3 mg / ml of sodium butyrate (SB) to induce the cells, where the induction time is 2 days; (3) Determination of cell viability and quantity: Use the trypan blue exclusion method to evaluate the cell viability, and use an automatic cell counter (Biorad TC 10 TM) To determine the number of cells; (4) Cell treatment and fixation: After washing the cells in 1× phosphate buffered saline (PBS), the cells are dropped onto a teflon-coated slide at a density of approximately 4000 cells per well, and the cells on the slide are fixed using acetone for 10 minutes. Finally, the fixed slide is dried and stored at -20 °C for subsequent use. (5) Antibody-antigen binding: The fixed and stored slide is taken out, and the serum sample to be tested is incubated with the cells on the slide to bind the antibodies in the serum to the EBV antigen (present on Raji cells), obtaining an antibody titer sample. Exemplary and non-limiting, when preparing an antibody titer sample corresponding to the VCA antigen, the specific operation method is as follows: (1) Cell acquisition and culture: The Raji cell line (CCL-86) is purchased from the American Type Culture Collection (ATCC), and the Raji cells are cultured in RPMI-1640 medium containing 10% fetal bovine serum (FBS), and the cells are placed in suspension culture at 37 °C and 5% CO2. (2) Cell viability and quantity assessment: Trypan blue exclusion method is used to evaluate the viability of the cells, and an automatic cell counter (Biorad TC 10 TM ) is used to determine the number of cells. (3) Cell treatment and slide preparation: The cells are washed in 1× phosphate buffered saline (PBS) to remove the culture medium and other impurities, and the washed cells are dropped onto a teflon-coated slide at a density of approximately 4000 cells per well for subsequent experimental operations; (4) Cell fixation and storage: The cells on the slide are fixed using acetone for 10 minutes to maintain the cell morphology and structure for subsequent staining and observation, and then the fixed slide is dried and stored at -20 °C to preserve the cell sample for a long time to ensure the stability and reliability of subsequent experimental results. (5) Antibody-antigen binding: The fixed and stored slide is taken out, and the serum sample to be tested is incubated with the cells on the slide to bind the antibodies in the serum to the VCA antigen (present on B958 cells), obtaining an antibody titer sample.

[0096] Specifically, the acquisition module 100 includes: a preset unit and a collection unit; the preset unit is configured to preset the exposure parameters of the image acquisition device, where the exposure parameters include an exposure value and an exposure time; the collection unit is configured to perform multiple image acquisitions on the antibody titer sample based on the exposure parameters, and stack multiple images with different exposures to obtain an image sample set.

[0097] It is understandable that after obtaining the antibody titer sample, the antibody titer sample needs to be diluted and the image of the diluted sample needs to be captured, wherein the specific method of image capture is: pre-setting the exposure parameters of the image acquisition device, wherein the exposure parameters include exposure value and exposure time; based on the exposure parameters, multiple images of the antibody titer sample are captured, and multiple images with different exposures are stacked to obtain an image sample set.

[0098] It should be noted that the image acquisition device may be, but is not limited to, a EUROstar IIIPlus microscope and a Basler's Ace sensor camera.

[0099] It should be noted that in single image recognition, the cell boundary may be unstable due to poor image quality or complex cell morphology, and false positives and false negatives may occur due to noise in the image or overlap between cells. Therefore, it is necessary to obtain images of multiple antibody titer samples. Different exposure parameters can be used to obtain images of titer samples, and multiple images with different exposures can be stacked to obtain an image sample set.

[0100] In this way, by performing image acquisition on the acquired antibody titer samples, it is convenient to perform instance segmentation on the acquired images to determine the cell boundaries in the image sample set, thereby facilitating the subsequent rapid, accurate and objective prediction of the antibody titer value.

[0101] The segmentation module 200 is used to perform instance segmentation on each cell in the image sample set according to the cellpose algorithm to obtain a cell set image;

[0102] Specifically, the segmentation module 200 includes: a segmentation unit, a first processing unit, a second processing unit and a third processing unit; the segmentation unit is used to perform instance segmentation on each cell in the image sample set according to the cellpose algorithm to obtain an instance segmentation result; the first processing unit is used to eliminate cells that do not meet a preset cell area threshold in the instance segmentation result to obtain a first processing result; the second processing unit is used to count the number of detections of each cell in the first processing result, and retain cells that meet a preset number of detections to obtain a second processing result; the third processing unit is used to eliminate overlapping cells in the second processing result to obtain a cell collection image.

[0103] It can be understood that after obtaining the image sample set, it is necessary to predict the boundaries of cells, which is convenient for subsequent determination of positive or negative cells. Specifically: First, perform instance segmentation on each cell in the image sample set according to the cellpose algorithm to obtain the instance segmentation result. Among them, the cellpose algorithm can, but is not limited to, select the cyto3 pre-trained model to perform cell instance segmentation on each image in the image sample set one by one. At the same time, when performing cell segmentation, parameters such as cell diameter and watershed value can be continuously adjusted to obtain a better instance segmentation result; Second, eliminate the cells in the instance segmentation result that do not meet the preset cell area threshold to obtain the first processing result. Among them, the cell area threshold can be preset, including the minimum cell area threshold and the maximum cell area threshold. By comparing the area of each cell in the instance segmentation result with the preset cell area threshold, cells outside the reasonable cell size range can be excluded; Third, count the number of times each cell is detected in the first processing result and retain the cells that meet the preset detection times to obtain the second processing result. Among them, in an ideal situation, the positions and regions of cells should be exactly the same. Therefore, count the number of times each cell is detected in the first processing result. When the detection times are greater than the preset detection times, it can be considered that the cells detected in this region are valid, and then the second processing result is obtained; Finally, eliminate the overlapping cells in the second processing result to obtain the cell set image. Among them, since there will be multiple detection results at the same position in the second processing result, it is necessary to exclude overlapping cells so that there is only one detected cell at each position, and finally obtain the cell set image.

[0104] It should be noted that the non-maximum suppression (NMS) method can be used, but is not limited to, to exclude overlapping cells. Specifically, the non-maximum suppression method will sort the cells from large to small, first select the cells with larger areas, and traverse the remaining cell set to calculate the intersection over union (IoU) with the currently selected cell. When the IoU is greater than the threshold, the cell is excluded from the remaining cell set, thus ensuring that there is only one detected cell at each position.

[0105] In this way, by performing instance segmentation on each cell in the image sample set through the cellpose algorithm, the cell boundaries of each cell in the image sample set can be determined, which is convenient for subsequent determination of whether the cell is a positive cell or a negative cell, and further convenient for subsequent rapid, accurate and objective prediction of the antibody titer value.

[0106] The prediction module 300 is configured to input the cell set image into a preset titer prediction model to obtain a corresponding feature map, extract the corresponding regional features of the feature map, and respectively extract the corresponding statistical features from the positive cell feature map and the negative cell feature map, and predict the titer value of the antibody based on the regional features and the statistical features, wherein the feature map includes a positive cell feature map, a negative cell feature map, and a background feature map, and the titer prediction model includes a UNet positive cell segmenter, a deep learning feature extractor, a statistical feature extractor, and an omics titer regression predictor.

[0107] It can be understood that after obtaining the cell set image, the positive cells and negative cells can be determined. Specifically, the cell set image can be input into a preset titer prediction model. The schematic diagram of the prediction process of the titer prediction model is as Figure 2 shown. Through the UNet positive cell segmenter in the titer prediction model, a positive cell feature map, a negative cell feature map, and a background feature map can be obtained. The feature map is also the semantic segmentation result.

[0108] It should be noted that the UNet positive cell segmenter can be trained but is not limited to using the UNet architecture, wherein the UNet network structure uses ResNet18 as an encoder. By way of example but not limitation, the training process of the UNet positive cell segmenter is as follows: first, a training image sample set is prepared in advance; second, since there is an obvious difference in brightness values ​​between the background image, negative cell image and positive cell image in the training image sample set, that is, the positive cell image brightness value > negative cell image brightness value > background area brightness value, based on this feature, the average brightness value in each cell area in the training image sample set is extracted as the brightness representative value of the cell, and the brightness representative values ​​of all cells are sorted to obtain a brightness representative value sorting curve with the horizontal axis being the sorted cell number and the vertical axis being the brightness value; third, the brightness sorting curve is analyzed by an inflection point detection algorithm, and then the curve is extracted. The inflection point represents the potential dividing point between positive cells and negative cells, wherein the inflection point detection algorithm may be, but is not limited to, KneeLocator in the knee library of python. When there is an obvious inflection point in the curve, the brightness value corresponding to the inflection point can be used as the judgment threshold of the positive cell, and according to the determined judgment threshold, the cell instance with a brightness value higher than the threshold is marked as a positive cell, otherwise it is a negative cell, so that the positive cell feature map and the negative cell feature map can be preliminarily determined; the preliminarily determined positive cell feature map and negative cell feature map are used as the target of initialization training of the UNet positive cell segmentor, and the training image sample set is input into the UNet positive cell segmentor, and the UNet positive cell segmentor uses the softmax function to correspond to the probability distribution S of the three categories of the background image, the negative cell feature map and the positive cell feature map for each cell in the training image sample set. b , S n and S p , and then based on the probability distribution S b , S n and S p Determine the corresponding feature map.

[0109] It should be noted that in order to keep the semantic segmentation result of the UNet positive cell segmenter differentiable, we do not use the argmax function to obtain the segmentation result of each pixel, but use softmax to obtain the classification probability map S of the three regions of background, negative cells and positive cells. b , S n and S p , and there is S b +S n +S p =1, the corresponding feature map can be determined based on the classification probability map.

[0110] It should be noted that UNet can be replaced by other semantic segmentation models.

[0111] It should be noted that the positive cell feature map and negative cell feature map obtained through inflection point detection are only roughly estimated images. Therefore, in order to ensure segmentation accuracy, it is necessary to further train the UNet positive cell segmenter.

[0112] It can be understood that after obtaining the feature maps, the weighted average features will be extracted from each feature map (positive cell feature map, negative cell feature map, and background feature map) through the deep learning feature extractor in the titer prediction model. At the same time, the global average pooling features will be extracted, and the weighted average features and global average pooling features will be concatenated to obtain the regional features, and the regional features will be reduced to a fixed dimension (32 dimensions) through a multi-layer perceptron.

[0113] It can be understood that after obtaining the feature maps, it is also necessary to extract the corresponding statistical features (7 dimensions) from the positive cell feature map and negative cell feature map through the statistical feature extractor in the titer prediction model. Among them, the statistical features include the average fluorescence intensity and standard deviation of fluorescence intensity corresponding to the positive cell feature map, the average fluorescence intensity and standard deviation of fluorescence intensity corresponding to the negative cell feature map, the overall average fluorescence intensity and standard deviation of fluorescence intensity; the statistical features are input into the omics titer prediction model to obtain the titer value of the antibody. Among them, the calculation formula of the statistical features is specifically:

[0114]

[0115]

[0116] In the formula, mean p is the average fluorescence intensity of the positive cell feature map, V is the average brightness image, S p is the distribution probability of positive cells, std p is the standard deviation of fluorescence intensity of the positive cell feature map, mean n is the average fluorescence intensity of the negative cell feature map, S n is the distribution probability of negative cells, std n is the standard deviation of fluorescence intensity of the negative cell feature map, mean p+n is the overall average fluorescence intensity, std p+n is the overall standard deviation of fluorescence intensity.

[0117] In this way, the positive cell feature map and the negative cell feature map in the image sample set can be quickly segmented by the UNet positive cell segmenter, and the corresponding statistical features in the positive cell feature map and the negative cell feature map can be calculated respectively, which is convenient for quickly, accurately and objectively predicting the titer value of the antibody based on the omics titer prediction model.

[0118] It can be understood that after obtaining the regional features and statistical features, the regional features and statistical features need to be concatenated, and the titer regression predictor in the titer prediction model is used for regression prediction to predict the titer value of the antibody. Among them, the titer regression predictor is a multi-layer perceptron with dropout.

[0119] It should be noted that the titer prediction model is weakly supervised and trained through a multi-task learning framework. By simultaneously optimizing the two tasks of cell region segmentation and ordinal regression, the performance of the model is improved. During the training process, a variety of data augmentation methods are adopted to increase the diversity of data. The Adam optimizer and the cosine annealing learning rate scheduling strategy are used to optimize the model parameters. The data imbalance problem is handled by weighted random sampling, and a memory bank is used to store features and prediction results to achieve the effect of global optimization.

[0120] It should be noted that the titer prediction model includes two main loss functions, namely the mask segmentation loss and the ordinal regression loss. Among them, the mask segmentation loss uses the cross-entropy loss function to optimize the cell region segmentation, and the ordinal regression loss uses the global ordinal loss to ensure that the prediction result satisfies the ordinal relationship of the titer. The calculation formula of the loss function is: Total_Loss = w1 * Mask_Loss + w2 * Ordinal_Loss, where Total_Loss is the total loss value, Mask_Loss and Ordinal_Loss are the cross-entropy loss value and the ordinal regression loss value respectively, and w1 and w2 are weight coefficients used to balance the importance of the two tasks.

[0121] It should be noted that when training the titer prediction model, data augmentation needs to be performed on the training data set. The means of data augmentation include but are not limited to: random rotation (±30 degrees), random cropping, and scaling while maintaining the aspect ratio, etc.

[0122] It should be noted that the training process is as follows: First, in terms of the optimizer, the Adam optimizer with excellent performance was selected, and its initial learning rate was set to 0.0005 to ensure that the model can converge smoothly and quickly in the initial stage of training. To further improve the training effect, a cosine annealing learning rate scheduling strategy was introduced, which can dynamically adjust the learning rate according to the training progress, so as to give the model the most appropriate update step size at different stages of training. In addition, special attention was paid to the selection of the batch size during training, and the batch size was set to 16. This setting not only ensures the training efficiency but also ensures that the model can fully learn the internal characteristics of the data. At the same time, 100 training epochs were set to ensure that the titer prediction model has enough time to fully fit the data and achieve stable performance. To address the common problem of data imbalance, a weighted random sampling method was adopted. By giving different sampling weights to different categories of data, the data distribution was effectively balanced, and overfitting of the model to majority-class data was avoided. Finally, to further improve the generalization ability and global optimization effect of the model, a MemoryBank mechanism was introduced. This mechanism can store the features and prediction results during the training process, providing more context information for the model and helping the model to perform more accurate optimization globally.

[0123] It should be noted that it is also necessary to evaluate the two trained models using the validation dataset, and measure the model performance through indicators such as the confusion matrix, accuracy, Kappa coefficient, Pearson correlation coefficient, and R2 score. The confusion matrix can evaluate the prediction performance of different titers, and the higher the accuracy, Kappa coefficient, Pearson correlation coefficient, and R2 score, the better. Adjust the training strategies of the two models according to the indicators, including parameters such as data balance and learning rate, until the two models with the best performance are obtained.

[0124] In the embodiment of the present application, by collecting images of the obtained antibody titer samples, it is convenient to perform instance segmentation on the collected images subsequently to determine the cell boundaries in the image sample set, and further convenient to quickly, accurately, and objectively predict the titer value of the antibody subsequently; by using the cellpose algorithm to perform instance segmentation on each cell in the image sample set, the cell boundaries of each cell in the image sample set can be determined, which is convenient to determine whether it is a positive cell or a negative cell subsequently, and further convenient to quickly, accurately, and objectively predict the titer value of the antibody subsequently; by using the titer prediction model, the positive cell feature map and negative cell feature map in the image sample set can be quickly segmented, and the titer is predicted by combining regional features and statistical features, which not only realizes the rapid, accurate, and objective evaluation of the antibody titer but also reduces the cost of sample preparation.

[0125] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separated, that is, they may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the method of this embodiment.

[0126] The specific embodiments described above further elaborate on the purpose, technical solutions, and beneficial effects of this application. It should be understood that the above are only specific embodiments of this application and are not used to limit the protection scope of this application.

[0127] It is particularly pointed out that for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of this application shall be included within the protection scope of this application.

Claims

1. A method for predicting antibody titer based on a titer prediction model, characterized in that: include: Capturing images of the acquired antibody titer samples to generate an image sample set; Perform instance segmentation on each cell in the image sample set according to the cellpose algorithm to obtain a cell set image; The cell collection image is input into a preset titer prediction model to obtain a corresponding feature map, the regional features corresponding to the feature map are extracted, and the corresponding statistical features are extracted from the positive cell feature map and the negative cell feature map respectively, and the titer value of the antibody is predicted based on the regional features and the statistical features, wherein the feature map includes a positive cell feature map, a negative cell feature map and a background feature map, and the titer prediction model includes a UNet positive cell segmentor, a deep learning feature extractor, a statistical feature extractor and an omics titer regression predictor.

2. The method for predicting antibody titer based on a titer prediction model according to claim 1, characterized in that: The method of collecting images of the acquired antibody titer samples to generate an image sample set is specifically as follows: Presetting exposure parameters of the image acquisition device, wherein the exposure parameters include exposure value and exposure time; A plurality of images of the antibody titer sample are collected based on the exposure parameters, and a plurality of images with different exposures are stacked to obtain an image sample set.

3. The antibody titer prediction method based on the titer prediction model according to claim 1, characterized in that: The cell pose algorithm is used to perform instance segmentation on each cell in the image sample set to obtain a cell set image, specifically: Perform instance segmentation on each cell in the image sample set according to the cellpose algorithm to obtain an instance segmentation result; Eliminate cells that do not meet a preset cell area threshold in the instance segmentation result to obtain a first processing result; Counting the number of times each cell is detected in the first processing result, and retaining cells that meet a preset number of detections to obtain a second processing result; The overlapping cells in the second processing result are eliminated to obtain a cell collection image.

4. The method for predicting antibody titer based on a titer prediction model according to claim 1, characterized in that: The statistical features include the average fluorescence intensity and standard deviation of the fluorescence intensity corresponding to the positive cell characteristic graph, the average fluorescence intensity and standard deviation of the fluorescence intensity corresponding to the negative cell characteristic graph, and the overall average fluorescence intensity and standard deviation of the fluorescence intensity.

5. The method for predicting antibody titer based on the titer prediction model according to claim 4, characterized in that: The calculation formula of the statistical characteristics is specifically: In the formula, mean p is the average fluorescence intensity of the positive cell feature map, V is the average brightness image, S p is the distribution probability of positive cells, std p is the standard deviation of the fluorescence intensity of the positive cell feature map, mean n is the average fluorescence intensity of the negative cell feature map, S n is the distribution probability of negative cells, std n is the standard deviation of the fluorescence intensity of the negative cell feature map, mean p+n is the overall mean fluorescence intensity, std p+n is the standard deviation of the overall fluorescence intensity.

6. An antibody titer prediction system based on a titer prediction model, characterized in that: include: Acquisition module, segmentation module and prediction module; The acquisition module is used to collect images of the acquired antibody titer samples to generate an image sample set; The segmentation module is used to perform instance segmentation on each cell in the image sample set according to the cellpose algorithm to obtain a cell set image; The prediction module is used to input the cell collection image into a preset titer prediction model to obtain a corresponding feature map, extract the regional features corresponding to the feature map, and respectively extract corresponding statistical features from the positive cell feature map and the negative cell feature map, and predict the titer value of the antibody based on the regional features and the statistical features, wherein the feature map includes a positive cell feature map, a negative cell feature map and a background feature map, and the titer prediction model includes a UNet positive cell segmentor, a deep learning feature extractor, a statistical feature extractor and an omics titer regression predictor.

7. The antibody titer prediction system based on the titer prediction model according to claim 6, characterized in that: The acquisition module includes: a preset unit and a collection unit; The preset unit is used to preset the exposure parameters of the image acquisition device, wherein the exposure parameters include the exposure value and the exposure time; The acquisition unit is used to acquire multiple images of the antibody titer sample based on the exposure parameters, and stack multiple images with different exposures to obtain an image sample set.

8. The antibody titer prediction system based on the titer prediction model according to claim 6, characterized in that: The segmentation module includes: a segmentation unit, a first processing unit, a second processing unit and a third processing unit; The segmentation unit is used to perform instance segmentation on each cell in the image sample set according to the cellpose algorithm to obtain an instance segmentation result; The first processing unit is used to remove cells that do not meet a preset cell area threshold in the instance segmentation result to obtain a first processing result; The second processing unit is used to count the number of times each cell is detected in the first processing result, and retain cells that meet a preset number of detections to obtain a second processing result; The third processing unit is used to eliminate overlapping cells in the second processing result to obtain a cell collection image.

9. The antibody titer prediction system based on the titer prediction model according to claim 6, characterized in that: The statistical features include the average fluorescence intensity and standard deviation of the fluorescence intensity corresponding to the positive cell characteristic graph, the average fluorescence intensity and standard deviation of the fluorescence intensity corresponding to the negative cell characteristic graph, and the overall average fluorescence intensity and standard deviation of the fluorescence intensity.

10. The antibody titer prediction system based on the titer prediction model according to claim 9, characterized in that: The calculation formula of the statistical feature is specifically: In the formula, mean p is the average fluorescence intensity of the positive cell feature map, V is the average brightness image, S p is the distribution probability of positive cells, std p is the standard deviation of the fluorescence intensity of the positive cell feature map, mean n is the average fluorescence intensity of the negative cell feature map, S n is the distribution probability of negative cells, std n is the standard deviation of the fluorescence intensity of the negative cell feature map, mean p+n is the overall mean fluorescence intensity, std p+n is the standard deviation of the overall fluorescence intensity.

Images