Method for evaluating immune function in peripheral blood based on tertiary structure of immune

Abstract

This invention discloses a peripheral blood immune function scoring method based on the three-tiered immune structure, belonging to the field of immune function detection technology. This invention addresses the problem that existing scoring methods rely solely on simple index aggregation or mean calculation, failing to consider the weight differences between different immune levels and the intrinsic correlations between indicators, resulting in low assessment accuracy. By optimizing the weight allocation process of the gradient-enhanced machine learning algorithm, introducing population classification, multi-dimensional labels, and a basic weight library retrieval mechanism, combined with validation strategies, the invention improves the targeting and accuracy of weight allocation, addressing its limitations in adapting to specific populations and lacking dynamic expansion. Furthermore, by constructing a dynamic prediction model through linear regression, it establishes a correlation model between the score and disease incidence risk, clarifies the early warning and traceability process, and achieves integrated immune function monitoring, prediction, early warning, and traceability, enhancing clinical guidance value and providing support for precise dynamic management and clinical intervention.

Description

Technical Field

[0001] This invention relates to the field of immune function detection technology, specifically a peripheral blood immune function scoring method based on the three-tiered immune structure. Background Technology

[0002] The immune system is the core of the body's immune defense and maintenance of homeostasis. It has a sophisticated three-tiered structure, consisting of physical and chemical barriers (the first line of defense), innate immune response (the second line of defense), and adaptive immune response (the third line of defense). These three work together to form a complete immune defense network. Peripheral blood is an important carrier for communication between the immune system and all tissues and organs in the body. Changes in its immune-related indicators can directly reflect the overall immune function status of the body. Therefore, immune function assessment methods based on peripheral blood testing have important value in clinical diagnosis, health management, and disease prognosis.

[0003] Currently, existing methods for assessing peripheral blood immune function have many limitations. For example, existing scoring methods only use simple index superposition or mean calculation, without considering the weight differences of different immune levels and the intrinsic correlation between indicators, resulting in low assessment accuracy.

[0004] Therefore, to meet current needs, a peripheral blood immune function scoring method based on the three-tiered immune structure is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide a peripheral blood immune function scoring method based on the three-tiered immune structure. By optimizing the weight allocation process of the gradient-enhanced machine learning algorithm, introducing population classification, multi-dimensional labels, and basic weight library retrieval mechanisms, and combining validation strategies, the method improves the targeting and accuracy of weight allocation, addressing the pain points of its inability to adapt to specific populations and lack of dynamic expansion. Furthermore, by constructing a dynamic prediction model through linear regression, a correlation model between the score and disease incidence risk is established, clarifying the early warning and traceability process, and realizing the integration of immune function monitoring, prediction, early warning, and traceability. This enhances the clinical guidance value, provides support for precise dynamic management and clinical intervention, and solves the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A peripheral blood immune function scoring method based on the three-tiered immune structure includes the following steps:

[0008] Step 1: Construct a peripheral blood immune assessment index system based on the three-tiered immune structure, which includes the first line of defense, the second line of defense, and the third line of defense, with specific peripheral blood assessment indicators set for each line of defense;

[0009] Step 2: Collect peripheral venous blood from the subjects and use real-time quantitative PCR to quantitatively detect each assessment indicator and obtain the actual detection value of each indicator;

[0010] Step 3: Using the gradient enhancement machine learning algorithm, combined with clinical sample data, weights are assigned to each indicator in the three-level indicator system, the weight coefficients of each indicator and each immune defense are determined, and the weight coefficients are optimized for different populations to form a personalized weight scheme.

[0011] Step 4: Based on the standardized scores of each indicator and their corresponding weight coefficients, calculate the comprehensive immune function score using the weighted summation method, and calculate the sub-scores for the three lines of immune defense;

[0012] Step 5: Based on the comprehensive scoring results, peripheral blood immune function is divided into multiple levels to provide guidance for clinical diagnosis and health management.

[0013] Furthermore, it also includes the following steps:

[0014] Step 6: Collect peripheral blood from the subjects periodically, repeat steps 2-5, and obtain the comprehensive immune function score and sub-scores at different time points;

[0015] Construct dynamic monitoring curves to analyze the changing trends of immune function;

[0016] Based on clinical symptoms, personalized suggestions for immune regulation or treatment adjustments are provided to the subjects.

[0017] Further, step 3: Using a gradient-enhancing machine learning algorithm, combined with clinical sample data, weights are allocated to each indicator in the three-level indicator system, including the following steps:

[0018] Clinical sample data were collected, including peripheral blood test values ​​of various indicators and clinical immune function assessment results of the subjects. The clinical immune function assessment results were determined by clinicians based on the subjects' infection frequency, disease recovery speed and immune-related complications, and were divided into four levels: normal, mildly deficient, moderately deficient and severely deficient.

[0019] The subjects were divided into four basic categories: healthy people, immunodeficient people, cancer people, and people with autoimmune diseases. Multidimensional characteristic information of the subjects was collected, including age, gender, history of underlying diseases and medication, to form a multidimensional labeling system.

[0020] The classification information and feature labels are integrated to form training datasets and validation datasets;

[0021] Construct the input feature matrix X and output label vector Y of the XGBoost model, and set the hyperparameters of the XGBoost model;

[0022] In this system, each row of the input feature matrix X corresponds to a subject sample, each column corresponds to a specific detection value of a peripheral blood assessment indicator, and the output label vector Y corresponds to the clinical immune function assessment level. The learning rate eta is 0.1, the maximum tree depth max_depth is 5, the number of iterations n_estimators is 100, and the subsample ratio subsample is 0.8.

[0023] The training dataset was input into the XGBoost model, and the clinical immune function assessment level was used as the supervision label for training.

[0024] During training, the model calculates the importance score of each indicator based on the frequency of its occurrence in the decision tree split nodes and the reduction in impurity.

[0025] The weighted sum of the importance scores of this indicator in all trees is divided by the total number of splits to obtain the standardized weight coefficient.

[0026] For different population groups, the weight coefficients are optimized separately, and a transfer learning strategy is adopted. Using the basic weight vector of healthy people as the initial benchmark, the XGBoost model is fine-tuned for immunodeficient people, cancer patients and autoimmune disease patients using training samples of the corresponding populations, and the weights of each indicator are recalculated.

[0027] The accuracy of the weighted coefficient prediction was evaluated using a validation dataset. The validation results were compared with the clinical immune function assessment results, and the Spearman correlation coefficient between the weighted score and the clinical assessment results was calculated.

[0028] If the correlation coefficient is lower than the preset threshold range, adjust the hyperparameters of XGBoost and retrain until the requirements are met.

[0029] If the weighted score is consistent with the clinical assessment results, it indicates that the modulation direction is correct, thus forming a personalized weight allocation scheme.

[0030] Further, step 3: Optimize the weighting coefficients for different groups of people, including the following steps:

[0031] The generated population-specific weighting schemes are validated at multiple levels. The prediction accuracy of each population weighting scheme within the population is tested using reserved validation set data.

[0032] Cross-population validation was conducted to verify the specificity of the weighting scheme;

[0033] Newly enrolled subjects were selected and assessed using both traditional and this method. The results of the two assessments were compared to verify its clinical applicability and accuracy.

[0034] If it is found that the characteristics of newly enrolled subjects cannot be classified into any existing population category, and their immune indicators show unique regularity, the new population identification process will be automatically triggered.

[0035] Through cluster analysis and expert evaluation, we can confirm whether there are new population subtypes. If they are confirmed, we can include the population in the model training scope, establish an independent basic weight vector for it, and form a complete weight scheme library that is dynamically modulated according to individual characteristics.

[0036] Further, step 6: Construct a dynamic monitoring curve to analyze the changing trends of immune function, including the following steps:

[0037] We collected immune function-related scores and clinical baseline information of the subjects at different time points, and used a linear regression algorithm to fit the monitoring curves to establish a dynamic prediction model.

[0038] Input the latest scoring data to predict the trend and fluctuation range of immune function changes within a specified future time period;

[0039] Collect clinical sample data, including disease incidence corresponding to different immune function scores;

[0040] By using logistic regression analysis, a correlation model between the scoring results and the risk of developing immune-related diseases is established, and the disease incidence probability for the corresponding scoring interval is output.

[0041] By combining the predicted trend, the probability of disease incidence, and the warning threshold, an early warning will be automatically triggered when the predicted trend value is lower than or higher than the preset threshold.

[0042] Furthermore, combining predicted trends, disease risk probabilities, and early warning thresholds, the following steps are included:

[0043] Set early warning thresholds, including thresholds for predicted trend values ​​of immune function and thresholds for the probability of disease incidence;

[0044] Set up corresponding early warning methods according to the graded early warning standards;

[0045] The system automatically records the trigger time, warning level, core trigger indicators, target audience, and intervention feedback results for each warning, creating a warning traceability archive.

[0046] Further, step 1: Constructing a peripheral blood immune assessment index system based on the three-tiered immune structure, including the following steps:

[0047] To construct the first line of defense indicators, soluble indicators in peripheral blood that are directly related to mucosal barrier function were selected, including secretory immunoglobulin A and mucosa-associated epithelial cytokines.

[0048] To construct indicators for the second line of defense, we selected innate immune cells and related cytokines in peripheral blood, including the percentage and activity of natural killer cells, macrophage markers, and innate immune cytokines.

[0049] To construct the third line of defense indicators, adaptive immune cell subsets and related molecules in peripheral blood were selected, including the percentages of CD3+ T cells, CD4+ T cells, and CD8+ T cells, the CD4+ / CD8+ ratio, the percentage of B cells, immunoglobulins, and the T cell receptor diversity index.

[0050] Furthermore, step 2 also includes the following steps:

[0051] By collecting peripheral blood samples from healthy subjects of different ages and genders, specific peripheral blood assessment indicators were detected.

[0052] The normal reference range for each indicator is calculated, and the actual test values ​​of each indicator are converted into standardized scores using standardized formulas to eliminate the dimensional differences between different indicators and form a health reference database.

[0053] Furthermore, the process of periodically obtaining comprehensive and sub-item scores of immune function at different time points also includes the following steps:

[0054] Obtain the sub-score data of the three lines of immune defense of the subject at adjacent test nodes, and extract the timestamp of the previous test and the timestamp of the current test;

[0055] Then, based on the previous detection timestamp and the current detection timestamp, the actual time span of adjacent test nodes is calculated, and the signed sub-item score deviation of the corresponding immune defense in adjacent test nodes is extracted based on the sub-item score data.

[0056] Based on the actual time span and the preset time decay rate constant, a stress decay factor is calculated to characterize the degree of decay of the effects of acute physiological stress.

[0057] Based on the marked sub-item scoring deviation, the actual time span, the preset physiological buffer time constant, the basal metabolic compensation rate of the specific defense line, and the stress attenuation factor, the initial comprehensive immune function score obtained in this test is dynamically compensated in two directions. This filters out spurious fluctuations caused by non-substantial immune function decline or transient pseudo-hyperactivity, and calculates the calibrated comprehensive score. The corresponding calculation formula is as follows:

[0058]

[0059] in, This indicates the overall score after calibration; This represents the initial, uncalibrated comprehensive immune function score calculated using the weighted summation method. This represents the preset fluctuation adjustment constant; This indicates the hierarchical number of the immune defense line, specifically taking the value of a positive integer 1, 2, or 3; Indicates the first The first detection corresponds to the first Sub-score data for the immune defense line; Indicates the first The first detection corresponds to the first Sub-score data for the immune defense line; Corresponding to the marked itemized scoring deviation; Corresponding to the actual time span; Corresponding to the preset physiological buffer time constant; Corresponding to the The basal metabolic compensation rate of a specific line of defense of the immune system represents the expected basal fluctuation of normal scores per unit of day. Corresponding to the preset time decay rate constant; Corresponding to the stress attenuation factor;

[0060] Finally, based on the calculated calibrated comprehensive score, the dynamic monitoring curve of the subject's immune function is reset and updated; when the calibrated comprehensive score is determined to cross the preset risk defense threshold, the targeted immune regulation intervention control process for abnormal immune defense is automatically triggered, and a multi-dimensional immune traceability early warning report is output to the corresponding clinician port.

[0061] Furthermore, after determining the weight coefficients for each indicator and each immune defense line, an adaptive balancing mechanism for cross-defense weights is included to correct the data masking effect caused by imbalances in the defense cascade. This mechanism specifically includes the following steps:

[0062] Obtain the sub-scores of adjacent immune defenses in the three lines of immune defense, and extract the absolute difference between the sub-scores of adjacent immune defenses;

[0063] Then, based on the absolute difference and the preset full score standard value of the defense line, the relative deviation ratio between the adjacent immune defense lines is calculated;

[0064] The relative deviation ratio is then compared with a preset cascade imbalance threshold. If the relative deviation ratio is greater than the preset cascade imbalance threshold, it is determined that the current subject has an internal cascade imbalance in the immune defense network, and the cross-defense weight adaptive balancing mechanism is automatically triggered.

[0065] The specific logic for implementing the cross-defense line weight adaptive balancing mechanism includes:

[0066] First, extract the core specific peripheral blood assessment indicators that cause the highest relative deviation rate;

[0067] Then, a generative adversarial network is constructed, and the core specific peripheral blood assessment indicators, the relative deviation ratio, and the initial personalized weighting scheme are simultaneously input into the generative adversarial network;

[0068] Using the generator model contained within the generative adversarial network, candidate inhibition decay factors with different decay step sizes are generated for the core specific peripheral blood assessment indicators.

[0069] Simultaneously, using the discriminator model contained within the generative adversarial network, and based on the real balanced clinical weight distribution retrieved from the health reference database, the candidate inhibition decay factor is judged for goodness of fit. Through multiple adversarial game iterations between the generator model and the discriminator model, until the discriminator model can no longer distinguish the difference between the adjusted weight and the real balanced clinical weight, the optimal inhibition decay factor is output.

[0070] Based on the optimal inhibition attenuation factor, the weight coefficients corresponding to the core specific peripheral blood assessment indicators are adjusted downward for the second time, while the weight coefficients of the remaining specific peripheral blood assessment indicators that have not caused deviation are adjusted upward proportionally, thereby updating the personalized weight scheme.

[0071] Finally, the weighted summation step is re-executed based on the updated personalized weighting scheme to expose potential immunodeficiency states and recalculate the comprehensive immune function score. At the same time, a cascade imbalance depth warning instruction is generated based on the recalculated comprehensive immune function score. The cascade imbalance depth warning instruction and the underlying immunodeficiency state data are synchronously sent to the corresponding clinician terminal interface to forcibly trigger a special medical intervention operation for hidden immune imbalance.

[0072] Compared with the prior art, the beneficial effects of the present invention are:

[0073] 1. By optimizing the weight allocation process of the gradient boosting machine learning algorithm, introducing different population categories, multi-dimensional label systems, and basic weight library retrieval mechanisms, and combining iterative optimization and dynamic modulation strategies, the relevance and accuracy of weight allocation are significantly improved. Through multi-level validation, cross-population cross-validation, and comparative validation with traditional evaluation methods, the accuracy, specificity, and clinical applicability of personalized weight schemes for each population are effectively guaranteed, solving the technical pain points of existing technologies that weight allocation cannot be adapted to special populations and lacks dynamic expansion capabilities.

[0074] 2. A dynamic prediction model is constructed using a linear regression algorithm to accurately predict future trends and fluctuations in immune function; a correlation model between scores and disease incidence risk is established to quantify the probability of disease; and clear warning thresholds, tiered warning methods, and the process for constructing warning traceability files are defined. This achieves integrated dynamic monitoring of immune function, risk prediction, tiered warning, and full-process traceability, solving the pain points of existing technologies such as simple monitoring, lack of trend prediction, and imperfect warnings. It enhances the clinical guidance value of the scoring method and provides strong support for precise dynamic management of immune function and clinical intervention adjustments. Attached Figure Description

[0075] Figure 1 This is a flowchart of the peripheral blood immune function scoring method based on the three-level immune structure of the present invention. Detailed Implementation

[0076] 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.

[0077] To address the technical issue of low accuracy in existing scoring methods that rely solely on simple index aggregation or mean calculation without considering weight differences across different immune levels and the inherent correlations between indicators, please refer to [link to relevant documentation]. Figure 1 This embodiment provides the following technical solution:

[0078] A peripheral blood immune function scoring method based on the three-tiered immune structure includes the following steps:

[0079] Step 1: Construct a peripheral blood immune assessment index system based on the three-tiered immune structure, which includes the first line of defense (e.g., physiochemical barrier), the second line of defense (e.g., innate immunity), and the third line of defense (e.g., adaptive immunity). Specific peripheral blood assessment indicators are set for each line of defense; this includes the following steps:

[0080] To construct the first line of defense indicators, soluble indicators directly related to mucosal barrier function in peripheral blood were selected, including secretory immunoglobulin A (sIgA) and mucosa-associated epithelial cytokines (IL-22 and IL-17). These indicators directly reflect the integrity and defense capability of the physicochemical barrier. Among them, sIgA is the core defense molecule of the mucosal barrier, while IL-22 and IL-17 can promote the proliferation and repair of mucosal epithelial cells and enhance barrier function.

[0081] The second line of defense indicators were constructed by selecting innate immune cells and related cytokines in peripheral blood, including the percentage and activity of natural killer cells (NK cells), macrophage markers (CD68), and innate immune cytokines (TNF-α, IL-6). The indicators reflect the rapid response capability of innate immunity and the efficiency of pathogen clearance. Among them, the number and activity of NK cells directly affect the early pathogen killing effect, while TNF-α and IL-6 can mediate inflammatory responses and recruit innate immune cells to participate in defense.

[0082] To construct the third line of defense indicators, adaptive immune cell subsets and related molecules in peripheral blood were selected, including the percentages of CD3+ T cells, CD4+ T cells, and CD8+ T cells, the CD4+ / CD8+ ratio, the percentage of B cells, immunoglobulins (IgG, IgM), and the T cell receptor (TCR) diversity index. These indicators reflect the specific recognition, proliferation, differentiation, and immune memory capabilities of adaptive immunity. Among them, the TCR diversity index can effectively assess the recognition ability of adaptive immunity, and the CD4+ / CD8+ ratio reflects the balance of T cell subsets and is a core evaluation indicator of adaptive immune function.

[0083] Step 2: Collect peripheral venous blood from the subjects and use real-time quantitative PCR to quantitatively detect various assessment indicators, obtaining the actual detection value of each indicator; collect peripheral blood samples from healthy subjects of different ages and genders to detect specific peripheral blood assessment indicators; calculate the normal reference range for each indicator, such as mean ± 2 standard deviations, and use a standardized formula to convert the actual detection values ​​of each indicator into standardized scores, such as a range of 0-10 points, eliminating dimensional differences between different indicators and forming a health reference database.

[0084] Step 3: Using a gradient-enhanced machine learning algorithm, combined with clinical sample data (such as peripheral blood indicators and clinical immune function assessment results from healthy individuals, immunodeficient patients, cancer patients, and patients with autoimmune diseases), weights are assigned to each indicator in the three-level indicator system. The weight coefficients for each indicator and each immune defense line are determined, and the weight coefficients are optimized for different populations to form a personalized weighting scheme. This includes the following steps:

[0085] Clinical sample data were collected, including peripheral blood test values ​​of various indicators and clinical immune function assessment results of the subjects. The clinical immune function assessment results were determined by clinicians based on the subjects' infection frequency, disease recovery speed and immune-related complications, and were divided into four levels: normal, mildly deficient, moderately deficient and severely deficient.

[0086] The participants were divided into four basic categories: healthy individuals, immunodeficient individuals, cancer patients, and individuals with autoimmune diseases. Multidimensional characteristic information was collected from the participants, including age, gender, history of underlying diseases, and medication use, forming a multidimensional labeling system. Among them, healthy individuals were defined as those with no recent history of infection, no chronic diseases, and whose immune function indicators were all within the normal range; immunodeficient individuals included patients with primary immunodeficiency diseases and those with secondary immunodeficiency caused by radiotherapy, chemotherapy, organ transplantation, or HIV infection; cancer patients were further subdivided into solid tumor patients and hematologic malignancy patients, and key information such as tumor stage and whether they were receiving immunotherapy was recorded; the autoimmune disease group covered common autoimmune diseases such as rheumatoid arthritis, systemic lupus erythematosus, and inflammatory bowel disease, and disease activity scores were recorded. Multidimensional characteristic information included demographic characteristics (age and gender), immune status characteristics (degree of immune aging, state of immune exhaustion, and level of inflammation), and treatment intervention characteristics (recording whether the participant was receiving immunomodulatory therapy, the type of medication used, and the duration of treatment).

[0087] The classification information and feature labels are integrated to form training datasets and validation datasets;

[0088] Construct the input feature matrix X and output label vector Y of the XGBoost model, and set the hyperparameters of the XGBoost model;

[0089] In this system, each row of the input feature matrix X corresponds to a subject sample, and each column corresponds to a specific detection value of a peripheral blood assessment indicator. The output label vector Y corresponds to the clinical immune function assessment level. The learning rate eta is 0.1, the maximum tree depth max_depth is 5, the number of iterations n_estimators is 100, and the subsample ratio subsample is 0.8. For example, taking the first line of defense as an example, the input features include the concentration of secretory immunoglobulin A and the concentration of mucosa-associated epithelial cytokines. The second line of defense includes the percentage and activity of natural killer cells and the proportion of macrophage marker CD14+ cells. The third line of defense includes the percentage of CD3+ T cells, the percentage of CD4+ T cells, the percentage of CD8+ T cells, the CD4+ / CD8+ ratio, the percentage of B cells, and the concentration of immunoglobulin G.

[0090] The training dataset is input into the XGBoost model, and training is performed using clinical immune function assessment levels as supervisory labels. During training, the model calculates the importance score of each indicator based on its frequency of occurrence in decision tree split nodes and the reduction in impurity. The importance scores of the indicator in all trees are weighted and summed, and then divided by the total number of splits to obtain the standardized weight coefficients. The weight coefficients range from 0 to 1, and the sum of the weight coefficients of all indicators is 1. For example, for a cancer patient receiving PD-1 inhibitor therapy, the baseline weights of the cancer population are first retrieved. After the model identifies the characteristic of receiving immunotherapy, it analyzes the correlation between this characteristic and the baseline weights. It is concluded that during immunotherapy, the importance of T cell exhaustion-related markers, such as PD-1 expression level and Tim-3 expression level, is significantly increased. Therefore, the weight of this indicator is increased, while the weights of other indicators are decreased accordingly.

[0091] For example, taking the third line of defense as an example, after model training, the weight coefficients for the percentage of CD4+ T cells are 0.28, the weight coefficient for the percentage of CD8+ T cells is 0.22, the weight coefficient for the CD4+ / CD8+ ratio is 0.18, the weight coefficient for the percentage of CD3+ T cells is 0.16, the weight coefficient for the percentage of B cells is 0.10, and the weight coefficient for immunoglobulin G concentration is 0.06. The sum of the weight coefficients of all indicators within each line of defense is 1.

[0092] For different population groups, weight coefficients were optimized separately using a transfer learning strategy. Using the basic weight vector of healthy individuals as the initial baseline, the XGBoost model was fine-tuned using training samples from the respective populations (immune-deficient, cancer, and autoimmune disease) to recalculate the weights of each indicator. For example, for patients recovering from chemotherapy for cancer, the weight coefficient for the percentage of CD4+ T cells was increased to 0.35, while the weight coefficient for the percentage of B cells was decreased to 0.05 to reflect the predominantly cellular immunosuppression clinical characteristics of this group. The weight coefficient of each immune defense line is the sum of the weight coefficients of all indicators within that defense line. For example, the weight coefficient for the first line of defense is 0.20, the second line of defense is 0.35, and the third line of defense is 0.45, with the sum of the weight coefficients of the three lines of defense being 1.

[0093] The accuracy of the weighted coefficient prediction was evaluated using a validation dataset. The validation results were compared with the clinical immune function assessment results, and the Spearman correlation coefficient between the weighted score and the clinical assessment results was calculated.

[0094] If the correlation coefficient is lower than the preset threshold range, such as 0.75, adjust the hyperparameters of XGBoost, such as increasing the number of iterations or adjusting the tree depth, and retrain until the requirements are met.

[0095] If the weighted score is consistent with the clinical assessment results, it indicates that the modulation direction is correct, and a personalized weight allocation scheme is formed. The final weight allocation principle is: adaptive immunity (third line of defense) weight > innate immunity (second line of defense) weight > physical and chemical barrier (first line of defense) weight. Among them, the weight of adaptive immunity accounts for 40%-50%, the weight of innate immunity accounts for 30%-35%, and the weight of physical and chemical barrier accounts for 15%-25%, which conforms to the synergistic effect of the three-level immune structure and highlights the core value of the specific defense and memory function of adaptive immunity.

[0096] In one embodiment, 320 subjects admitted to the Department of Clinical Immunology of a tertiary hospital were used as the sample subjects, including 80 healthy individuals undergoing physical examinations, 80 immunodeficiency patients, 80 cancer patients, and 80 patients with autoimmune diseases.

[0097] Peripheral venous blood was collected from each subject, and the values ​​of 15 immune indicators in the peripheral blood were recorded for each subject. The first line of defense indicators included: secretory immunoglobulin A concentration and mucosa-associated epithelial cytokine concentration. The second line of defense indicators included: percentage of natural killer cells, natural killer cell activity, proportion of CD14+ macrophage markers, and innate immune cytokines (IL-6, TNF-α). The third line of defense (adaptive immunity) indicators included: percentage of CD3+ T cells, percentage of CD4+ T cells, percentage of CD8+ T cells, CD4+ / CD8+ ratio, percentage of B cells, and immunoglobulin G concentration.

[0098] An XGBoost base model was constructed using 80 healthy individuals as the training set. Hyperparameters were set as follows: learning rate eta = 0.1, maximum tree depth max_depth = 5, number of iterations n_estimators = 100, and subsample ratio subsample = 0.8. The depth decay coefficient γ was set to 0.10. These hyperparameters were optimized through grid search, multiple comparative experiments, and backtesting with clinical samples, representing the optimal combination of parameters that balances model fitting accuracy and generalization ability.

[0099] The model training employs a decision tree node depth decay correction rule. For each feature, the frequency of occurrence, average gain, and corresponding depth decay factor are statistically analyzed at shallow nodes (depth ≤ 2), mid-level nodes, and deep nodes (depth ≥ 4). The corrected contribution value of each indicator is obtained by summing "frequency of occurrence × average gain × depth decay factor". The basic importance score is then calculated as the ratio of each indicator's contribution value to the total contribution value of all indicators. After training, the basic importance scores for each immune indicator are extracted from the model.

[0100] Taking the third line of defense as an example, the basic importance scores for each indicator are calculated as follows:

[0101] Regarding the percentage of CD4+ T cells, this feature appeared 85 times in the decision tree splitting nodes; among them, 32 appeared in shallow nodes (depth ≤ 2), with an average gain of 0.045 and an average depth decay factor of approximately 0.95; 18 appeared in deep nodes (depth ≥ 4), with an average gain of 0.018 and an average depth decay factor of approximately 0.45. The contribution value after depth decay factor correction is:

[0102] 32×0.045×0.95+35×0.030×0.75+18×0.018×0.45=2.302.

[0103] Similarly, other indicators were calculated: CD8+ T cell percentage contribution value 1.850, CD3+ T cell percentage contribution value 1.420, CD4+ / CD8+ ratio contribution value 1.180, B cell percentage contribution value 0.850, and immunoglobulin G concentration contribution value 0.560.

[0104] The total contribution value was 8.162, obtained by summing the contribution values ​​of all indicators. The basic importance scores of each indicator were as follows: CD4+ T cell percentage 0.282, CD8+ T cell percentage 0.227, CD3+ T cell percentage 0.174, CD4+ / CD8+ ratio 0.145, B cell percentage 0.104, and immunoglobulin G concentration 0.069. The sum of the scores was 1.001 (approximately equal to 1).

[0105] The normalized universal calculation formula within the immune defense line is adopted, applicable to all populations and weight conversion of all immune defense indicators. Its expression is:

[0106]

[0107] In the formula, This is represented as the final weighting coefficient for a single indicator; This is represented as the importance score after adjustment for each individual indicator; It is represented as the sum of the importance scores of all features within the same defense line after correction; This represents the fixed total weight of the corresponding immune defense line.

[0108] For healthy individuals, without introducing population adjustment coefficients, the basic importance scores are directly summed = 1.001 ≈ 1. Combined with the fixed total weight of 0.45 for the third line of defense, the original baseline weights of each indicator of the third line of defense in healthy individuals are calculated item by item. Using the above formula, the following values ​​are obtained: CD4+T: 0.282 / 1 × 0.45 = 0.127; CD8+T: 0.227 × 0.45 = 0.102; CD3+T: 0.174 × 0.45 = 0.078; CD4+ / CD8+: 0.145 × 0.45 = 0.065; B cells: 0.104 × 0.45 = 0.047; Immunoglobulin G: 0.069 × 0.45 = 0.031. The sum of the six weights equals 0.45, which gives the baseline weights of each indicator of the third line of defense in healthy individuals.

[0109] This three-tiered immune structure includes a first line of defense, a second line of defense, and a third line of defense, fully covering 15 peripheral blood immune indicators. The fixed weight coefficients for the three lines of defense are pre-set as follows: 0.20 for the first line of defense, 0.35 for the second line of defense, and 0.45 for the third line of defense, with a total weight of 1. This weighting ratio is determined based on the proportion of physiological contributions to the three-tiered immune structure in the human body and the statistical comprehensive analysis of the variance contribution rate of immune indicators in a large sample of healthy individuals, and has a clear immunological and statistical basis.

[0110] The first line of defense includes the following indicators: secretory immunoglobulin A, mucosa-associated epithelial cytokines IL-17 and IL-22. Using the same XGBoost model and depth attenuation correction rule as the third line of defense, the corrected contribution values ​​of each indicator were calculated to be 1.025, 0.682, and 0.413, respectively, with a total contribution of 2.12. The baseline importance scores for each indicator were calculated to be 0.483, 0.322, and 0.195, respectively, with the sum of these three scores approximately equal to 1. Applying the normalization formula and multiplying by the defense weight of 0.20, the health baseline weight for the first line of defense was obtained.

[0111] The second line of defense includes indicators such as the percentage and activity of natural killer cells, the macrophage marker CD68, and the innate immune cytokines TNF-α and IL-6. Using the same unified model rules, the corrected contribution values ​​for each indicator are 1.562, 1.218, 0.935, and 0.726, respectively, with a total contribution of 4.441. The sum of the baseline importance scores is approximately 1. Multiplying this by the defense line weight of 0.35 yields the health baseline weight for the second line of defense.

[0112] The above normalized general calculation formula ensures that the sum of the weights of each indicator within each defense line is strictly equal to the fixed weight of that defense line, and the final sum of the weight coefficients of all 15 indicators is strictly 1.

[0113] For the cancer population, the population-specific correction coefficient was set to 1.25. The population-specific correction coefficient was determined by statistical fitting based on the deviation of the mean values ​​of immune indicators between various populations and healthy populations, the deviation rate of the distribution of clinical immune function levels, and the differences in the distribution of the importance of model features. Among them, the correction coefficient of 1.25 for the cancer population is derived from the fact that the overall adaptive immune indicator features of cancer patients are significantly more important than those of healthy populations by 25%, and was adjusted after statistical fitting of multiple batches of clinical samples and correlation verification of the validation set.

[0114] The general rule for population-adjusted importance score is: population-adjusted importance score = baseline importance score of healthy population × population-specific adjustment coefficient.

[0115] Multiply the baseline importance score output by the healthy population model by a correction factor to obtain the corrected importance score for the cancer population:

[0116] The percentages of CD4+ T cells are 0.282 × 1.25 = 0.353; CD8+ T cells are 0.227 × 1.25 = 0.284; CD3+ T cells are 0.174 × 1.25 = 0.218; the CD4+ / CD8+ ratio is 0.145 × 1.25 = 0.181; the percentage of B cells is 0.104 × 1.25 = 0.130; and the immunoglobulin G concentration is 0.069 × 1.25 = 0.086. The sum of the adjusted importance scores for each indicator is 1.252.

[0117] The weighting coefficients for each line of defense are set as follows: 0.20 for the first line of defense, 0.35 for the second line of defense, and 0.45 for the third line of defense. For features within the third line of defense, the corrected importance scores are first normalized (so that the sum of the weights within that line of defense equals 0.45). The sum of the corrected importance scores for the third line of defense is 1.252, therefore the normalization factor is 1.252. The final weighting coefficients for each indicator are: CD4+ T cell percentage 0.353 / 1.252 × 0.45 = 0.127; CD8+ T cell percentage 0.2. 84 / 1.252×0.45=0.102; CD3+ T cell percentage 0.218 / 1.252×0.45=0.078; CD4+ / CD8+ ratio 0.181 / 1.252×0.45=0.065; B cell percentage 0.130 / 1.252×0.45=0.047; Immunoglobulin G concentration 0.086 / 1.252×0.45=0.031; The sum of the weights of the six indicators within the third line of defense is 0.450, which meets the requirements for the weight allocation of the defense line.

[0118] Similarly, the weight coefficients of each indicator within the first and second lines of defense are calculated, and the sum of the weight coefficients of all 15 indicators is 1.

[0119] For patients undergoing chemotherapy for cancer, the above-corrected importance score for cancer patients is used as the benchmark, and the above-mentioned normalized benchmark weight for cancer is used as the initial base. The correction coefficient is not recalculated. Based on the weight allocation scheme for healthy patients, a slight adjustment is made: the weight coefficient of CD4+ T cell percentage is increased to 0.142 (approximately 12%), and the weight coefficient of B cell percentage is decreased to 0.035 (approximately 25%) to reflect the clinical characteristics of this type of patient, which is mainly characterized by cellular immunosuppression. The remaining indicators are proportionally distributed to the weight difference, and the sum of the weights of all indicators within the third line of defense remains at 0.45 after the adjustment.

[0120] For patients undergoing chemotherapy for cancer, the baseline weights of healthy individuals after tumor correction and normalization were used as the initial values. Fine-tuning rules were formulated based on the clinical immunological characteristics centered on cellular immunosuppression after chemotherapy. Weight redistribution was carried out only within a single line of defense, keeping the total weight of the single line of defense unchanged at 0.45. The weights of core cellular immune indicators were increased, the weights of humoral immune-related indicators were decreased proportionally, and the remaining indicators were adjusted using a proportional adaptive fine-tuning method. The adjustment range was taken from the statistical mean of the attenuation range of clinical immune indicators in chemotherapy patients.

[0121] For individuals in the active phase of autoimmune diseases, the autoimmune corrected importance score is obtained by multiplying the basic health importance score by the autoimmune-specific correction coefficient. The autoimmune baseline weight is then normalized. The weight coefficient for the percentage of B cells is increased to 0.058 (approximately 23%), the weight coefficient for immunoglobulin G concentration is increased to 0.042 (approximately 35%), and the weight coefficient for the percentage of CD4+ T cells is decreased to 0.110 (approximately 13%). After the adjustment, the sum of the weights of these defense indicators remains at 0.45.

[0122] The active phase of autoimmune diseases is characterized by abnormal B cell activation, elevated immunoglobulins, and CD4+ T cell subset imbalance. The weight adjustment range is determined based on the statistical values ​​of the differences in immune indicators between the active and remission phases. The adjustment process also follows the rule of constant total weight of the defense line and proportional complementary allocation of internal indicators.

[0123] For immunocompromised individuals (such as HIV patients), an HIV-specific correction coefficient and a baseline health importance score were used to calculate the corrected importance score. This score was then normalized to obtain the HIV baseline weight. Based on this baseline weight, the weight coefficient for the percentage of CD4+ T cells was increased to 0.145, and the weight coefficient for the percentage of CD8+ T cells was increased to 0.115. Other indicators were adjusted accordingly to maintain the total weight of this line of defense at 0.45. HIV immunocompromised individuals exhibit typical immune characteristics of progressive depletion of CD4+ T cells and compensatory increases in CD8+ T cells. The weight adjustment range was determined based on statistical analysis of longitudinal monitoring data on peripheral blood T cell subsets from HIV patients, maintaining the total weight of the third line of defense at 0.45 throughout the adjustment process.

[0124] The predictive accuracy of the weighting coefficients was evaluated using a validation set of 64 cases. For each subject in the validation set, standardized values ​​of 15 immune indicators were collected, and these values ​​were weighted and summed according to the aforementioned weighting scheme for the corresponding population to calculate the subject's comprehensive immune function score, denoted as [score here]. .

[0125] Two clinical experts independently assessed the subjects' clinical immune function levels using a blinded method, based on their clinical symptoms, history of infection, complications, laboratory tests, and previous treatment history, according to a pre-defined four-level standard (e.g., 1 = normal immunity, 2 = mild deficiency, 3 = moderate deficiency, 4 = severe deficiency). The consensus assessment result of the two clinical experts was taken as the final clinical immune function level. If the assessment results are inconsistent, the determination shall be made by a third expert.

[0126] The overall scores of the 64 subjects in the validation set Sort by rank according to numerical value to obtain the corresponding rank sequence. Clinical immune function level Sort by rank from highest to lowest to obtain the corresponding rank sequence. ; Calculate the Spearman correlation coefficient between the two sets of rank sequences. The formula is as follows:

[0127]

[0128] In the formula, Represented as the correlation coefficient; This is expressed as the size of the validation set. , which is represented as the first Example: The difference between the subject's rating rank and clinical grade rank.

[0129] For correlation coefficient Perform a two-sided significance test and set the significance level. .

[0130] In this embodiment, the Spearman correlation coefficient obtained on the validation set is 0.83, which is verified. The value is greater than the preset threshold of 0.75, indicating a significant positive correlation between the comprehensive immune function score and the immune function level assessed by clinical experts. The weight allocation scheme has good clinical consistency and predictive accuracy, verifying the effectiveness of the weight allocation scheme of the present invention.

[0131] An assessment of a cancer patient undergoing chemotherapy was conducted using the aforementioned weighting scheme: the patient's CD4+ T cell percentage was 28%; CD8+ T cell percentage was 35%; CD3+ T cell percentage was 55%; CD4+ / CD8+ ratio was 0.80; B cell percentage was 8%; and immunoglobulin G concentration was 10.2 g / L. Based on the cancer population weighting scheme, the third line of defense scored 62 points. Combining the scores of the first and second lines of defense, the overall score was 58 points, indicating moderate immunodeficiency.

[0132] Based on the patient's infection history (two respiratory infections within the past 3 months) and recovery speed (delayed postoperative wound healing), the clinicians comprehensively assessed the patient as having a moderately low level of illness, which was consistent with the assessment and verified the effectiveness of the weighting scheme.

[0133] The beneficial effects achieved by the above methods are as follows: By introducing a depth attenuation factor, a population-specific compensation coefficient, and a defense line weight allocation mechanism, the weight ratios of different immune defense lines and their internal indicators can be dynamically adjusted according to the actual needs of clinical immune assessment. This allows the scoring results to more accurately reflect the true immune status of different populations, including healthy individuals, cancer patients, immunocompromised patients, and patients with autoimmune diseases. Compared to traditional scoring models based on uniform weights, this method significantly improves the consistency between the scoring and clinical assessment results through shallow node weight enhancement and population-specific correction. Simultaneously, the defense line cascade balance constraint ensures the medical rationality of the weight allocation, effectively avoiding the problem of one defense line being overweighted and masking abnormalities in other defense lines. This helps clinicians accurately identify immune weaknesses and guide intervention decisions.

[0134] For different population groups, including healthy individuals, cancer patients, and immunodeficiency patients, the weighting coefficients are optimized to form a personalized weighting allocation scheme, thereby improving the relevance and accuracy of the scoring method.

[0135] A quantitative scoring standard and weight conversion relationship for immune indicators were established, and normal reference ranges for each peripheral blood immune indicator were created. The actual test values ​​of each indicator were converted into standardized individual scores of 0-10 points according to a linear normalization rule. Then, using population-specific indicator weight coefficients, the scores of each immune defense line were calculated by summing the individual indicator scores multiplied by their corresponding indicator weights. Finally, the comprehensive immune function score was obtained by summing the scores. The comprehensive score level ranges were predefined, such as: 0-40 points for severe immunodeficiency, 41-60 points for moderate immunodeficiency, 61-80 points for mild immunodeficiency, and 81-100 points for good immune function. The scoring range standards and the scores and weights of each indicator have a fixed conversion correspondence, which can completely reproduce the entire scoring calculation process.

[0136] The generated population-specific weighting schemes are validated at multiple levels. Using reserved validation set data, the prediction accuracy of each population weighting scheme within its respective population is examined. For example, the prediction accuracy for each population is required to reach over 85%, and for rare populations with small sample sizes, the accuracy requirement can be appropriately relaxed to 80%. Cross-population cross-validation is also performed to verify the specificity of the weighting schemes. For example, the weighting scheme for cancer patients is applied to healthy individuals to observe whether there are any abnormal deviations in the scoring results. If the scores of healthy individuals are significantly lower after using the cancer weighting, it indicates that the weighting scheme has good specificity and can effectively distinguish between different populations.

[0137] Newly enrolled subjects were selected and assessed using both traditional and this method. The results of the two assessments were compared to verify its clinical applicability and accuracy. If the characteristics of newly enrolled subjects could not be classified into any existing population category, and their immune indicators showed unique regularities, the new population identification process was automatically triggered. Through cluster analysis and expert evaluation, the existence of new population subtypes was confirmed. If confirmed, the population was included in the model training scope, and an independent basic weight vector was established for them, forming a complete weight scheme library that is dynamically modulated according to individual characteristics. This scheme library includes: basic weight vectors for healthy individuals, cancer patients, immunodeficient individuals, and autoimmune disease patients; subtype weight adjustment rules for different age groups (children, adults, and the elderly); weight modulation parameters for different treatment stages (during chemotherapy, during immunotherapy, and during recovery); and characteristic weight patterns for different immune states (immunosenescence, immune exhaustion, and inflammatory states).

[0138] Furthermore, in the above embodiments, taking a tumor population (80 cases in total, 56 in the training set and 24 in the validation set) as an example, the fixed weight coefficients of each immune indicator in the tumor population, the standardized scoring rules of the unified 0-10 score indicators, and the comprehensive scoring level range standard were used to calculate the scores for the 24 tumor patients in the validation set. First, the percentage values ​​of peripheral blood T cells, B cells, and other cells were converted into individual standardized scores, and then the weights of the tumor population-specific indicators were substituted into the scores for weighted summation to obtain the comprehensive immune function score, which was then compared with the immune function level assessed by clinical experts. Statistical analysis showed that 21 out of the 24 cases were correctly identified, with a prediction accuracy rate of 87.5%, which is greater than the preset requirement of 85%. Specifically, among the 8 patients clinically assessed as severely immunocompromised, all scores fell within the severe range (0-40 points); among the 10 patients with moderate immunocompromised, 9 scores fell within the moderate range (41-60 points), and 1 score was 58 points (borderline moderate / mild); among the 6 patients with mild immunocompromised, 5 scores fell within the mild range (61-80 points), and 1 score was 82 points (high; upon review, this patient was receiving immune-enhancing therapy). The predictive accuracy within this population met the requirements.

[0139] Taking an immunodeficient population (80 cases) as an example, using a specific weighting scheme, the same indicator score conversion rules, and scoring grading standards, the prediction accuracy was calculated for 24 patients in the validation set. The accuracy rate was 83.3%, close to but slightly below the standard of 85%. The main reason for the lower accuracy rate is that this population includes two subtypes: primary immunodeficiency and secondary immunodeficiency, with differences in their immune characteristics. Therefore, this population was further subdivided: the accuracy rate for primary immunodeficiency patients (12 cases) was 91.7%, and the accuracy rate for secondary immunodeficiency patients (12 cases) was 75.0%. For secondary immunodeficiency patients, the weighting scheme was further refined, such as distinguishing between HIV-related immunodeficiency and post-transplant immunosuppression, which improved the subsequent prediction accuracy rate to 87.5%.

[0140] For rare populations with small sample sizes, in this example, there are only 5 cases of specific subtypes of autoimmune diseases such as IgG4-related diseases among patients, and the validation set has fewer than 3 cases. The accuracy requirement is relaxed to 80%, and leave-one-out cross-validation is used for evaluation. The actual accuracy rate is 80%.

[0141] To test the specificity of the weighting schemes for different population groups, cross-population validation was conducted. The weighting scheme for the cancer population was applied to a validation set of healthy individuals, and their comprehensive immune function scores were calculated using 24 healthy subjects. The results showed that the average score of the healthy individuals using the cancer weighting scheme was 52.3, which falls within the moderately low range, while the average score using the healthy population weighting scheme was 86.7, which falls within the good immune range. The difference of 34.4 points indicates a significantly abnormally low score, suggesting that the cancer weighting scheme underestimates the immune function of the healthy individuals. This validates that the weighting scheme has good specificity and can effectively distinguish between different population groups.

[0142] Similarly, the weighting scheme for autoimmune diseases was applied to the validation set of cancer patients to calculate their comprehensive scores. The results showed that the average score for cancer patients using the autoimmune disease weighting scheme was 44.2 points, while the average score using the cancer weighting scheme was 38.5 points, a difference of 5.7 points. Although there was a difference, the magnitude was small, which is attributed to the fact that both groups of patients have immune abnormalities, and the weighting schemes have some overlap in their emphasis on specific characteristics.

[0143] When the immunodeficiency weighting scheme was applied to the validation set of healthy individuals, the average score of the healthy individuals using the immunodeficiency weighting scheme was 58.6, which was 28.1 points lower than the 86.7 score using the health weighting scheme, also showing a significant specific difference.

[0144] Fifty newly enrolled subjects were selected and assessed using both the traditional assessment method based on thresholds of single immune indicators and the personalized weighting scheme of this method. The results of the two assessments were then compared.

[0145] Take a newly enrolled tumor immunotherapy patient as an example: The patient is a 62-year-old male who received PD-1 inhibitor treatment for 3 months after surgery for lung adenocarcinoma.

[0146] The results of the traditional assessment method were as follows: CD4+ T cell count 320 cells / μl (slightly below the lower limit of normal), CD8+ T cell count 280 cells / μl, and CD4+ / CD8+ ratio 1.14 (normal). The overall assessment was mild immunodeficiency.

[0147] This method uses a weighting scheme based on the population category (tumor population + immunotherapy subtype), that is: the weight of CD4+ T cell percentage is increased to 0.35, and the weight of B cell percentage is decreased to 0.05. The calculated comprehensive score is 54 points, which is judged as moderate immunodeficiency.

[0148] Based on the patient's infection history (two lung infections within the past four months) and immune-related adverse reactions (mild rash, hypothyroidism), the clinician assessed the patient as having moderate hypothyroidism. The results of this method were consistent with the clinical assessment, while traditional methods tend to underestimate the severity.

[0149] Statistical analysis of the assessment results of 50 newly enrolled subjects showed that the consistency between this method and the traditional method was 68% (34 / 50), of which the consistency between this method and the clinical assessment was 84% ​​(42 / 50), while the consistency between the traditional method and the clinical assessment was 72% (36 / 50). The clinical accuracy and practicality of this method are superior to those of the traditional method.

[0150] In clinical applications, if the characteristics of newly enrolled subjects cannot be classified into any existing population category (healthy people, cancer patients, immunodeficiency patients, patients with autoimmune diseases), and their immune indicators show unique regularity, the new population identification process is automatically triggered.

[0151] For example, a subject is a relapsed / refractory B-cell lymphoma patient who has been receiving long-term CAR-T cell therapy. Their immune characteristics are: almost undetectable B cells (CD19+ cells <1%), significantly low CD4+ T cell count (absolute count 180 cells / μl), but near-normal CD8+ T cell count, and a significantly elevated PD-1+ T cell proportion (45%). This type of patient cannot be easily categorized into existing cancer populations; their immune indicators exhibit unique patterns. The system automatically triggers a new population identification process.

[0152] First, cluster analysis was performed on the subject's multidimensional characteristics. The K-means algorithm (K=3) was used to compare the subject's feature vector with the cluster centers of existing population groups. The results showed that the Euclidean distance between the subject and the four existing population groups was greater than a preset threshold (1.5 standard deviations), suggesting a possible new subtype. Subsequently, an expert evaluation process was initiated, with two chief hematologists and one immunology expert assessing the case. The experts confirmed that the patient, after CAR-T therapy, possessed a unique immunophenotype of B-cell aplasia accompanied by T-cell depletion, distinct from both conventional cancer patients and patients with primary immunodeficiency, thus belonging to a new subtype.

[0153] After confirming the existence of a new subtype, it was included in the model training scope. At least 30 samples of patients from this subtype were collected as the training set. The clinical immune function assessment level of this subtype was used as the supervision label, and the XGBoost model was retrained using the method in step 3, outputting a base weight vector specific to this subtype. Taking patients after CAR-T therapy as an example, the characteristics of the trained base weight vector are: the weight coefficient of the B cell percentage was further reduced from 0.05 in the conventional tumor population to 0.01 (almost not involved in the scoring); the PD-1+ T cell ratio was introduced as a new feature into the model with a weight coefficient of 0.18; and the CD4+ T cell percentage weight was increased to 0.32. Through continuous optimization and expansion, a complete weight scheme library including the following content was formed:

[0154] Basic weight vectors for various population groups:

[0155] The basic weight vector for healthy individuals was obtained by training with 80 healthy subjects, with the third line of defense accounting for 45% of the weight. Among them, the weight of CD4+ T cell percentage was 0.28; the weight of CD8+ T cell percentage was 0.22; the weight of CD4+ / CD8+ ratio was 0.18; the weight of CD3+ T cell percentage was 0.16; the weight of B cell percentage was 0.10; and the weight of immunoglobulin G concentration was 0.06.

[0156] The basic weight vector for cancer patients was slightly adjusted based on the healthy population, with the weight of the third line of defense increasing to 48%; among them, the weight of CD4+ T cells percentage was increased to 0.32, and the weight of B cells percentage was decreased to 0.06.

[0157] Tumor + Immunotherapy Subtype Weighting Vector: CD4+ T cell percentage weight 0.35; PD-1+ T cell percentage weight 0.15 (new); B cell percentage weight 0.03.

[0158] Tumor subtype weighting vector after chemotherapy: CD4+ T cell percentage weight 0.34; CD8+ T cell percentage weight 0.24; B cell percentage weight 0.04.

[0159] Basic weighted vector for immunodeficient populations: CD4+ T cell percentage weight 0.32; CD8+ T cell percentage weight 0.24; CD4+ / CD8+ ratio weight 0.16.

[0160] HIV-associated immunodeficiency subtype weight vector: CD4+ T cell percentage weight 0.40; viral load-related feature weight 0.12 (new).

[0161] Basic weighted vector for autoimmune disease population: B cell percentage weight 0.20; immunoglobulin G concentration weight 0.14; CD4+ / CD8+ ratio weight 0.12.

[0162] Subtype weighting vector for active autoimmune diseases: B cell percentage weight 0.25; inflammatory factor weight 0.15 (new).

[0163] Subtype weighting rules for different age groups

[0164] Children (<18 years old): The weighting coefficient for the first line of defense is increased to 0.25 (due to the development of the mucosal immune system), and the weighting coefficient for the third line of defense is decreased to 0.40.

[0165] Adults (18-65 years old): Basic weight vectors for various population groups are used.

[0166] For older adults (≥66 years old): the weighting coefficient for the third line of defense is lowered to 0.38, while the weighting coefficient for the second line of defense is raised to 0.38.

[0167] Weighted modulation parameters at different treatment stages

[0168] During chemotherapy: the weighting of CD4+ T cell percentage is increased by 10%, and the weighting of B cell percentage is decreased by 20%.

[0169] During immunotherapy: the weight of PD-1 / PD-L1 related features increased by 0.05-0.10.

[0170] Recovery period: The basic weight vector is gradually regressed, and the modulation parameters are regressed monthly with a decay coefficient of 0.85.

[0171] Feature weighting patterns for different immune states

[0172] Immunoresenescence status (CD28-CD57+ T cell ratio >15%): The weighting coefficient of the third line of defense was lowered by 0.05, and the weighting coefficient of the second line of defense was increased by 0.05.

[0173] Immune exhaustion status (PD-1+ T cell ratio >25%): The weight of exhaustion-related features increases by 0.10.

[0174] High inflammatory state (CRP>10mg / L): The weight of inflammatory factors is increased by 50%, and the weight of other defense lines is temporarily suppressed.

[0175] After the weighted scheme library is established, the system will periodically re-validate the weighted schemes in the library; collect the newly added clinical data in the quarter and calculate the real-time prediction accuracy of each weighted scheme; if the prediction accuracy of a certain scheme is lower than the preset threshold twice in a row (such as lower than 80%), the re-optimization process of the scheme will be triggered, the model will be fine-tuned using the latest clinical samples, and the weight vector will be updated.

[0176] Meanwhile, the weighted scheme library supports manual input; clinical experts can manually adjust the weight coefficients or add new schemes based on clinical experience and research results, which are then incorporated into the scheme library after system verification, achieving dual optimization through expert knowledge and data-driven approaches.

[0177] The beneficial effects achieved by the above are as follows: By establishing differentiated weighting schemes for healthy individuals, cancer patients, immunodeficiency patients, and patients with autoimmune diseases, and by introducing internal validation, cross-population cross-validation, and new population identification mechanisms, the weighting allocation can be dynamically adjusted according to the immune characteristics and treatment stages of different populations, significantly improving the consistency between scores and clinical assessment results; at the same time, through the identification of new populations and the dynamic expansion of the weighting scheme library, a precise and scalable immune status assessment tool is provided for clinical practice.

[0178] Accordingly, by optimizing the weight allocation process of the gradient enhancement machine learning algorithm, introducing different population categories, multi-dimensional label systems, and basic weight library retrieval mechanisms, and combining iterative optimization and dynamic modulation strategies, the relevance and accuracy of weight allocation have been significantly improved. Through multi-level validation, cross-population cross-validation, and comparative validation with traditional evaluation methods, the accuracy, specificity, and clinical applicability of personalized weight schemes for each population have been effectively guaranteed, solving the technical pain points of existing technologies that weight allocation cannot be adapted to special populations and lacks dynamic expansion capabilities.

[0179] Step 4: Based on the standardized scores and corresponding weight coefficients of each indicator, calculate the comprehensive immune function score using the weighted summation method, and calculate the sub-scores of the three lines of immune defense to assess the strength and synergistic effect of immune function at each level; the standardized scores of each indicator are calculated as follows: for each peripheral blood assessment indicator, a reference range database based on healthy population (n≥200) is established in advance, and the mean and standard deviation of the indicator are calculated.

[0180] For positive indicators (higher values ​​indicate stronger immune function, such as the percentage of CD4+ T cells), the standardized score formula is: Standardized score = (actual test value - mean of healthy population) / standard deviation of healthy population, with a cutoff range of -3 to 3, and then linearly mapped to 0-100 points.

[0181] For negative indicators (higher values ​​indicate weaker immune function), first take their reciprocal or convert them using the method of mean of healthy population - actual test value / standard deviation.

[0182] The comprehensive immune function score is calculated using a weighted summation method. The formula is: Comprehensive score = (Standardized score of each indicator × Weight coefficient of the indicator) × 100.

[0183] The scores for each of the three lines of immune defense are calculated separately: the score for the first line of defense = (standardized score of each indicator in the first line of defense × weight coefficient of the indicator) × 100, and the same applies to the second and third lines of defense; the final comprehensive score ranges from 0 to 100, where 0-40 points indicate low immunity, 41-70 points indicate moderate immunity, and 71-100 points indicate good immunity.

[0184] The three lines of defense also use a 0-100 point scale to identify specific weak points in the immune system.

[0185] Step 5: Based on the comprehensive score results, peripheral blood immune function is divided into multiple levels, clarifying the clinical significance of each level to provide guidance for clinical diagnosis and health management. In one embodiment, it is assumed that peripheral blood immune function is divided into 5 levels, for example: Level 1 (Extremely Strong Immune Function): Comprehensive score ≥ 9 points, all indicators are within the normal reference range and at the optimal level, the three lines of immune defense work well together, the body's defense ability is strong, and the risk of disease is extremely low. Level 2 (Good Immune Function): 7 points ≤ comprehensive score < 9 points, most indicators are within the normal reference range, some indicators are slightly abnormal, the three lines of immune defense work well together, the body's defense ability is relatively strong, and no special intervention is required. Level 3 (Normal Immune Function): 5 points ≤ comprehensive score < 7 points, all indicators are within the normal reference range, the three lines of immune defense function is stable, the body's defense ability is normal, and the current healthy state needs to be maintained. Level 4 (Weak Immune Function): 3 points ≤ comprehensive score < 5 points, some indicators are abnormal (below the normal reference range), at least one line of immune defense function is weakened, the body's defense ability is reduced, and it is susceptible to disease, requiring immune regulation. Level 5 (Severely Impaired Immune Function): Overall score <3 points, multiple indicators are significantly abnormal, two or more lines of immune defense are severely impaired, the body's defense ability is extremely poor, and serious infections or immune-related diseases are likely to occur, requiring immediate clinical intervention.

[0186] Step 6: Periodically collect peripheral blood from subjects, repeat steps 2-5, and obtain comprehensive immune function scores and sub-scores at different time points; construct dynamic monitoring curves to analyze the changing trends of immune function; combine clinical symptoms to provide subjects with personalized immune conditioning suggestions or treatment plan adjustments, achieving dynamic management of immune function and enhancing the clinical guidance value of the scoring method; including the following steps:

[0187] We collected immune function-related scores and clinical baseline information from subjects at different time points, and used a linear regression algorithm to fit the monitoring curves to establish a dynamic prediction model. We input the latest score data to predict the trend and fluctuation range of immune function changes within a specified future period, such as 3-6 months. We collected clinical sample data, including the incidence of diseases corresponding to different immune function scores. Through logistic regression analysis, we established a correlation model between the score results and the risk of immune-related diseases, and output the probability of disease incidence for the corresponding score range.

[0188] Combining predicted trends, disease risk probability, and early warning thresholds, an early warning is automatically triggered when the predicted trend value is lower than or higher than a preset threshold, such as below 5 points or when the disease risk probability is higher than 30%. This includes the following steps:

[0189] Set early warning thresholds, including immune function prediction trend value thresholds (e.g., 5 points, normal threshold) and disease incidence risk probability thresholds (e.g., 30%, risk threshold). Based on the graded early warning standards, set corresponding early warning methods. For example: Level 1 Early Warning (Mild Risk): Predicted trend value 4-5 points, and disease incidence probability 20%-30%; the early warning method is that the system automatically generates a mild early warning prompt and pushes it to the subject's personal health portal, such as a mobile APP or health management platform, simultaneously sending basic immune conditioning suggestions such as dietary and lifestyle adjustments, without requiring clinical physician intervention. Level 2 Early Warning (Moderate Risk): Predicted trend value 3-4 points (low immune function trend), or disease incidence probability 30%-50%; the early warning method is that the system generates a moderate early warning prompt and simultaneously pushes it to the subject and corresponding clinical physician's portal, indicating the core reason for the early warning, such as: a certain immune defense line sub-score is low. The physician, based on the subject's clinical baseline information, provides a personalized conditioning plan within 3 working days. Level 3 Warning (Severe Risk): Predicted trend value < 3 points (severely low immune function trend), or probability of disease incidence > 50%; The warning method is that the system immediately triggers a severe warning, which is simultaneously pushed to the subjects, clinicians, and nurses, and an audio-visual reminder is issued. Physicians need to complete the clinical assessment of the subjects within 24 hours, formulate an emergency intervention plan such as medication, further examination, and track the intervention effect; The system automatically records the trigger time, warning level, core trigger indicators, warning recipients, and intervention feedback results of each warning, forming a warning traceability file for subsequent model optimization and clinical review.

[0190] The beneficial effects achieved by the above content are as follows: a dynamic prediction model is constructed using a linear regression algorithm to accurately predict the future trend and fluctuation range of immune function; a correlation model between the score and the risk of disease incidence is established to quantify the probability of disease incidence; and clear warning thresholds, graded warning methods, and the process for constructing warning traceability files are defined, realizing the integration of dynamic monitoring of immune function, risk prediction, graded warning, and full traceability. This solves the pain points of existing technologies, such as simple monitoring, lack of trend prediction, and imperfect warnings, and enhances the clinical guidance value of the scoring method, providing strong support for precise dynamic management of immune function and clinical intervention adjustments.

[0191] Working principle: By constructing an assessment system covering three lines of immune defense and corresponding specific indicators for each line of defense, peripheral venous blood is collected and the indicators are detected. Then, through gradient-enhanced machine learning algorithms, combined with clinical sample data, population classification, and multi-dimensional labeling systems, basic weights are retrieved and iteratively optimized and validated multiple times to form a personalized weight scheme. A comprehensive score and sub-scores are calculated, and immune function levels are classified. Samples are collected regularly for repeated testing to construct a dynamic monitoring curve. Through linear regression and logistic regression, immune function trend prediction, disease risk quantification, and graded early warning are achieved. Personalized intervention suggestions are provided in conjunction with clinical symptoms, realizing accurate assessment and dynamic management of immune function throughout the entire process.

[0192] Furthermore, the process of periodically obtaining comprehensive and sub-item scores of immune function at different time points also includes the following steps:

[0193] Obtain the sub-score data of the three lines of immune defense of the subject at adjacent test nodes, and extract the timestamp of the previous test and the timestamp of the current test;

[0194] Then, based on the previous detection timestamp and the current detection timestamp, the actual time span of adjacent test nodes is calculated, and the signed sub-item score deviation of the corresponding immune defense in adjacent test nodes is extracted based on the sub-item score data.

[0195] Based on the actual time span and the preset time decay rate constant, a stress decay factor is calculated to characterize the degree of decay of the effects of acute physiological stress.

[0196] Based on the marked sub-item scoring deviation, the actual time span, the preset physiological buffer time constant, the basal metabolic compensation rate of the specific defense line, and the stress attenuation factor, the initial comprehensive immune function score obtained in this test is dynamically compensated in two directions. This filters out spurious fluctuations caused by non-substantial immune function decline or transient pseudo-hyperactivity, and calculates the calibrated comprehensive score. The corresponding calculation formula is as follows:

[0197]

[0198] in, This indicates the overall score after calibration; This represents the initial, uncalibrated comprehensive immune function score calculated using the weighted summation method. This represents the preset fluctuation adjustment constant; This indicates the hierarchical number of the immune defense line, specifically taking the value of a positive integer 1, 2, or 3; Indicates the first The first detection corresponds to the first Sub-score data for the immune defense line; Indicates the first The first detection corresponds to the first Sub-score data for the immune defense line; Corresponding to the marked itemized scoring deviation; Corresponding to the actual time span; Corresponding to the preset physiological buffer time constant; Corresponding to the The basal metabolic compensation rate of a specific line of defense of the immune system represents the expected basal fluctuation of normal scores per unit of day. Corresponding to the preset time decay rate constant; Corresponding to the stress attenuation factor;

[0199] Finally, based on the calculated calibrated comprehensive score, the dynamic monitoring curve of the subject's immune function is reset and updated; when the calibrated comprehensive score is determined to cross the preset risk defense threshold, the targeted immune regulation intervention control process for abnormal immune defense is automatically triggered, and a multi-dimensional immune traceability early warning report is output to the corresponding clinician port.

[0200] In this embodiment, to eliminate evaluation errors caused by irregular sampling and acute physiological stress in subjects, the steps for periodically obtaining comprehensive immune function scores and sub-scores at different time points specifically include the following steps: obtaining sub-score data of the three lines of immune defense for subjects at adjacent test nodes, and extracting the timestamps of the previous test and the current test; then calculating the actual time span of adjacent test nodes based on the previous test timestamp and the current test timestamp. Based on the sub-score data, the signed sub-score deviations of the corresponding immune defense lines in adjacent test nodes are extracted. Understandably, the system extracts precise physical time data from two consecutive venous blood samples taken from the subject using a clock module and calculates the difference to obtain the actual time span. Simultaneously, historical scores for each immune defense line stored in memory are retrieved, subtracted from the current score for the corresponding defense line, and the sign direction is preserved to obtain the signed sub-score bias. .

[0201] Furthermore, based on the actual time span and the preset time decay rate constant The stress attenuation factor, used to characterize the degree of decay of the effects of acute physiological stress, was calculated. Based on the aforementioned marked item scoring deviations... Actual time span Preset physiological buffer time constant Basal metabolic compensation rate of specific defense lines and stress attenuation factor The initial comprehensive score of immune function obtained from this test Two-way dynamic compensation is performed to filter out spurious fluctuations caused by non-substantial immune function decline or transient pseudo-hyperactivity. The calibrated comprehensive score is then calculated using the aforementioned dynamic compensation formula. .

[0202] Regarding the above dynamic compensation formula, we will now provide a detailed breakdown of the meaning, principle, and source of each parameter it contains:

[0203] Calibrated overall score The meaning is the final immune assessment score after the system algorithm filters out interfering variables; the principle is that it serves as the sole basis for determining whether to trigger the early warning mechanism and draw the dynamic monitoring curve, and it can truly reflect the immune homeostasis of the body after bidirectional compensation; the source is the dimensionless value calculated by the system.

[0204] Initial immune function comprehensive score The meaning is the original baseline score output after real-time quantitative PCR detection and weighted summation of the peripheral blood sample; the principle is to serve as the baseline starting value for dynamic compensation calculation; the source is the dimensionless value output by the system detection and weighted summation module.

[0205] Fluctuation adjustment constant This refers to the global scaling factor that controls the overall compensation strength; its principle is to prevent the calculated value of the compensation term in the formula from being too large, which would lead to an excessively high overall score after calibration. It produces reverse distortion; the source is an empirical constant pre-set based on a large-scale clinical sample retrospective test in the system backend, which is a dimensionless value.

[0206] Hierarchical sequence of immune defenses : Meaning is the number of the subject's three-level immune structure; the principle is to limit the scope of the summation function, corresponding to the first, second and third lines of defense; the source is the system's preset set of positive integers, with values ​​of 1, 2 and 3.

[0207] Signed item scoring bias This refers to the true magnitude and direction of the score change between two consecutive tests for a single line of immune defense; the principle is that if a subject experiences a transiently inflated score due to acute stress (i.e., the score of the first test is falsely high), then the score of the second test is falsely high. The first detection corresponds to the first Sub-score data of the immune defense line Greater than the The first detection corresponds to the first Sub-score data of the immune defense line If the formula compensation term is positive, the system performs subtraction to filter out pseudo-hyperactivity interference; conversely, if the score is transiently reduced due to short-term stress (i.e., the first...), the system performs subtraction to filter out pseudo-hyperactivity interference. The first detection corresponds to the first Sub-score data of the immune defense line Less than the The first detection corresponds to the first Sub-score data of the immune defense line When the formula compensation term is negative, the system executes the addition logic of negative times negative equals positive, compensating the non-substantial score drop back to the baseline; the source is obtained by the data processing unit by subtracting the two periods of test records and strictly preserving the sign direction, and is a dimensionless value.

[0208] Actual time span The meaning is the length of the time interval between two consecutive peripheral blood samples taken from a subject; the principle is that the system determines whether the score change is caused by acute stress or substantial degeneration. The shorter the time interval, the higher the probability that it is caused by stress; the source is the conversion of the time difference between the previous test timestamp and the current test timestamp extracted by the system clock module, and the dimension is days.

[0209] Physiological buffer time constant This refers to the body's underlying tolerance delay to abnormal stimuli within a very short period of time; the principle is that the time smoothing term in the denominator avoids issues arising from the actual time span. When the value approaches zero (e.g., continuous high-frequency blood sampling), the denominator becomes zero, leading to a system singularity overflow. This also indicates that the changes in the immune system within a very short period of time are not infinitely drastic, which is consistent with the biological homeostasis buffering law. The source is a small positive real constant (e.g., with a value of 0.01 to 0.05) set based on system limit tests, with the dimension of days.

[0210] Basal metabolic compensation rate of specific defense lines : Meaning the first The expected mean of the inherent score fluctuation of the first line of defense within a unit of time; the principle is that the first line of defense is metabolized quickly, while the third line of defense is metabolized very slowly. Based on this parameter, a compensatory weight is expected to be assigned to different lines of defense that matches their biological metabolic patterns; the source is obtained by the system through linear regression analysis of long-term longitudinal monitoring data in the health reference database, and the dimension is the ratio of the dimensionless value to the number of days.

[0211] Time decay rate constant : Meaning is a fixed parameter that controls the rate at which the stress effect fades over time; the principle is to calibrate the slope of the time function of the body's self-repair function; the source is a preset constant in clinical pharmacokinetics and physiological statistics, and the dimension is the reciprocal of the number of days.

[0212] Stress attenuation factor The meaning is a proportional variable that characterizes the interference effect of acute physiological stress on immune scores, which decreases exponentially over time; the principle is to construct a negative exponential decay model using the natural logarithm base, when the actual time span... When the stress decay factor approaches 0, it is relatively small. When the value approaches 1, the system applies maximum intensity of bidirectional compensation; when the actual time span... When the stress decay factor approaches infinity, When the score approaches zero, the compensation item becomes invalid, and the system determines that the current score represents a substantial change; the source is based on the actual time span. With time decay rate constant The result, obtained through exponential function calculation, is a dimensionless value.

[0213] Finally, a comprehensive score based on calculation and calibration. The dynamic monitoring curve of the subject's immune function was reset and updated; the calibrated comprehensive score was then determined. When a preset risk defense threshold is exceeded, a targeted immune regulation intervention control process against the abnormal immune defense is automatically triggered, and a multi-dimensional immune tracing and early warning report is output to the corresponding clinician. Specifically, the data processor compares the calibrated comprehensive score. If the comparison result with the data in the threshold memory is lower than the low-risk defense threshold, a trigger level signal is generated to the intervention control module to execute the instruction.

[0214] Furthermore, after determining the weight coefficients for each indicator and each immune defense line, an adaptive balancing mechanism for cross-defense weights is included to correct the data masking effect caused by imbalances in the defense cascade. This mechanism specifically includes the following steps:

[0215] Obtain the sub-scores of adjacent immune defenses in the three lines of immune defense, and extract the absolute difference between the sub-scores of adjacent immune defenses;

[0216] Then, based on the absolute difference and the preset full score standard value of the defense line, the relative deviation ratio between the adjacent immune defense lines is calculated;

[0217] The relative deviation ratio is then compared with a preset cascade imbalance threshold. If the relative deviation ratio is greater than the preset cascade imbalance threshold, it is determined that the current subject has an internal cascade imbalance in the immune defense network, and the cross-defense weight adaptive balancing mechanism is automatically triggered.

[0218] The specific logic for implementing the cross-defense line weight adaptive balancing mechanism includes:

[0219] First, extract the core specific peripheral blood assessment indicators that cause the highest relative deviation rate;

[0220] Then, a generative adversarial network is constructed, and the core specific peripheral blood assessment indicators, the relative deviation ratio, and the initial personalized weighting scheme are simultaneously input into the generative adversarial network;

[0221] Using the generator model contained within the generative adversarial network, candidate inhibition decay factors with different decay step sizes are generated for the core specific peripheral blood assessment indicators.

[0222] Simultaneously, using the discriminator model contained within the generative adversarial network, and based on the real balanced clinical weight distribution retrieved from the health reference database, the candidate inhibition decay factor is judged for goodness of fit. Through multiple adversarial game iterations between the generator model and the discriminator model, until the discriminator model can no longer distinguish the difference between the adjusted weight and the real balanced clinical weight, the optimal inhibition decay factor is output.

[0223] Based on the optimal inhibition attenuation factor, the weight coefficients corresponding to the core specific peripheral blood assessment indicators are adjusted downward for the second time, while the weight coefficients of the remaining specific peripheral blood assessment indicators that have not caused deviation are adjusted upward proportionally, thereby updating the personalized weight scheme.

[0224] Finally, the weighted summation step is re-executed based on the updated personalized weighting scheme to expose potential immunodeficiency states and recalculate the comprehensive immune function score. At the same time, a cascade imbalance depth warning instruction is generated based on the recalculated comprehensive immune function score. The cascade imbalance depth warning instruction and the underlying immunodeficiency state data are synchronously sent to the corresponding clinician terminal interface to forcibly trigger a special medical intervention operation for hidden immune imbalance.

[0225] In this embodiment, after determining the weight coefficients of each indicator and each immune defense line, an adaptive balancing mechanism for cross-defense line weights is also included to correct the data masking effect caused by the imbalance of the defense line cascade. Specifically, the following steps are included:

[0226] The system acquires the sub-scores of adjacent immune defenses within the three lines of defense and extracts the absolute differences between these scores. Based on these absolute differences and preset full-score standards for each defense line, it calculates the relative deviation ratio between adjacent immune defenses. This relative deviation ratio is then compared to a preset cascade imbalance threshold. If the relative deviation ratio exceeds the preset threshold, an internal cascade imbalance within the current subject's immune defense network is identified, and an adaptive balancing mechanism for cross-defense weights is automatically triggered. Essentially, the calculation module performs a logic gate-level comparison of the relative deviation ratio with a preset constant. This logic operation identifies whether there is a data masking phenomenon in the system data where abnormally high scores in local defense lines mask overall immune deficiencies.

[0227] Furthermore, the specific logic for implementing the cross-defense weight adaptive balancing mechanism includes: first, extracting the core-specific peripheral blood assessment index that results in the highest relative deviation rate; then constructing a generative adversarial network (GAN), and simultaneously inputting the core-specific peripheral blood assessment index, the relative deviation rate, and the initial personalized weight scheme into the GAN.

[0228] In the specific network structure organization and data flow processing, the generator model consists of a feedforward neural network containing an input layer, multiple hidden layers, and an output layer. Its input layer receives combined feature data including a tensor of the current score of the core-specific peripheral blood assessment indicator and a scalar of the relative deviation ratio. To conform to the mathematical law of weight decay, the generator output layer uses the sigmoid activation function, mapping values ​​strictly limited to... The candidate suppression decay factor variable is defined within the interval. The discriminator model is configured as a multilayer perceptron (MLP) structure to perform a binary classification task. The input sample vector it receives is divided into two groups: the first group is the current one-dimensional weight vector modulated by the candidate suppression decay factor, and the second group is the standard one-dimensional weight distribution vector within the healthy reference database. The discriminator performs cross-entropy loss calculation and backpropagates the gradient parameters. When the number of iterations meets the set threshold or the difference in the loss function converges to within the allowable error range, iterative training stops, the network parameters output by the generator at this moment are locked, and the corresponding output value is fixed as the optimal suppression decay factor.

[0229] Specifically, the data processor then performs dynamic weight reconstruction on the core-specific peripheral blood assessment indicators based on the optimal suppression and decay factor. To ensure the normalization of the global weight coefficients, the system adopts the following strict adaptive balancing reassignment formula to update the personalized weight scheme:

[0230] Formula for adjusting the weight of core indicators:

[0231]

[0232] Formula for proportionally adjusting remaining indicators:

[0233]

[0234] Regarding the above weight reassignment formula, we will now provide a detailed breakdown of the meaning, principle, and source of each parameter it contains:

[0235] Optimal suppression attenuation factor The term refers to the attenuation multiplier of the generative adversarial network's output targeting the core bias indicator; the principle is based on a realistically balanced clinical weight distribution, forcibly suppressing abnormally high weights that mask immune deficiencies; the source is the output scalar of the generator model after Sigmoid activation, with dimensions of... Dimensionless values ​​within an interval.

[0236] Initial core indicator weights The meaning is the original weight coefficient of the core specific peripheral blood assessment indicator before the warning is triggered; the principle is to serve as the target base to be attenuated; the source is the initial personalized weighting scheme.

[0237] Updated core metric weights The meaning is the new weight after dimensionality reduction by the attenuation factor; the principle is to reduce the weight ratio of the inflated indicator in the total score and expose the underlying defects; the source is obtained by multiplying the formula.

[0238] Initial Residual Index Weights The meaning is the original weight of other specific peripheral blood assessment indicators that have not triggered abnormal deviation ratios; the principle is to serve as the base that needs to be compensated and amplified; the source is the initial personalized weighting scheme.

[0239] Updated residual indicator weights This means that after deducting the new weight share of the core indicators, the remaining weight space is redistributed according to the original internal ratio, resulting in a new weight. The principle is to ensure that the sum of the weights of all indicators in the system always remains constant. To maintain the mathematical completeness of the weighted summation algorithm; the source is the formula calculated by the above proportional upward adjustment.

[0240] Finally, based on the updated personalized weighting scheme, the weighted summation step is re-executed to recalculate and output the current true comprehensive immune function score. The system transmits this recalculated comprehensive immune function score to the comparator. If an alarm condition is triggered, a cascaded imbalance deep warning command data packet is generated according to the communication protocol. This data packet and the underlying immune deficiency state source data it carries are sent to the clinician's terminal, forcibly triggering a specific medical intervention operation for occult immune imbalance.

[0241] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, or article that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, or article.

[0242] 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. A method for scoring immune function in peripheral blood based on tertiary structure of immunity, characterized by, Includes the following steps: Step 1: Construct a peripheral blood immune assessment index system based on the three-tiered immune structure, which includes the first line of defense, the second line of defense, and the third line of defense, with specific peripheral blood assessment indicators set for each line of defense; Step 2: Collect peripheral venous blood from the subjects and use real-time quantitative PCR to quantitatively detect each assessment indicator and obtain the actual detection value of each indicator; Step 3: Using the gradient enhancement machine learning algorithm, combined with clinical sample data, weights are assigned to each indicator in the three-level indicator system, the weight coefficients of each indicator and each immune defense are determined, and the weight coefficients are optimized for different populations to form a personalized weight scheme. Step 4: Based on the standardized scores of each indicator and their corresponding weight coefficients, calculate the comprehensive immune function score using the weighted summation method, and calculate the sub-scores for the three lines of immune defense; Step 5: Based on the comprehensive scoring results, peripheral blood immune function is divided into multiple levels to provide guidance for clinical diagnosis and health management; Step 1: Constructing a peripheral blood immune assessment index system based on the three-tiered immune structure, including the following steps: To construct the first line of defense indicators, soluble indicators in peripheral blood that are directly related to mucosal barrier function were selected, including secretory immunoglobulin A and mucosa-associated epithelial cytokines. To construct indicators for the second line of defense, we selected innate immune cells and related cytokines in peripheral blood, including the percentage and activity of natural killer cells, macrophage markers, and innate immune cytokines. To construct the third line of defense indicators, adaptive immune cell subsets and related molecules in peripheral blood were selected, including the percentages of CD3+ T cells, CD4+ T cells, and CD8+ T cells, the CD4+ / CD8+ ratio, the percentage of B cells, immunoglobulins, and the T cell receptor diversity index.

2. The peripheral blood immune function scoring method based on the tertiary structure of immunity according to claim 1, characterized in that, It also includes the following steps: Step 6: Collect peripheral blood from the subjects periodically, repeat steps 2-5, and obtain the comprehensive immune function score and sub-scores at different time points; Construct dynamic monitoring curves to analyze the changing trends of immune function; Based on clinical symptoms, personalized suggestions for immune regulation or treatment adjustments are provided to the subjects.

3. The peripheral blood immune function scoring method based on the tertiary structure of immunity according to claim 1, characterized in that, Step 3: Using a gradient-enhancing machine learning algorithm, combined with clinical sample data, weights are assigned to each indicator in the three-level indicator system, including the following steps: Clinical sample data were collected, including peripheral blood test values ​​of various indicators and clinical immune function assessment results of the subjects. The clinical immune function assessment results were determined by clinicians based on the subjects' infection frequency, disease recovery speed and immune-related complications, and were divided into four levels: normal, mildly deficient, moderately deficient and severely deficient. The subjects were divided into four basic categories: healthy people, immunodeficient people, cancer people, and people with autoimmune diseases. Multidimensional characteristic information of the subjects was collected, including age, gender, history of underlying diseases and medication, to form a multidimensional labeling system. The classification information and feature labels are integrated to form training datasets and validation datasets; Construct the input feature matrix X and output label vector Y of the XGBoost model, and set the hyperparameters of the XGBoost model; In this model, each row of the input feature matrix X corresponds to a subject sample, each column corresponds to a specific detection value of a peripheral blood assessment indicator, and the output label vector Y corresponds to the clinical immune function assessment level. The learning rate eta is 0.1, the maximum tree depth max_depth is 5, the number of iterations n_estimators is 100, and the subsample ratio subsample is 0.

8. The training dataset is input into the XGBoost model, and the clinical immune function assessment level is used as the supervision label for training. During training, the model calculates the importance score of each indicator based on the frequency of its occurrence in the decision tree split nodes and the reduction in impurity. The weighted sum of the importance scores of this indicator in all trees is divided by the total number of splits to obtain the standardized weight coefficient. For different population groups, the weight coefficients are optimized separately, and a transfer learning strategy is adopted. Using the basic weight vector of healthy people as the initial benchmark, the XGBoost model is fine-tuned for immunodeficient people, cancer patients and autoimmune disease patients using training samples of the corresponding populations, and the weights of each indicator are recalculated. The accuracy of the weighted coefficient prediction was evaluated using a validation dataset. The validation results were compared with the clinical immune function assessment results, and the Spearman correlation coefficient between the weighted score and the clinical assessment results was calculated. If the correlation coefficient is lower than the preset threshold range, adjust the hyperparameters of XGBoost and retrain until the requirements are met. If the weighted score is consistent with the clinical assessment results, it indicates that the modulation direction is correct, thus forming a personalized weight allocation scheme.

4. The peripheral blood immune function scoring method based on the tertiary structure of immunity according to claim 3, characterized in that, Step 3: Optimize the weighting coefficients for different groups of people, including the following steps: The generated population-specific weighting schemes are validated at multiple levels. The prediction accuracy of each population weighting scheme within the population is tested using reserved validation set data. Cross-population validation was conducted to verify the specificity of the weighting scheme; Newly enrolled subjects were selected and assessed using both traditional and this method. The results of the two assessments were compared to verify its clinical applicability and accuracy. If it is found that the characteristics of newly enrolled subjects cannot be classified into any existing population category, and their immune indicators show unique regularity, the new population identification process will be automatically triggered. Through cluster analysis and expert evaluation, we can confirm whether there are new population subtypes. If they are confirmed, we can include the population in the model training scope, establish an independent basic weight vector for it, and form a complete weight scheme library that is dynamically modulated according to individual characteristics.

5. The peripheral blood immune function scoring method based on the tertiary structure of immunity according to claim 2, characterized in that, Step 6: Construct dynamic monitoring curves to analyze the changing trends of immune function, including the following steps: We collected immune function-related scores and clinical baseline information of the subjects at different time points, and used a linear regression algorithm to fit the monitoring curves to establish a dynamic prediction model. Input the latest scoring data to predict the trend and fluctuation range of immune function changes within a specified future time period; Collect clinical sample data, including disease incidence corresponding to different immune function scores; By using logistic regression analysis, a correlation model between the scoring results and the risk of developing immune-related diseases is established, and the disease incidence probability for the corresponding scoring interval is output. By combining the predicted trend, the probability of disease incidence, and the warning threshold, an early warning will be automatically triggered when the predicted trend value is lower than or higher than the preset threshold.

6. The peripheral blood immune function scoring method based on the tertiary structure of immunity according to claim 5, characterized in that, Combining predicted trends, disease risk probability, and early warning thresholds, the following steps are included: Set early warning thresholds, including thresholds for predicted trend values ​​of immune function and thresholds for the probability of disease incidence; Set up corresponding early warning methods according to the graded early warning standards; The system automatically records the trigger time, warning level, core trigger indicators, target audience, and intervention feedback results for each warning, creating a warning traceability archive.

7. The peripheral blood immune function scoring method based on the tertiary structure of immunity according to claim 1, characterized in that, Step 2 also includes the following steps: By collecting peripheral blood samples from healthy subjects of different ages and genders, specific peripheral blood assessment indicators were detected. The normal reference range for each indicator is calculated, and the actual test values ​​of each indicator are converted into standardized scores using standardized formulas to eliminate the dimensional differences between different indicators and form a health reference database.

8. The peripheral blood immune function scoring method based on the tertiary immune structure according to claim 2, characterized in that the step of periodically obtaining the comprehensive immune function score and sub-scores at different time points further includes the following steps: Obtain the sub-score data of the three lines of immune defense of the subject at adjacent test nodes, and extract the timestamp of the previous test and the timestamp of the current test; Then, based on the previous detection timestamp and the current detection timestamp, the actual time span of adjacent test nodes is calculated, and the signed sub-item score deviation of the corresponding immune defense in adjacent test nodes is extracted based on the sub-item score data. Based on the actual time span and the preset time decay rate constant, a stress decay factor is calculated to characterize the degree of decay of the effects of acute physiological stress. Based on the signed sub-item scoring deviation, the actual time span, the preset physiological buffer time constant, the basal metabolic compensation rate of the specific defense line, and the stress attenuation factor, the initial comprehensive immune function score obtained in this test is dynamically compensated in both directions. This filters out spurious fluctuations caused by non-substantial immune function decline or transient pseudo-hyperactivity, and calculates the calibrated comprehensive score. The corresponding calculation formula is as follows: ； in, This indicates the overall score after calibration; This represents the initial, uncalibrated comprehensive immune function score calculated using the weighted summation method. This represents the preset fluctuation adjustment constant; This indicates the hierarchical number of the immune defense line, specifically taking the value of a positive integer 1, 2, or 3; Indicates the first The first detection corresponds to the first Sub-score data for the immune defense line; Indicates the first The first detection corresponds to the first Sub-score data for the immune defense line; Corresponding to the marked sub-item scoring deviation; Corresponding to the actual time span; Corresponding to the preset physiological buffer time constant; Corresponding to the The basal metabolic compensation rate of a specific line of defense of the immune system represents the expected basal fluctuation of normal scores per unit of day. Corresponding to the preset time decay rate constant; Corresponding to the stress attenuation factor; Finally, based on the calculated calibrated comprehensive score, the dynamic monitoring curve of the subject's immune function is reset and updated; when the calibrated comprehensive score is determined to cross the preset risk defense threshold, the targeted immune regulation intervention control process for abnormal immune defense is automatically triggered, and a multi-dimensional immune traceability early warning report is output to the corresponding clinician port.

9. The peripheral blood immune function scoring method based on the three-tiered immune structure according to claim 1, characterized in that, after determining the weight coefficient of each indicator and each immune defense line, it further includes a cross-defense line weight adaptive balancing mechanism to correct the data masking effect caused by the imbalance of the defense line cascade, specifically including the following steps: Obtain the sub-scores of adjacent immune defenses in the three lines of immune defense, and extract the absolute difference between the sub-scores of adjacent immune defenses; Based on the absolute difference and the preset full score standard value of the defense line, the relative deviation ratio between the adjacent immune defense lines is calculated. The relative deviation ratio is then compared with a preset cascade imbalance threshold. If the relative deviation ratio is greater than the preset cascade imbalance threshold, it is determined that the current subject has an internal cascade imbalance in the immune defense network, and the cross-defense weight adaptive balancing mechanism is automatically triggered. The specific logic for implementing the cross-defense line weight adaptive balancing mechanism includes: First, extract the core specific peripheral blood assessment indicators that cause the highest relative deviation rate; Then, a generative adversarial network is constructed, and the core specific peripheral blood assessment indicators, the relative deviation ratio, and the initial personalized weighting scheme are simultaneously input into the generative adversarial network; Using the generator model contained within the generative adversarial network, candidate inhibition decay factors with different decay step sizes are generated for the core specific peripheral blood assessment indicators. Simultaneously, using the discriminator model contained within the generative adversarial network, and based on the real balanced clinical weight distribution retrieved from the health reference database, the candidate inhibition decay factor is judged for goodness of fit. Through multiple adversarial game iterations between the generator model and the discriminator model, until the discriminator model can no longer distinguish the difference between the adjusted weight and the real balanced clinical weight, the optimal inhibition decay factor is output. Based on the optimal inhibition attenuation factor, the weight coefficients corresponding to the core specific peripheral blood assessment indicators are adjusted downward for the second time, while the weight coefficients of the remaining specific peripheral blood assessment indicators that have not caused deviation are adjusted upward proportionally, thereby updating the personalized weight scheme. Finally, the weighted summation step is re-executed based on the updated personalized weighting scheme to expose potential immunodeficiency states and recalculate the comprehensive immune function score. At the same time, a cascade imbalance depth warning instruction is generated based on the recalculated comprehensive immune function score. The cascade imbalance depth warning instruction and the underlying immunodeficiency state data are synchronously sent to the corresponding clinician terminal interface to forcibly trigger a special medical intervention operation for hidden immune imbalance.

Images