Machine learning method for rapid preoperative prediction of pain improvement index after vertebroplasty
The support vector machine model predicts the pain after vertebroplasty, solves the problem of inaccurate postoperative pain assessment, and realizes the formulation of scientific analgesic plans to reduce the risk of patients' pain and complications, and promotes rehabilitation.
Patent Information
- Application Number
- CN202410408057.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-07
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-04-07
AI Technical Summary
After vertebroplasty, patients often experience residual pain in the lower back, which affects early functional exercise and rehabilitation after the operation. The existing pain assessment methods are difficult to accurately predict the improvement of postoperative pain, resulting in lag in analgesic treatment, increasing the patient's pain and Complication risk.
A type of support vector machine (OCSVM) method is used to predict postoperative pain based on patient clinical data such as gender, age, injury site, admission pain index, etc., and adjust preoperative and intraoperative analgesic plans through training models to ensure The patient's perioperative pain is within the tolerable range.
Effectively predict the improvement of pain after surgery, formulate scientific analgesic plans for clinicians to reduce patients' anxiety, improve comfort, promote early functional exercise and rapid recovery, and reduce the risk of complications caused by pain.
Smart Images

Figure CN118230954B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the fields of medical informatics, biostatistics and pain management, and specifically relates to a machine learning method for quickly predicting the pain improvement index after vertebroplasty before surgery. Background Art
[0002] Percutaneous vertebroplasty (PVP) has been recognized for its effectiveness in treating osteoporotic vertebral compression fractures and can effectively reduce patients' pain. Although literature reports that the pain relief rate after PVP is as high as 78%-95%, residual pain in the back and waist after PVP is not uncommon in clinical work, with a minimum incidence of about 5% and a maximum of 22%, which seriously affects the patient's quality of life after surgery. For orthopedic patients, early postoperative functional exercise is one of the important factors to promote rehabilitation, but postoperative pain often becomes the biggest obstacle to early functional exercise and thus promote rehabilitation. With the continuous development of medicine and nursing, pain as the fifth vital sign has attracted more and more attention. Timely and effective analgesia during the perioperative period can relieve patients' pain and reduce the occurrence of postoperative complications, which is crucial to preventing the progressive development of osteoporosis and promoting rehabilitation.
[0003] As the concept of rapid recovery continues to deepen in clinical work, medical workers are paying more and more attention to pain. At present, the methods for evaluating postoperative pain in patients in clinical practice include visual analogue scale (VAS), numerical rating scale (NRS), verbal rating scale (VRS) and Faces pain scale-revised (FPS-R). Usually, clinicians give patients analgesia based on the pain score of the patient after surgery. When the patient's pain index is > 5 points, symptomatic treatment is given. The stress response caused by pain often increases the pain of patients and causes harm. Therefore, there is an urgent need for clinical methods to predict the improvement of postoperative pain in patients before surgery. By formulating intraoperative and postoperative analgesia plans before surgery, effective pain control can be achieved, the comfort of surgical patients can be improved, complications caused by pain can be reduced, and patients can be encouraged to cooperate with functional exercises early to accelerate their recovery. Machine learning methods have emerged on the medical stage. Domestic and foreign scholars have conducted a lot of research on machine learning methods in orthopedic clinical medicine. Studies have shown that machine learning methods can be used in orthopedic clinical medical treatment. Summary of the invention
[0004] In order to solve the above technical problems, the present invention provides a machine learning method for preoperative rapid prediction of the pain improvement index after vertebroplasty. The clinical data of the patient are used, based on gender, age, number of injured parts (thoracic vertebrae, lumbar vertebrae), injury time, admission pain index, hospitalization days, postoperative pain score and pain improvement, etc., and a one-class support vector machine (OCSVM) method is used to predict the postoperative pain condition of the patient before the PVP surgery. The research results provide guidance for clinicians to formulate analgesia plans before surgery, and also provide a basis for formulating analgesia plans during and after surgery. An effective analgesia plan can control the patient's pain throughout the perioperative period to a level that the patient can tolerate, that is, mild pain, which helps to reduce the patient's anxiety, increase comfort, and is of great significance for the patient to actively cooperate with treatment, perform functional exercises at an early stage, and promote rapid recovery.
[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:
[0006] A machine learning method for preoperatively quickly predicting the pain improvement index after vertebroplasty includes the following steps:
[0007] Step 1: Collect the patient's gender, age, number of injured parts (thoracic vertebrae, lumbar vertebrae), injury time, pain upon admission, and length of hospital stay, and use the above clinical data as training samples;
[0008] Step 2: Check and process missing values and outliers. Convert non-numerical features (such as gender) to numerical types and normalize numerical features to better fit the model.
[0009] Step 3: Select a type of support vector machine as the anomaly monitoring model, determine the radial basis function (RBF) as the kernel function, and calculate the kernel parameters through the optimization method;
[0010] Step 4: Use the patient's clinical data to train a support vector machine model;
[0011] Step 5: Use the trained model to predict the patient's postoperative pain improvement;
[0012] Step 6: Adjust treatment plans based on the predicted results, such as patients who may not experience significant pain improvement after surgery and may need more intensive care or pain management.
[0013] Furthermore, the support vector machine model in step 3 is expressed as:
[0014] Considering clinical data as training samples, in order to obtain the hyperplane, the optimization problem of one-class support vector machine OCSVM can be described as:
[0015] (1)
[0016] In the formula, Indicates that the variable is The minimum value of the function; represents the normal vector of the hyperplane; represents the offset of the hyperplane; [ ⋅ ] T Represents the transpose of a matrix; δ ∈ ( 0 , 1 ] Indicates the maximum misclassification rate of training samples; is the number of training samples; From 1 to integer variable of ; is the slack variable; is a constraint condition; Represents a nonlinear mapping that projects raw data into a high-dimensional feature space; Represents the i-th sample element, also called the support vector;
[0017] Enter the first and second Lagrange multipliers into formula (1) , for:
[0018] (2)
[0019] In the formula, represents the objective function to be optimized after the Lagrange multiplier is introduced; are the first and second Lagrange multipliers and slack variables corresponding to the i-th sample, respectively.
[0020] According to the Karush-Kuch-Tucker (KKT) condition, the following formula can be obtained:
[0021] (3)
[0022] Substituting (3) into (2) yields:
[0023] (4)
[0024] Substituting equation (4) into equation (2) yields the dual form of the elementary optimization problem:
[0025] (5)
[0026] in, α = [ α 1 , α 2 , ⋯ , α N ] T ; is the kernel matrix calculated from the training samples, , represents the kernel function; subscript and Represents the index of the kernel matrix element; From 1 to An integer variable.
[0027] By solving the above dual problem, we get the Lagrange multiplier, calculate the normal vector, and obtain the hyperplane with the optimal classification interval.
[0028] The kernel function in step 3 combines the feature map and the inner product operation, and the selected kernel function Mercer conditions must be met. Since the radial basis kernel function RBF has good generalization ability and fast convergence, it requires fewer parameters and improves the effective classification rate of the patient's postoperative pain improvement. Therefore, the radial basis kernel function RBF is used as the nonlinear mapping function, as shown in formula (6):
[0029] (6)
[0030] Wherein, exp represents the exponential function; is the bandwidth of the Gaussian kernel; represents the two-norm. and represents the i-th sample element and the j-th sample element; the decision function f can be expressed as:
[0031] (7)
[0032] in, represents the sample element; sign represents the sign function; represents the index of the support vector; is the support vector; It can be solved by the following formula:
[0033] (8)
[0034] Among them, the sample The Lagrange multiplier of is not 0. For the sample ,when When , it means that the postoperative pain state of the patient is mild pain; when When , it means that the patient's pain status after surgery is moderate or severe.
[0035] The kernel parameters include two, namely, And the maximum misclassification rate of training samples .in, To avoid the trial and error method of selecting kernel parameters , this paper establishes the objective function by maximizing the difference between the average distance of the farthest element and the average distance of the nearest element, and solves the optimal kernel parameter. The expression is as follows:
[0036] (9)
[0037] In the formula, is the objective function; Indicates the maximum value.
[0038] Therefore, the RBF kernel function is:
[0039] (10)
[0040] The Gaussian kernel bandwidth in the radial basis kernel function RBF is obtained by solving formula (10): .
[0041] At present, both deterministic algorithms and metaheuristic algorithms can solve the objective function, that is, formula (10). The Trees Social Relations Optimization Algorithm (TSR) is a metaheuristic algorithm proposed by Alimoradi M. et al. in 2021. The Trees Social Relations Optimization Algorithm TSR is based on the hierarchical relationship of trees in the forest and simulates the collective living behavior of trees. Each tree represents a solution and the algorithm is used to find the optimal solution. Since TSR has fewer parameters, fast convergence speed, and high solution accuracy, this algorithm is used to solve the kernel parameters in the RBF kernel function to avoid blindly selecting kernel parameters.
[0042] Compared with conventional postoperative pain index assessment methods, the OCSVM method disclosed in the present invention for preoperatively predicting the improvement of postoperative pain in patients has the following advantages:
[0043] 1) The OCSVM method can predict the improvement of postoperative pain in patients before surgery, provide guidance for intraoperative and postoperative anesthesia medication regimens, and promote the physical and mental recovery of patients;
[0044] 2) The computational complexity of the OCSVM method is relatively low and the solution speed is fast, which is conducive to clinical application;
[0045] 3) The OCSVM method can effectively utilize clinical data to handle complex classification problems such as high dimensionality, small samples, and nonlinearity, and is very suitable for predicting patients' postoperative pain improvement.
[0046] In summary, the OCSVM method disclosed in the present invention can quickly predict the improvement of postoperative pain in patients before PVP surgery, and provide a reliable basis for anesthesiologists to select intraoperative anesthesia methods, formulate intraoperative medication plans and postoperative analgesia plans. At the same time, it is of great significance to give patients preoperative health education, reduce anxiety and fear, and promote patient recovery. And ensure that the patient's pain is always at a tolerable level throughout the perioperative period, and improve the patient's quality of life after surgery. This technology can be applied to various medical institutions, such as hospitals, clinics and rehabilitation centers, and has broad market prospects for implementation and transfer. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 This is a flow chart of a machine learning method for quickly predicting the pain improvement index after vertebroplasty before surgery according to the present invention. DETAILED DESCRIPTION
[0048] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.
[0049] Classic PVP is performed through bilateral pedicle puncture. However, compared with conventional unilateral PVP, bilateral PVP can increase the distribution effect and stability of bone cement, but it will increase the risk of puncture and cause greater pain to patients. Some patients cannot even tolerate the pain, which makes the operation unable to proceed smoothly. At the same time, the pain caused by fracture will affect the patient's cooperation with the position required by the operating physician, and the repeated puncture will increase the patient's psychological and physical pain. Patients may suffer from tachycardia and hypertension due to pain during PVP. At the same time, pain may increase the risk of surgery and lead to a sharp decline in patient satisfaction. In addition, there will be a high frequency of intraoperative hypotension, which increases the risk of surgery. Therefore, in order to ensure the smooth progress of the operation and improve the patient's comfort and cooperation, it is of great significance to implement preoperative prediction of the improvement of postoperative pain in patients. The present invention divides the patient's postoperative pain improvement into good and poor, and realizes the preoperative prediction of the patient's postoperative pain improvement, which is a binary classification problem. The present invention uses the patient's clinical data based on gender, age, number of injured parts (thoracic vertebrae, lumbar vertebrae), injury time, admission pain, and postoperative pain score, and uses a machine learning method to predict the postoperative pain improvement of the patient before performing PVP surgery.
[0050] At present, the commonly used machine learning classification methods include support vector machines, extreme learning machines, decision trees, etc. Since OCSVM can use kernel techniques to solve nonlinear classification problems based on small samples, the present invention uses the OCSVM model to predict the postoperative pain of patients before performing PVP surgery (the calculation process is shown in Figure 1The results of this study provide guidance for clinicians to formulate analgesia plans before surgery, and also provide a basis for formulating analgesia plans during and after surgery. An effective analgesia plan can control the pain of patients throughout the perioperative period to a level that the patients can tolerate, that is, mild pain, which helps to reduce patients' anxiety, increase comfort, and is of great significance for patients to actively cooperate with treatment, conduct functional exercises at an early stage, and promote rapid recovery.
[0051] The machine learning method for quickly predicting the pain improvement index after vertebroplasty before surgery of the present invention comprises the following steps:
[0052] Step 1: Collect the patient's gender, age, number of injured parts (thoracic vertebrae, lumbar vertebrae), injury time, pain upon admission, and length of hospital stay, and use the above clinical data as training samples;
[0053] Step 2: Check and process missing values and outliers. Convert non-numerical features (such as gender) to numerical types and normalize numerical features to better fit the model.
[0054] Step 3: Select a type of support vector machine as the anomaly monitoring model, determine the radial basis function (RBF) as the kernel function, and calculate the kernel parameters through the optimization method;
[0055] Step 4: Use the patient's clinical data to train a support vector machine model;
[0056] Step 5: Use the trained model to predict the patient's postoperative pain improvement;
[0057] Step 6: Adjust treatment plans based on the predicted results, such as patients who may not experience significant pain improvement after surgery and may need more intensive care or pain management.
[0058] The support vector machine model in step 3 is expressed as:
[0059] Considering clinical data as training samples, in order to obtain the hyperplane, the optimization problem of one-class support vector machine OCSVM can be described as:
[0060] (1)
[0061] In the formula, Indicates that the variable is The minimum value of the function; represents the normal vector of the hyperplane; represents the offset of the hyperplane; [ ⋅ ] T Represents the transpose of a matrix; δ ∈ ( 0 , 1 ] Indicates the maximum misclassification rate of training samples; is the number of training samples; From 1 to integer variable of ; is the slack variable; is a constraint condition; Represents a nonlinear mapping that projects raw data into a high-dimensional feature space; represents the i-th sample element, also called the support vector; enter the first and second Lagrange multipliers into formula (1) , for:
[0062] (2)
[0063] In the formula, represents the objective function to be optimized after the Lagrange multiplier is introduced; are the first and second Lagrange multipliers and slack variables corresponding to the i-th sample, respectively.
[0064] According to the Karush-Kuch-Tucker (KKT) condition, the following formula can be obtained:
[0065] (3)
[0066] Substituting (3) into (2) yields:
[0067] (4)
[0068] Substituting equation (4) into equation (2) yields the dual form of the elementary optimization problem:
[0069] (5)
[0070] in, α = [ α 1 , α 2 , ⋯ , α N ] T ; is the kernel matrix calculated from the training samples, , Represents kernel function; index of kernel matrix element; subscript and Represents the index of the kernel matrix element; From 1 to An integer variable.
[0071] By solving the above dual problem, we get the Lagrange multiplier, calculate the normal vector, and obtain the hyperplane with the optimal classification interval.
[0072] The kernel function in step 3 combines the feature map and the inner product operation, and the selected kernel function The Mercer condition must be met. Since the radial basis kernel function RBF has good generalization ability and fast convergence, it requires fewer parameters and improves the effective classification rate of the patient's postoperative pain improvement. Therefore, the radial basis kernel function RBF is used as the nonlinear mapping function, as shown in formula (6).
[0073] (6)
[0074] Wherein, exp represents the exponential function; is the bandwidth of the Gaussian kernel; represents the two-norm. and represents the i-th sample element and the j-th sample element, also known as the support vector; the decision function f is expressed as:
[0075] (7)
[0076] in, represents the sample element; sign represents the sign function; represents the index of the support vector; It can be solved by the following formula.
[0077] (8)
[0078] Among them, the sample The Lagrange multiplier of is not 0. For the sample ,when When , it means that the postoperative pain state of the patient is mild pain; when When , it means that the patient's pain status after surgery is moderate or severe.
[0079] The kernel parameters in step 3 include two, namely, And the maximum misclassification rate of training sample elements .in, The given hyperparameters are set for the experiment. In order to avoid the trial and error method to select the kernel parameters , this paper establishes the objective function by maximizing the difference between the average distance of the farthest element and the average distance of the nearest element, and solves the optimal kernel parameter. The expression is as follows:
[0080] (9)
[0081] In the formula, is the objective function; Indicates the maximum value.
[0082] Therefore, the RBF kernel function is:
[0083] (10)
[0084] At present, both deterministic algorithms and metaheuristic algorithms can solve the objective function, that is, formula (10). The Trees Social Relations Optimization Algorithm (TSR) is a metaheuristic algorithm proposed by Alimoradi M. et al. in 2021. The Trees Social Relations Optimization Algorithm TSR is based on the hierarchical relationship of trees in the forest and simulates the collective living behavior of trees. Each tree represents a solution and the algorithm is used to find the optimal solution. Since TSR has fewer parameters, fast convergence speed, and high solution accuracy, this algorithm is used to solve the kernel parameters in the RBF kernel function to avoid blindly selecting kernel parameters.
[0085] In step 4, the clinical data of the patients, including gender, age, number of injured parts (thoracic vertebrae, lumbar vertebrae), injury time, admission pain index, hospitalization days, and postoperative pain score, are used to train a class support vector machine model, and ten-fold cross validation is used to evaluate the stability and accuracy of the model.
[0086] The OCSVM method was used to analyze the risk factors of vertebroplasty, explore the application of the OCSVM model in predicting postoperative pain in patients before PVP surgery, and reduce the pain of patients with thoracolumbar osteoporotic vertebral compression fractures (OVCF) by formulating a reasonable anesthesia plan. In this section, the logistic classification method, extreme learning machine, decision tree, and discriminant analysis method were used to train and classify samples of patients with osteoporotic compression fractures, and the results were compared with those calculated by the OCSVM method, as shown in Table 1.
[0087] Table 1. Comparison of prediction errors of different machine learning methods
[0088]
[0089] Table 1 shows that the logistic classification method was used to train clinical samples of 102 patients with osteoporotic compression fractures. The accuracy of 18 samples for PVP before surgery to predict the improvement of postoperative pain in patients was 88.89%. The extreme learning machine was used to predict the preoperative pain index of patients with osteoporotic compression fractures. When the number of hidden units was 10, the classification accuracy was 77.78%. The random forest model constructed on the training data was used to predict the improvement of postoperative pain in some patients before surgery, and the classification accuracy was 88.89%. The classification accuracy of discriminant analysis for predicting the improvement of postoperative pain in patients before PVP was 83.33%. According to the results in the table, compared with the logistic classification method, extreme learning machine, random forest and discriminant analysis, the one-class support vector machine OCSVM has the highest classification accuracy for preoperative prediction of postoperative pain improvement in patients.
[0090] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A machine learning method for rapid preoperative prediction of pain improvement index after vertebroplasty, characterized in that: The following steps are involved: Step 1: Collect clinical data, including patient gender, age, number of injured parts, injury time, admission pain index, and hospitalization days; Step 2: Preprocess the clinical data, check and process missing values and outliers, convert non-numeric features into numerical features, and normalize the numerical features; Step 3: Select a type of support vector machine model as the anomaly monitoring model, determine the radial basis function RBF as the kernel function, and calculate the kernel parameters through the optimization method; Step 4: Use the preprocessed clinical data as training samples to train a class of support vector machine models; Step 5: Use the trained one-class support vector machine model to predict the patient's postoperative pain improvement; Step 6: Adjust the treatment plan based on the predicted results; Wherein, the step 3 comprises: The optimization problem of a class of support vector machine models is described as: (1) In the formula, Indicates that the variable is The minimum value of the function; represents the normal vector of the hyperplane; represents the offset of the hyperplane; Represents the transpose of a matrix; Indicates the maximum misclassification rate of training samples; is the number of training samples; From 1 to integer variable of ; is the slack variable; is a constraint condition; Represents a nonlinear mapping that projects raw data into a high-dimensional feature space; Represents the i-th sample element, also called the support vector; Introducing the first and second Lagrange multipliers into formula (1) , for: (2) In the formula, represents the objective function to be optimized after the Lagrange multiplier is introduced; are the first and second Lagrange multipliers and slack variables corresponding to the i-th sample respectively; According to the KKT conditions, we can conclude that: (3) Substituting formula (3) into formula (2), we get: (4) Substituting equation (4) into equation (2) yields the dual form of the elementary optimization problem: (5) in, ; represents the kernel matrix calculated from the training samples, , represents the kernel function; subscript and Represents the index of the kernel matrix element; From 1 to integer variable; by solving the above dual problem, we can obtain the Lagrange multiplier, calculate the normal vector, and obtain the hyperplane with the optimal classification interval; The radial basis kernel function RBF is used as the nonlinear mapping function. (6) Where, exp represents the exponential function; is the Gaussian kernel bandwidth; represents the two-norm, represents the jth sample element; the decision function f is expressed as: (7) in, represents the sample element; sign represents the sign function; represents the index of the support vector; is the support vector; Solved by the following formula: (8) Among them, the i-th training sample The Lagrange multiplier is not 0. For all training samples ,when When , it means that the postoperative pain state of the patient is mild pain; when When , it means that the patient's pain status after surgery is moderate or severe.
2. The machine learning method for rapid preoperative prediction of postoperative pain improvement index after vertebroplasty according to claim 1, characterized in that: In step 3, the kernel parameters include the Gaussian kernel bandwidth in the radial basis kernel function RBF And the maximum misclassification rate of training samples ,in, For the experimental setting, the optimal kernel parameters of the kernel function are solved by establishing the objective function to maximize the difference between the average distance of the farthest element and the average distance of the nearest element: (9) In the formula, is the objective function; Indicates the maximum value; Therefore, the radial basis kernel function RBF is: (10) By using the tree-based social relationship optimization algorithm TSR to solve formula (10), the Gaussian kernel bandwidth in the radial basis kernel function RBF is obtained .
3. The machine learning method for rapid preoperative prediction of postoperative pain improvement index after vertebroplasty according to claim 1, characterized in that: In step 4, the preprocessed clinical data is used as training samples to train a class of support vector machine models, and K-fold cross validation is used to evaluate the stability and accuracy of the model.
