Construction method for prediction model after vertebral basilar artery occlusion operation
Through LASSO regression screening clinical parameters and combining multiple machine learning algorithms to build prediction models, the problem of insufficient accuracy of VBAO prognostic evaluation system in complex cases is solved, personalized prediction and long-term effectiveness of the system is achieved, and the accuracy of the evaluation is improved. Sex and reliability.
Patent Information
- Application Number
- CN202411431361.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-14
- Publication Date
- 2025-01-14
AI Technical Summary
The existing prognostic evaluation system for vertebral basal artery occlusion (VBAO) is insufficiently accurate when facing complex cases and the individual differentiated needs cannot be met, resulting in insufficient objectivity and repeatability of the evaluation results, which makes it difficult to meet the needs of precision medicine.
The clinical parameters are screened by LASSO regression method, and a variety of prediction models are constructed in combination with machine learning algorithms such as logistic regression, random forest, gradient hoist, etc. Through cross-validation and hyperparameter tuning, the optimal prediction model that meets the preset conditions is obtained, and the Iterative training optimizes model performance.
It improves the accuracy and reliability of prognostic evaluation, can provide personalized predictions based on patient individual data, enhances the system's generalization ability, adapts to different patient situations, and is effective in clinical practice for a long time.
Smart Images
Figure CN119314676A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of medical data analysis and prediction, and in particular to a method for constructing a prediction model for postoperative vertebral basilar artery occlusion. Background Art
[0002] Vertebrobasilar Artery Occlusion (VBAO) is a rare but serious type of ischemic stroke, mainly caused by the blockage of the vertebral artery and basilar artery supplying the back of the brain, resulting in interruption of blood flow in the brainstem, cerebellum, occipital lobe and parts of the temporal lobe, which in turn leads to a series of neurological dysfunctions. These disorders include but are not limited to changes in consciousness, syncope, balance disorders, vertigo, diplopia, dysphagia, limb weakness or numbness, and even coma or death. Although VBAO accounts for only 1% to 5% of all ischemic strokes, once it occurs, the prognosis is extremely severe, and it is one of the common causes of disability and death in patients.
[0003] At present, the treatment methods for VBAO mainly include drug therapy and interventional therapy. Endovascular treatment (EVT), as an emerging minimally invasive treatment method, has been widely used in the treatment of VBAO in recent years. EVT directly removes or resolves the blockage in the artery through interventional means (such as mechanical thrombectomy, intravascular stent implantation, etc.), restores blood flow to the back of the brain, and reduces further damage to brain tissue, thereby significantly improving the patient's survival rate and neurological recovery. However, although EVT technology has gradually become an important means of treating VBAO, its treatment effect is often affected by many factors, including but not limited to the patient's baseline characteristics (such as age, medical history, initial disease severity, etc.), treatment timing (such as the time window after occlusion occurs), treatment technology (such as the type of thrombectomy device, surgical proficiency, etc.) and the patient's tolerance to treatment. Therefore, how to accurately evaluate the prognosis after EVT treatment based on the individual characteristics of the patient has become an important issue that needs to be solved in clinical treatment.
[0004] In current clinical practice, clinicians mainly rely on personal experience and intuition to conduct prognostic assessments for VBAO patients. Although experienced doctors can make relatively accurate judgments on the prognosis of certain patients based on years of experience and case accumulation, this evaluation method based on subjective judgment has obvious limitations, which are mainly reflected in the following aspects: First, the objectivity and repeatability of the evaluation results are insufficient. Different doctors make different prognostic judgments on the same patient's condition, resulting in large individual differences in treatment decisions; second, it is difficult for doctors' subjective judgments to fully consider a variety of potential influencing factors, especially when facing complex conditions, and it is easy to ignore certain key parameters, thus affecting the accuracy and reliability of the evaluation; third, with the continuous development of medical technology and the differentiation of individual characteristics of patients, the method of relying on a single experience for prognostic assessment can no longer meet the needs of precision medicine.
[0005] In order to overcome the above problems, in recent years, the medical community has begun to introduce clinical scoring systems based on data and algorithms to help doctors better evaluate the prognosis of VBAO patients. These scoring systems collect clinical parameters of patients, such as age, gender, underlying diseases, imaging examination results, treatment methods, etc., and combine a large amount of case data and statistical analysis models to quantitatively evaluate the prognosis of patients after EVT. This type of scoring system reduces the dependence of doctors' personal experience on prognostic judgments through quantitative analysis, thereby improving the objectivity, repeatability and universal applicability of the evaluation results.
[0006] However, although the existing scoring systems can provide a certain reference value for prognosis assessment, they still have some significant limitations. First, most scoring systems can identify a limited number of parameters, especially when facing complex cases, some key factors affecting prognosis are not included in the evaluation scope, resulting in insufficient accuracy of the scoring results in some cases. Secondly, some scoring systems have different applicability between different patient groups or different medical institutions, especially when used across regions or populations, the generalization ability of the scoring system is limited.
[0007] Therefore, how to build a prognostic evaluation system that can integrate multiple clinical parameters and improve the accuracy and reliability of system predictions has become a technical problem that needs to be solved urgently. Summary of the invention
[0008] The main purpose of the present invention is to provide a method for constructing a prediction model for postoperative vertebrobasilar artery occlusion, aiming to construct a prognostic evaluation system that can integrate multiple clinical parameters and improve the accuracy and reliability of system prediction.
[0009] In order to achieve the above object, the present invention proposes a method for constructing a prediction model for postoperative vertebral basilar artery occlusion, comprising the following steps: Obtain clinical parameters; The clinical parameters are screened by LASSO regression method to obtain clinical variables related to prognosis; According to the clinical variables, a variety of different prediction models are constructed through a variety of different machine learning algorithms, and each prediction model is comprehensively evaluated through evaluation indicators to obtain evaluation results that meet preset conditions, and the corresponding model is set as the prediction model, and the prediction model is trained and iterated according to the next clinical parameter.
[0010] In one embodiment of the present application, the clinical parameters include: at least one of demographic characteristics, cerebrovascular disease risk factors, baseline NIHSS score, pc-ASPECTS score, preoperative IVT status, occlusion site, reperfusion status, collateral status and time measurement data.
[0011] In one embodiment of the present application, the screening of the clinical parameters by the LASSO regression method also includes: Data cleaning was performed for outlier treatment and / or missing value filling for clinical parameters.
[0012] In one embodiment of the present application, the cleaned data is also normalized.
[0013] In one embodiment of the present application, the multiple algorithms include at least one of logistic regression, random forest, and gradient boosting machine.
[0014] In one embodiment of the present application, comprehensively evaluating each prediction model using evaluation indicators includes the following steps: S1: Cross-validation of different prediction models; S2: Perform hyperparameter tuning on different prediction models that have completed cross-validation; S3: After the hyperparameter tuning is completed, the results of different prediction models under the predetermined indicators are obtained.
[0015] In one embodiment of the present application, the predetermined indicator includes: at least one of the area under the ROC curve, accuracy, sensitivity, and specificity.
[0016] By adopting the above technical solution, by integrating multiple clinical parameters and using LASSO regression for feature screening, the model is ensured to focus on key predictive variables, reduce the interference of irrelevant or redundant information on the model, and thus improve the prediction accuracy and reliability of the model. By applying multiple machine learning algorithms such as logistic regression, random forest, gradient boosting machine, etc. for modeling, the advantages of each algorithm are utilized to finally select the optimal prediction model to ensure the optimal performance of the system. The system can provide personalized prognosis predictions based on the individual clinical data of patients, which helps doctors adjust treatment plans according to specific circumstances and realize personalized medicine. Through iterative training, the model can be continuously optimized according to new data to ensure that it still has good generalization ability when facing different patients, so that the system is effective in clinical practice for a long time. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] The present invention is described in detail below with reference to specific embodiments and accompanying drawings, wherein: Figure 1 It is a schematic structural diagram of the first embodiment of the present invention. DETAILED DESCRIPTION
[0018] In order to make the purpose, technical solution and advantages of the present invention clearer, the present invention is described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the following specific embodiments are only used to explain the present invention and do not constitute a limitation to the present invention.
[0019] like Figure 1 As shown, in order to achieve the above-mentioned purpose, the present invention proposes a method for constructing a prediction model for postoperative vertebral basilar artery occlusion, comprising the following steps: Obtain clinical parameters; The clinical parameters are screened by LASSO regression method to obtain clinical variables related to prognosis; According to the clinical variables, a variety of different prediction models are constructed through a variety of different machine learning algorithms, and each prediction model is comprehensively evaluated through evaluation indicators to obtain evaluation results that meet preset conditions, and the corresponding model is set as the prediction model, and the prediction model is trained and iterated according to the next clinical parameter.
[0020] Specifically, comprehensive clinical parameters are obtained from the cases of patients with VBAO (vertebrobasilar artery occlusion). The required parameters include demographic characteristics: such as age, gender, etc. Cerebrovascular risk factors: such as hypertension, diabetes, smoking history, etc. Neurological function score: baseline NIHSS (National Institutes of Health Stroke Scale) score, pc-ASPECTS (Posterior Circulation Alberta Stroke Score). Treatment information: preoperative intravenous thrombolytic therapy (IVT) status, postoperative reperfusion status (blood flow recovery after surgery). Lesion site: such as the specific location of the occluded artery, the status of collateral circulation. Time data: onset time, treatment window period. The above parameters are used as input data for the model, which comprehensively covers the patient's basic situation, medical history and postoperative recovery.
[0021] Due to the large number of clinical parameters obtained, in order to avoid the influence of redundant or irrelevant features on the model, LASSO regression (Least Absolute Shrinkage and Selection Operator) was used for feature screening. LASSO regression has the function of automatically selecting variables, which can screen out variables that have no significant effect on prognosis and retain clinical variables that are closely related to prognosis.
[0022] All clinical parameters were input into the LASSO regression model. Through the LASSO regularization process, the regression coefficient was gradually reduced, and finally the variables with significant impact on prognosis were screened out. The screened variables were used as input for subsequent model training to ensure that the model was more streamlined and focused on the core influencing factors.
[0023] After screening out clinical variables that are significantly correlated with prognosis, different prediction models are constructed using a variety of machine learning algorithms. These algorithms may include at least one of: logistic regression, random forest, gradient boosting machine, and support vector machine. Among them, logistic regression: used to establish a binary classification model to evaluate whether the patient has a good prognosis. Random forest: prediction is made by integrating multiple decision tree models, which is suitable for processing complex nonlinear data relationships. Gradient boosting machine (GBM): a strong prediction model is constructed by optimizing weak models, with high prediction accuracy. Support vector machine (SVM): data is classified by constructing a hyperplane, which is suitable for binary classification problems. Since the above machine learning algorithms are all existing technologies, they will not be described one by one here. For different machine learning algorithms, multiple prediction models are constructed in turn, and each model is trained according to the input clinical variables.
[0024] In order to evaluate the quality of each prediction model, the performance of the model is comprehensively evaluated through commonly used evaluation indicators. Evaluation indicators include the area under the ROC curve (AUC): used to measure the classification ability of the model. The closer the value is to 1, the better the model performance. Accuracy: the proportion of correct classifications in the model prediction results. Sensitivity: the model's ability to identify positive classes (good prognosis). Specificity: the model's ability to identify negative classes (poor prognosis).
[0025] Based on these evaluation indicators, all constructed models are compared and the model with the best performance is selected.
[0026] After the best prediction model is determined through model evaluation, it is set as the final prognosis prediction model. Subsequently, the model is trained and iterated through further clinical parameter input to continuously optimize the model's prediction performance. During the training process, the system continuously updates the model based on the newly input data so that it can adapt to more diverse patient conditions.
[0027] The following are some examples: Suppose there is a 65-year-old male patient with a history of hypertension, a baseline NIHSS score of 8, a pc-ASPECTS score of 7, who received intravenous thrombolysis before surgery, the occlusion site is located in the proximal segment of the vertebral artery, the reperfusion status is partially recovered (TICI score 2b), and the time from onset to treatment is 4 hours. After LASSO regression screening, the system identified the patient's age, baseline NIHSS score, pc-ASPECTS score, and reperfusion status as the main predictor variables. Next, the system built a model through multiple algorithms such as random forest and gradient boosting machine. After evaluation, the LR logistic regression model was finally selected because its AUC value was 0.815, showing high accuracy and sensitivity.
[0028] After the doctor inputs the patient's relevant data on the interface, the system uses the trained random forest model to predict the prognosis, showing that the patient has a good prognosis and has recovered good neurological function. The doctor adjusts the postoperative management and rehabilitation plan based on the prediction results.
[0029] By adopting the above technical solution, by integrating multiple clinical parameters and using LASSO regression for feature screening, the model is ensured to focus on key predictive variables, reduce the interference of irrelevant or redundant information on the model, and thus improve the prediction accuracy and reliability of the model. By applying multiple machine learning algorithms such as logistic regression, random forest, gradient boosting machine, etc. for modeling, the advantages of each algorithm are utilized to finally select the optimal prediction model to ensure the optimal performance of the system. The system can provide personalized prognosis predictions based on the individual clinical data of patients, which helps doctors adjust treatment plans according to specific circumstances and realize personalized medicine. Through iterative training, the model can be continuously optimized according to new data to ensure that it still has good generalization ability when facing different patients, so that the system is effective in clinical practice for a long time.
[0030] In one embodiment of the present application, the clinical parameters include: at least one of demographic characteristics, cerebrovascular disease risk factors, baseline NIHSS score, pc-ASPECTS score, preoperative IVT status, occlusion site, reperfusion status, collateral status and time measurement data.
[0031] Specifically, demographic characteristics refer to the basic demographic information of patients, which are basic variables that affect the occurrence, development and prognosis of the disease, including age, gender and other data. Demographic characteristics are basic information that is easier to obtain and can provide a basic reference for the differences between different populations in the early training of the model. Age is an important risk factor for stroke, and older patients are associated with worse prognosis, while gender differences affect prognosis in different pathological mechanisms.
[0032] Cerebrovascular disease risk factors refer to factors that are closely related to the risk of cerebrovascular disease and can help the model identify high-risk patient groups. Included data include hypertension, diabetes, smoking history, and hyperlipidemia. Cerebrovascular disease risk factors help the model identify patients with a higher risk of stroke, especially when accompanied by multiple chronic diseases, with a worse prognosis. By integrating these risk factors, the model can more accurately assess patients' postoperative recovery and mortality risks.
[0033] The baseline NIHSS score is used to assess the degree of neurological impairment in stroke patients. The baseline score refers to the initial score when the patient is admitted to the hospital. The higher the value, the more severe the neurological impairment. The score covers multiple aspects such as consciousness, visual field, language, and motor function. The baseline NIHSS score is an indicator that directly reflects the severity of the patient's condition and has important reference value for evaluating postoperative neurological recovery. The higher the score, the worse the postoperative prognosis. The addition of this score enables the model to make stratified predictions based on the degree of neurological impairment.
[0034] The pc-ASPECTS score is mainly used to assess the degree of damage in posterior circulation stroke. The higher the score, the smaller the ischemic area in the posterior circulation brain region on imaging, the milder the brain tissue damage, and the better the prognosis. The pc-ASPECTS score is based on CT image evaluation and can provide the model with objective quantitative indicators of preoperative brain tissue damage in patients. By combining imaging data, the model can make a more accurate judgment on the patient's prognosis.
[0035] Preoperative IVT refers to whether the patient has received intravenous thrombolysis (IVT) before surgery. It includes whether IVT was performed and the duration of thrombolysis. Preoperative IVT can significantly affect the patient's prognosis. Timely thrombolytic therapy helps to restore some blood flow early, reduce the area of cerebral ischemia, and thus improve postoperative prognosis. By integrating preoperative IVT information, the model can more accurately evaluate prognosis.
[0036] The occlusion site refers to the specific location where the arterial blockage occurs, including different parts of the vertebral artery and the basilar artery. Such as the proximal and distal vertebral artery, the proximal basilar artery or the basilar artery trunk. The occlusion site is an important factor that directly affects the prognosis. Different occlusion sites have different effects on the blood supply to the brain and the affected areas, which in turn affects the patient's functional recovery. Through this parameter, the model can predict the damaged brain area and the corresponding prognosis based on the occlusion site.
[0037] Reperfusion status refers to the recovery of blood flow after surgery, that is, whether the blood vessels have restored sufficient blood supply after surgery. The TICI score (Thrombolysis In Cerebral Infarction Score) measures blood flow recovery, which is divided into 0 to 3 points. TICI 3 indicates complete reperfusion and TICI 0 indicates no reperfusion. Reperfusion status is one of the core indicators for predicting postoperative prognosis. Successful restoration of blood flow during surgery means that most brain tissue can avoid further damage, indicating better neurological recovery. Reperfusion status combined with other parameters can significantly improve the accuracy of prognosis prediction.
[0038] Collateral status refers to the state of collateral circulation, that is, whether there are other blood vessels around the occluded blood vessels to provide compensatory blood supply. It can be divided into good, partial or no collateral circulation. Collateral circulation can provide partial blood flow to brain tissue when the artery is occluded, reducing the degree of ischemia. Therefore, good collateral status is associated with better prognosis. This parameter can help the model more accurately predict the prognosis of patients with incomplete reperfusion.
[0039] Time measurement data refers to indicators related to the time of stroke occurrence and treatment, and is closely related to the patient's postoperative prognosis. It includes the time from onset to hospitalization, the time from onset to surgery, and the time from admission to surgery. The time window is one of the key factors in determining the prognosis, especially the early intervention after stroke is effective. The earlier the treatment, the less damage to the brain tissue and the better the prognosis. Incorporating these time parameters into the model can help predict postoperative outcomes in different time windows.
[0040] In one embodiment of the present application, the screening of the clinical parameters by the LASSO regression method also includes: Data cleaning was performed for outlier treatment and / or missing value filling for clinical parameters.
[0041] Specifically, outliers refer to values that appear to be abnormally out of the group in the data set and are significantly different from other data. Outliers are caused by data entry errors, measurement errors, or extreme individual differences. In the process of machine learning modeling, outliers will interfere with the model and lead to inaccurate prediction results. Therefore, outliers need to be processed before LASSO regression. By setting the normal value range, obvious errors or extreme values are directly deleted from the data set. If the outliers are caused by extreme individual differences, adjust them to a relatively reasonable range. If the outliers reflect real extreme case situations and are clinically significant, they can be retained, but their potential impact needs to be recorded.
[0042] In actual clinical data, some parameters of some patients are not collected or recorded. If these missing data are not processed, the effect of LASSO regression and subsequent training will be affected. For continuous data, the mean or median of the variable can be used to fill the missing values. For categorical variables, the mode can be used to fill in. Use other related variables to build a regression model to predict missing values. For example, if the missing data is age data, it can be estimated through a regression model based on other relevant information of the patient (such as medical history). Using multiple models for multiple interpolations and then averaging the results can better preserve the authenticity and uncertainty of the data. If a variable has too much missing data and is difficult to fill in, or its importance is not high, you can choose to delete the variable or delete the sample where the missing data is located.
[0043] In one embodiment of the present application, the cleaned data is also normalized.
[0044] Specifically, when the scales of features vary greatly, the gradient descent algorithm will search along the direction of the larger feature, causing the optimization process to oscillate and converge slowly. After normalization, the scales of the features become consistent, and the gradient descent can move more stably toward the minimum, thereby accelerating data convergence.
[0045] In one embodiment of the present application, the multiple algorithms include at least one of logistic regression, random forest, and gradient boosting machine.
[0046] Specifically, logistic regression: used to build a binary classification model to assess whether a patient has a good prognosis. Random forest: predicts by integrating multiple decision tree models, suitable for processing complex nonlinear data relationships. Gradient boosting machine (GBM): builds a strong prediction model by optimizing weak models, with high prediction accuracy.
[0047] In one embodiment of the present application, comprehensively evaluating each prediction model using evaluation indicators includes the following steps: S1: Cross-validation of different prediction models; S2: Perform hyperparameter tuning on different prediction models that have completed cross-validation; S3: After the hyperparameter tuning is completed, the results of different prediction models under the predetermined indicators are obtained.
[0048] Specifically, the original dataset is divided into K subsets of equal size (usually K=5 or K=10), of which K-1 subsets are used to train the model and the remaining 1 subset is used to verify the model.
[0049] Different subsets are alternately selected as validation sets, and the remaining subsets are used as training sets, and repeated K times. Each time the model is trained, the model is evaluated on a different subset of the data.
[0050] Collect the evaluation results of K validations and calculate the average performance indicators of the model (such as accuracy, AUC, sensitivity, specificity, etc.) to obtain the comprehensive performance of the model. Cross-validation aims to test the performance of the model on new data through training and validation of different data subsets, reduce the risk of data overfitting, and ensure the stability of the model.
[0051] Determine the hyperparameters of the model to be tuned and their possible value ranges, such as the number of trees in the random forest model (n_estimators), the maximum depth of each tree (max_depth), etc.
[0052] By traversing all possible hyperparameter combinations, we find the best hyperparameter combination. Under each hyperparameter combination, we train the model and evaluate its performance through cross-validation, and record the performance indicators.
[0053] Based on the cross-validation evaluation results, the best performing hyperparameter combination is selected and used for subsequent model training. By adjusting the model's hyperparameters, it performs best under the predetermined evaluation indicators (such as accuracy, AUC), thereby improving the accuracy and reliability of the prediction model.
[0054] After completing cross-validation and hyperparameter tuning, the model will generate performance under various evaluation indicators. Common evaluation indicators include accuracy, area under the ROC curve (AUC), sensitivity, specificity, etc. Select appropriate evaluation indicators according to task requirements. For example, accuracy indicates the proportion of correct predictions made by the model, AUC indicates the model's ability to distinguish between positive and negative samples, sensitivity (recall) measures the model's ability to identify positive samples, and specificity measures the model's ability to identify negative samples. Under the tuned hyperparameters, the model is run on the complete dataset to obtain the final values of each predetermined evaluation indicator.
[0055] All prediction models are ranked according to the performance of the evaluation indicators. The model with the best performance in the main evaluation indicators (such as AUC and accuracy) is selected as the final prediction model. Through the predetermined evaluation indicators, the performance of the model in different dimensions is comprehensively measured to ensure that the selected model not only has high accuracy, but also can effectively identify positive and negative samples in different situations.
[0056] By adopting the above technical solution and cross-validation, the model is trained and verified on different data subsets, which effectively reduces the risk of overfitting and makes the model more robust on new data. Through hyperparameter tuning, the internal parameters of different models can be adjusted according to their characteristics, the model performance can be optimized, and the accuracy and reliability of predictions can be improved. Through multi-dimensional evaluation indicators (such as AUC, accuracy, sensitivity, specificity, etc.), it can ensure that the selected model performs best in different tasks.
[0057] In one embodiment of the present application, the predetermined indicator includes: at least one of the area under the ROC curve, accuracy, sensitivity, and specificity.
[0058] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. All equivalent structural changes made by using the contents of the present invention specification and drawings under the inventive concept of the present invention, or directly / indirectly applied in other related technical fields are included in the patent protection scope of the present invention.
Claims
1. A method for constructing a prediction model for postoperative vertebral basilar artery occlusion, characterized in that: The following steps are involved: Obtain clinical parameters; The clinical parameters are screened by LASSO regression method to obtain clinical variables related to prognosis; According to the clinical variables, a variety of different prediction models are constructed through a variety of different machine learning algorithms, and each prediction model is comprehensively evaluated through evaluation indicators to obtain evaluation results that meet preset conditions, and the corresponding model is set as the prediction model, and the prediction model is trained and iterated according to the next clinical parameter.
2. The method for constructing a prediction model for postoperative vertebrobasilar artery occlusion according to claim 1, characterized in that: The clinical parameters include at least one of demographic characteristics, cerebrovascular disease risk factors, baseline NIHSS score, pc-ASPECTS score, preoperative IVT status, occlusion site, reperfusion status, collateral status and time measurement data.
3. The method for constructing a prediction model for postoperative vertebrobasilar artery occlusion according to claim 1, characterized in that: The screening of the clinical parameters by LASSO regression method also includes: Data cleaning was performed for outlier treatment and / or missing value filling for clinical parameters.
4. The method for constructing a prediction model for postoperative vertebrobasilar artery occlusion according to claim 3, characterized in that: It also includes normalizing the cleaned data.
5. The method for constructing a prediction model for postoperative vertebrobasilar artery occlusion according to claim 1, characterized in that: The multiple algorithms include at least one of logistic regression, random forest, and gradient boosting machine.
6. The method for constructing a prediction model for postoperative vertebrobasilar artery occlusion according to claim 1, characterized in that: Comprehensively evaluate each prediction model through evaluation indicators The following steps are involved: S1: Cross-validation of different prediction models; S2: Perform hyperparameter tuning on different prediction models that have completed cross-validation; S3: After the hyperparameter tuning is completed, the results of different prediction models under the predetermined indicators are obtained.
7. The method for constructing a prediction model for postoperative vertebrobasilar artery occlusion according to claim 6, characterized in that: The predetermined index includes at least one of the area under the ROC curve, accuracy, sensitivity, and specificity.