Thoracic postoperative neuropathic pain risk prediction method and system based on machine learning

Abstract

The invention discloses a thoracic postoperative neuropathic pain risk prediction method and system based on machine learning, and relates to the technical field of medical information. According to the method, multiple advanced machine learning models and optimization technologies are comprehensively applied, so that the accuracy of postoperative neuropathic pain risk prediction is remarkably improved; pain data at multiple time points before operation, during operation and after operation are comprehensively considered, dynamic updating and iterative prediction can be achieved, and postoperative analgesia management is optimized in time; a set of complete automatic solution is provided, whole-process systematization of data acquisition, risk prediction and result judgment is realized, and powerful data support is provided for clinical decision making. The combination of the method and the system not only improves the accuracy and reliability of prediction, but also optimizes the postoperative pain management, and improves the postoperative life quality of the patient.

Description

Technical Field

[0001] The present invention relates to the field of medical information technology, and in particular to a method and system for predicting the risk of postoperative neuropathic pain in thoracic surgery based on machine learning. Background Art

[0002] Surgery plays an important role in the treatment of thoracic organ diseases. Persistent postoperative pain (PPSP) is a common complication of many surgical operations, and thoracic surgery is one of the most painful surgical operations.

[0003] Neuropathic pain caused by nerve damage has been repeatedly considered to be the main cause of PPSP. Studies have shown that the prevalence of possible or confirmed neuropathic pain in patients with persistent pain after chest and breast surgery is as high as 66% and 68%. Postoperative neuropathic pain seriously affects the patient's postoperative life, often accompanied by depression, anxiety, and damage to sleep, social function, and quality of life, and even loss of daily living ability. Neuropathic pain has poor relief effect, short duration of treatment response, and common and unbearable treatment side effects. Therefore, prediction and prevention are the most important links in the prevention and treatment of postoperative neuropathic pain.

[0004] At present, relevant research on existing technologies mainly focuses on the establishment of prediction models for chronic pain after thoracic and knee replacement surgery and the analysis of risk factors for neuropathic pain after thoracic tumor surgery. Most of the models use logistic regression, and there is a lack of risk prediction methods for postoperative neuropathic pain after thoracic tumor surgery. Summary of the invention

[0005] The purpose of the present invention is to provide a method and system for predicting the risk of postoperative neuropathic pain in thoracic surgery based on machine learning, which can effectively predict the risk of postoperative neuropathic pain in patients and provide clinical medical staff with accurate prevention and intervention plans.

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

[0007] A risk prediction method for postoperative neuropathic pain in thoracic surgery based on machine learning includes:

[0008] Based on time conditions, pain data of patients with thoracic tumors are obtained and preprocessed to obtain risk factors; the pain data includes preoperative pain data, intraoperative pain data, and postoperative pain data;

[0009] The risk factors are input into the risk prediction model for postoperative neuropathic pain, and the risk prediction results are output;

[0010] Determine whether the risk prediction result meets the safety threshold; if so, complete the postoperative pain risk prediction; otherwise, update the risk factor and iterate again.

[0011] Furthermore, the preoperative pain data include the patient's age, gender, BMI, whether he or she has undergone radiotherapy or chemotherapy before surgery, drinking history, smoking history, chest surgery history, marital status, occupation, education level, ASA grade, DN4 scale to assess whether he or she has neuropathic pain before surgery, and whether he or she has used hypnotic drugs before surgery;

[0012] The intraoperative pain data include surgical method, surgical name, surgical time, intraoperative blood loss, number of chest drainage tubes placed, and surgeon information;

[0013] The postoperative pain data included NRS score, DN4 scale score, degree of incision infection, I-DN4 scale score, analgesic medication instructions, and chest drainage tube placement time.

[0014] Furthermore, the time conditions include the day of surgery, 1 day after surgery, 3 days after surgery, 7 days after surgery, 1 month after surgery and 3 months after surgery; when the time conditions are different, the risk factors input into the postoperative neuropathic pain risk prediction model are different.

[0015] Furthermore, the preprocessing includes:

[0016] Based on the normal distribution, statistical analysis is performed on the pain data to obtain statistical analysis results; based on the statistical analysis results, the statistical significance of the pain data is calculated;

[0017] Based on statistical significance, the pain data are screened to obtain the screened pain data;

[0018] The screened pain data were screened for variable characteristics to obtain risk factors.

[0019] Furthermore, the postoperative neuropathic pain risk prediction model includes a global feature risk prediction module, a numerical feature risk prediction module and a core feature risk prediction module; the global feature risk prediction module can adopt any one of the SVM model, RF model, LGBM model, LR model, KNN model, and XGBoost model; the numerical feature risk prediction module can adopt any one of the SVM model, RF model, LGBM model, LR model, KNN model, and XGBoost model; the core feature risk prediction module can adopt any one of the SVM model, RF model, LGBM model, LR model, KNN model, and XGBoost model.

[0020] Furthermore, the process of constructing the postoperative neuropathic pain risk prediction model includes:

[0021] Obtain historical risk factors, divide them using K-fold cross validation, and obtain training historical risk data and verification historical risk data;

[0022] The training historical risk data were input into the SVM model, RF model, LGBM model, LR model, KNN model and XGBoost model respectively, and grid search and random search were used for iterative training to obtain each model after initial training;

[0023] Input the verification historical risk data into each initially trained model and calculate the corresponding AUC index;

[0024] Based on the AUC index and model characteristics, three models were selected as the global feature risk prediction module, the numerical feature risk prediction module and the core feature risk prediction module to complete the construction of the postoperative neuropathic pain risk prediction model.

[0025] Furthermore, the processing process of the postoperative neuropathic pain risk prediction model includes:

[0026] Classify risk factors to obtain numerical risk factors and core risk factors;

[0027] Input the risk factors into the global characteristic risk prediction module, and output the global risk prediction value;

[0028] Inputting the numerical risk factors into the numerical characteristic risk prediction module, and outputting the numerical risk prediction value;

[0029] The core risk factors are input into the core characteristic risk prediction module, and the core risk prediction value is output;

[0030] Setting risk prediction weight parameters through weighted average method;

[0031] Based on the risk prediction weight parameters, the global risk prediction value, the numerical risk prediction value and the core risk prediction value, the risk prediction value is calculated to obtain the risk prediction result.

[0032] Furthermore, the risk prediction result is judged whether it meets the safety threshold; if so, the postoperative pain risk prediction is completed; otherwise, the risk factor is updated and iterated again. Including:

[0033] Determine whether the risk prediction results meet the safety threshold; if the analgesic medication is maintained, complete the postoperative pain risk prediction;

[0034] Otherwise, it is determined whether the patient's rehabilitation treatment time meets the time conditions; if so, the analgesic drug prescription is adjusted, the postoperative pain data is updated, and the preoperative pain data, intraoperative pain data and updated postoperative pain data are input into the postoperative neuropathic pain risk prediction model for iteration; otherwise, the current iteration is stopped, the analgesic drug prescription is adjusted, and the next rehabilitation treatment time is waited.

[0035] A machine learning-based thoracic tumor postoperative pain risk prediction system, including:

[0036] The pain data acquisition and preprocessing module is used to acquire pain data of patients with thoracic tumors and preprocess them to obtain risk factors;

[0037] Postoperative neuropathic pain risk prediction module, used to process risk factors and output risk prediction results;

[0038] The risk prediction result judgment module is used to determine whether the risk prediction result meets the safety threshold; if so, the postoperative pain risk prediction is completed; otherwise, the risk factors are updated and re-iterated.

[0039] Furthermore, the postoperative neuropathic pain risk prediction module includes:

[0040] A risk factor classification unit is used to classify risk factors to obtain numerical risk factors and core risk factors;

[0041] A global characteristic risk prediction unit is used to process risk factors and obtain a global risk prediction value;

[0042] A numerical characteristic risk prediction unit is used to process numerical risk factors to obtain numerical risk prediction values;

[0043] The core characteristic risk prediction unit is used to process the core risk factors and obtain the core risk prediction value;

[0044] The risk prediction result calculation unit is used to set the risk prediction weight parameter; based on the risk prediction weight parameter, the global risk prediction value, the numerical risk prediction value and the core risk prediction value, the risk prediction value is calculated to obtain the risk prediction result.

[0045] The beneficial effects of the present invention are:

[0046] This method significantly improves the accuracy of postoperative neuropathic pain risk prediction by comprehensively applying multiple machine learning models and using techniques such as K-fold cross validation, grid search, and random search for model training and optimization. At the same time, this method takes into account pain data at multiple time points before, during, and after surgery, and can comprehensively assess the patient's risk factors, and dynamically update data for iterative predictions, timely adjust treatment plans, and optimize postoperative analgesia management.

[0047] This system includes a pain data acquisition preprocessing module, a postoperative neuropathic pain risk prediction module and a risk prediction result judgment module, which realizes the automation and systematization of the entire process from data collection to risk prediction and then to result judgment. By classifying and processing risk factors and combining the advantages of different models, the accuracy and reliability of the prediction are further improved, providing strong support for clinical decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0049] Figure 1 A flow chart of a method in an embodiment of the present invention;

[0050] Figure 2 : is a ROC curve diagram of different models in the embodiment of the present invention;

[0051] Figure 3 1 is a system structure diagram in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention.

[0053] See also Figure 1 The present embodiment provides a method for predicting the risk of postoperative neuropathic pain in thoracic surgery based on machine learning, including:

[0054] S1. Based on time conditions, pain data of patients with thoracic tumors are obtained and preprocessed to obtain risk factors; the pain data includes preoperative pain data, intraoperative pain data, and postoperative pain data;

[0055] The preoperative pain data include the patient's age, gender, BMI, whether or not they have undergone radiotherapy or chemotherapy before surgery, drinking history, smoking history, chest surgery history, marital status, occupation, education level, ASA grade, DN4 scale to assess whether or not there is neuropathic pain before surgery, and whether or not hypnotic drugs are used before surgery; preoperative pain data are collected one day before the patient's surgery.

[0056] The intraoperative pain data includes the surgical method, surgical name, surgical time, intraoperative blood loss, number of chest drainage tubes placed, and surgeon information; the surgical method is laparoscopic or open, and the surgeon information includes the surgeon's working time and surgical volume. The intraoperative pain data is obtained through the hand numbness system.

[0057] The postoperative pain data were divided into 1 day after surgery, 3 days after surgery, 7 days after surgery, 1 month after surgery and 3 months after surgery, including NRS score, DN4 scale score, degree of incision infection, I-DN4 scale score, analgesic medication orders, and chest drainage tube placement time.

[0058] The time conditions include the day of surgery, 1 day after surgery, 3 days after surgery, 7 days after surgery, 1 month after surgery and 3 months after surgery; when the time conditions are different, the risk factors input into the postoperative neuropathic pain risk prediction model are different.

[0059] The pre-processing comprises:

[0060] S1-1. Based on normal distribution, statistical analysis is performed on the pain data to obtain statistical analysis results, that is, normal distribution (mean standard deviation), determine the pain data that meet the normal distribution and the pain data that do not meet the normal distribution; based on the statistical analysis results, calculate the statistical significance of the pain data. For the pain data that meet the normal distribution, use the chi-square test or Fish's exact test to calculate the corresponding statistical significance P; for the pain data that do not meet the normal distribution, use the t-test or Mann Whitney rank test to calculate the corresponding statistical significance P;

[0061] S1-2. Screen the pain data based on statistical significance; determine whether the statistical significance P is less than 0.2, if so, retain the corresponding pain data; otherwise, remove the corresponding pain data; obtain the screened pain data;

[0062] S1-3. Use the Lasso regression method to screen the variable characteristics of the screened pain data and obtain risk factors.

[0063] S2, inputting the risk factors into the risk prediction model for postoperative neuropathic pain, and outputting the risk prediction results;

[0064] The postoperative neuropathic pain risk prediction model includes a global feature risk prediction module, a numerical feature risk prediction module and a core feature risk prediction module; the global feature risk prediction module can adopt any one of the SVM model, RF model, LGBM model, LR model, KNN model, and XGBoost model; the numerical feature risk prediction module can adopt any one of the SVM model, RF model, LGBM model, LR model, KNN model, and XGBoost model; the core feature risk prediction module can adopt any one of the SVM model, RF model, LGBM model, LR model, KNN model, and XGBoost model.

[0065] The process of constructing the postoperative neuropathic pain risk prediction model includes:

[0066] S201, obtaining historical risk factors, dividing the historical risk factors using K-fold cross validation, and obtaining training historical risk data and verification historical risk data;

[0067] S202, inputting the training historical risk data into the SVM model, RF model, LGBM model, LR model, KNN model and XGBoost model respectively, and performing iterative training using grid search and random search to obtain each initially trained model;

[0068] S203, respectively input the verified historical risk data into each initially trained model, and calculate the corresponding AUC index; the AUC index is the area of ​​the ROC curve.

[0069] S204. Based on the AUC index and the characteristics of each model in S202 (i.e., the type of data that the model is good at processing), select the model with the largest AUC corresponding to the data types of mixed data (i.e., risk factors), numerical data (i.e., numerical risk factors), and core data (core risk factors), and use them as the global feature risk prediction module, numerical feature risk prediction module, and core feature risk prediction module, respectively, to complete the construction of the postoperative neuropathic pain risk prediction model.

[0070] In this embodiment, pain data of 670 patients with thoracic tumors are selected as training data. As shown in Table 1, the training data are divided into a training set (517 cases) and a test set (130 cases) at a ratio of 8:2.

[0071] Table 1 Training data parameter table

[0072]

[0073] Pearson chi-square test, t-test, and Mann-Whitney U rank test were used for univariate analysis of the training set and validation set, and P < 0.2 was considered statistically significant. It can be seen that 14 factors, including education level, preoperative radiotherapy, preoperative chemotherapy, preoperative continuous pain duration, surgical method, surgical resection site, single-hole or multi-hole, surgeon, postoperative acute pain, postoperative wound infection, operation time, number of chest drainage tubes placed, chest drainage tube placement time, and intraoperative blood loss, are related to the occurrence of postoperative neuropathic pain in thoracic patients.

[0074] Lasso regression was used to screen the data with P < 0.2, and the corresponding risk factors were obtained, including duration of pain, surgical method, name of surgery, surgical resection site, single-hole or multi-hole, number of chest drainage tubes, number of days of chest drainage tube placement, intraoperative blood loss, surgeon, postoperative acute pain, and wound infection.

[0075] The risk factors were input into the SVM model (support vector machine), RF model (random forest), LGBM model (light gradient boosting), LR model (logistic regression), KNN model (nearest neighbor classification), and XGBoost model (extreme gradient boosting) to calculate the corresponding AUC indicators. Figure 2 As shown in the figure, according to the ROC curve, the AUC index of each model is calculated. The RF model has the highest AUC index, reaching 0.85; followed by the XGBoost model, with an AUC index of 0.83; the LGBM model has an AUC index of 0.82; the LR model and the SVM model have an AUC index of 0.79; the KNN model has the lowest AUC index, with an AUC index of 0.69. Figure 2 In the above equation, TPR is the true positive rate and FPR is the false positive rate.

[0076] Since the feature data types adapted by the RF model and the LGBM model are mixed type data (risk factors), the feature data types adapted by the XGBoost model are numerical dominant data (numerical risk factors), the feature data types adapted by the LR model are low-dimensional linear sensitive features (core risk factors), and the AUC index of the RF model is greater than that of the LGBM model, the global feature risk prediction module adopts the RF model, the numerical feature risk prediction module adopts the XGBoost model, and the core feature risk prediction module adopts the LR model.

[0077] The processing process of the postoperative neuropathic pain risk prediction model includes:

[0078] S211. Classify the risk factors to obtain numerical risk factors and core risk factors; numerical risk factors are data with numerical values ​​(time, age, value, etc.), and core risk factors are data directly related to postoperative neuropathic pain. For example, numerical risk factors include age, BMI, operation time, intraoperative blood loss, number of chest drainage tubes placed, NRS score, DN4 scale score, I-DN4 scale score, and chest drainage tube placement time. Core risk factors include DN4 scale assessment of preoperative neuropathic pain, NRS score, DN4 scale score, I-DN4 scale score, and intraoperative blood loss.

[0079] S212: Input the risk factors into the global characteristic risk prediction module, and output a global risk prediction value ;

[0080] S213: Input the numerical risk factors into the numerical feature risk prediction module, and output the numerical risk prediction value ;

[0081] S214: Input the core risk factors into the core characteristic risk prediction module, and output the core risk prediction value ;

[0082] S215. Setting risk prediction weight parameters by weighted average method; risk prediction weight parameters include global risk prediction weight , numerical risk prediction weight , Core risk prediction weight .in, or ,and .

[0083] S216, based on risk prediction weight parameters, global risk prediction value , numerical risk prediction value , core risk prediction value , calculate the risk prediction value and get the risk prediction result , the corresponding formula is:

[0084] .

[0085] S3. Determine whether the risk prediction result meets the safety threshold; if so, complete the postoperative pain risk prediction; otherwise, update the risk factor and iterate again.

[0086] The specific process of S3 is as follows:

[0087] Determine whether the risk prediction results meet the safety threshold; if the analgesic medication is maintained, complete the postoperative pain risk prediction;

[0088] Otherwise, it is determined whether the patient's rehabilitation treatment time meets the time conditions, that is, whether the rehabilitation treatment time falls on the day of surgery, 1 day after surgery, 3 days after surgery, 7 days after surgery, 1 month after surgery, or 3 months after surgery; if so, the analgesic drug order is adjusted, the postoperative pain data is updated, and the preoperative pain data, intraoperative pain data, and updated postoperative pain data are input into the postoperative neuropathic pain risk prediction model for iteration; otherwise, the current iteration is stopped, the analgesic drug order is adjusted, and the next rehabilitation treatment time is waited for.

[0089] When the patient's rehabilitation treatment time is the day of surgery, the pain data are preoperative pain data and intraoperative pain data; when the patient's rehabilitation treatment time is 1 day after surgery, the pain data are preoperative pain data, intraoperative pain data, and postoperative pain data 1 day after surgery. When the patient's rehabilitation treatment time is 3 days after surgery, the pain data are preoperative pain data, intraoperative pain data, and first postoperative pain data; the first postoperative pain data includes postoperative pain data corresponding to 1 day after surgery and 3 days after surgery. When the patient's rehabilitation treatment time is 7 days after surgery, the pain data are preoperative pain data, intraoperative pain data, and second postoperative pain data; the second postoperative pain data includes postoperative pain data corresponding to 1 day after surgery, 3 days after surgery, and 7 days after surgery. When the patient's rehabilitation treatment time is 1 month, the pain data are preoperative pain data, intraoperative pain data, and third postoperative pain data; the third postoperative pain data includes postoperative pain data corresponding to 1 day after surgery, 3 days after surgery, 7 days after surgery, and 1 month after surgery. When the patient's rehabilitation treatment time is 3 months after the operation, the pain data includes the preoperative pain data, the intraoperative pain data and the fourth postoperative pain data; the fourth postoperative pain data includes the postoperative pain data corresponding to the 1st day after operation, the 3rd day after operation, the 7th day after operation, the 1st month after operation and the 3rd month after operation.

[0090] The time for postoperative pain risk prediction (rehabilitation treatment time) for patients with thoracic tumors includes the day of surgery, 1 day after surgery, 3 days after surgery, 7 days after surgery, 1 month after surgery, and 3 months after surgery, that is, the iterative time for patients to undergo postoperative pain risk prediction.

[0091] This method significantly improves the accuracy of postoperative neuropathic pain risk prediction by comprehensively applying multiple machine learning models and using techniques such as K-fold cross validation, grid search, and random search for model training and optimization. At the same time, this method takes into account pain data at multiple time points before, during, and after surgery, and can comprehensively assess the patient's risk factors, and dynamically update data for iterative predictions, timely adjust treatment plans, and optimize postoperative analgesia management.

[0092] like Figure 3 As shown, a machine learning-based thoracic tumor postoperative pain risk prediction system includes:

[0093] The pain data acquisition and preprocessing module is used to acquire pain data of patients with thoracic tumors and preprocess them to obtain risk factors;

[0094] Postoperative neuropathic pain risk prediction module, used to process risk factors and output risk prediction results;

[0095] The postoperative neuropathic pain risk prediction module includes:

[0096] A risk factor classification unit is used to classify risk factors to obtain numerical risk factors and core risk factors;

[0097] A global characteristic risk prediction unit is used to process risk factors and obtain a global risk prediction value;

[0098] A numerical characteristic risk prediction unit is used to process numerical risk factors to obtain numerical risk prediction values;

[0099] The core characteristic risk prediction unit is used to process the core risk factors and obtain the core risk prediction value;

[0100] The risk prediction result calculation unit is used to set the risk prediction weight parameter; based on the risk prediction weight parameter, the global risk prediction value, the numerical risk prediction value and the core risk prediction value, the risk prediction value is calculated to obtain the risk prediction result.

[0101] The risk prediction result judgment module is used to determine whether the risk prediction result meets the safety threshold; if so, the postoperative pain risk prediction is completed; otherwise, the risk factors are updated and re-iterated.

[0102] This system includes a pain data acquisition preprocessing module, a postoperative neuropathic pain risk prediction module and a risk prediction result judgment module, which realizes the automation and systematization of the entire process from data collection to risk prediction and then to result judgment. By classifying and processing risk factors and combining the advantages of different models, the accuracy and reliability of the prediction are further improved, providing strong support for clinical decision-making.

[0103] In summary, the present invention significantly improves the accuracy of risk prediction of postoperative neuropathic pain by comprehensively using a variety of advanced machine learning models and optimization techniques; comprehensively considers pain data at multiple time points before, during and after surgery, can dynamically update and iterate predictions, and timely optimize postoperative analgesia management; provides a complete set of automated solutions, realizes the systematization of the entire process of data collection, risk prediction and result judgment, and provides strong data support for clinical decision-making. The combination of this method and system not only improves the accuracy and reliability of predictions, but also optimizes postoperative pain management and improves the patient's postoperative quality of life.

[0104] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for predicting the risk of postoperative neuropathic pain in thoracic surgery based on machine learning, characterized in that: include: Based on time conditions, pain data of thoracic tumor patients were obtained and preprocessed to obtain risk factors; The pain data includes preoperative pain data, intraoperative pain data and postoperative pain data; The risk factors are input into the risk prediction model for postoperative neuropathic pain, and the risk prediction results are output; Determine whether the risk prediction results meet the safety threshold; If yes, then complete the postoperative pain risk prediction; Otherwise, update the risk factors and iterate again; The postoperative neuropathic pain risk prediction model includes a global feature risk prediction module, a numerical feature risk prediction module and a core feature risk prediction module; The global feature risk prediction module can adopt any one of the SVM model, RF model, LGBM model, LR model, KNN model, and XGBoost model; the numerical feature risk prediction module can adopt any one of the SVM model, RF model, LGBM model, LR model, KNN model, and XGBoost model; the core feature risk prediction module can adopt any one of the SVM model, RF model, LGBM model, LR model, KNN model, and XGBoost model; The processing process of the postoperative neuropathic pain risk prediction model includes: Classify risk factors to obtain numerical risk factors and core risk factors; Input the risk factors into the global characteristic risk prediction module, and output the global risk prediction value; Inputting the numerical risk factors into the numerical characteristic risk prediction module, and outputting the numerical risk prediction value; The core risk factors are input into the core characteristic risk prediction module, and the core risk prediction value is output; Setting risk prediction weight parameters through weighted average method; Based on the risk prediction weight parameters, global risk prediction value, numerical risk prediction value and core risk prediction value, the risk prediction value is calculated to obtain the risk prediction result.

2. The method for predicting the risk of postoperative neuropathic pain in thoracic surgery based on machine learning according to claim 1, characterized in that: The preoperative pain data include the patient's age, gender, BMI, whether he or she has undergone radiotherapy or chemotherapy before surgery, drinking history, smoking history, chest surgery history, marital status, occupation, education level, ASA grade, DN4 scale to assess whether he or she has neuropathic pain before surgery, and whether he or she has used hypnotic drugs before surgery; The intraoperative pain data include surgical method, surgical name, surgical time, intraoperative blood loss, number of chest drainage tubes placed, and surgeon information; The postoperative pain data included NRS score, DN4 scale score, degree of incision infection, I-DN4 scale score, analgesic medication instructions, and chest drainage tube placement time.

3. The method for predicting the risk of postoperative neuropathic pain in thoracic surgery based on machine learning according to claim 1, characterized in that: The time conditions include the day of surgery, 1 day after surgery, 3 days after surgery, 7 days after surgery, 1 month after surgery and 3 months after surgery; when the time conditions are different, the risk factors input into the postoperative neuropathic pain risk prediction model are different.

4. The method for predicting the risk of postoperative neuropathic pain in thoracic surgery based on machine learning according to claim 1, characterized in that: The pre-processing comprises: Based on the normal distribution, statistical analysis is performed on the pain data to obtain statistical analysis results; based on the statistical analysis results, the statistical significance of the pain data is calculated; Based on statistical significance, the pain data are screened to obtain the screened pain data; The screened pain data were screened for variable characteristics to obtain risk factors.

5. The method for predicting the risk of postoperative neuropathic pain in thoracic surgery based on machine learning according to claim 1, characterized in that: The process of constructing the postoperative neuropathic pain risk prediction model includes: Obtain historical risk factors, divide them using K-fold cross validation, and obtain training historical risk data and verification historical risk data; The training historical risk data were input into the SVM model, RF model, LGBM model, LR model, KNN model and XGBoost model respectively, and grid search and random search were used for iterative training to obtain each model after initial training; Input the verification historical risk data into each initially trained model and calculate the corresponding AUC index; Based on the AUC index and model characteristics, three models were selected as the global feature risk prediction module, the numerical feature risk prediction module and the core feature risk prediction module to complete the construction of the postoperative neuropathic pain risk prediction model.

6. The method for predicting the risk of postoperative neuropathic pain in thoracic surgery based on machine learning according to claim 1 or 3, characterized in that: The step of determining whether the risk prediction result meets the safety threshold; if so, completing the postoperative pain risk prediction; Otherwise, update the risk factors and iterate again, including: Determine whether the risk prediction results meet the safety threshold; if the analgesic medication is maintained, complete the postoperative pain risk prediction; Otherwise, it is determined whether the patient's rehabilitation treatment time meets the time conditions; if so, the analgesic drug prescription is adjusted, the postoperative pain data is updated, and the preoperative pain data, intraoperative pain data and updated postoperative pain data are input into the postoperative neuropathic pain risk prediction model for iteration; otherwise, the current iteration is stopped and the analgesic drug prescription is adjusted.

7. A system for predicting the risk of postoperative neuropathic pain in thoracic surgery based on machine learning, used to implement the method for predicting the risk of postoperative neuropathic pain in thoracic surgery based on machine learning according to any one of claims 1 to 6, characterized in that: include: The pain data acquisition and preprocessing module is used to acquire pain data of patients with thoracic tumors and preprocess them to obtain risk factors; Postoperative neuropathic pain risk prediction module, used to process risk factors and output risk prediction results; The risk prediction result judgment module is used to judge whether the risk prediction result meets the safety threshold; if so, the postoperative pain risk prediction is completed; Otherwise, update the risk factors and iterate again.

8. The machine learning-based thoracic postoperative neuropathic pain risk prediction system according to claim 7, characterized in that: The postoperative neuropathic pain risk prediction module includes: A risk factor classification unit is used to classify risk factors to obtain numerical risk factors and core risk factors; A global characteristic risk prediction unit is used to process risk factors and obtain a global risk prediction value; A numerical characteristic risk prediction unit is used to process numerical risk factors to obtain numerical risk prediction values; The core characteristic risk prediction unit is used to process the core risk factors and obtain the core risk prediction value; The risk prediction result calculation unit is used to set the risk prediction weight parameter; based on the risk prediction weight parameter, the global risk prediction value, the numerical risk prediction value and the core risk prediction value, the risk prediction value is calculated to obtain the risk prediction result.

Images