A postoperative ch method and system for ch based on multiple factors

Abstract

The application discloses a postoperative CH prediction method and system for hyperhidrosis based on multiple factors, and the method comprises the following steps: acquiring multiple groups of postoperative historical data of patients, including target variables and multiple characteristic variables which have influence on postoperative CH of hyperhidrosis, and pre-processing the historical data; representing the historical data of each patient in the form of a multi-dimensional feature vector, and dividing a training set and a verification set; constructing a Gaussian regression model, and selecting a radial basis kernel and white noise kernel as a covariance function; optimizing hyperparameters of the covariance function on the training set by maximizing a log-likelihood function; and inputting a current patient feature vector into the optimized model, and outputting a prediction result. The method can take into account small sample adaptability, multiple factor interaction analysis and ordered classification prediction, and provides a scientific basis for clinical decision-making.

Description

Technical Field

[0001] This invention relates to the field of medical data analysis technology, and in particular to a method and system for predicting postoperative hemorrhage (CH) in palmar hyperhidrosis based on multiple factors. Background Technology

[0002] Primary palmar hyperhidrosis (PPH) is an autonomic nervous system disorder characterized by excessive sweating of the palms, which can severely impact a patient's quality of life and mental health. Endoscopic thoracoscopic sympathectomy (ETS) is an effective treatment for moderate to severe PPH, but postoperative compensatory hyperhidrosis (CH) is the most common complication, occurring in 30%-90% of cases. CH manifests as compensatory sweating in non-surgical areas such as the trunk and thighs, and its severity directly affects patient satisfaction with the surgical outcome. While international research has made progress in optimizing surgical techniques (such as T3 and T4 segment selection) in recent years, individualized prediction of postoperative CH in PPH still relies on logistic regression analysis of independent factors, with independent factors constructed using nomograms. However, this predictive method still faces the following challenges:

[0003] Complexity of multifactor interactions: The occurrence of CH is closely related to factors such as patient age, gender, preoperative sweating level, psychological state, and smoking behavior. Traditional statistical models are difficult to capture the nonlinearity and interaction effects between variables.

[0004] Limitations of small sample data: Single-center clinical studies have limited sample sizes, and conventional machine learning models (such as SVM and random forest) are prone to overfitting, resulting in insufficient generalization ability.

[0005] Low accuracy in grade prediction: Existing methods often simplify CH into a binary classification (present / absent), ignoring the ordinal characteristics of its clinical grading (absent, mild, moderate, severe), which cannot meet the needs of precise intervention.

[0006] Insufficient integration of dynamic data: Dynamic variables such as lifestyle habits (e.g., smoking levels) are not included in the prediction system, affecting the timeliness of the model. Summary of the Invention

[0007] To address the shortcomings of the existing technologies, the technical problem to be solved by this invention is to propose a multi-factor-based method and system for predicting postoperative hemorrhage (CH) in palmar hyperhidrosis, which can take into account small sample adaptability, multi-factor interaction analysis, and ordered classification prediction, thus providing a scientific basis for clinical decision-making.

[0008] One technical solution adopted in this invention is to provide a multifactorial method and system for predicting postoperative hemorrhage (CH) in palmar hyperhidrosis, the method comprising the following steps:

[0009] S1: Obtain postoperative historical data from multiple groups of patients, including target variables and various characteristic variables that affect the presence of CH after hand hyperhidrosis surgery, and preprocess the historical data;

[0010] S2: Represent each patient's historical data in the form of a multidimensional feature vector and divide it into a training set and a validation set;

[0011] S3: Construct a Gaussian regression model, and select the radial basis kernel and white noise kernel as the covariance function;

[0012] S4: Optimize the hyperparameters of the covariance function on the training set by maximizing the log-likelihood function;

[0013] S5: Input the current patient feature vector into the optimized model and output the prediction result.

[0014] Furthermore, the method also includes the following steps:

[0015] S6: Generate a clinical decision report based on the prediction results.

[0016] Furthermore, the historical data includes static and dynamic variables. The static variables include age, gender, surgical segment, and preoperative severity of palmar hyperhidrosis. The dynamic variables include postoperative time, anxiety score, and smoking level. The target variable is the patient's postoperative hemorrhage severity, including four levels: none, mild, moderate, and severe. The preprocessing of the historical data involves the following sub-steps:

[0017] S11: Z-score standardization of the continuous variables of age and postoperative time;

[0018] S12: Perform KNN interpolation on the categorical variables of gender and surgical segment.

[0019] Furthermore, the specific representation of the feature vector is as follows:

[0020]

[0021] Where X i This represents the feature vector of the i-th patient. The values ​​represent the i-th patient's age, gender, surgical segment, preoperative severity of hand sweating, postoperative time, anxiety score, and smoking level, in that order.

[0022] Furthermore, the covariance function is specifically expressed as follows:

[0023]

[0024] Where K represents the feature vector X i and X jThe covariance value;

[0025] X i X j Let represent the feature vectors of the i-th and j-th patients, with a dimension of 7;

[0026] σ f It represents the signal variance, used to control the amplitude range of the model output;

[0027] l represents the length scale, used to determine the similarity decay rate in the feature space;

[0028] σ n This represents the noise variance, used to represent the random error in observed data.

[0029] δ ij Denotes the Kronecker Delta function, where δ = j. ij =1, otherwise 0.

[0030] Furthermore, step S4 includes the following sub-steps:

[0031] S41: Define the log-likelihood function as:

[0032]

[0033] Among them, Y T Represents the transpose of the target variable vector Y; θ = {σ f ,l,σ n} represents the three hyperparameters of the covariance function; I represents the identity matrix, whose dimension is consistent with the number of training samples n; K represents the covariance matrix;

[0034] S42: Calculate the log-likelihood pair θ = {σ f ,l,σ n The partial derivatives of};

[0035] S43: Maximize the log-likelihood function using the conjugate gradient algorithm until the log-likelihood pairs θ = {σ} f ,l,σ n The convergence is achieved, and the hyperparameters of the converged covariance function are obtained.

[0036] Furthermore, step S5 includes the following sub-steps:

[0037] S51: Input the current patient feature vector and calculate the current patient's predicted distribution:

[0038] P(Y * |X * (X, Y) = N(μ) * , ∑ * );

[0039]

[0040] Wherein K(X) * , X) represents the covariance vector between the current patient feature vector and the training sample X, with a dimension of 1×n;

[0041] K(X, X) represents the covariance matrix of the training sample X itself;

[0042] K(X * X * ) represents the covariance scalar of the current patient's feature vector itself;

[0043] Y * This indicates the predicted degree of postoperative hemorrhage in the current patient;

[0044] P(Y * |X * (X, Y) represents the predicted probability distribution of the postoperative CH degree of the current patient;

[0045] N represents a normal distribution, used to describe Y * Distribution pattern;

[0046] μ * This represents the predicted mean of the postoperative CH severity in current patients; I represents the identity matrix;

[0047] S52: Based on the distribution of the target variables in the training set, a preset discretization threshold range is determined;

[0048] S53: Based on the discretization threshold range, the predicted mean of the postoperative CH degree of the current patient is discretized and mapped to obtain the postoperative CH degree of the current patient.

[0049] Furthermore, step S6 includes the following sub-steps:

[0050] S61: Calculate the 95% confidence interval based on the predicted covariance;

[0051] S62: Calculate the SHAP value of each feature variable or the SHAP interaction value of several feature variables;

[0052] S63: Output a clinical decision report based on the 95% confidence interval, SHAP value, and SHAP interaction value.

[0053] To solve the above-mentioned technical problems, the second technical solution adopted by the present invention is: to provide a multifactorial palmar hyperhidrosis postoperative CH prediction system, the system comprising:

[0054] The data acquisition module is used to acquire postoperative historical data from multiple groups of patients, including target variables and various characteristic variables that affect the presence of CH after hand hyperhidrosis surgery, and to preprocess the historical data.

[0055] The data partitioning module is used to represent each patient's historical data in the form of a multidimensional feature vector and to partition the data into training and validation sets.

[0056] The model building module is used to build a Gaussian regression model, selecting the radial basis kernel and white noise kernel as the covariance function;

[0057] The model training module is used to optimize the hyperparameters of the covariance function on the training set by maximizing the log-likelihood function.

[0058] The prediction module inputs the current patient feature vector into the optimized model and outputs the prediction result.

[0059] Furthermore, the system also includes a decision reporting module for generating clinical decision reports based on the prediction results.

[0060] The multifactor-based method and system for predicting postoperative hemorrhage (CH) in palmar hyperhidrosis of the present invention has at least the following beneficial effects: 1. By combining radial basis function (RBF) with covariance function as the covariance function, it implicitly captures the nonlinear relationships and complex interactions of features such as surgical segment, smoking level, and anxiety score, without the need for manual pre-setting of interaction terms. The accuracy of interaction effect detection on the test set is improved by 25%, which is significantly better than the prediction accuracy of the traditional logistic regression model, effectively solving the problem of multifactorial complexity modeling; 2. Based on the Gaussian process regression framework, the covariance matrix is ​​used to suppress noise overfitting, further improving the prediction accuracy, and its prediction effect is significantly better than that of the traditional model. Attached Figure Description

[0061] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0062] Figure 1 This is a flowchart of one embodiment of the multifactor-based method for predicting postoperative hemorrhage (CH) in palmar hyperhidrosis according to the present invention.

[0063] Figure 2 For the present invention Figure 1 The sub-flowchart of step S1.

[0064] Figure 3 For the present invention Figure 1 The sub-flowchart of step S4 in the middle.

[0065] Figure 4 For the present invention Figure 1 Sub-flowchart of step S5 in the middle.

[0066] Figure 5 For the present invention Figure 1 The sub-flowchart of step S6 in the middle.

[0067] Figure 6 This is a structural block diagram of one embodiment of the multifactor-based palmar hyperhidrosis postoperative CH prediction system of the present invention. Detailed Implementation

[0068] The invention will now be further described with reference to the accompanying drawings.

[0069] Please see Figure 1 This is a flowchart of an embodiment of the multifactor-based method for predicting postoperative hemorrhage (CH) in palmar hyperhidrosis according to the present invention. This embodiment may specifically include the following steps:

[0070] S1: Obtain postoperative historical data from multiple groups of patients, including target variables and various characteristic variables that affect the presence of CH after hand hyperhidrosis surgery, and preprocess the historical data.

[0071] Specifically, in this embodiment, the data source can be a combined database of follow-up cases of palmar hyperhidrosis from several hospitals, such as a hospital's electronic medical record system, from which postoperative historical data of several patients can be collected. This historical data may include a target variable (i.e., the degree of postoperative palmar hyperhidrosis) and characteristic variables (characteristic data that affect postoperative palmar hyperhidrosis ...

[0072] Please see Figure 2 The historical data includes static and dynamic variables. The static variables include age, gender, surgical segment, and preoperative severity of palmar hyperhidrosis. The dynamic variables include postoperative time, anxiety score, and smoking level. The target variable is the patient's postoperative hemorrhage severity, including four levels: none, mild, moderate, and severe. This step S1 may include the following sub-steps:

[0073] S11: Z-score standardization of the continuous variables of age and postoperative time;

[0074] S12: Perform KNN interpolation on the categorical variables of gender and surgical segment.

[0075] Specifically, in this embodiment, historical data may include seven feature variables: age, gender, surgical segment, preoperative severity of palmar hyperhidrosis, postoperative time, anxiety score, and smoking level. Among these, age, gender, surgical segment (T3 or T4), and preoperative severity of palmar hyperhidrosis (mild, moderate, or severe) are static variables, while postoperative time, anxiety score, and smoking level are dynamic variables. The target variable represents the patient's postoperative hemorrhage severity, which can be divided into four levels: none, mild, moderate, and severe. To meet the requirements of the model input and eliminate semantic ambiguity of non-numerical data, some features in the aforementioned historical data need to be encoded. For example, gender (male or female) and surgical segment (T3 or T4) can be one-hot encoded, while the target feature and smoking level can be ordered classification encoded (levels 0-4). In addition, for outlier data, continuous variables such as age and postoperative time are Z-score standardized to remove outliers. Z-score standardization is a conventional data standardization method, and its specific steps will not be elaborated here. For other missing data, KNN imputation can be used. KNN imputation is also a conventional data preprocessing method, and its specific steps will not be elaborated here. This allows us to obtain complete historical data after processing. The historical data is then stratified according to the target feature, i.e., the degree of postoperative hemorrhage (CH), and divided according to a preset ratio, such as 7:3 in this embodiment, to obtain the training set and validation set.

[0076] S2: Represent each patient's historical data in the form of a multidimensional feature vector and divide it into a training set and a validation set.

[0077] In some implementations, the feature vector is specifically represented as follows:

[0078]

[0079] Where X i This represents the feature vector of the i-th patient. The values ​​represent the i-th patient's age, gender, surgical segment, preoperative severity of hand sweating, postoperative time, anxiety score, and smoking level, in that order.

[0080] Specifically, in this embodiment, the historical data of each patient is represented in the form of a multi-dimensional feature vector, which facilitates the input and calculation of the subsequent model. Since the historical data of this scheme includes the seven different types of feature data mentioned above, namely age, gender, surgical segment, preoperative severity of hand sweating, postoperative time, anxiety score and smoking level, the feature vector of each patient can be represented as a 1×7 dimensional vector, and each element in the vector represents a feature data type.

[0081] S3: Construct a Gaussian regression model and select the radial basis kernel and white noise kernel as covariance functions.

[0082] Specifically, current research on postoperative hemorrhage (CH) in palmar hyperhidrosis is limited to studies of independent factors influencing CH. Whether interactions between feature variables can have a combined effect on CH remains unclear. Therefore, this embodiment uses a Gaussian regression model as the prediction model. This model can implicitly model the nonlinear interactions between feature data, avoiding the need for manually constructed interaction terms, and it can provide a reliable assessment of the interaction effects of feature data. Furthermore, due to limitations in sample data size (e.g., obtaining historical data by combining follow-up databases of several hospitals for palmar hyperhidrosis in this embodiment), the sample size is small. The Gaussian regression model is also suitable for nonlinear problems with small samples. When constructing this prediction model, the appropriate kernel function must first be selected based on the type of feature data. Since this implementation includes seven feature data types, including not only continuous feature data such as age and postoperative time, but also observational data such as anxiety scores and smoking levels, a radial basis function kernel that can capture smooth and continuous data patterns and a white noise kernel that can eliminate subjective biases introduced by patients in anxiety scores and smoking levels can be selected. The radial basis function kernel and the white noise kernel are combined as the covariance function.

[0083] In some embodiments, the covariance function is specifically represented as:

[0084]

[0085] Where K represents the feature vector X i and X j The covariance value;

[0086] X i X j Let represent the feature vectors of the i-th and j-th patients, with a dimension of 7;

[0087] σ f It represents the signal variance, used to control the amplitude range of the model output;

[0088] l represents the length scale, used to determine the similarity decay rate in the feature space;

[0089] σ n This represents the noise variance, used to represent the random error in observed data.

[0090] δ ij Denotes the Kronecker Delta function, where δ = j. ij =1, otherwise 0.

[0091] Specifically, in this embodiment, a covariance function is constructed based on the radial basis function kernel and the white noise kernel. Simultaneously, the hyperparameter θ = {σ} of this covariance function is...f ,l,σ n} is initialized, and its initial values ​​are σ. f =1.0, l=1.0, σ n =0.1; its constraint range is σ f ∈[0.1,10.0],l∈[0.1,10.0],σ n ∈0.01,1. It should be noted that the hyperparameter θ = {σ} mentioned above... f ,l,σ n The initial value of} can be adaptively selected based on the actual model.

[0092] S4: Optimize the hyperparameters of the covariance function on the training set by maximizing the log-likelihood function.

[0093] Specifically, in this embodiment, the hyperparameters are optimized by maximizing the logistic likelihood function using either the conjugate algorithm or the L-BFGS-B algorithm.

[0094] In some implementations, please refer to Figure 3 Step S4 may include the following sub-steps:

[0095] S41: Define the log-likelihood function as:

[0096]

[0097] Among them, Y T Represents the transpose of the target variable vector Y; θ = {σ f ,l,σ n} represents the three hyperparameters of the covariance function; I represents the identity matrix, whose dimension is consistent with the number of training samples n; K represents the covariance matrix;

[0098] S42: Calculate the log-likelihood pair θ = {σ f ,l,σ n The partial derivatives of};

[0099] S43: Maximize the log-likelihood function using the conjugate gradient algorithm until the log-likelihood pairs θ = {σ} f ,l,σ n The convergence is achieved, and the hyperparameters of the converged covariance function are obtained.

[0100] Specifically, the conjugate gradient algorithm is used to maximize the log-likelihood function until the log-likelihood pairs θ = σ. f ,l,σ n Once convergence is achieved, the hyperparameters of the converged covariance function can be substituted into the covariance function described above to obtain the optimized prediction model.

[0101] S5: Input the current patient feature vector into the optimized model and output the prediction result.

[0102] In some embodiments, please refer to Figure 4 Step S5 may include the following sub-steps:

[0103] S51: Input the current patient feature vector and calculate the current patient's predicted distribution:

[0104] P(Y * |X * (X, Y) = N(μ) * , ∑ * );

[0105]

[0106] Wherein K(X) * , X) represents the covariance vector between the current patient feature vector and the training sample X, with a dimension of 1×n;

[0107] K(X, X) represents the covariance matrix of the training sample X itself;

[0108] K(X * X * ) represents the covariance scalar of the current patient's feature vector itself;

[0109] Y * This indicates the predicted degree of postoperative hemorrhage in the current patient;

[0110] P(Y * |X * (X, Y) represents the predicted probability distribution of the postoperative CH degree of the current patient;

[0111] N represents a normal distribution, used to describe Y * Distribution pattern;

[0112] μ * This represents the predicted mean of the postoperative CH severity in current patients; I represents the identity matrix;

[0113] ∑ * This represents the predictive covariance of the current patient's postoperative CH severity.

[0114] S52: Based on the distribution of the target variables in the training set, a preset discretization threshold range is determined;

[0115] S53: Based on the discretization threshold range, the predicted mean of the postoperative CH degree of the current patient is discretized and mapped to obtain the postoperative CH degree of the current patient.

[0116] Specifically, in this implementation, the feature vector of the current patient can be directly input into the prediction model to obtain the postoperative CH severity of the current patient. Simultaneously, after predicting the postoperative CH severity of the current patient, since this value is in covariance form, it needs to be discretized. This process first requires pre-setting a discretization threshold range, for example, by setting a discretization threshold rule, i.e.

[0117] If μ * If <0.5, the predicted CH value is 0 (no CH after surgery);

[0118] If 0.5≤μ * If <1.5, the predicted CH value is 1 (mild postoperative CH);

[0119] If 1.5 ≤ μ * If the value is <2.5, the predicted CH value is 2 (moderate CH after surgery);

[0120] If μ * If the value is greater than 2.5, the predicted CH value is 3 (postoperative severe CH).

[0121] In addition, the above-mentioned setting of the discretization threshold is only the setting in this embodiment. The discretization threshold can also be dynamically adjusted according to the actual situation. For example, when the CH level distribution in the training set is unbalanced (e.g., very few heavy samples), the quantile method can be used: according to the distribution of the predicted mean in the training set, the quantile point is taken as the threshold.

[0122] This embodiment may also include the following steps:

[0123] S6: Generate a clinical decision report based on the prediction results.

[0124] In some embodiments, please refer to Figure 5 Step S6 may also include the following sub-steps:

[0125] S61: Calculate the 95% confidence interval based on the predicted covariance;

[0126] S62: Calculate the SHAP value of each feature variable or the SHAP interaction value of several feature variables;

[0127] S63: Output a clinical decision report based on the 95% confidence interval, SHAP value, and SHAP interaction value.

[0128] Specifically, in this embodiment, after obtaining the prediction results, a clinical decision report that can guide clinical decision-making can be generated based on the prediction results. First, a 95% confidence interval can be calculated based on the prediction covariance. The purpose of this is to quantify the uncertainty of the prediction results, and the specific calculation method can be as follows:

[0129]

[0130] With μ * =1.8, For example, based on the initial threshold [0.5, 1.5, 2.5], since 1.5 < 1.8 < 2.5, the predicted CH level is level 2, i.e. mild, and its confidence interval is 1.8 ± 1.96 × 0.4 = [1.02, 2.58], which suggests that the actual predicted CH level may be level 1 or level 2.

[0131] The aforementioned SHAP value and SHAP interaction value can be calculated by weighting the marginal contribution of the feature variable in all possible subset combinations, or by weighting the marginal contribution of the feature variable combination in all possible subset combinations. This allows for the quantification of the contribution of each feature variable or feature variable combination to the predicted value, thereby enhancing the interpretability of the model.

[0132] Based on the above method, the process of predicting postoperative hemorrhage (CH) and generating a clinical decision report for patients is illustrated below. For example, for a 28-year-old female patient (pre-selected T3 surgical segment, heavy smoker, anxiety score 85), the postoperative CH prediction process would be as follows:

[0133] Input the standardized data: X = [28, 0, 0, 3, 6, 85, 4];

[0134] Obtain the model prediction result μ * =2.8, Therefore, the 95% confidence interval is [1.6, 4.0];

[0135] The prediction results were mapped to the severity level of CH: using a discretized threshold range [0.5, 1.5, 2.5], the predicted level of CH for this patient after surgery was 3, which is severe CH after surgery;

[0136] Based on the prediction results, clinical recommendations are generated, marking the patient as "high-risk uncertainty" and suggesting multidisciplinary consultation and smoking cessation intervention.

[0137] Please see Figure 6 This is a structural block diagram of one embodiment of the multifactorial palmar hyperhidrosis postoperative hemorrhage (CH) prediction system of the present invention. The multifactorial palmar hyperhidrosis postoperative CH prediction system of this embodiment is used to implement the multifactorial palmar hyperhidrosis postoperative CH prediction method described in the above embodiment. Specifically, the multifactorial palmar hyperhidrosis postoperative CH prediction system of this embodiment includes a data acquisition module 100, a data partitioning module 200, a model building module 300, a model training module 400, and a prediction module 500. Wherein:

[0138] The data acquisition module 100 is used to acquire postoperative historical data of multiple groups of patients, including target variables and various characteristic variables that affect the presence of CH after hand hyperhidrosis, and to preprocess the historical data.

[0139] The data partitioning module 200 is used to represent the historical data of each patient in the form of a multidimensional feature vector and to partition the data into a training set and a validation set.

[0140] Model building module 300 is used to build a Gaussian regression model, selecting radial basis kernel and white noise kernel as covariance functions;

[0141] The model training module 400 is used to optimize the hyperparameters of the covariance function on the training set by maximizing the log-likelihood function.

[0142] The prediction module 500 is used to input the current patient feature vector into the optimized model and output the prediction result.

[0143] In some embodiments, the system may further include a decision reporting module 600 for generating clinical decision reports based on prediction results.

[0144] This invention utilizes a combination of radial basis function (RBF) and covariance function as the covariance function to implicitly capture the nonlinear relationships and complex interactions of features such as surgical segments, smoking levels, and anxiety scores, without requiring manual pre-setting of interaction terms. The accuracy of interaction effect detection on the test set is improved by 25%, significantly enhancing the prediction accuracy compared to traditional logistic regression models, effectively solving the challenge of modeling complex multi-factor scenarios. Furthermore, this solution, based on a Gaussian process regression framework, further improves prediction accuracy by suppressing noise overfitting through the covariance matrix, resulting in significantly better prediction performance than traditional models.

[0145] The above description merely illustrates preferred embodiments of the present invention and is quite specific and detailed; however, it should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of the present invention. Therefore, the scope of protection of this invention should be determined by the appended claims.

Claims

1. A multifactorial method for predicting postoperative hemorrhage (CH) in palmar hyperhidrosis, characterized in that, The method includes the following steps: S1: Obtain postoperative historical data from multiple groups of patients, including target variables and various characteristic variables that affect the presence of CH after hand hyperhidrosis surgery, and preprocess the historical data; S2: Represent each patient's historical data in the form of a multidimensional feature vector and divide it into a training set and a validation set; S3: Construct a Gaussian regression model, and select the radial basis kernel and white noise kernel as the covariance function; S4: Optimize the hyperparameters of the covariance function on the training set by maximizing the log-likelihood function; S5: Input the current patient feature vector into the optimized model and output the prediction result; S6: Based on the prediction results, generate a clinical decision report. This step includes the following sub-steps: S61: Calculate the 95% confidence interval based on the predicted covariance; S62: Calculate the SHAP value of each feature variable or the SHAP interaction value of several feature variables; S63: Output a clinical decision report based on the 95% confidence interval, SHAP value, and SHAP interaction value.

2. The multifactorial method for predicting postoperative hemorrhage (CH) in palmar hyperhidrosis as described in claim 1, characterized in that, The historical data includes static and dynamic variables. Static variables include age, gender, surgical segment, and preoperative palmar hyperhidrosis severity. Dynamic variables include postoperative time, anxiety score, and smoking level. The target variable is the patient's postoperative hemorrhage severity, categorized into four levels: none, mild, moderate, and severe. The preprocessing of the historical data involves the following sub-steps: S11: Z-score standardization of the continuous variables of age and postoperative time; S12: Perform KNN interpolation on the categorical variables of gender and surgical segment.

3. The multifactorial method for predicting postoperative hemorrhage (CH) in palmar hyperhidrosis as described in claim 2, characterized in that, The specific representation of the feature vector is as follows: ； in This represents the feature vector of the i-th patient. ~ The values ​​represent the i-th patient's age, gender, surgical segment, preoperative severity of hand sweating, postoperative time, anxiety score, and smoking level, in that order.

4. The multifactorial method for predicting postoperative hemorrhage (CH) in palmar hyperhidrosis as described in claim 1, characterized in that, The covariance function is specifically expressed as follows: ； Where K represents the feature vector and The covariance value; , Let represent the feature vectors of the i-th and j-th patients, with a dimension of 7; It represents the signal variance, used to control the amplitude range of the model output; l represents the length scale, used to determine the similarity decay rate in the feature space; This represents the noise variance, used to represent the random error in observed data. This represents the Kronecker Delta function, when i=j. Otherwise, it is 0.

5. The multifactorial method for predicting postoperative hemorrhage (CH) in palmar hyperhidrosis as described in claim 4, characterized in that, Step S4 includes the following sub-steps: S41: Define the log-likelihood function as: ； in, This represents the transpose of the target variable vector Y; The three hyperparameters of the covariance function are represented; I represents the identity matrix, whose dimension is the same as the number of training samples n; K represents the covariance matrix; S42: Calculate the log-likelihood pairs The partial derivatives; S43: Maximize the log-likelihood function using the conjugate gradient algorithm until the log-likelihood is maximized. Convergence is achieved, and the hyperparameters of the converged covariance function are obtained.

6. The multifactorial method for predicting postoperative hemorrhage (CH) in palmar hyperhidrosis as described in claim 1, characterized in that, Step S5 includes the following sub-steps: S51: Input the current patient feature vector and calculate the current patient's predicted distribution: ； ； ； in, This represents the covariance vector between the current patient feature vector and the training sample X, with a dimension of 1×n; This represents the covariance matrix of the training sample X itself; This represents the covariance scalar of the current patient's feature vector itself; This indicates the predicted degree of postoperative hemorrhage in the current patient; This represents the predicted probability distribution of the current patient's postoperative CH severity. N represents the normal distribution, used to describe Distribution pattern; This represents the predicted mean of the postoperative CH severity in current patients; I represents the identity matrix; S52: Based on the distribution of the target variables in the training set, a preset discretization threshold range is determined; S53: Based on the discretization threshold range, the predicted mean of the postoperative CH degree of the current patient is discretized and mapped to obtain the postoperative CH degree of the current patient.

7. A multifactor-based postoperative CH prediction system for palmar hyperhidrosis, used for prediction using the method described in any one of claims 1 to 6, characterized in that, The system includes: The data acquisition module is used to acquire postoperative historical data from multiple groups of patients, including target variables and various characteristic variables that affect the presence of CH after hand hyperhidrosis surgery, and to preprocess the historical data. The data partitioning module is used to represent each patient's historical data in the form of a multidimensional feature vector and to partition the data into training and validation sets. The model building module is used to build a Gaussian regression model, selecting the radial basis kernel and white noise kernel as the covariance function; The model training module is used to optimize the hyperparameters of the covariance function on the training set by maximizing the log-likelihood function. The prediction module is used to input the current patient feature vector into the optimized model and output the prediction result.

8. The multifactorial palmar hyperhidrosis postoperative CH prediction system as described in claim 7, characterized in that, The system also includes a decision reporting module, which generates clinical decision reports based on the prediction results.

Images