Automatic analysis and risk assessment system and method for peripheral blood endothelial progenitor cells

Abstract

The invention relates to the technical field of biomedicine, in particular to an automatic analysis and risk assessment system and method for peripheral blood endothelial progenitor cells, and the method comprises the following steps: collecting a peripheral blood sample and clinical data of a subject; the method comprises the following steps: enriching CD34 + cells from a peripheral blood sample based on microfluidic-immunomagnetic bead enrichment, carrying out VEGFR2 and DAPI immunofluorescent staining on the enriched cells, and obtaining a multi-channel fluorescent digital image; analyzing the multi-channel fluorescent digital image, identifying and counting VEGFR < 2 + > and DAPI < + > double-positive cells to obtain an EPC counting result; inputting the EPC counting result and clinical data of the subject into a multi-modal AI risk assessment model to obtain a risk assessment result; the health management suggestion information is formulated based on the risk assessment result, so that the model can have better suitability with the target subject, and the accuracy of risk assessment is improved.

Description

Technical Field

[0001] This invention relates to the field of biomedical technology, specifically to an automated analysis and risk assessment system and method for peripheral blood endothelial progenitor cells. Background Technology

[0002] Vascular health is the cornerstone of the overall cardiovascular system, and its damage is the initiating factor in the development of cardiovascular and cerebrovascular diseases. Endothelial progenitor cells (EPCs) are key cells for vascular endothelial repair and regeneration. Their quantity and function in peripheral blood have been widely recognized as important biomarkers for assessing vascular endothelial health and predicting the risk of various cardiovascular and cerebrovascular diseases such as atherosclerosis, coronary heart disease, and stroke. However, current EPC enrichment procedures are cumbersome, counting accuracy is poor, and it is difficult to combine EPC biomarkers with clinical data to assess vascular risk, thus affecting the accuracy of risk assessment.

[0003] Therefore, we propose an automated analysis and risk assessment system and method for peripheral blood endothelial progenitor cells to address the above-mentioned problems. Summary of the Invention

[0004] The purpose of this invention is to provide an automated analysis and risk assessment system and method for peripheral blood endothelial progenitor cells to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an automated analysis and risk assessment system and method for peripheral blood endothelial progenitor cells, the method comprising the following steps: Collect peripheral blood samples and clinical data from the subjects; CD34+ cells were enriched from peripheral blood samples using microfluidic immunomagnetic bead enrichment. The enriched cells were then subjected to immunofluorescence staining with VEGFR2 and DAPI, and multi-channel fluorescence digital images were acquired. Multichannel fluorescence digital images were analyzed to identify and count double-positive cells of VEGFR2+ and DAPI+, and EPC counting results were obtained. The EPC count results and the examinee's clinical data are input into a multimodal AI risk assessment model to obtain risk assessment results. The multimodal AI risk assessment model includes multiple risk assessment models. Health management recommendations are then developed based on the risk assessment results.

[0006] Preferably, the steps of enriching CD34+ cells from peripheral blood samples based on microfluidic-immunomagnetic bead enrichment, performing VEGFR2 and DAPI immunofluorescence staining on the enriched cells, and acquiring multi-channel fluorescence digital images include: The enriched CD34+ cells were centrifuged and smeared to prepare monolayer cell samples. The monolayer cell sample was sequentially fixed and permeabilized; Cells were immunofluorescence stained using a primary antibody against VEGFR2 and the corresponding fluorescently labeled secondary antibody, and the cell nuclei were stained with DAPI dye. The stained samples were scanned using a fully automated fluorescence microscope, and digital images of the DAPI and VEGFR2 channels covering the entire sample area were acquired to obtain multi-channel fluorescence digital images.

[0007] Preferably, the step of analyzing multi-channel fluorescence digital images to identify and count double-positive cells of VEGFR2+ and DAPI+ to obtain EPC counting results includes: The acquired DAPI channel image and VEGFR2 channel image are preprocessed and registered. The DAPI channel image is processed based on a deep learning segmentation model to identify and segment all cell nucleus instances and obtain the geometric attributes and spatial location of each cell nucleus. For each identified cell nucleus, a region of interest was defined at the corresponding location in the VEGFR2 channel image, and VEGFR2 positive signals within that region were identified. Based on preset logical rules, each cell nucleus is judged to be positive twice. The rules require that there is a VEGFR2 positive signal in the region of interest corresponding to the cell nucleus that is spatially co-located with the cell nucleus and whose fluorescence intensity exceeds a preset threshold. All cells identified as double-positive were counted to obtain the endothelial progenitor cell count.

[0008] Preferably, the step of inputting the EPC counting results and the examinee's clinical data into the multimodal AI risk assessment model to obtain the risk assessment results includes: Obtain the EPC count results and clinical data of the subjects, and perform feature processing to obtain standardized feature vectors; Subjects are assigned to a predetermined population subgroup based on at least one clinical data point; Acquire historical clinical datasets from multiple historical subjects and construct a risk assessment model based on the population subgroup to which the subjects belong; Based on the risk assessment model, the standardized feature vector is processed to generate risk assessment results, which include at least one of vascular age, endothelial repair capacity index, and long-term cardiovascular and cerebrovascular disease risk.

[0009] Preferably, the step of classifying subjects into predetermined population subgroups based on at least one clinical data point includes: The clinical characteristics corresponding to at least one clinical data of the subject are obtained, and the clinical characteristics are input into the pre-stratification unit to obtain the identification of the population subgroup. The pre-stratification unit divides the subject into a predetermined population subgroup based on predefined rules, and different population subgroups are associated with different risk assessment models.

[0010] Preferably, the step of obtaining historical clinical datasets of multiple historical subjects and constructing a risk assessment model based on the population subgroup to which the subjects belong includes: Obtain historical clinical datasets from multiple historical subjects, where each historical clinical dataset contains multiple clinical characteristic variables; Each clinical feature variable is treated as a network node, and multiple network nodes are connected to form a feature network. Acquire clinical data of the target subjects and extract their corresponding target clinical feature variables. Using the target clinical feature variables as seeds, extract a set of clinical feature variables from the feature network as feature variable clusters. Based on the characteristic variable clusters, historical subject subgroups consistent with the population subgroups to which the subjects belonged were identified from the historical clinical dataset; The risk assessment model for the corresponding target subject is obtained by using the characteristic variable groups of historical subject subgroups as historical datasets.

[0011] Preferably, the step of developing health management recommendations based on risk assessment results includes: Obtain the risk assessment results for the examinee and match the risk assessment results with a pre-set medical knowledge base; Based on the matching results, the corresponding parameterized suggestion template is invoked, and health management suggestions are generated based on the examinee's clinical data.

[0012] An automated analysis and risk assessment system for peripheral blood endothelial progenitor cells, applied to any of the automated analysis and risk assessment methods for peripheral blood endothelial progenitor cells described above, including: The data collection module is used to collect peripheral blood samples and clinical data from the examinee. The imaging module is used to enrich CD34+ cells from peripheral blood samples based on microfluidic-immunomagnetic bead enrichment, perform immunofluorescence staining of the enriched cells with VEGFR2 and DAPI, and acquire multi-channel fluorescence digital images. The counting module is used to analyze multi-channel fluorescence digital images, identify and count double-positive cells of VEGFR2+ and DAPI+, and obtain EPC counting results; The risk assessment module is used to input EPC count results and the examinee's clinical data into a multimodal AI risk assessment model to obtain risk assessment results. The multimodal AI risk assessment model includes multiple risk assessment models. Based on the risk assessment results, health management recommendations are formulated.

[0013] Compared with the prior art, the beneficial effects of the present invention are: 1. By employing AI image recognition based on deep learning, the subjectivity and fatigue errors of manual microscope counting are avoided, and highly sensitive and accurate counting of EPCs, which are present in extremely low amounts in peripheral blood, is achieved, thereby improving the efficiency of cell detection and thus improving the efficiency of vascular risk assessment. 2. Cellular-level EPC biomarkers are deeply integrated with macroscopic clinical indicators through a multimodal AI model to achieve comprehensive analysis of multi-dimensional information. Feature clusters are dynamically matched based on real-time data of the subjects, and historical data corresponding to the target subject's subgroup are acquired to train the risk assessment model. This allows the model to have better fit with the target subjects, improving the accuracy of risk assessment. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a system structure block diagram of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Example Please see Figures 1 to 2 This invention provides a technical solution for an automated analysis and risk assessment system and method for peripheral blood endothelial progenitor cells: the automated analysis and risk assessment method for peripheral blood endothelial progenitor cells includes the following steps: S1: Collect peripheral blood samples and clinical data from the examinee; S2: CD34+ cells were enriched from peripheral blood samples based on microfluidic-immunomagnetic bead enrichment. The enriched cells were then subjected to immunofluorescence staining with VEGFR2 and DAPI, and multi-channel fluorescence digital images were acquired. The steps for enriching CD34+ cells from peripheral blood samples using microfluidic immunomagnetic bead enrichment, followed by immunofluorescence staining of the enriched cells with VEGFR2 and DAPI, and acquisition of multi-channel fluorescence digital images, include: centrifuging and smearing the enriched CD34+ cells to prepare a monolayer cell sample; fixing and permeabilizing the monolayer cell sample sequentially; performing immunofluorescence staining on the cells using a primary antibody against VEGFR2 and a corresponding fluorescently labeled secondary antibody, while simultaneously staining the cell nuclei with DAPI dye; and scanning the stained sample using a fully automated fluorescence microscope to acquire digital images of the DAPI and VEGFR2 channels covering the entire sample area to obtain multi-channel fluorescence digital images. Specifically, this method combines CD34 antibody-coated immunomagnetic beads with microfluidic chip technology. Anticoagulated peripheral blood samples from the subject are mixed with CD34+ immunomagnetic beads and incubated to allow CD34+ cells (including EPCs) to bind to the beads. The mixture is then injected into a microfluidic chip, which integrates microchannels with a specific structure and an externally applied magnetic field. Under the influence of the magnetic field, CD34+ cells bound to the beads are specifically captured in a specific area of ​​the chip, while unbound cells are eluted with buffer. Finally, the enriched CD34+ cells are collected and output by removing the magnetic field or altering the flow path. Microfluidic technology enables precise control of fluids and cells, improving the recovery rate and enrichment purity of CD34+ cells, while requiring less sample volume and reagents.

[0018] Enriched CD34+ cell suspension was collected from the outlet of the microfluidic chip and quantitatively added to the sample chamber of a cell centrifuge smear machine. The cells were then spread monolayer evenly onto a clean glass slide or a dedicated cytology slide using centrifugation (e.g., 500-800 rpm for 5 minutes). After smearing, the slide was briefly air-dried at room temperature to allow cell adhesion, preventing cell overlap and improving the accuracy of subsequent image recognition. An appropriate amount of cell fixative (e.g., 4% paraformaldehyde phosphate buffer) was added to the smear area and fixed at room temperature for 10-15 minutes to fix cell morphology and internal antigen structures, preventing damage or loss during subsequent staining steps. The slide was then gently rinsed three times with phosphate-buffered saline (PBS) to thoroughly remove any residual fixative. Add a cell membrane permeability agent (such as 0.1%-0.5% Triton X-100 in PBS solution) and incubate at room temperature for 5-10 minutes. This creates micropores in the cell membrane, allowing subsequent macromolecular antibodies to enter the cell and bind to the target antigen. Gently rinse the slide three times with PBS to remove the permeability agent and increase cell membrane permeability. Add a PBS blocking solution containing 1%-5% bovine serum albumin (BSA) to the smear area and incubate at room temperature for 30 minutes to block non-specific binding sites, reduce background staining by subsequent antibodies, and improve the signal-to-noise ratio. Discard the blocking solution and directly add anti-VEGFR2 primary antibody (usually a mouse or rabbit monoclonal antibody) diluted with the blocking solution. Place the slide in a humidified chamber and incubate at room temperature or 4°C for 1-2 hours. Wash thoroughly three times with PBS-Tween buffer for 5 minutes each time to completely remove unbound primary antibody. Add a fluorescently labeled secondary antibody diluted with the blocking solution and targeting the primary antibody species (e.g., Alexa Fluor for mouse primary antibody). 488-labeled goat anti-mouse IgG was used. Simultaneously, DAPI dye was added to counterstain the cell nuclei. The slide was incubated in a humidified chamber at room temperature in the dark for 45-60 minutes, followed by three thorough washes with PBS-Tween buffer in the dark to remove unbound secondary antibody and DAPI. A drop of anti-fluorescence quenching mounting medium was added to the smear area, and a coverslip was placed on top, avoiding air bubbles. The edges of the coverslip were sealed with nail polish or a special mounting adhesive. DAPI dye was then added to stain the cell nuclei. The prepared slide was placed on the stage of an automated motorized fluorescence microscope, and the scanning program was preset in the control software. For the objective lens, a 20x or 40x objective lens was selected to achieve a balance between throughput and resolution. The fluorescence channels included the DAPI channel and the FITC / GFP channel. The excitation / emission wavelengths (e.g., ~358 nm / ~461 nm) were set for the DAPI channel to collect the blue fluorescence of the cell nuclei. Set the excitation / emission wavelengths (e.g., ~488 nm / ~525 nm) for the FITC / GFP channels to collect the green fluorescence corresponding to VEGFR2. Define the cell smear area on the slide as the scanning range.The panoramic scanning function seamlessly stitches images from multiple fields of view into a single high-resolution digital image. The microscope automatically completes the entire process of focusing, filter switching, exposure, platform movement, and image acquisition and storage. Ultimately, two well-aligned high-resolution digital images of the same area are obtained: a DAPI channel image (cell nucleus) and a VEGFR2 channel image (membrane / plasma signal). These multi-channel fluorescence digital images are then directly transmitted to the subsequent AI image recognition and counting module.

[0019] S3: Analyze the multi-channel fluorescence digital images to identify and count double-positive cells of VEGFR2+ and DAPI+, and obtain the EPC counting results; The steps for analyzing multi-channel fluorescence digital images to identify and count double-positive cells of VEGFR2+ and DAPI+, and obtaining EPC counting results, include: preprocessing and registering the acquired DAPI channel images and VEGFR2 channel images; processing the DAPI channel images based on a deep learning segmentation model to identify and segment all cell nucleus instances, and obtaining the geometric attributes and spatial location of each cell nucleus; defining a region of interest (ROI) at the corresponding location in the VEGFR2 channel image for each identified cell nucleus, and identifying VEGFR2 positive signals within that region; determining double positivity for each cell nucleus based on preset logical rules, wherein the rules require that a VEGFR2 positive signal exists within the ROI corresponding to the cell nucleus, which is spatially co-located with the cell nucleus and has a fluorescence intensity exceeding a preset threshold; and counting all cells determined to be double-positive to obtain the endothelial progenitor cell (EPC) counting results. It should be noted that the region of interest (ROI) refers to a circular or rectangular region defined by a radius calculated based on the cell nucleus size, centered on the cell nucleus centroid; the deep learning segmentation model is either U-Net or Mask R-CNN architecture; the logical rules for double-positive determination include: the average fluorescence intensity of the VEGFR2 positive signal region exceeds the positive threshold statistically derived from negative control samples, and the signal region is spatially enclosed or adjacent to the cell nucleus; after identifying cell nucleus instances, a filtering step based on cell nucleus morphology and area is included to exclude interference from cell debris or overlapping nuclei; combining deep learning instance segmentation with logical judgment based on biological knowledge improves the high specificity and sensitivity of rare EPC identification under high background and cellular heterogeneity. Specifically, panoramic images of the DAPI channel (nuclear signal) and VEGFR2 channel (membrane / plasma signal) are acquired from an automated microscope system. Gaussian filtering or median filtering algorithms are applied to both channels to reduce noise interference. Simultaneously, non-uniformity correction of background fluorescence intensity is performed to improve the signal-to-noise ratio. Since the two channels may have slight pixel-level offsets due to optical paths, image registration algorithms (such as feature point-based or phase-correlation-based methods) are used to precisely align the two channels, ensuring the accuracy of subsequent co-localization analysis. The preprocessed DAPI channel image is input into a pre-trained deep learning segmentation model (e.g., U-Net or Mask R-CNN). The model outputs a binary mask image, where each identified nucleus is an independent, precisely labeled connected region. The system performs connected component analysis on the mask image, assigning a unique ID to each individual nucleus and calculating its geometric properties, such as centroid coordinates, area, and perimeter. For each nucleus identified in the previous step, a predetermined radius (e.g., 1.5-2 times the equivalent radius of the nucleus) is extended outward from its centroid to define a "region of interest" (ROI) on the VEGFR2 channel image. Within this ROI, an adaptive thresholding algorithm or another lightweight model trained for cytoplasmic segmentation is used to identify continuous regions where the VEGFR2 fluorescence signal is higher than the background. For each identified VEGFR2 signal region, its average fluorescence intensity, total area, shape factor, and other features are calculated. This process is repeated for each identified nucleus (DAPI+ object), and the existence of a valid VEGFR2 signal region within its corresponding ROI is checked. A cell is considered a VEGFR2+ / DAPI+ double-positive EPC if it simultaneously meets the following conditions: the cell must be an intact, morphologically normal D... API-positive objects (non-fragmented, non-overlapping nuclei) must have an average fluorescence intensity exceeding a preset, validated positive threshold for the VEGFR2 signal region identified within their corresponding ROI. Furthermore, the VEGFR2 signal region must have sufficient spatial overlap or close proximity to the cell nucleus (e.g., the centroid of the VEGFR2 signal is located within a ring-shaped region surrounding the cell nucleus, or its area is proportionally proportional to the area of ​​the cell nucleus). The absolute count of all cells meeting these double-positive criteria is then calculated. Based on the known scan area and sample volume, the absolute count is converted to the number of EPCs per milliliter of peripheral blood, automatically generating a structured count report including: total EPC count, cell density, analyzed image area, and confidence index. This result is then directly output to the downstream multimodal AI risk assessment module.

[0020] S4: Input the EPC count results and the examinee's clinical data into the multimodal AI risk assessment model to obtain the risk assessment results. The multimodal AI risk assessment model includes multiple risk assessment models. Based on the risk assessment results, develop health management recommendations. The steps for inputting EPC count results and subject clinical data into a multimodal AI risk assessment model to obtain risk assessment results include: acquiring the subject's EPC count results and clinical data, and performing feature processing to obtain a standardized feature vector; classifying the subject into a predetermined population subgroup based on at least one clinical data point; acquiring historical clinical datasets of multiple historical subjects, and constructing a risk assessment model based on the population subgroup to which the subject belongs; processing the standardized feature vector based on the risk assessment model to generate a risk assessment result, wherein the risk assessment result includes at least one of vascular age, endothelial repair capacity index, and long-term cardiovascular and cerebrovascular disease risk.

[0021] The steps of acquiring historical clinical datasets of multiple historical subjects and constructing a risk assessment model based on the subgroup of the population to which the subject belongs include: acquiring historical clinical datasets of multiple historical subjects, wherein the historical clinical datasets contain multiple clinical feature variables; treating each clinical feature variable as a network node and connecting multiple network nodes to obtain a feature network; acquiring the clinical data of the target subject and extracting its corresponding target clinical feature variables, using the target clinical feature variables as seeds, extracting a set of clinical feature variables from the feature network as a feature variable cluster; dividing the historical subject subgroup from the historical clinical dataset according to the feature variable cluster, which is consistent with the subgroup of the population to which the subject belongs; and using the feature variable cluster of the historical subgroup as the historical dataset to train the risk assessment model for the corresponding target subject. It should be noted that in the feature network, seed nodes corresponding to the target clinical feature variables are located; neighbor nodes with a correlation strength exceeding a preset threshold with the seed nodes are found and aggregated to form feature variable clusters; the specific content of dividing historical subject subgroups is as follows: calculating the similarity between the feature profile of each subject in the historical dataset on the feature variable cluster and the feature profile of the target subject; the top K historical subjects with the highest similarity, or all historical subjects with similarity exceeding the threshold, are divided into a dynamic similar population subgroup. The risk assessment model is a neural network, random forest, or gradient boosting decision tree model, and is no longer... Instead of "one population, one model," the approach leans towards "one person, one model" (trained by matching the most similar data and features). The subgrouping of the population is not based on fixed medical labels (such as "diabetes"), but on data similarity as shown in the feature network. Since each model is trained on a tightly interconnected "feature variable cluster," the feature cluster is dynamically matched based on the real-time data of the examinee. The risk assessment model is trained by acquiring historical data corresponding to the population subgroup of the target examinee, which makes the model more adaptable to the target examinee and improves the accuracy of risk assessment.

[0022] The steps of classifying subjects into predetermined population subgroups based on at least one clinical data point include: obtaining clinical characteristics corresponding to at least one clinical data point of the subject, inputting the clinical characteristics into a pre-stratification unit to obtain the identifier of the population subgroup, wherein the pre-stratification unit classifies subjects into a predetermined population subgroup based on predefined rules, and different population subgroups are associated with different risk assessment models. Specifically, the key clinical features may include at least one of the following: diabetes diagnosis status or glycated hemoglobin value; age or age segment; baseline cardiovascular disease risk score or confirmed cardiovascular disease status; chronic kidney disease diagnosis status or estimated glomerular filtration rate. For example, if the subject's glycated hemoglobin value is greater than or equal to 6.5%, or has a history of diabetes diagnosis, they are classified into the diabetes subgroup; if they do not meet the above conditions, and the subject's age is greater than or equal to 65 years, they are classified into the advanced age risk subgroup. The pre-stratification unit is a pre-trained machine learning classification model that is trained using key clinical features of historical populations and their corresponding optimal subgroup labels. The predetermined population subgroups may include at least: a diabetes subgroup, an advanced age risk subgroup, and a general adult subgroup. Based on the subgroup identifier output by the pre-stratification unit, the corresponding dedicated risk assessment model is called from the model library to assess the subject, wherein the model library stores risk assessment models corresponding to each population subgroup.

[0023] The steps for developing health management recommendations based on risk assessment results include: obtaining the risk assessment results for the examinee; matching the risk assessment results with a pre-set medical knowledge base, wherein the medical knowledge base stores rules that map specific risk conditions to at least one health management recommendation item; calling the corresponding parameterized recommendation template according to the matching results; and generating health management recommendations based on the examinee's clinical data, wherein the content of the health management recommendations includes at least one of the following: lifestyle intervention, dietary recommendations, exercise prescriptions, medical consultation recommendations, and follow-up plans. Specifically, multiple personalized health management recommendations can be prioritized based on the severity or urgency of the risk assessment results. For example, the structured data received from the AI ​​risk assessment module shows a vascular age of 52 years, an endothelial repair capacity index of 65 (score 0-100), a 10-year CVD risk of 8.5%, and risk markers such as elevated LDL-C, low EPC count, and stage 1 hypertension. Analyzing these results identifies abnormal risk dimensions and key abnormal clinical indicators. A structured medical knowledge base stores intervention strategies based on the latest clinical guidelines and evidence-based medicine. The core structure of the knowledge base is a "condition-action" rule pair, which matches the parsed abnormal dimensions with the "conditions" in the knowledge base. For example, the knowledge base rules can be: if vascular age > calendar age + 5 years and LDL-C > target value, then trigger suggestions A01 and A02; if the endothelial repair capacity index < 75, then trigger suggestion IDs B01 and B02; if the 10-year CVD risk ≥ 7.5% and smoking history = yes, then trigger suggestion ID C01; if systolic blood pressure is consistently ≥ 140 mmHg, then trigger suggestion IDs D01 and D02. Each suggestion ID corresponds to one or more parameterized text templates. The examinee's specific data (such as actual values ​​and target values) are entered into the template to generate natural language suggestions. For example, the template for suggestion ID A01 is: "Your LDL-C is [current LDL-C value] mmol / L, which is higher than the recommended target value [target value] mmol / L. It is recommended to take dietary control measures (such as reducing saturated fat intake) and consider starting statin therapy under the guidance of a doctor." All triggered suggestions are intelligently sorted according to the severity of the risk. For example, suggestions related to "diagnosed CVD subgroup" and "high risk" (such as medication and immediate medical attention) will be ranked before lifestyle suggestions (such as increasing exercise). The medical knowledge base is a rule database that stores multiple "risk condition-suggestion template" mapping relationships. Through the medical knowledge base, corresponding health management suggestions can be given based on the examinee's clinical data and risk assessment results, thereby improving the suitability of health management suggestions to the examinee.

[0024] By employing AI image recognition based on deep learning, the subjectivity and fatigue errors of manual microscope counting are avoided, and highly sensitive and accurate counting of EPCs, which are present in extremely low amounts in peripheral blood, is achieved, thereby improving the efficiency of cell detection and thus improving the efficiency of vascular risk assessment. By deeply integrating cellular-level EPC biomarkers with macroscopic clinical indicators through a multimodal AI model, a comprehensive assessment of multidimensional information can be achieved. A dedicated risk assessment model library based on different population subgroups (such as diabetic patients and the elderly) is constructed, and a pre-stratification mechanism is used for intelligent routing to match the most suitable model to each subject, thereby improving the accuracy of risk assessment.

[0025] An automated analysis and risk assessment system for peripheral blood endothelial progenitor cells, applied to any of the automated analysis and risk assessment methods for peripheral blood endothelial progenitor cells described above, including: The data collection module is used to collect peripheral blood samples and clinical data from the examinee. The imaging module is used to enrich CD34+ cells from peripheral blood samples based on microfluidic-immunomagnetic bead enrichment, perform immunofluorescence staining of the enriched cells with VEGFR2 and DAPI, and acquire multi-channel fluorescence digital images. The counting module is used to analyze multi-channel fluorescence digital images, identify and count double-positive cells of VEGFR2+ and DAPI+, and obtain EPC counting results; The risk assessment module is used to input EPC count results and the examinee's clinical data into a multimodal AI risk assessment model to obtain risk assessment results. The multimodal AI risk assessment model includes multiple risk assessment models. Based on the risk assessment results, health management recommendations are formulated.

[0026] This invention first utilizes CD34+ immunomagnetic beads combined with microfluidic technology to efficiently and highly enrich target cells from peripheral blood. Then, cell labeling and imaging are performed using automated immunofluorescence staining (VEGFR2 / DAPI) and fully automated fluorescence microscopy. Next, an artificial intelligence image recognition algorithm is used to automatically analyze the scanned images and accurately count double-positive (VEGFR2+ / DAPI+) EPCs. Finally, a multimodal AI risk assessment model is constructed, combining the EPC counting results with clinical data such as the subject's blood lipids, blood pressure, history of diabetes, and smoking history to automatically assess their vascular age, vascular endothelial self-repair capacity, and long-term cardiovascular and cerebrovascular disease risk, and generate personalized health management recommendations.

[0027] This system integrates sample pretreatment, cell separation, staining, imaging, counting, and risk assessment into a coherent workflow, significantly reducing manual operations and improving efficiency. The combination of microfluidic technology and immunomagnetic beads enhances cell capture efficiency and purity; AI image recognition avoids subjective errors, making it particularly suitable for the precise counting of rare cells. A multimodal AI model facilitates the transformation from biomarkers to personalized health risks, providing strong data support for early intervention and health management. The system automates and automates the entire process from "a tube of blood" to "a comprehensive risk assessment report," providing a precise and convenient assessment tool for vascular health. Microfluidic-immunomagnetic bead technology improves cell recovery rate and enrichment purity. The deep learning-based AI image recognition algorithm reduces the subjectivity and fatigue errors associated with manual microscopic counting, especially for EPCs (epidermal cells) present in extremely low concentrations in peripheral blood, enabling fatigue-free, highly consistent, and highly sensitive accurate identification and counting. The multimodal AI model achieves multi-dimensional information fusion, resulting in more comprehensive assessment results.

[0028] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0029] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. Method for automated analysis and risk assessment of peripheral blood endothelial progenitor cells, characterized in that, The method comprises the following steps: Collecting a peripheral blood sample of a subject and clinical data; Enriching CD34+ cells from the peripheral blood sample based on microfluidic-immunomagnetic bead enrichment, performing immunofluorescence staining of VEGFR2 and DAPI on the enriched cells, and acquiring a multi-channel fluorescence digital image; Analyzing the multi-channel fluorescence digital image, identifying and counting double-positive cells of VEGFR2+ and DAPI+, and obtaining an EPC count result; Inputting the EPC count result and the clinical data of the subject into a multi-modal AI risk assessment model to obtain a risk assessment result, wherein the multi-modal AI risk assessment model comprises a plurality of risk assessment models; and formulating health management recommendation information based on the risk assessment result.

2. The method of automated analysis and risk assessment of peripheral blood endothelial progenitor cells according to claim 1, characterized in that: The step of enriching CD34+ cells from the peripheral blood sample based on microfluidic-immunomagnetic bead enrichment, performing immunofluorescence staining of VEGFR2 and DAPI on the enriched cells, and acquiring a multi-channel fluorescence digital image comprises: Performing centrifugal smearing on the enriched CD34+ cells to prepare a single-layer cell sample; Performing fixation and permeabilization treatment on the single-layer cell sample in sequence; Performing immunofluorescence staining on the cells using a primary antibody for VEGFR2 and a corresponding fluorescently labeled secondary antibody, and simultaneously staining the cell nuclei using DAPI dye; Performing full-automatic fluorescence microscope scanning on the stained sample to acquire a multi-channel fluorescence digital image comprising a DAPI channel and a VEGFR2 channel digital image covering the entire sample area.

3. The method of automated analysis and risk assessment of peripheral blood endothelial progenitor cells according to claim 2, characterized in that: The step of analyzing the multi-channel fluorescence digital image, identifying and counting double-positive cells of VEGFR2+ and DAPI+, and obtaining an EPC count result comprises: Performing preprocessing and registration alignment on the acquired DAPI channel image and VEGFR2 channel image; Processing the DAPI channel image based on a deep learning segmentation model to identify and segment all cell nucleus instances, and acquiring the geometric properties and spatial positions of each cell nucleus; For each identified cell nucleus, defining a region of interest at the corresponding position of the VEGFR2 channel image, and identifying VEGFR2 positive signals in the region; Performing double-positive judgment on each cell nucleus based on a preset logical rule, wherein the rule requires that there is a VEGFR2 positive signal that is spatially co-located with the cell nucleus and has a fluorescence intensity exceeding a preset threshold in the region of interest corresponding to the cell nucleus; Counting all cells determined to be double-positive to obtain an endothelial progenitor cell count result.

4. The method of automated analysis and risk assessment of peripheral blood endothelial progenitor cells according to claim 1, characterized in that: The step of inputting the EPC count result and the clinical data of the subject into a multi-modal AI risk assessment model to obtain a risk assessment result comprises: Acquiring the EPC count result and the clinical data of the subject, and performing feature processing to obtain a standardized feature vector; Dividing the subject into a predetermined population subgroup based on at least one clinical data; Acquiring a historical clinical data set of a plurality of historical subjects, and constructing a risk assessment model based on the population subgroup to which the subject belongs; and The standardized feature vector is processed based on the risk assessment model to generate a risk assessment result, the risk assessment result including at least one of blood vessel age, endothelial repair capacity index, and long-term cardiovascular and cerebrovascular disease risk.

5. The method of automated analysis and risk assessment of peripheral blood endothelial progenitor cells according to claim 4, characterized in that: The step of dividing the subject into a predetermined population subgroup based on at least one clinical data includes: Obtaining the clinical features corresponding to the at least one clinical data of the subject, inputting the clinical features into the pre-hierarchical classifier to obtain the identification of the population subgroup, wherein the pre-hierarchical classifier is based on a pre-defined rule to divide the subject into a predetermined population subgroup, and different population subgroups are associated with different risk assessment models.

6. The method of automated analysis and risk assessment of peripheral blood endothelial progenitor cells according to claim 4, characterized in that: The step of obtaining a historical clinical data set of a plurality of historical subjects and constructing a risk assessment model based on the population subgroup to which the subject belongs includes: Obtaining a historical clinical data set of a plurality of historical subjects, wherein the historical clinical data set contains a plurality of clinical feature variables; Connecting a plurality of network nodes to obtain a feature network, each clinical feature variable being a network node; Obtaining the clinical data of the target subject and extracting the target clinical feature variables corresponding thereto, taking the target clinical feature variables as seeds, and extracting a clinical feature variable set from the feature network as a feature variable group; According to the feature variable group, the historical subject subgroup consistent with the population subgroup to which the subject belongs is divided from the historical clinical data set; Training the feature variable group of the historical subject subgroup as a historical data set to obtain a risk assessment model corresponding to the target subject.

7. The method of automated analysis and risk assessment of peripheral blood endothelial progenitor cells according to claim 1, characterized in that: The step of formulating health management recommendation information based on the risk assessment result includes: Obtaining the risk assessment result of the subject, matching the risk assessment result with a pre-installed medical knowledge base; According to the matching result, calling the corresponding parameterized recommendation template, and generating health management recommendations based on the clinical data of the subject.

8. A system for automated analysis and risk assessment of peripheral blood endothelial progenitor cells, applying the method for automated analysis and risk assessment of peripheral blood endothelial progenitor cells according to any one of claims 1 to 7, characterized in that It includes: The acquisition module is used for acquiring peripheral blood samples and clinical data of the subject; The imaging module is used for enriching CD34+ cells from the peripheral blood sample based on microfluidic-immunomagnetic bead enrichment, performing VEGFR2 and DAPI immunofluorescence staining on the enriched cells, and obtaining a multi-channel fluorescence digital image; The counting module is used for analyzing the multi-channel fluorescence digital image, identifying and counting VEGFR2+ and DAPI+ double-positive cells, and obtaining an EPC count result; The risk assessment module is used for inputting the EPC count result and the clinical data of the subject into a multi-modal AI risk assessment model to obtain a risk assessment result, wherein the multi-modal AI risk assessment model includes a plurality of risk assessment models; and formulating health management recommendation information based on the risk assessment result.

Images