Method for identifying Tfh cell subpopulation based on generative AI and application of method in evaluating R-CHOP treatment effect of DLBCL patient

Abstract

The invention discloses a method for recognizing a Tfh cell subset based on generative AI and application of the method in evaluating the R-CHOP treatment effect of a DLBCL patient. The Tfh cell subset is a cell subset with the cell surface marker defined as CD4 + CXCR5 + PD-1-on the protein level; the higher the proportion of CD4 + CXCR5 + PD-1-Tfh cells is, the better the prognosis effect of a patient is, and the longer the lifetime is, it is proved that the CD4 + CXCR5 + PD-1-Tfh cells can be used for evaluation of the R-CHOP treatment effect and prognosis of DLBCL, an adaptive-based adversarial generation model is constructed, generation of multiple immunofluorescent staining images can be achieved based on a monochromatic DAPI staining image, and the DLBCL R-CHOP treatment effect and prognosis evaluation can be achieved. Recognition of Tfh cell subpopulation CD4 + CXCR5 + PD-1-cells is achieved, and the Tfh cell subpopulation CD4 + CXCR5 + PD-1-cells have great application prospects in the biological field and medicine convertible research.

Description

Technical Field

[0001] The present invention relates to the technical field of tumor prognosis assessment, and specifically to a method for identifying Tfh cell subpopulations based on generative AI, and also to an application of Tfh cell subpopulations in evaluating the efficacy of R-CHOP treatment in DLBCL patients. Background Art

[0002] Diffuse large B cell lymphoma (DLBCL) is the most common and most prevalent subtype of non-Hodgkin lymphoma (NHL), accounting for 30% to 35% of all NHL cases. DLBCL is a group of heterogeneous tumors with great differences in clinical manifestations, tissue morphology and prognosis. The emergence of rituximab has brought the cure rate of DLBCL to 60% to 80%, but one-third of patients still have a poor prognosis. It is of great significance to detect patients with poor prognosis early and adjust the treatment strategy in time to prolong the patient's survival.

[0003] The development of molecular pathology and targeted therapy has revealed many prognostic evaluation factors for DLBCL. Currently, the existing clinical prognostic factors for survival include: international prognostic index (IPI) score, MYC and BCL-2, cell origin, gene mutation, genetic risk prediction model, circulating tumor DNA (ct-DNA), the ratio of the absolutelymphocyte count to the absolute monocyte count (ALC / AMC) in peripheral blood, programmed death-1 ligand (PD-L1), serum soluble interleukin-2 receptor (sIL-2R) and other indicators. However, most indicators lack independent validation and are limited in clinical use. Among them, the most widely used and guideline-approved is the IPI prognostic scoring system, which has been widely used in the prognosis of patients with invasive NHL since 1993. However, in recent years, with the continuous progress of targeted immunotherapy, rituximab combined with chemotherapy has improved the prognosis of DLBCL patients, and the status of IPI score in the prognostic system has been constantly impacted. Some scholars have proposed the modified international prognostic index (National Comprehensive Cancer Network-internationalprognostic index, NCCN-IPI) and GELTAMO-IPI (G-IPI) stratification method based on the IPI score. Compared with IPI, it has higher predictive efficiency, but it still cannot fully identify high-risk groups. IPI and its derived scoring system provide simple and reliable prognostic information for patients at diagnosis and are the benchmark for measuring all additional prognostic factors. However, IPI has not had a substantial impact on treatment so far. There is currently no data to support changes in treatment strategies based on IPI scores, and it is impossible to predict responses to new drug treatments.

[0004] Therefore, it is urgent to establish a more complete DLBCL prognosis assessment system to truly divide the high-risk population, and at the same time fully utilize the advantages of targeted therapy, adopt the most appropriate treatment plan to improve the patient's remission rate, reduce drug resistance and prolong the patient's survival, and guide clinical physicians to give patients more precise individualized treatment. Summary of the invention

[0005] In view of this, the purpose of the present invention is to provide the application of Tfh cell subpopulations in the evaluation of the therapeutic efficacy of R-CHOP treatment for DLBCL patients or in the screening of clinical therapeutic drugs; the second purpose of the present invention is to provide the application of reagents for detecting Tfh cell subpopulations in the preparation of products for evaluating the therapeutic efficacy of R-CHOP treatment for DLBCL patients; the third purpose of the present invention is to provide a method for identifying Tfh cell subpopulations based on generative AI.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] 1. Application of Tfh cell subsets in evaluating the efficacy of R-CHOP treatment in DLBCL patients, characterized in that the Tfh cell subsets are defined as CD4 + CXCR5 + PD-1 - cell subpopulations.

[0008] Tfh cell subsets CD4 + CXCR5 + PD-1 - The higher the proportion in the total cells, the better the R-CHOP treatment efficacy and the longer the prognosis and survival of DLBCL patients.

[0009] 2. Application of reagents for detecting Tfh cell subsets in the preparation of products for evaluating the efficacy of R-CHOP treatment in DLBCL patients.

[0010] Preferably, the reagent for detecting Tfh cell subsets includes one or more immune reagents of CD3E, CD3D, CD3G, CD4, BCL6, CXCR5, EBI3, NMB, IFNG or PDCD1.

[0011] Preferably, the immunoreagent is used for immunofluorescence / immunohistochemistry antibody or immunocolloidal gold labeling.

[0012] Preferably, the immune reagent includes a colorimetric marker molecular probe that specifically recognizes CD4, CXCR5 and PD-1.

[0013] Preferably, the color developer-labeled molecular probes are CD4 molecular probes, CXCR5 molecular probes and PD-1 molecular probes respectively labeled with different fluorescent markers.

[0014] Preferably, the product of the present invention further comprises a dewaxing agent, a gradient ethanol solution, a blocking solution, a dye, a DAPI stain and a TBST buffer.

[0015] The present invention preferably detects Tfh cell subsets CD4 + CXCR5+ PD-1 - The higher the proportion in the total cells, the better the R-CHOP treatment efficacy and the longer the prognosis and survival of DLBCL patients.

[0016] 3. A method for identifying Tfh cell subsets based on generative AI, comprising the following processing steps: In order to achieve accurate generation of multiple immunofluorescence staining images mIHC, a conditional generative adversarial network is used as the baseline model, in which ResNet-6Blocks is used as the generator and a 5-layer PatchGAN is used as the discriminator, that is, each image is divided into patches and input into the discriminator, and then the overall results are averaged as the discriminator result; at the same time, an adaptive supervised PatchNCE loss function is introduced, and based on the patch input, the distance between positive and negative samples is evaluated, so that similar samples are judged as the same sample area as much as possible, so as to ensure that the generated image is highly consistent with the real image in terms of pixels.

[0017] Preferably, the present invention uses the structural similarity index (SSIM) and the peak signal-to-noise ratio (PSNR) to evaluate the quality of the generated image. Assuming that x and y are the generated image and the real image, respectively, SSIM and PSNR are defined as follows:

[0018]

[0019] Where μ and σ represent the mean and variance of the image respectively, and σ xy is the covariance between image x and image y, c1, c2 are constants;

[0020]

[0021] Where MAX is the maximum value of the image pixel value, m and n are the number of pixels of image x and image y.

[0022] The present invention has a good vertical application scenario in clinical application, targeting the Tfh cell subset CD4 + CXCR5 + PD-1 - To identify cells, we built an adaptive adversarial generation model, which can generate multiple staining images based on single-color DAPI staining images, and can also identify Tfh cell subsets CD4 + CXCR5 + PD-1 - Cell recognition. Effective expansion of CD4 + CXCR5 + PD-1 - Adjunctive applications of cells in tumor progression.

[0023] The beneficial effects of the present invention are as follows: The present invention discloses the application of Tfh cell subpopulations in evaluating the therapeutic efficacy of R-CHOP in DLBCL patients. The present invention adopts single-cell RNA sequencing technology, external database data and multiple immunofluorescence (mIF) detection technology to analyze the characteristics and spatial distribution characteristics of new subpopulations of follicular helper T cells (Tfh) in DLBCL and immune microenvironment, and finds that in tumor samples of relapsed patients, the proportion of Tfh cell subpopulations with high expression of CD3E, CD3D, CD3G, CD4, BCL6, CXCR5, EBI3 or NMB and low expression of IFNG or PDCD1 genes is significantly reduced, suggesting that this Tfh cell subpopulation may be related to poor tumor prognosis. Further in-depth analysis of the CD4 Tfh subpopulation in the microenvironment of tumor tissue samples of DLBCL patients who respond and do not respond to R-CHOP treatment was performed. + CXCR5 + PD-1 - The proportion of tumor cells interacts with the space and finds that CD4 + CXCR5 + PD-1 - The higher the proportion of Tfh cells, the better the prognosis of the patient and the longer the survival period. + CXCR5 + PD-1 - Tfh cells can be used for R-CHOP treatment efficacy and prognosis evaluation of DLBCL. We constructed an adaptive adversarial generation model, which can generate multiple immunofluorescence staining images based on single-color DAPI staining images, and can also generate Tfh cell subsets CD4 + CXCR5 + PD-1 - Cell recognition. Effective expansion of CD4 + CXCR5 + PD-1 - The auxiliary application of cells in tumor progression lowers the experimental threshold of cell-assisted diagnosis in clinical diagnosis, making it have great application prospects in the biological field and medical translational research. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to make the purpose, technical solution and beneficial effects of the present invention clearer, the present invention provides the following drawings for illustration:

[0025] Figure 1 Flow chart for sequencing library data analysis;

[0026] Figure 2, A is the UMAP analysis of immune cell subsets in DLBCL tumor tissue scRNA-seq; B is the gene annotation diagram of cell subsets; C is the heat map of differentially expressed genes in different subgroups; D is the definition of T cell subsets and the heat map of differentially expressed genes;

[0027] Figure 3 , A is the relationship between Tfh cell subsets and poor tumor prognosis; B is the relationship between the proportion of Tfh cell subsets and long-term tumor prognosis, q1, q2, q3, q4 represent different regional segments, and the quartiles represent 25%, 50%, 75%, and 100% respectively;

[0028] Figure 4 , A is the tissue staining image and CD4 + CXCR5 + PD-1 - Tfh cell marker image; B is CD4 + CXCR5 + PD-1 - The proportion of Tfh cells in total cells in the R-CHOP response group was higher than that in the R-CHOP non-response group; C is CD4 + CXCR5 + PD-1 - The proportion of Tfh cells in total cells and prognostic analysis;

[0029] Figure 5 , A is CD4 + CXCR5 + PD-1 - Analysis of spatial distribution of Tfh cells and their interaction with tumor cells; B is CD4 + CXCR5 + PD-1 - Distribution of Tfh cells in the R-CHOP response and R-CHOP non-response patient groups; C is CD4 + CXCR5 + PD-1 - Relationship between Tfh cell distribution and prognosis of DLBCL patients.

[0030] Figure 6 , the network composition and generation results of multiple immunofluorescence staining images; a is a schematic diagram of the mIHC image prediction network; the generation network consists of a generator and a discriminator; the generator uses the ResNet-6 block structure to generate mIHC images, while the discriminator uses the modality discrimination module based on PatchGAN; b is the use of the generated mIHC images to predict CD4 + Performance demonstration of classifying TFH1 cells; the top shows the CD4 +Confusion matrix for TFH1 cell classification; the confusion matrix for CD4 + Various classification indicators of TFH1 cells, including accuracy, precision, recall, and F1 score; Figure c is a visualization of the generated results; three cases are shown here; in each column, the top image is a DAPI staining image, the middle image is the corresponding mIHC image, and the bottom image is a synthetic mIHC image generated using our proposed method (DAPI: blue; CD20: red; CD4: magenta; CD8: cyan; PD-1: yellow; CXCR5: green, scale bar is 200μm. DETAILED DESCRIPTION

[0031] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.

[0032] Example 1

[0033] 1. Dissociate fresh DLBCL tumor tissue samples to isolate single cell suspension. The specific steps are as follows:

[0034] (1) After surgery, the excised tumor tissue was immersed in PBS and moved to a clean bench. The tissue was cut into 5 mm thick pieces using a sterile scalpel. 3 Size: about the size of a soybean;

[0035] (2) Wash the minced tissue 1-2 times with PBS and store the tissue in tissue protection solution (130-100-008, Miltenyi Biotec), making sure that the sample is submerged in the protection solution;

[0036] (3) Refer to the tissue dissociation kit (130-095-929, Miltenyi Biotec) operating procedures to dissociate tumor tissue into single-cell suspension; sample requirements: cell viability ≥ 90%, uniform cell size, diameter between 7-40 μm, cell clusters and debris impurities < 5%, double cell rate < 0.09%, cell number ≥ 100,000, and no reverse transcription inhibitors are present in the cell suspension.

[0037] 2. Library construction and sequencing. The specific steps are as follows:

[0038] (1) 10x single-cell RNA sequencing technology labels cDNA fragments. Gel beads containing barcode information are first combined with a mixture of cells and enzymes, and then wrapped by oil surfactant droplets located in the microfluidic "double cross" connection. The droplets are very small and can only accommodate one gel bead. After the droplets flow into the reservoir and are collected, the gel beads dissolve to release the primer sequence, reverse transcribe the cDNA fragments, and perform PCR amplification using cDNA as a template. The products containing barcode information in each droplet are mixed to construct a sequencing library.

[0039] (2) Library construction: First, the cDNA is fragmented into fragments of about 200-300 bp, and then the sequencing adapter P5 and sequencing primer R1 are added to the traditional second-generation sequencing library construction process. Finally, PCR amplification is performed to obtain the DNA library.

[0040] (3) Cluster generation and sequencing: The library is quantified and the qualified library is placed in the cBot for bridge amplification. After the cluster is generated, it is sequenced using a sequencer. The principle is that dNTPs labeled with four fluorescent groups, A, T, G, and C, are added in each cycle. According to the AT and GC pairing, the corresponding dNTPs are bound to the template DNA chain by DNA polymerase, and other unbound dNTPs are washed away. The binding position releases a fluorescent signal, which is captured by the computer and converted accordingly, thereby obtaining the base information at that position.

[0041] 3. Data analysis process Figure 1 As shown, the sample clinical information and survival data from an external database were combined for verification.

[0042] First, a total of 9 fresh DLBCL tumor tissue samples were collected from Sichuan Cancer Hospital for initial diagnosis and recurrence. Single-cell RNA sequencing technology was used to perform scRNA-seq UMAP analysis of DLBCL tumor tissues. The results are shown in Figure 2 , A; Gene annotation of cell subpopulations, the results are shown in Figure 2 , B; Heat map results of differentially expressed genes in different subgroups are shown in Figure 2 , C; T cell subset definition and differential gene heat map results are shown in Figure 2 , D; Reference genes for single-cell RNA sequencing of new Tfh cell subpopulations are shown in Table 1.

[0043] Table 1. Reference genes for new Tfh cell subpopulations sequenced in single cells

[0044]

[0045]

[0046] By analyzing the single-cell sequencing data (8,000 to 10,000 cells were collected and tested for each sample on average), it was found that in the tumor samples of recurrent patients, the proportion of Tfh cell subpopulations with high expression of CD3E, CD3D, CD3G, CD4, BCL6, CXCR5, EBI3 or NMB and low expression of IFNG or PDCD1 genes decreased significantly, suggesting that this Tfh cell subpopulation may be related to poor tumor prognosis ( Figure 3 , A), further validation and comparison of RNA bulk sequencing data of newly diagnosed DLBCL tumor tissue samples from the external public database TCGA also found that in newly diagnosed DLBCL patients, the higher the proportion of secondary Tfh cell subsets in the tumor tissue microenvironment, the better the long-term prognosis of the patients ( Figure 3 , B).

[0047] Example 2

[0048] Tumor tissue samples of DLBCL patients who responded to R-CHOP treatment in Sichuan Cancer Hospital (a total of 14 cases) and those who did not respond (a total of 18 cases) were obtained. Tissue HE staining was used to preliminarily determine the spatial regional characteristics of the tissues, group them and number them, and a follow-up cohort study was conducted. Multiple fluorescent staining was used to stain the tumor tissue samples for molecular markers CD20, CD4, CXCR5, PD-1, and DAPI (Novo-Light TSA, Shanghai Yuanxi Biotechnology Co., Ltd.), and a fluorescent imaging scanner (3DHISTECH) was used for panoramic scanning of the slices. The specific steps are as follows:

[0049] (1) Use Leica bond III stainer to perform dewaxing, antigen repair and staining steps. The antigen repair condition is ER2 antigen repair solution at 100 degrees for 20 minutes. After each step between the following (2)-(5), wash with TBST 3-5 times;

[0050] (2) Blocking: 3% hydrogen peroxide (Peroxide Block) for 10 min, then block with blocking solution for 10 min at room temperature, and wash with TBST 3-5 times;

[0051] (3) Antibody dilution and staining: Wash 3-5 times with TBST after each step between the ab steps;

[0052] CD20:

[0053] a. Primary antibody incubation: CD20 antibody stock solution, incubate at room temperature for 60 minutes.

[0054] b. Secondary antibody incubation: Incubate with Leica DS9800 staining system secondary antibody (Post Primary / Polymer) at room temperature for 10 min.

[0055] c. Incubate with fluorescent dye Novo-Light TSA620 at 1:100 at room temperature for 10 minutes.

[0056] PD-1:

[0057] a. Primary antibody incubation: dilute PD-1 antibody in antibody diluent / blocking solution at a concentration of 1:400 and incubate at room temperature for 60 minutes;

[0058] b. Secondary antibody incubation: Incubate with Leica DS9800 staining system secondary antibody (Post Primary / Polymer) at room temperature for 10 min.

[0059] c. Incubate with fluorescent dye Novo-Light TSA570 at 1:100 at room temperature for 10 minutes.

[0060] CXCR5:

[0061] a. Primary antibody incubation: dilute CXCR5 antibody in antibody diluent / blocking buffer at a concentration of 1:200 and incubate at room temperature for 30 minutes.

[0062] b. Secondary antibody incubation: Leica DS9800 staining system secondary antibody (Polymer) incubated at room temperature for 10 minutes.

[0063] c. Incubate with fluorescent dye Novo-Light TSA 520 at 1:100 at room temperature for 10 min.

[0064] CD4:

[0065] a. Primary antibody incubation: dilute CD4 antibody in antibody diluent / blocking buffer at a concentration of 1:400 and incubate at room temperature for 30 minutes.

[0066] b. Secondary antibody incubation: Leica DS9800 staining system secondary antibody (Polymer) incubated at room temperature for 10 minutes.

[0067] (4) Incubate with DAPI for 10 min to stain the nucleus

[0068] (5) Seal the slides with anti-fluorescence quenching agent and take pictures.

[0069] In this example, the antibodies for markers CD20, CD4, CXCR5, and PD-1 are as follows:

[0070] The primary antibody for CD20 labeling was rabbit anti-human CD20 monoclonal antibody (CM-0221, Shanghai Jiehao Biotechnology Co., Ltd.), and the secondary antibody was Leica DS9800 staining system secondary antibody (Post Primary / Polymer).

[0071] The primary antibody for CD4 marker was rabbit anti-human CD4 monoclonal antibody (ab133616, Abcam), and the secondary antibody was Leica DS9800 staining system secondary antibody (Post Primary / Polymer).

[0072] The primary antibody for CXCR5 labeling was rabbit anti-human CXCR5 monoclonal antibody (72712S, CST), and the secondary antibody was Leica DS9800 staining system secondary antibody (Post Primary / Polymer).

[0073] The primary antibody for PD-1 labeling was rabbit anti-human PD-1 monoclonal antibody (10377-MM23, Sino biological), and the secondary antibody was Leica DS9800 staining system secondary antibody (Post Primary / Polymer).

[0074] Tissue staining images and CD4 + CXCR5 + PD-1 - The iconic images of TFH cells are as follows Figure 4 , as shown in A; the High plex FL module of the digital case analysis software Halo (Indica Lab) was used to classify single cells in the tissue, and the results showed that CD4 + CXCR5 + PD-1 - The proportion of Tfh cells in total cells in the R-CHOP response group was higher than that in the R-CHOP non-response group, which successfully predicted the effect of R-CHOP treatment on DLBCL patients. Figure 4 , as shown in B; CD4 + CXCR5 + PD-1 - The proportion of Tfh cells in total cells was analyzed for prognosis, and the results showed that CD4 + CXCR5 + PD-1 - The prognosis of DLBCL patients with a high proportion of Tfh cells was better. Figure 4 , as shown in C.

[0075] The spatial analysis module of digital case analysis software Halo (Indica Lab) was used to analyze CD4 + CXCR5 + PD-1 - Analysis of the spatial distribution of Tfh cells and their interaction with tumor cells, with 20 μm as the cell interaction distance ( Figure 5 , A), The results showed that CD4 + CXCR5+ PD-1 - Tfh cells were more abundant in the R-CHOP response group than in the R-CHOP non-response group, and successfully predicted the effect of R-CHOP treatment on DLBCL patients. Figure 5 , B); CD4 + CXCR5 + PD-1 - Tfh cells were used for prognostic analysis, and the results showed that CD4 + CXCR5 + PD-1 - The more Tfh cells there are, the better the prognosis of DLBCL patients is. The prognosis of DLBCL patients can be successfully predicted. Figure 5 , as shown in C.

[0076] Example 3

[0077] Generation of multiplex immunofluorescence staining images mIHC and Tfh cell subsets CD4 + CXCR5 + PD-1 - Identification

[0078] Due to the complexity of multiple immunofluorescence staining technology and the high cost of materials and time, it is difficult to directly apply it in clinical scenarios. Therefore, in order to solve this problem, this study innovatively applied artificial intelligence methods to realize the generation of multiple immunofluorescence staining images mIHC images, and at the same time constructed and evaluated the generated mIHC images for CD4 + CXCR5 + PD-1 - The specific implementation is as follows.

[0079] 1. Construction of training data set and image preprocessing.

[0080] To train our model, we used 30 DLBCL samples, each containing six different IHC staining images. Each sample was then divided into two parts: one was a single-color DAPI image as the input of the model; the other was a six-channel (DAPI, CD20, CD4, CD8, PD-1, CXCR5) merged mIHC image as the label for generating output. Subsequently, we carefully aligned the images of each sample pair and used the clustering constrained attention multi-instance learning (CLAM) method to identify cell regions and segment image patches. This process generated 40,000 image patches, each with a resolution of 1024×1024 pixels. Among them, 32,000 (80%) images were used for the training set, and the remaining 8,000 (20%) images were reserved as the test set.

[0081] 2. Model construction.

[0082] In order to achieve accurate generation of multiple immunofluorescence staining images mIHC, we used a more robust conditional generative adversarial network (pyramid pix2pix) as the baseline model ( Figure 6 , a), where ResNet-6Blocks is used as the generator, and a 5-layer PatchGAN is used as the discriminator. Each image is divided into patches and input into the discriminator, and then the overall results are averaged as the discriminator result, as shown in Figure 6 , as shown in a. At the same time, we introduced the adaptive supervised PatchNCE (ASP) loss function, and based on the patch input, we evaluated the distance between positive and negative samples, so that similar samples are judged as the same sample area as much as possible, to ensure that the generated image is highly consistent with the real image in terms of pixels.

[0083] 3. Model training.

[0084] After building the generative model, we deployed the model on an RTX 4090 GPU for training, with a batch size of 2 and 50 epochs. We used the Adam optimizer, linear decay scheduler, and set the initial learning rate to 2×10 -4 The entire training process took 4 days.

[0085] 4. Evaluation of the model.

[0086] In order to evaluate the performance of the model, we used the structural similarity index (SSIM) and peak signal-to-noise ratio (PSNR) to evaluate the quality of the generated images. Assuming x and y are the generated image and the real image respectively, SSIM and PSNR are defined as follows:

[0087]

[0088] Where μ and σ represent the mean and variance of the image respectively, and σ xy is the covariance between image x and image y, c1 and c2 are constants.

[0089]

[0090] Where MAX is the maximum value of the image pixel value, m and n are the number of pixels of image x and image y.

[0091] In addition, to evaluate the model's ability to identify CD4 +To evaluate the performance of TFH1 cells, we invited experienced doctors to manually annotate 2,000 real mIHC images and generated images and divide them into two categories: + Images of TFH1 cells and those without CD4 + We then used classification metrics to evaluate the performance of the 2000 generated mIHC images. First, we calculated the confusion matrix, which included the following: True positives (TP, CD4 + TFH1 cells exist in both real and generated images), true negatives (TN, CD4 + TFH1 cells), false positives (FPs, CD4 + TFH1 cells but not present in the real image) and false negatives (FN, CD4 + TFH1 cells but not identified in the generated images). We then calculated the accuracy Recall Accuracy and F1 score

[0092] 5. Results display.

[0093] The model was trained on about 32,000 pairs of paired image patches and showed robust performance, achieving a structural similarity index (SSIM) of 0.69 and a peak signal-to-noise ratio (PSNR) of 24.83 on the test dataset (Methods, N=8,000). + In terms of the ability assessment of TFH1 cells, the results showed that the model was effective in recognizing CD4 + TFH1 cells showed a precision of 0.86 and a recall of 0.68 (Methods, Figure 6 , b). Figure 6 , c shows that the generated mIHC image can effectively detect CD4 + This suggests that our model can effectively generate cells that can accurately recognize CD4 + mIHC images of TFH1 cells.

[0094] The above-described embodiments are only preferred embodiments for fully illustrating the present invention, and the protection scope of the present invention is not limited thereto. Equivalent substitutions or changes made by those skilled in the art based on the present invention are within the protection scope of the present invention. The protection scope of the present invention shall be subject to the claims.

Claims

1. Application of Tfh cell subsets in evaluating the efficacy of R-CHOP treatment in DLBCL patients, characterized in that: The Tfh cell subset is defined as CD4 + CXCR5 + PD-1 - cell subpopulations.

2. The use according to claim 1, wherein the Tfh cell subset CD4 + CXCR5 + PD-1 - The higher the proportion in the total cells, the better the R-CHOP treatment efficacy and the longer the prognosis and survival of DLBCL patients.

3. Application of reagents for detecting Tfh cell subsets in the preparation of products for evaluating the efficacy of R-CHOP treatment in DLBCL patients.

4. The use according to claim 3, characterized in that: The reagents for detecting Tfh cell subpopulations include one or more immune reagents of CD3E, CD3D, CD3G, CD4, BCL6, CXCR5, EBI3, NMB, IFNG or PDCD1.

5. The use according to claim 3, characterized in that: The immunological reagent is used for immunofluorescence / immunohistochemistry antibody or immunocolloidal gold labeling.

6. The use according to claim 4, characterized in that: The immune reagent includes a colorimetric marker molecular probe that specifically recognizes CD4, CXCR5 and PD-1.

7. The use according to claim 6, characterized in that: The color developer-labeled molecular probes are CD4 molecular probe, CXCR5 molecular probe and PD-1 molecular probe respectively labeled with different fluorescence.

8. The use according to claims 3-7, characterized in that: The product also includes a dewaxing agent, a gradient ethanol solution, a blocking solution, a dye, a DAPI stain, and a TBST buffer.

9. The use according to claim 3 to 7, wherein the Tfh cell subset CD4 + CXCR5 + PD-1 - The higher the proportion in the total cells, the better the R-CHOP treatment efficacy and the longer the prognosis and survival of DLBCL patients.

10. A method for identifying Tfh cell subsets based on generative AI, characterized in that: The process includes the following steps: In order to achieve accurate generation of multiple immunofluorescence staining images mIHC, a conditional generative adversarial network is used as the baseline model, in which ResNet-6Blocks is used as the generator and a 5-layer PatchGAN is used as the discriminator. Each image is divided into patches and input into the discriminator, and then the overall results are averaged as the discriminator result; at the same time, an adaptive supervised PatchNCE loss function is introduced, and based on the patch input, the distance between positive and negative samples is evaluated, so that similar samples are judged as the same sample area as much as possible, to ensure that the generated image is highly consistent with the real image in terms of pixels.

11. The method according to claim 10, characterized in that The structural similarity index (SSIM) and peak signal-to-noise ratio (PSNR) are used to evaluate the quality of the generated image. Assuming that x and y are the generated image and the real image respectively, SSIM and PSNR are defined as follows: Where μ and σ represent the mean and variance of the image respectively, and σ xy is the covariance between image x and image y, c1, c2 are constants; Where MAX is the maximum value of the image pixel value, m and n are the number of pixels of image x and image y.

Images