Method for constructing six-channel dialectical classification model based on NEI system and application thereof

Abstract

The method comprises the following steps: (1) acquiring ChE, SCFA and NE detection data of patients with different six-channel syndrome types, dividing the data of any syndrome type into a training set and a test set, and standardizing the data of the training set; (2) standardized data of the training set construct a classification model through a machine learning algorithm, the model satisfies: logit (P) = a + b1 * ChE + b2 * SCFA + b3 * NE, P represents the probability of a certain six-meridian syndrome type, and a, b1, b2 and b3 are parameters to be optimized; using logistic regression or a neural network to train the model, and training to obtain parameters; and (3) verifying, evaluating and optimizing the model of any syndrome type by using the test set to obtain a classification model. According to the method, the corresponding mapping relation between the three subsystems of the NEI and the six-channel syndrome type is established for the first time, ChE, NE and SCFA serve as quantification bases, accuracy and repeatability are high, and the technical problem that in the prior art, six-channel dialectical classification lacks quantitative indexes is solved.

Description

Technical Field

[0001] The present invention belongs to the interdisciplinary research field of traditional Chinese medicine and western medicine, and specifically relates to a method for constructing a six-channel syndrome differentiation classification model based on the three major systems of neuro-endocrine-immune (NEI) and its application. Background Art

[0002] In Traditional Chinese Medicine (TCM), syndrome differentiation refers to the identification of the nature, cause, location, and pathogenesis of a disease through a comprehensive analysis of the patient's symptoms, signs, and medical history. This is typically done through observation, auscultation, inquiry, and pulse diagnosis. This information is then comprehensively analyzed based on TCM theory to identify the nature, cause, location, and pathogenesis of the disease. Finally, the disease is categorized into specific syndrome types, such as "wind-cold cold syndrome" or "liver depression and qi stagnation syndrome."

[0003] Six-channel syndrome differentiation categorizes illnesses into six stages: Taiyang, Yangming, Shaoyang, Taiyin, Shaoyin, and Jueyin. Each stage has its own unique pathogenesis and clinical manifestations, reflecting the pathological changes and treatment principles of the disease at different stages.

[0004] Currently, TCM Six-Jing Syndrome Differentiation relies on empirical judgment and lacks modern quantitative standards. This subjectivity leads to individual differences in diagnostic results, making objectivity and standardization difficult to achieve.

[0005] Traditional Chinese Medicine's Five Elements theory emphasizes mutual generation and mutual restraint, reflecting the principles of systems theory. Modern medicine's "neuro-endocrine-immune system" (NEI) also embodies the principles of systems regulation.

[0006] While modern medicine's neuro-endocrine-immune (NEI) system shares similarities in its regulatory principles with Traditional Chinese Medicine (TCM) systems theory, a complete correspondence between the NEI system and the Six Meridians has yet to be established. Traditional methods struggle to effectively integrate TCM philosophy with modern medical theory, resulting in limited diagnostic objectivity and accuracy.

[0007] In existing technologies, the integration of TCM syndrome differentiation with modern medical testing methods is insufficient, and there is a lack of effective quantitative indicators and data model support. This makes TCM syndrome differentiation difficult to be widely accepted and applied in the modern medical system.

[0008] It is necessary to combine the Six-Channel Syndrome Differentiation of Traditional Chinese Medicine with the NEI system of modern medicine, develop a method that can quantify the results of dialectics, and construct a Six-Channel Syndrome Differentiation classification model based on the NEI system to provide an accurate, objective and quantifiable basis for TCM dialectics. Summary of the Invention

[0009] The purpose of this invention is to combine Traditional Chinese Medicine, neuroscience, endocrinology, and immunology, and to establish a corresponding Traditional Chinese Medicine Six-Channel Syndrome Differentiation system by integrating the physiological states of neurotransmitters and hormones such as acetylcholine, norepinephrine, and thyroxine with the NEI system of modern medicine. The NEI system is used to quantify the Six-Channel Syndrome Differentiation of Traditional Chinese Medicine, and based on data from the three representative energy indicators of cholinesterase, norepinephrine, and total intestinal short-chain fatty acid value, the objectivity of Traditional Chinese Medicine Syndrome Differentiation is achieved. This optimizes clinical diagnosis, combines the Six-Channel theory, provides a more accurate disease classification model, and improves the targeted nature of treatment plans.

[0010] The technical solution adopted in the present invention is:

[0011] The present invention provides a method for constructing a Six-Jing Syndrome Differentiation classification model based on the NEI system, wherein the method establishes a matching relationship between the Six-Jing Syndrome Differentiation and the physiological detection data of the NEI system, and the method comprises the following steps:

[0012] (1) Obtain the ChE, SCFA, and NE test data of patients with different six-channel syndromes, annotate and classify the data according to the six-channel syndromes, divide the data of any syndrome into training and test sets, and preprocess the training set data by cleaning, processing missing values, and standardizing the raw data to obtain the standardized data of ChE, SCFA, and NE;

[0013] The missing value processing is to fill or eliminate the missing values with the mean value;

[0014] The standardization process is to perform Z-score standardization (mean variance normalization) on the data after cleaning and missing value processing:

[0015] The formula for Z-score standardization is:

[0016] Where X is the original data value;

[0017] μ is the mean of the data set;

[0018] σ is the standard deviation of the data set;

[0019] X' is the normalized data value.

[0020] (2) The standardized data of the training set is used to construct a classification model through a machine learning algorithm, and the model satisfies:

[0021] logit(P)=a+b1×ChE+b2×SCFA+b3×NE

[0022] P represents the probability of a certain Six Meridians syndrome type, that is, the probability of any of the Taiyang, Yangming, Shaoyang, Taiyin, Shaoyin, and Jueyin syndrome types;

[0023] ChE is the normalized data for cholinesterase level (mg / L).

[0024] SCFA is the normalized value of the total acidity of intestinal short-chain fatty acids (mmol / L).

[0025] NE is the normalized data for the concentration of norepinephrine (ng / mL).

[0026] a, b1, b2, b3 are the parameters to be optimized;

[0027] Use logistic regression or neural network to train the model and obtain the classification model parameters a, b1, b2, b3 of the syndrome type;

[0028] Preferably, in the logistic regression algorithm, the parameters (a, b1, b2, b3) are solved using maximum likelihood estimation or gradient descent method.

[0029] The training set is generally 80% of the data volume, and the test set is generally 20% of the data volume.

[0030] (3) The data of the test set is standardized, and the model of any syndrome type is verified, evaluated and optimized to obtain a classification model.

[0031] The test set is subjected to standardization processing, which refers to standardizing the data of the test set according to the standardization parameters of the training set to obtain the ChE, SCFA, and NE standardized data of the test set.

[0032] Furthermore, the data of the training set were standardized using Z-score to obtain the mean μ and standard deviation σ. The data of the test set were Z-score standardized according to the μ and σ of the training set to obtain the standardized data of ChE, SCFA, and NE of the test set, which were then used for verification and evaluation of the classification model.

[0033] Furthermore, in step (3), the validation, evaluation, and optimization steps are as follows: using the standardized data of the test set to calculate the accuracy, precision, and specificity, draw the ROC curve, calculate the AUC, and perform validation and evaluation on the model, requiring the AUC and accuracy of the classification model to meet the preset requirements; if the AUC and accuracy of the model do not meet the preset requirements, re-iterate the training model until the AUC and accuracy of the classification model meet the preset requirements.

[0034] Generally speaking, the preset requirements for AUC and accuracy of classification models are:

[0035] AUC ≥ 0.70 and accuracy ≥ 0.70, preferably AUC ≥ 0.80 and accuracy ≥ 0.75;

[0036] Furthermore, in step (3), if conditions permit, independent sample data from other medical institutions or different populations are collected as a data set, and the same model is applied to the external data set for verification and evaluation;

[0037] Compare the predicted results with the actual dialectical results to verify the robustness of the model in different samples and environments;

[0038] If the external validation results are close to the internal validation results, it further proves the generalizability and practicality of the model.

[0039] Furthermore, the present invention may also include step (4), continuously collecting ChE, SCFA, and NE test data of patients with different six-channel syndromes, and regularly updating and correcting model parameters and thresholds to ensure their long-term applicability and accuracy.

[0040] The present invention has been found through research:

[0041] The cholinergic system is closely related to Jueyin syndrome and Shaoyang syndrome, and the detection indicator is cholinesterase (ChE);

[0042] The adrenergic system is closely related to Shaoyin syndrome and Taiyang syndrome, and the detection indicator is norepinephrine (NE);

[0043] The thyrotropin system is closely related to Taiyin syndrome and Yangming syndrome, and the detection indicator is the total intestinal short-chain fatty acid (SCFA).

[0044] The reasons for selecting these three as the judgment subjects are: cholinesterase (ChE) reflects the acetylcholine (ACh) level of the cholinergic system. Overexcitement of the cholinergic system manifests as ganglion excitation, muscle spasms, sweating, increased glandular secretion, etc., which is closely related to Jueyin syndrome; enhanced activity of the cholinergic system manifests as bitter fullness in the chest and flanks, alternating chills and fever, irritability and vomiting, etc., which is closely related to Shaoyang syndrome.

[0045] Norepinephrine (NE) levels reflect the adrenergic system (Shaoyin, Taiyang). Insufficient norepinephrine secretion leads to decreased blood pressure, fatigue, and drowsiness, which is closely related to Shaoyin syndrome. Increased norepinephrine secretion manifests as peripheral vasoconstriction, pale skin, and aversion to cold, which is closely related to Taiyang syndrome.

[0046] The level of total short-chain fatty acid (SCFA) in the intestine is related to thyroxine (T3, T4), the core hormone of the thyroxine system (Taiyin and Yangming). Reduced thyroxine secretion and low total acid level are manifested as abdominal distension, loss of appetite, loose stools, fatigue, etc., which are closely related to Taiyin syndrome; increased thyroxine level and high total acid level increase the body's heat production, which are manifested as high fever, profuse sweating, dry skin, etc., which are closely related to Yangming syndrome.

[0047] Based on the above reasons, the present invention selects three physiological indicators, namely cholinesterase (ChE), norepinephrine (NE) and total intestinal short-chain fatty acid value (SCFA), as the core basis for judging the six meridian syndromes. Although from a physiological point of view, the correlation between cholinesterase (ChE) and Jueyin and Shaoyang syndromes is more significant, norepinephrine (NE) is more closely related to Shaoyin and Taiyang syndromes, and SCFA can better reflect the characteristics of Taiyin and Yangming syndromes, it must be pointed out that there is a complex interaction between the three major systems of neuro-endocrine-immunity. This interaction means that a single indicator is not enough to independently determine the six meridian syndromes, but all three indicators need to be considered comprehensively. Therefore, when constructing the model, the judgment of each six meridian syndrome is based on the linear combination of the three indicators ChE, NE and SCFA, and the accuracy and reliability of the judgment are improved through comprehensive analysis of the model. This method of multi-indicator comprehensive judgment ensures the comprehensiveness and scientificity of the six meridian syndrome judgment, and avoids the deviation that may be caused by relying on only a single indicator.

[0048] The present invention also provides a method for calculating the probability of six meridian syndrome types based on the six meridian syndrome differentiation classification model constructed by the NEI system, the method being:

[0049] After the ChE, SCFA, and NE test data of the test samples are standardized, they are input into the Six-Channel Syndrome Differentiation classification model. Each syndrome type corresponds to an independent classification model, and the model outputs the probability P of the syndrome type. If P ≥ the set threshold, the syndrome type is judged to be positive (the probability of the syndrome type being positive is high); if P < the set threshold, the syndrome type is judged to be negative (the probability of the syndrome type being negative is high).

[0050] Set the threshold to 0.3-0.7, preferably 0.5;

[0051] Furthermore, each syndrome type corresponds to its own set threshold value, which can be the same or different. Generally speaking, the set threshold value P is generally preferably 0.5, but the size can also be adjusted to adjust the sensitivity and specificity. For example, to improve the detection rate (sensitivity), the threshold value can be appropriately lowered from 0.5 (such as 0.4, 0.3, etc.); if you want to improve the specificity, you can increase the threshold value (such as 0.6, 0.7, etc.).

[0052] Furthermore, the ChE, SCFA, and NE detection data of the sample to be tested are standardized according to the standardization parameters of the training set. Specifically, Z-score standardization can be performed according to μ and σ of the training set to obtain the ChE, SCFA, and NE standardized data of the sample to be tested, and then input into the classification model for prediction calculation.

[0053] The present invention also provides the application of the six-channel syndrome differentiation classification model constructed based on the NEI system in non-diagnostic auxiliary judgment of six-channel syndrome types.

[0054] The present invention also provides a six-channel syndrome differentiation classification system based on the NEI system, the system comprising:

[0055] The data processing module is used to obtain ChE, SCFA, and NE test data from patients with different six-channel syndromes and perform preprocessing: cleaning, missing value processing, and standardization of the original data to obtain ChE, SCFA, and NE standardized data; annotating and classifying the data according to the six-channel syndrome type, and dividing the data of any syndrome type into training and test sets;

[0056] The model training module is used to train the model using logistic regression or neural network data.

[0057] The model satisfies:

[0058] logit(P)=a+b1×ChE+b2×SCFA+b3×NE

[0059] P represents the probability of a certain Six Meridians syndrome type, that is, the probability of any of the Taiyang, Yangming, Shaoyang, Taiyin, Shaoyin, and Jueyin syndrome types;

[0060] ChE is the normalized data of cholinesterase level in mg / L;

[0061] SCFA is the standardized data of the total acid value of intestinal short-chain fatty acids, with the unit of mmol / L;

[0062] NE is the normalized data of norepinephrine concentration, in ng / mL;

[0063] a, b1, b2, b3 are the parameters to be optimized;

[0064] The classification model parameters a, b1, b2, and b3 of this syndrome are obtained through machine algorithm training;

[0065] Model validation module: used to input test set data into the model of any syndrome type, perform validation evaluation and parameter optimization, and obtain a classification model with high accuracy and specificity;

[0066] Six-Jing Dialectical Classification Module:

[0067] It is used to standardize the ChE, SCFA, and NE test data of the sample to be tested and input them into the Six-Channel Syndrome Differentiation classification model. Each syndrome type corresponds to an independent classification model, and the model outputs the probability P of the syndrome type. If P ≥ the set threshold, the syndrome type is judged to be positive, and if P < the set threshold, the syndrome type is judged to be negative.

[0068] The beneficial effects of the present invention are:

[0069] (1) Objective quantification: This invention establishes for the first time the corresponding mapping relationship between the three major subsystems of NEI and the six meridian syndrome types, using three biochemical indicators (ChE, NE, and SCFA) as the basis for quantification, thus solving the technical problem that the dialectical diagnosis in the existing technology mainly relies on the subjective experience of physicians and lacks quantitative indicators.

[0070] (2) High accuracy and repeatability:

[0071] By obtaining the optimal classification threshold through artificial intelligence algorithms, the impact of individual differences on syndrome differentiation is reduced. The accuracy, precision, specificity, and recall rates are all high, with an AUC of over 0.8.

[0072] (3) The system of the present invention continuously collects clinical NEI system data and six-channel dialectical data, regularly updates and corrects model parameters and thresholds, and ensures the long-term applicability and accuracy of the classification model.

[0073] The present invention selects blood biochemical clinical data cholinesterase, norepinephrine and fecal metabolite SCFA to reflect the acetylcholine (ACh) level of the cholinergic system, the adrenergic system level and the thyroxine system level, respectively, which correspond to the three major systems of neuro-endocrine-immune system. The three biochemical indicators of cholinesterase, NE and SCFA are used to achieve objective quantification, construct a six-channel syndrome differentiation classification model, optimize the classification threshold through artificial intelligence algorithm, realize the objectivity and quantification of TCM syndrome differentiation, provide a scientific basis for clinical diagnosis, and provide technical support for the further integration of TCM and modern medicine.

[0074] It should be emphasized that the present invention is non-diagnostic in nature. This method is mainly used to study the correspondence between physiological data and six-channel syndrome differentiation. The results obtained are intermediate data and can be further used for clinical reference. It does not directly lead to the final diagnosis of the disease. Clinical medical behavior should be completed under the guidance of a practicing physician. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 This is the ROC curve of the "Yangming disease" binary classification model in Example 1 on the test set.

[0076] Figure 2 This is the ROC curve of the “Yangming disease” binary classification model of Example 1 on external data.

[0077] Figure 3 This is a box plot comparison of ChE, NE, and SCFA indicators in the training set (40 cases) and validation set (10 cases).

[0078] Figure 4 Box plot comparison of ChE, NE, and SCFA indicators between the training set (40 cases) and external cases (20 cases). DETAILED DESCRIPTION

[0079] The technical solution of the present invention is further described below with reference to specific embodiments, but the protection scope of the present invention is not limited thereto:

[0080] Example 1

[0081] (1) Collect data and label it

[0082] Data Collection: Standardized clinical testing methods are used to obtain data on ChE, SCFA, NE, and other indicators in patient blood or other body fluids. The collection process is strictly carried out in accordance with national and industry standards to ensure data accuracy and reproducibility.

[0083] Labeling: Based on the theory of "Six-Jing Syndrome Differentiation" in Traditional Chinese Medicine, several senior Chinese medicine practitioners conducted comprehensive diagnosis on the patients and labeled the six-Jing syndrome types. The collected cases are shown in the following table:

[0084]

[0085] (2) Data preprocessing: Clean the raw data, handle missing values, and standardize them to ensure the quality of the training data entering the model. For missing values, mean filling or elimination can be used.

[0086] Data partitioning: The samples are divided into a training set (80%) and a test set (20%).

[0087] The box plot comparison of ChE, NE, and SCFA indicators of the training set (40 cases) and the validation set (10 cases) is shown in the figure. Figure 3 shown.

[0088] Clinical data standardization processing:

[0089] In order to eliminate the dimensional differences of each indicator and the influence of extreme values, it is necessary to standardize the cleaned data and use Z-score standardization (mean variance normalization):

[0090] The formula is:

[0091] Where X is the original data value;

[0092] μ is the mean of the data set;

[0093] σ is the standard deviation of the data set;

[0094] X' is the normalized data value.

[0095] Taking the collected clinical data as an example: 80% of the above cases are selected as the training set, that is, the first 40 cases (P001-P040) are used for illustration.

[0096] ChE (mg / L)

[0097] Sample size: 40

[0098] Mean μ1=6.25

[0099] Standard deviation σ1=1.24

[0100] Range [4.60, 8.25]

[0101] NE (ng / mL)

[0102] Sample size: 40

[0103] Mean μ2 = 0.81

[0104] Standard deviation σ2=0.30

[0105] Range [0.42, 1.25]

[0106] SCFA (mmol / L)

[0107] Sample size: 40

[0108] Mean μ3 = 104.43

[0109] Standard deviation σ3=32.60

[0110] Range [65.00,165.00]

[0111] Standardized processing instructions

[0112] Before model training, the above training set statistics (μ, σ) are used to perform Z-score standardization on all samples (including the test set and subsequent new samples). The formula is:

[0113] Specifically:

[0114] ChE normalization:

[0115] NE standardization:

[0116] SCFA standardization:

[0117] If the indicator distribution is significantly skewed, the indicator can be transformed into log(x+1) before standardization, and then the above Z-score standardization can be applied to reduce the impact of outliers.

[0118] If you need to remove extremely outliers, after completing Z-score normalization, screen and remove samples with |z| > 3. For example, if two cases in the training set have normalized ChE values exceeding ±3, they can be excluded, reducing the remaining training sample size from 40 to 38. After these exclusions, μ and σ should be recalculated and normalized again to ensure model training data quality.

[0119] (3) Use logistic regression to train the model through maximum likelihood estimation or gradient descent to minimize the difference between the model's prediction results for patients in the training set and the actual syndrome type labels, and solve the parameters to obtain a, b1, b2, and b3.

[0120] Each syndrome type corresponds to a set of coefficients a, b1, b2, and b3. After substituting the respective parameters, the classification model of the respective syndrome type is obtained.

[0121] (4) For the clinical samples to be tested, the ChE, SCFA, and NE data of the clinical samples are standardized according to the μ and σ corresponding to the training set, and then input into the classification model of any syndrome type. The model outputs the probability P value of the syndrome type. If P ≥ the corresponding set threshold, the syndrome type is judged to be positive, otherwise it is judged to be negative.

[0122] Determine the threshold

[0123] Usually logit(P)=log(P / 1-P)=0 is selected to correspond to P=0.5, thus obtaining the combined threshold relationship:

[0124] 0=a+b1×ChE+b2×SCFA+b3×NE

[0125] application

[0126] The patient's (ChE, SCFA, NE) standardized data (standardized according to the corresponding μ and σ of the training set) are substituted into the logit (P). If the result is > 0, it is judged as positive, otherwise it is negative.

[0127] (5) Verification: After the test set data is standardized according to the μ and σ corresponding to the training set, it is input into the trained model and the key parameters such as prediction accuracy, precision, specificity and AUC are statistically analyzed. If the effect is not good, the model structure or hyperparameters can be adjusted to solve again.

[0128] The following example uses the single label "whether or not it is Yangming disease" as an example to demonstrate how to perform logistic regression on 50 clinical samples, obtain model parameters (a, b1, b2, b3), and derive thresholds from them to construct a Yangming disease classification model. In actual clinical practice, if more syndrome types or multiple labels (such as "Tai Yang + Yangming") need to be determined, a one-vs-rest strategy (performing a regression for each syndrome type) or a multi-output neural network approach can be used.

[0129] I. Data collection and model training

[0130] 1. For Yangming diseases, collect data of cases labeled as "Yangming diseases" (i.e., cases with the label "Yangming" in the "Six Meridians Differentiation") and other syndrome types.

[0131] The ChE, SCFA, NE data and the Six Meridians syndrome type labels of the collected clinical samples are shown in the above table. Standardize them according to step (1) to obtain standardized data.

[0132] 2. Establish a label: whether it is "Yangming" ​​​​​​​​​​​​​​​​​​​​​​​​​7. Train to obtain model parameters: After training is complete, the regression coefficients and intercept of the model can be obtained. The regression coefficients b1, b2, and b3 correspond to the weights of the three features of ChE, SCFA, and NE, respectively, indicating the contribution of each feature to the prediction of the probability of "Yangming disease";

[0142] The intercept a represents the baseline value of the model's predicted probability of "Yangming disease" when all eigenvalues are zero.

[0143] In the example data, the regression coefficients and intercept are obtained as:

[0144] Regression coefficient: [0.032637580.733936650.01789112]

[0145] Intercept: [-106.53306157]

[0146] The specific meaning can be explained as follows: the log-odds function given by Logistic regression is:

[0147] logit(P)=a+b1×ChE+b2×SCFA+b3×NE

[0148] Substituting the values obtained above, we can write:

[0149] logit(P)=-106.5331+0.03264·ChE+0.73394·SCFA+0.01789·NE

[0150]

[0151] Where P is the probability of predicting "Yangming disease".

[0152] 2. Probability calculation of Yangming disease

[0153] For any record (i.e., given its ChE, SCFA, and NE values), a trained logistic regression model can be used for prediction. Specifically, these values are standardized and input into the model, which calculates the probability that the record belongs to the "Yangming disease" syndrome based on the previously trained parameters. The prediction probability formula given by the logistic regression model is:

[0154]

[0155] Substituting the coefficients in the above example, we can get the probability of predicting Yangming disease.

[0156] In practical applications, when predicting new samples, the probability value is also obtained by calling the prediction probability function of the model.

[0157] 3. “Yangming Disease” Threshold (Decision Boundary)

[0158] In logistic regression, the default classification threshold is usually P = 0.5. That is, when the predicted probability P ≥ 0.5, the sample is judged to be Yangming disease (Yangming = 1), otherwise it is judged as non-Yangming disease (= 0).

[0159] From the log-odds formula for binary classification, we know that P = 0.5 is equivalent to logit(p) = 0. In other words, the decision boundary satisfies:

[0160] a0+b1·ChE+b2·SCFA+b3·NE=0

[0161] For the parameter values calculated above, the decision boundary equation for "Yangming disease" is:

[0162] -106.5331+0.03264·ChE+0.73394·SCFA+0.01789·NE=0

[0163] It can be seen from the equation that in the three-dimensional space (ChE, SCFA, NE), this is a plane, that is, if the calculated value of the above linear combination is greater than 0, the corresponding predicted probability of the sample is >0.5, and it is judged to be "Yangming" positive; if it is less than 0, the corresponding probability is <0.5, and it is judged to be "Yangming" negative.

[0164] Therefore, for the "Yangming disease threshold," when the log-odds output of logistic regression is ≥ 0, or equivalently, when the output probability is ≥ 0.5, the disease is considered "Yangming." In practice, if more flexible adjustment of sensitivity and specificity is required, 0.5 can be replaced with other appropriate thresholds, such as 0.3 or 0.7, which will affect the range of cases ultimately classified as Yangming disease.

[0165] Substitute the ChE, SCFA, and NE of the new sample into the above formula and calculate. If the result is ≥0, the predicted probability is ≥0.5, and the sample is predicted to belong to "Yangming disease"; otherwise, it does not belong.

[0166] If you want to increase the detection rate (sensitivity) of "Yangming disease", you can lower the threshold appropriately from 0.5 (such as 0.4, 0.3, etc.); if you want to increase the specificity, you can increase the threshold (such as 0.6, 0.7, etc.).

[0167] 4. Verify data:

[0168] 1. Internal Verification:

[0169] Using the last 10 test set data, the patients' (ChE, SCFA, NE) were standardized according to the μ and σ corresponding to the training set, and then input into the model to calculate the predicted probability and corresponding classification results;

[0170] The consistency between the statistical prediction results and the clinical differentiation results of traditional Chinese medicine practitioners was calculated, and the following key indicators were calculated:

[0171] Confusion Matrix: Shows the number of true positives, false positives, true negatives, and false negatives;

[0172] Classification report: including accuracy, precision, recall, specificity, and F1-Score;

[0173] Draw the ROC curve and calculate the AUC score to evaluate the model's ability to distinguish syndromes such as "Yangming disease".

[0174] The specific verification method is as follows:

[0175] Model validation evaluation

[0176] After training any syndrome type model in the embodiment, the model performance is comprehensively evaluated according to the following process and indicators:

[0177] 1) Build confusion matrix and generate classification report

[0178] For the data of the test set (P041–P050), the constructed classification model is used to make predictions, and the four types of prediction results are statistically analyzed:

[0179] True positive (TP): The patient actually has this syndrome and is predicted to be positive.

[0180] False positive (FP): The syndrome is actually not the same but is predicted to be positive.

[0181] True negative (TN): The patient is actually not of the syndrome and is predicted to be negative.

[0182] False negative (FN): The actual syndrome is predicted to be negative

[0183] Based on the above data, calculate the accuracy, precision, recall, specificity, F1-Score and generate a classification report:

[0184] Accuracy

[0185]

[0186] Measures the proportion of overall predictions that were correct.

[0187] Precision

[0188]

[0189] Measures how many of the predicted positives are true positives.

[0190] Recall / Sensitivity

[0191]

[0192] A measure of how many positive tests were successfully detected.

[0193] Specificity

[0194]

[0195] A measure of how many were correctly ruled out when they were actually negative.

[0196] F1-Score

[0197]

[0198] The harmonic mean of precision and recall is considered comprehensively.

[0199] 2) Draw the ROC curve and calculate the AUC value

[0200] ROC curve: With the false positive rate (FPR=FP / (FP+TN)) as the horizontal axis and the true positive rate (recall rate) (TPR=Recall) as the vertical axis, curves are drawn under different thresholds.

[0201] AUC (Area Under Curve): The area under the ROC curve reflects the overall discrimination ability of the model at each threshold.

[0202] 3) Evaluation Process

[0203] Calculate the above indicators on the test set and generate a confusion matrix and classification report (accuracy, precision, recall, F1-score).

[0204] Use sklearn, R and other tools to draw ROC curves and calculate AUC.

[0205] If a certain indicator (such as AUC, accuracy) does not meet the preset requirements, the model iteration will begin:

[0206] The weights of features (ChE, NE, SCFA) can be re-examined, or interaction terms, nonlinear transformations, etc. can be introduced; the model type can also be adjusted (such as adding regularization, trying random forests or neural networks);

[0207] K-fold cross validation (K = 5 or 10) was used to select the optimal hyperparameters and evaluate stability;

[0208] If necessary, expand the sample or optimize data preprocessing (such as using logarithmic transformation or Winsorization).

[0209] 4) Preset requirements

[0210] Ideal criteria: AUC ≥ 0.80 and Accuracy ≥ 0.75;

[0211] Acceptable lower limits: AUC ≥ 0.70 and Accuracy ≥ 0.70.

[0212] When the preset requirements are not met, iterative optimization is required; when the preset requirements are met or exceeded, the model can be embedded in the clinical decision support system as an auxiliary tool for Chinese medicine differentiation.

[0213] In this embodiment, the ROC curve of the trained "Yangming disease" binary classification model on the test set is as follows: Figure 1 As shown, Figure 1 AUC = 0.82 is marked in the figure.

[0214] like Figure 1 As shown, this is the ROC curve of the "Yangming disease" binary classification model on the test set based on the example embodiment, with AUC = 0.82.

[0215] The orange solid line is the ROC curve of the model, showing the relationship between TPR (recall rate) and FPR (false positive rate) at different thresholds;

[0216] The red dotted line is the diagonal reference (ROC curve when random guessing).

[0217] It can be seen that the AUC of the model on the test set reaches 0.82, which exceeds the preset ideal standard AUC ≥ 0.80

[0218] In this example, the binary classification model predicts 9 consistent labels for the test set (10 cases) and 1 inconsistent label. The calculated accuracy is Accuracy = 90%, which meets the preset requirement of Accuracy ≥ 75%, and AUC = 0.82 ≥ 0.80.

[0219] 2. External Verification:

[0220] If conditions permit, collect independent sample data from other medical institutions or different populations and apply the same model to external datasets;

[0221] Compare the predicted results with the actual dialectical results to verify the robustness of the model in different samples and environments;

[0222] If the external validation results are close to the internal validation results, it further proves the generalizability and practicality of the model.

[0223] To verify the effectiveness of the model, 20 cases of medical records from a cooperative hospital were collected. The ChE, NE, SCFA data and Yangming syndrome labels are shown in the following table. The box plot comparison of ChE, NE, and SCFA indicators of the training set (40 cases) and external cases (20 cases) is shown in the following figure. Figure 4 shown.

[0224] The original test values (ChE, NE, SCFA) were first standardized by Z-score according to the training set mean μ and σ, and then substituted into the logistic regression equation to calculate the predicted probability P.

[0225] Prediction P ≥ 0.5 is positive (1), otherwise it is negative (0). The predicted probability results are shown in the following table:

[0226]

[0227] The results showed that samples 5 and 20 were "false negative" and "false positive" respectively, and the remaining 18 cases were predicted correctly, with an accuracy rate of Accuracy = 18 / 20 = 0.90.

[0228] The AUC of this data set was calculated by sklearn = 0.87 (> 0.85) ( Figure 2 ), meeting the preset requirements.

[0229] This proves that the model has good generalization ability on this external sample set.

[0230] 3. Clinical implementation and feedback

[0231] Internally and externally validated models can be embedded in clinical decision support systems to provide auxiliary syndrome differentiation opinions for TCM practitioners;

[0232] Continuously collect clinical feedback, regularly update and calibrate model parameters and thresholds to ensure their long-term applicability and accuracy.

Claims

1. A method for constructing a six-channel syndrome differentiation classification model based on the NEI system, characterized in that The following steps are involved: (1) Obtain the ChE, SCFA, and NE test data of patients with different six-channel syndromes, annotate and classify the data according to the six-channel syndromes, divide the data of any syndrome into training and test sets, and preprocess the training set data by cleaning, processing missing values, and standardizing the raw data to obtain the standardized data of ChE, SCFA, and NE; (2) The standardized data of the training set is used to construct a classification model through a machine learning algorithm, and the model satisfies: logit(P)=a+b1×ChE+b2×SCFA+b3×NE P represents the probability of a certain Six Meridians syndrome type, that is, the probability of any of the Taiyang, Yangming, Shaoyang, Taiyin, Shaoyin, and Jueyin syndrome types; ChE is the normalized data of cholinesterase level in mg / L; SCFA is the standardized data of the total acid value of intestinal short-chain fatty acids, with the unit of mmol / L; NE is the normalized data of norepinephrine concentration, in ng / mL; a, b1, b2, b3 are the parameters to be optimized; Use logistic regression or neural network to train the model and obtain the classification model parameters a, b1, b2, b3 of the syndrome type; (3) The data of the test set is standardized, and the model of any syndrome type is verified, evaluated and optimized to obtain a classification model.

2. The method according to claim 1, wherein In the step (1), the missing value processing is to fill or remove the missing values with the mean; the standardization processing is to normalize the mean and variance of the data after cleaning and missing value processing.

3. The method according to claim 1, wherein In step (2), in the logistic regression algorithm, the parameters a, b1, b2, and b3 are solved using maximum likelihood estimation or gradient descent method.

4. The method according to claim 1, wherein In step (3), the verification, evaluation and optimization steps are: using the data after the test set is standardized to calculate the accuracy, precision and specificity, draw the ROC curve, calculate the AUC, and verify and evaluate the model, requiring the AUC and accuracy of the classification model to meet the preset requirements; if the AUC and accuracy of the model do not meet the preset requirements, re-iterate the training model until the AUC and accuracy of the classification model meet the preset requirements.

5. The method according to claim 4, wherein The preset requirements for AUC and accuracy of the classification model are: AUC ≥ 0.70 and accuracy ≥ 0.

70.

6. The method according to claim 1, wherein in step (3), independent sample data from other medical institutions or different populations are collected as external data sets, and the same model is applied to the external data sets for verification and evaluation.

7. The method according to claim 1, comprising step (4): continuously collecting ChE, SCFA, and NE test data of patients with different six-channel syndrome types, regularly updating and correcting model parameters and thresholds to ensure the long-term applicability and accuracy of the classification model.

8. The six-channel syndrome differentiation classification model constructed by the method according to any one of claims 1 to 7.

9. A method for calculating the probability of six meridian syndrome types using the six meridian syndrome differentiation classification model as claimed in claim 8, characterized in that The method is: After the ChE, SCFA, and NE test data of the test samples are standardized, they are input into the Six-Channel Syndrome Differentiation classification model. Each syndrome type corresponds to an independent classification model, and the model outputs the probability P of the syndrome type. If P ≥ the set threshold, the syndrome type is judged to be positive, and if P < the set threshold, the syndrome type is judged to be negative.

10. A six-channel syndrome differentiation classification system based on the NEI system, characterized by: The system comprises: The data processing module is used to obtain ChE, SCFA, and NE test data from patients with different six-channel syndromes, annotate and classify the data according to the six-channel syndromes, and divide the data of any syndrome into training and test sets. The training set data is preprocessed by cleaning the raw data, processing missing values, and standardizing them to obtain standardized data of ChE, SCFA, and NE. The model training module is used to train the model using logistic regression or neural network with the standardized data of the training set. The model satisfies: logit(P)=a+b1×ChE+b2×SCFA+b3×NE P represents the probability of a certain Six Meridians syndrome type, that is, the probability of any of the Taiyang, Yangming, Shaoyang, Taiyin, Shaoyin, and Jueyin syndrome types; ChE is the normalized data of cholinesterase level in mg / L; SCFA is the standardized data of the total acid value of intestinal short-chain fatty acids, with the unit of mmol / L; NE is the normalized data of norepinephrine concentration, in ng / mL; a, b1, b2, b3 are the parameters to be optimized; The classification model parameters a, b1, b2, and b3 of this syndrome are obtained through machine algorithm training; Model validation module: used to input standardized test set data into the model of any syndrome type for validation evaluation and parameter optimization to obtain a classification model with high accuracy and specificity; Six-Jing Dialectical Classification Module: It is used to standardize the ChE, SCFA, and NE test data of the sample to be tested and input them into the Six-Channel Syndrome Differentiation classification model. Each syndrome type corresponds to an independent classification model, and the model outputs the probability P of the syndrome type. If P ≥ the set threshold, the syndrome type is judged to be positive, and if P < the set threshold, the syndrome type is judged to be negative.

Images