Construction method of minimally invasive surgery risk assessment model based on craniocerebral tumor

Abstract

The invention relates to the technical field of brain surgery, in particular to a construction method based on a craniocerebral tumor minimally invasive surgery risk assessment model. The method comprises the following steps: acquiring craniocerebral tumor minimally invasive surgery clinical data, including patient iconography data, surgery record data, postoperative complication data and patient basic data; performing data cleaning on the surgical clinical data, and performing standardization processing to unify a data format so as to obtain a standardized clinical data set; and extracting the size, position and shape of the tumor and the relationship between the tumor and surrounding tissues from the standardized clinical data set, and generating a tumor space structure model by using a three-dimensional reconstruction technology. According to the method, main factors influencing the surgical risk are identified through a statistical analysis method, including the contribution degree of each feature to the surgical risk and the interaction between the features, and it is ensured that the evaluation process is more objective and accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of brain surgery, and in particular to a method for constructing a risk assessment model based on minimally invasive surgery for cranial brain tumors. Background Art

[0002] Minimally invasive surgery for craniocerebral tumors is a complex and high-risk operation in the field of modern medicine, involving the cross-application of multiple disciplines. Due to the complex location of craniocerebral tumors, the close relationship between tumors and peripheral nerves and vascular tissues, and the individual differences of patients, risk assessment during surgery has become a very challenging link. Traditional surgical risk assessment mainly relies on the experience and intuitive judgment of surgeons, and lacks a systematic and quantitative assessment method. Especially when facing complex cases, surgeons often find it difficult to fully consider all factors affecting surgical risks, resulting in a high incidence of intraoperative and postoperative complications. In this context, in order to improve the scientificity and accuracy of surgical risk assessment, there is an urgent need for a data-driven intelligent assessment method that can integrate multi-dimensional patient data, quantify and analyze various influencing factors, and provide doctors with more reliable surgical decision support. This method should be able to efficiently process complex clinical data, automatically identify key factors affecting surgical risks, and predict possible complications during and after surgery, thereby optimizing surgical plans and reducing medical risks.

[0003] The traditional method of constructing a risk assessment model for minimally invasive surgery of cranial brain tumors often has the following problems: In the traditional surgical risk assessment process, doctors rely on data from multiple sources, such as imaging data, surgical records, and basic information of patients; these data usually have problems such as inconsistent formats, large data volumes, and scattered information. Manual integration of these data is not only inefficient, but also prone to information omissions or misunderstandings. Traditional risk assessment models are usually based on simple linear regression or empirical formulas, which are difficult to adapt to complex clinical situations and individual differences; for minimally invasive surgery of cranial brain tumors, there are multiple nonlinear and multidimensional risk factors involved, and the existing models are significantly insufficient in accuracy and robustness when dealing with these complex data. Summary of the invention

[0004] Based on this, it is necessary for the present invention to provide a method for constructing a risk assessment model for minimally invasive surgery of cranial brain tumors to solve at least one of the above technical problems.

[0005] To achieve the above purpose, a method for constructing a risk assessment model for minimally invasive surgery of craniocerebral tumors comprises the following steps: Step S1: Obtaining clinical data of minimally invasive surgery for craniocerebral tumors, including patient imaging data, surgical record data, postoperative complication data, and patient basic data; cleaning the surgical clinical data and performing standardization processing to unify the data format, thereby obtaining a standardized clinical data set; Step S2: extracting the size, location, shape and relationship of the tumor with surrounding tissues from the standardized clinical data set, and generating a tumor spatial structure model using three-dimensional reconstruction technology; extracting physiological characteristics based on the underlying disease and preoperative status from the standardized clinical data set, thereby obtaining physiological characteristic data; Step S3: Identify the main factors affecting surgical risk based on the tumor spatial structure model and physiological characteristic data by applying statistical analysis methods, thereby generating risk factor weight data, wherein the main factor identification includes the contribution analysis of each feature to surgical risk, the interaction between features and their impact on surgical risk, the significance test of each feature, and the screening of risk factors; Step S4: Generate a surgical risk assessment model based on a neural network model according to the risk factor weight data; optimize the risk assessment model according to the standardized clinical data set through cross-validation, and use an independent data set in the pre-set standardized clinical data set to evaluate the prediction performance and accuracy, thereby obtaining a surgical risk prediction model; Step S5: The surgical risk prediction model is embedded into the medical system through the API interface, and integrated with the hospital's existing electronic medical record system and image archiving and communication system to assist surgical decision-making.

[0006] The data cleaning step of the present invention removes invalid data and outliers to ensure the accuracy and integrity of the data; the standardization processing step unifies the format and unit of the data, so that data from different sources can be effectively integrated and analyzed, reducing the analysis errors caused by inconsistent data formats; high-quality standardized data sets can significantly improve the accuracy and efficiency of data analysis and model training in subsequent steps. Through three-dimensional reconstruction technology, the size and shape of the tumor and its relative position to surrounding important tissue structures (such as nerves and blood vessels) are intuitively displayed, providing an intuitive reference for surgical planning and risk assessment; based on the extraction of physiological characteristics of the underlying disease and preoperative status, it can more comprehensively reflect the health status of the patient, taking into account individual differences, and help to conduct personalized risk assessment; by integrating spatial structure and physiological characteristic data, a more accurate surgical risk assessment model can be constructed, so that the model can better reflect the complexity and diversity of surgical risks. By analyzing the contribution of each feature to surgical risk and the interaction between features, the influence of each factor can be quantified, thereby providing a more objective data basis for risk assessment; identifying factors that have a significant impact on surgical risk, helping doctors better understand the source of surgical risk, and optimizing surgical strategies and preventive measures; clarifying the weight and role of each factor, so that the model not only has predictive ability, but also can explain the source and cause of risk, which is helpful for clinicians to understand and apply. The neural network model can handle complex nonlinear relationships, adapt to multi-dimensional and highly complex clinical data, and improve the accuracy and robustness of surgical risk prediction; through cross-validation and model optimization, it avoids overfitting problems, improves the performance of the model on different data sets, and enhances the generalization ability of the model; the high predictive performance and reliability of the model ensure that it has a high reference value for the prediction and evaluation of surgical risks in real clinical environments. In general, the method of the present invention provides an efficient, reliable and integrated solution for the risk assessment of minimally invasive surgery for craniocerebral tumors through a multi-step systematic and data-driven process, from data acquisition, preprocessing to model construction and system integration. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments thereof made with reference to the following drawings: Figure 1 It is a schematic diagram of the steps of the method for constructing a risk assessment model for minimally invasive surgery of craniocerebral tumors according to the present invention; Figure 2 for Figure 1 Detailed step flow diagram of step S1; Figure 3 for Figure 1 Detailed step flow chart of step S2 in FIG. DETAILED DESCRIPTION

[0008] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0009] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0010] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0011] To achieve this, please refer to Figures 1 to 3 The present invention provides a method for constructing a risk assessment model for minimally invasive surgery of craniocerebral tumors, the method comprising the following steps: Step S1: Obtaining clinical data of minimally invasive surgery for craniocerebral tumors, including patient imaging data, surgical record data, postoperative complication data, and patient basic data; cleaning the surgical clinical data and performing standardization processing to unify the data format, thereby obtaining a standardized clinical data set; Step S2: extracting the size, location, shape and relationship of the tumor with surrounding tissues from the standardized clinical data set, and generating a tumor spatial structure model using three-dimensional reconstruction technology; extracting physiological characteristics based on the underlying disease and preoperative status from the standardized clinical data set, thereby obtaining physiological characteristic data; Step S3: Identify the main factors affecting surgical risk based on the tumor spatial structure model and physiological characteristic data by applying statistical analysis methods, thereby generating risk factor weight data, wherein the main factor identification includes the contribution analysis of each feature to surgical risk, the interaction between features and their impact on surgical risk, the significance test of each feature, and the screening of risk factors; Step S4: Generate a surgical risk assessment model based on a neural network model according to the risk factor weight data; optimize the risk assessment model according to the standardized clinical data set through cross-validation, and use an independent data set in the pre-set standardized clinical data set to evaluate the prediction performance and accuracy, thereby obtaining a surgical risk prediction model; Step S5: The surgical risk prediction model is embedded into the medical system through the API interface, and integrated with the hospital's existing electronic medical record system and image archiving and communication system to assist surgical decision-making.

[0012] In the embodiment of the present invention, reference Figure 1 The above is a schematic diagram of the steps of a method for constructing a risk assessment model for minimally invasive surgery of craniocerebral tumors according to the present invention. In this example, the method for constructing a risk assessment model for minimally invasive surgery of craniocerebral tumors comprises the following steps: Step S1: Obtaining clinical data of minimally invasive surgery for craniocerebral tumors, including patient imaging data, surgical record data, postoperative complication data, and patient basic data; cleaning the surgical clinical data and performing standardization processing to unify the data format, thereby obtaining a standardized clinical data set; The embodiment of the present invention obtains a variety of clinical data related to minimally invasive surgery for cranial brain tumors through the hospital's information system (such as an electronic medical record system and an image archiving and communication system). These data include the patient's imaging data (CT or MRI, etc.), surgical records, postoperative complication information, and the patient's basic information (such as age, gender, medical history, etc.). After acquiring the data, it is cleaned, including removing duplicates, processing missing data and outliers, to ensure the integrity and consistency of the data. Next, the cleaned data is standardized, such as converting all numerical data into a unified unit format, and classifying text data, ultimately forming a standardized clinical data set with a unified structure and consistent format for subsequent analysis.

[0013] Step S2: extracting the size, location, shape and relationship of the tumor with surrounding tissues from the standardized clinical data set, and generating a tumor spatial structure model using three-dimensional reconstruction technology; extracting physiological characteristics based on the underlying disease and preoperative status from the standardized clinical data set, thereby obtaining physiological characteristic data; The embodiment of the present invention uses tumor-related information in the image data to extract the size, location, shape and other characteristics of the tumor through a segmentation algorithm. For example, a U-Net model based on deep learning can be used to segment the tumor area, and a three-dimensional reconstruction technology (such as the Marching Cubes algorithm) can be used to construct a three-dimensional spatial structure model of the tumor. At the same time, the patient's physiological characteristics (such as blood pressure, blood sugar, heart rate, etc.) are extracted, with special attention paid to the data of the underlying disease and preoperative status, and these data are used to generate a physiological characteristic data set. In this way, the spatial model of the tumor and the patient's physiological characteristic data can be extracted synchronously to provide basic data for subsequent surgical risk assessment.

[0014] Step S3: Identify the main factors affecting surgical risk based on the tumor spatial structure model and physiological characteristic data by applying statistical analysis methods, thereby generating risk factor weight data, wherein the main factor identification includes the contribution analysis of each feature to surgical risk, the interaction between features and their impact on surgical risk, the significance test of each feature, and the screening of risk factors; In the embodiment of the present invention, based on the tumor spatial structure model and physiological characteristic data obtained in step S2, a statistical analysis method is applied to identify the main factors affecting surgical risk. The specific operation includes using multiple regression analysis or logistic regression analysis to calculate the contribution of each feature to surgical risk, and at the same time evaluating the impact of the interaction between features on surgical risk. For example, statistically significant risk factors are screened out through significance tests such as analysis of variance (ANOVA) or chi-square test. Finally, risk factor weight data are generated based on the contribution and interaction effect of each risk factor, and these data will be used as input for the subsequent risk assessment model.

[0015] Step S4: Generate a surgical risk assessment model based on a neural network model according to the risk factor weight data; optimize the risk assessment model according to the standardized clinical data set through cross-validation, and use an independent data set in the pre-set standardized clinical data set to evaluate the prediction performance and accuracy, thereby obtaining a surgical risk prediction model; The embodiment of the present invention constructs a surgical risk assessment model based on a neural network based on the risk factor weight data in step S3. For example, a multi-layer perceptron (MLP) model can be used, the input layer receives the risk factor weight data, the hidden layer performs nonlinear transformation through the ReLU activation function, and the output layer is used to predict the surgical risk probability. Then, the model is optimized by the cross-validation method to ensure the generalization ability of the model. During the model training and verification process, the independent test data set reserved in step S1 is used to evaluate the prediction performance and accuracy of the model, and finally a reliable surgical risk prediction model is obtained.

[0016] Step S5: The surgical risk prediction model is embedded into the medical system through the API interface, and integrated with the hospital's existing electronic medical record system and image archiving and communication system to assist surgical decision-making.

[0017] The surgical risk prediction model constructed in step S4 of the embodiment of the present invention is integrated into the hospital's existing medical information system through an API interface. The specific operation includes embedding the model into the electronic medical record system (EMR) and the picture archiving and communication system (PACS) to ensure that clinicians can directly call the risk prediction model when viewing patient data. For example, when a doctor queries a patient's preoperative data, the system automatically calls the model for risk assessment and generates a detailed surgical risk report to help doctors make more accurate judgments during the surgical decision-making process.

[0018] The data cleaning step of the present invention removes invalid data and outliers to ensure the accuracy and integrity of the data; the standardization processing step unifies the format and unit of the data, so that data from different sources can be effectively integrated and analyzed, reducing the analysis errors caused by inconsistent data formats; high-quality standardized data sets can significantly improve the accuracy and efficiency of data analysis and model training in subsequent steps. Through three-dimensional reconstruction technology, the size and shape of the tumor and its relative position to surrounding important tissue structures (such as nerves and blood vessels) are intuitively displayed, providing an intuitive reference for surgical planning and risk assessment; based on the extraction of physiological characteristics of the underlying disease and preoperative status, it can more comprehensively reflect the health status of the patient, taking into account individual differences, and help to conduct personalized risk assessment; by integrating spatial structure and physiological characteristic data, a more accurate surgical risk assessment model can be constructed, so that the model can better reflect the complexity and diversity of surgical risks. By analyzing the contribution of each feature to surgical risk and the interaction between features, the influence of each factor can be quantified, thereby providing a more objective data basis for risk assessment; identifying factors that have a significant impact on surgical risk, helping doctors better understand the source of surgical risk, and optimizing surgical strategies and preventive measures; clarifying the weight and role of each factor, so that the model not only has predictive ability, but also can explain the source and cause of risk, which is helpful for clinicians to understand and apply. The neural network model can handle complex nonlinear relationships, adapt to multi-dimensional and highly complex clinical data, and improve the accuracy and robustness of surgical risk prediction; through cross-validation and model optimization, it avoids overfitting problems, improves the performance of the model on different data sets, and enhances the generalization ability of the model; the high predictive performance and reliability of the model ensure that it has a high reference value for the prediction and evaluation of surgical risks in real clinical environments. In general, the method of the present invention provides an efficient, reliable and integrated solution for the risk assessment of minimally invasive surgery for craniocerebral tumors through a multi-step systematic and data-driven process, from data acquisition, preprocessing to model construction and system integration.

[0019] Preferably, step S1 comprises the following steps: Step S11: extracting basic data of patients with minimally invasive brain tumors through the electronic medical record system, including age, gender, weight, height, and medical history; and obtaining patient imaging data including MRI and CT through the image archiving and communication system; Step S12: collecting the operation time, operation method and operation difficulty rating information through the operation information management system to obtain the operation record data; obtaining the postoperative complication data from the postoperative follow-up system, including the complication type, severity and occurrence time; Step S13: combining the patient's imaging data, surgical record data, postoperative complication data, and basic patient data into clinical data of minimally invasive surgery for craniocerebral tumors; Step S14: Check the data integrity of the surgical clinical data and use the multiple interpolation method to handle missing data, so as to obtain complete clinical data; Step S15: performing anomaly detection and outlier processing based on the interquartile range on the complete clinical data, and converting the numerical data therein into the interval [0, 1], thereby obtaining a preprocessed clinical data set; Step S16: Performing data consistency check and repeatability check on the preprocessed clinical data set to obtain validity verification data; deleting or merging duplicate records and sorting the preprocessed clinical data set based on the validity verification data to obtain surgical clinical optimization data; Step S17: The surgical clinical optimization data is converted into a unified format based on date format, numerical precision and unit, and the same concepts are mapped to unified standard terms, thereby standardizing the clinical data set.

[0020] As an embodiment of the present invention, refer to Figure 2 As shown, Figure 1 Detailed step flow diagram of step S1 in the embodiment of the present invention, step S1 includes the following steps: Step S11: extracting basic data of patients with minimally invasive brain tumors through the electronic medical record system, including age, gender, weight, height, and medical history; and obtaining patient imaging data including MRI and CT through the image archiving and communication system; The embodiment of the present invention uses the hospital's electronic medical record system (EMR) to extract basic patient data related to minimally invasive surgery for craniocerebral tumors, including age, gender, weight, height, medical history, and other information. These data are filtered by time range and surgery type through the query function in the electronic medical record system. At the same time, the image archiving and communication system (PACS) is used to obtain the patient's imaging data, especially MRI and CT scan images. The image data is stored in DICOM format, and is extracted and stored in batches in the local database through the system's interface function for subsequent analysis and processing.

[0021] Step S12: collecting the operation time, operation method and operation difficulty rating information through the operation information management system to obtain the operation record data; obtaining the postoperative complication data from the postoperative follow-up system, including the complication type, severity and occurrence time; The embodiment of the present invention extracts detailed records of the operation through the Surgery Information System (SIS), including the time of the operation, the surgical method (such as transnasal approach, craniotomy, etc.) and the surgical difficulty rating information. The surgical difficulty rating is usually automatically generated based on multiple factors such as the operation time, surgical complexity and surgeon experience. After the data is extracted, the operation ID is combined with other information for matching, so that these data can be integrated with other patient data. At the same time, the postoperative complication data of the patient is obtained through the postoperative follow-up system, and key information such as the type, severity and occurrence time of the complication is recorded in detail.

[0022] Step S13: combining the patient's imaging data, surgical record data, postoperative complication data, and basic patient data into clinical data of minimally invasive surgery for craniocerebral tumors; After collecting imaging data, surgical record data, postoperative complication data, and basic patient data, the embodiment of the present invention merges these data into a unified clinical data set for minimally invasive surgery of craniocerebral tumors. This process matches and integrates data between different systems through patient ID to ensure that all data of each patient can correspond one to one. Strict format alignment and field matching are required during the data merging process to ensure that data from different sources remain consistent in structure, and finally form a complete clinical data set for surgery.

[0023] Step S14: Check the data integrity of the surgical clinical data and use the multiple interpolation method to handle missing data, so as to obtain complete clinical data; In order to ensure the integrity of the data, the embodiment of the present invention performs a data integrity check on the merged surgical clinical data. First, check whether there are missing values ​​in each field, and apply multiple interpolation methods to the missing data. For example, for continuous variables such as patient age or operation time, the regression interpolation method is used to predict missing values; for categorical data such as complication type, the mode interpolation method based on similar cases is used to fill in missing values. The application of multiple interpolation methods can effectively improve the integrity of the data and ultimately generate a complete clinical data set without missing items.

[0024] Step S15: performing anomaly detection and outlier processing based on the interquartile range on the complete clinical data, and converting the numerical data therein into the interval [0, 1], thereby obtaining a preprocessed clinical data set; The embodiment of the present invention applies an anomaly detection method based on the interquartile range to a complete clinical data set to identify outliers in the data. Specifically, the interquartile range (IQR) of each numerical variable is calculated, and data points that exceed 1.5 times the IQR range are marked as outliers. When processing outliers, it is determined whether to delete the outliers or perform reasonable interpolation depending on the specific situation. Next, the numerical data is converted to the interval range of [0, 1] through the maximum-minimum normalization method, so as to ensure that each numerical feature is subsequently analyzed at the same scale, and finally the preprocessed clinical data set is obtained.

[0025] Step S16: Performing data consistency check and repeatability check on the preprocessed clinical data set to obtain validity verification data; deleting or merging duplicate records and sorting the preprocessed clinical data set based on the validity verification data to obtain surgical clinical optimization data; After the preprocessing is completed, the embodiment of the present invention performs a consistency check on the data, mainly to verify whether the data field value conforms to the expected format. For example, whether the date field is stored in the specified format, whether the numerical data is kept within a reasonable range, and the text data is checked for repeatability to ensure that there are no inconsistencies in multiple records of the same patient. The repeatability check includes filtering out duplicate records of the same patient through patient ID matching and time sequence, and deleting or merging them. After completing the consistency check, the data is sorted in chronological order to obtain an optimized surgical clinical data set.

[0026] Step S17: The surgical clinical optimization data is converted into a unified format based on date format, numerical precision and unit, and the same concepts are mapped to unified standard terms, thereby standardizing the clinical data set.

[0027] After the data optimization is completed in the embodiment of the present invention, the last step is to convert the data format to ensure the consistency of the format of all data. The specific operations include unifying the date field into the YYYY-MM-DD format, unifying the precision of the numerical field to two decimal places, and ensuring that all numerical units are consistent, for example, weight is unified into kilograms (kg) and height is unified into meters (m). In addition, fields with the same concept are standardized, for example, mapping terms representing the same complications in different systems to a unified medical standard terminology set (such as ICD-10 coding). Through these operations, a standardized surgical clinical data set is finally generated to ensure the consistency and accuracy of the data in subsequent analysis and model building.

[0028] The present invention automatically extracts data through the electronic medical record system to ensure comprehensive coverage and accurate acquisition of data, reducing errors caused by manual operations; the data extracted from multiple systems covers the basic characteristics of patients and detailed imaging information, providing a solid data foundation for subsequent comprehensive analysis; the automated data extraction process can obtain the latest patient information more quickly, ensuring that the analysis is based on real-time data and improving the accuracy of decision-making. The detailed collection of surgical record data and postoperative complication data provides accurate indicators for evaluating surgical effects and postoperative risks; by tracking postoperative complications, postoperative risks can be identified and managed earlier, providing patients with better postoperative care and preventive measures; detailed surgical data and complication information can provide feedback to doctors to help improve future surgical plans and techniques. Data integration allows data from different sources to be analyzed uniformly, providing a comprehensive foundation for risk assessment and decision support; the formation of a comprehensive data set allows different types of data (such as imaging data and surgical records) to be analyzed simultaneously, which helps to discover potential associations and patterns; integrating scattered data into one data set simplifies the subsequent data processing process and improves the efficiency and accuracy of analysis. The integrity check ensures the comprehensiveness of the data set and avoids analysis bias caused by missing data; the multiple interpolation method can effectively fill in missing data, improve the integrity of the data set and the reliability of the analysis results; the reasonable handling of missing values ​​reduces the bias caused by incomplete data during data analysis and improves the accuracy of subsequent models. By detecting and processing outliers, the stability and reliability of the data set are ensured, and the negative impact of extreme values ​​on the analysis results is reduced; normalization processing enables data of different dimensions to be compared and analyzed on the same scale, eliminating the bias caused by different data units and ranges; normalized data is more suitable for training machine learning models, which helps to improve the convergence speed and prediction accuracy of the model. The consistency check ensures the uniformity of data between various fields and records, eliminating potential problems caused by inconsistent data; by deleting or merging duplicate records, redundant information in the data set is reduced, and the quality and analysis efficiency of the data are improved; the sorting processing based on time sequence enables the data to correctly reflect the time sequence in the analysis, avoiding misleading analysis caused by incorrect data sequence. Format conversion allows data to be seamlessly connected in different systems and tools, improving the compatibility and operability of data; terminology standardization ensures consistent interpretation of data in different analysis scenarios and avoids misunderstandings caused by terminology differences; standardized data sets can be more conveniently shared and used in different research and clinical applications, promoting the secondary development and utilization of data.

[0029] Preferably, step S17 comprises the following steps: Step S171: converting the surgical clinical optimization data into a format based on date format, numerical precision and unified units, and mapping the same concepts to unified standard terms, thereby obtaining a formatted standard data set; The embodiment of the present invention performs format conversion on surgical clinical optimization data. For the date format, the date is uniformly converted into the "YYYY-MM-DD" format to ensure the consistency of all time data. For numerical precision, all numerical data are standardized to two decimal places, such as the patient's weight is maintained to two decimal places. In terms of unit unification, for example, blood pressure data uniformly uses "mmHg" and height is uniformly "cm". For terms that express the same meaning in medical records, for example, different systems may use different terms to describe the same disease, and these terms are standardized and mapped using ICD-10 codes, and finally a formatted standard data set is generated to ensure data consistency and availability.

[0030] Step S172: performing desensitization processing on the formatted standard data set to obtain a desensitized clinical data set, wherein the desensitization processing specifically includes identifying and removing or encrypting the patient's personal identification information, removing image elements containing patient information from the imaging data in the formatted standard data set, and generating a unique research ID; After completing the format conversion, the embodiment of the present invention desensitizes the formatted standard data set to protect patient privacy. The first step of desensitization is to identify and remove all direct personal identification information (PII), such as name, ID number and other sensitive information. For certain information that needs to be retained but is still sensitive, such as the patient's date of birth, it is encrypted. For personal information in imaging data, such as patient information metadata embedded in MRI and CT images, specific tools are used to automatically remove this information, and a unique research ID is generated to replace the personal identifier. After these steps, a desensitized clinical data set is generated to ensure data security.

[0031] Step S173: establishing a mapping relationship according to the desensitized clinical data set and the formatted standard data set, and storing them securely, thereby obtaining desensitized mapping relationship data; The embodiment of the present invention generates a desensitized clinical data set and establishes a mapping relationship between it and the original formatted standard data set. The construction of the mapping relationship is achieved by using a unique research ID to ensure that the desensitized data can be associated with the standardized data without exposing any personal identity information. The mapping relationship data must be encrypted and stored to ensure the security of the data during storage and transmission. The desensitized mapping relationship data is stored in an encrypted database, and only authorized personnel can access this mapping information to ensure that patient privacy is fully protected.

[0032] Step S174: Implement data version control processing on the desensitized mapping relationship data to obtain a standardized clinical data set with version control, wherein the implementation of the data version control processing specifically includes assigning a unique version number to the data set at each processing stage, recording the data processing operations, timestamps and operator information of each version, implementing a data rollback mechanism and establishing a data processing log.

[0033] In order to manage data updates and historical records, the embodiment of the present invention implements version control processing on the desensitized mapping relationship data. After each data processing operation, the system automatically assigns a unique version number to the data set and records the data processing operation, timestamp, and operator information of that version. These records are stored through the log system in the database to ensure the transparency and traceability of the data processing process. In addition, the version control system allows the implementation of a data rollback mechanism, that is, if there is a problem with a certain version of the data, it can be quickly rolled back to the previous version to ensure the accuracy and reliability of data processing. Finally, a standardized clinical data set with version control is generated for subsequent training and verification of surgical risk assessment models.

[0034] The present invention can ensure that data from different sources have consistent structure and meaning through unified format and standard terminology, thereby avoiding ambiguity in data analysis and integration; standardized data format can facilitate the circulation of data between different analysis tools and systems, and reduce the conversion workload caused by format incompatibility; standardized data structure and terminology make it easier for clinical and scientific researchers to understand and use data, and improve data utilization efficiency. Through data desensitization, it is ensured that the patient's personal information will not be leaked, and the requirements of relevant laws and regulations (such as GDPR, HIPAA) on data privacy are met; the desensitized data can be used for multi-institutional or interdisciplinary research and cooperation without violating privacy regulations, and improve the sharing and versatility of data; generate a unique research ID, so that the data can still guarantee its uniqueness and traceability after desensitization, which is helpful for subsequent research and analysis. The establishment of a mapping relationship enables the desensitized data to be securely associated with the original data when needed even after the data is desensitized, ensuring the traceability and integrity of the data; through secure storage, the desensitized mapping data is prevented from being accessed or tampered with without authorization, ensuring the confidentiality and integrity of the data; the establishment and storage of the mapping relationship provides a basis for data recovery and updating, which is convenient for supplementing and correcting the data as needed in the future. Through version control and operation records, you can clearly understand the content, time and responsible person of each data modification, which helps to track the process of data evolution; data version control provides unique identification for different versions of data, which helps to quickly roll back to a stable version when problems occur in the data, ensuring data consistency and stability; detailed logs and version control records can meet audit requirements, ensure that the data processing process complies with relevant regulations and standards, and reduce legal and compliance risks.

[0035] Preferably, step S2 comprises the following steps: Step S21: extracting the size, location, shape of the tumor and its relationship with surrounding tissues from the standardized clinical data set, and generating a tumor spatial structure model using three-dimensional reconstruction technology; Step S22: numerically processing the tumor spatial structure model and extracting quantitative indicators of the tumor and key brain tissues, thereby obtaining tumor spatial characteristic data; Step S23: extracting the patient's basic disease information and the patient's preoperative status information from the standardized clinical data set, thereby obtaining basic disease data and preoperative status data; Step S24: performing feature engineering processing based on feature selection, feature combination and feature coding on the basic disease data and preoperative status data, thereby obtaining preliminary physiological feature data; Step S25: Perform dimensionality reduction processing on the preliminary physiological characteristic data, and perform feature fusion according to the tumor spatial characteristic data, so as to obtain physiological characteristic data.

[0036] As an embodiment of the present invention, refer to Figure 3 As shown, Figure 1 Detailed step flow diagram of step S2 in the embodiment of the present invention, step S2 includes the following steps: Step S21: extracting the size, location, shape of the tumor and its relationship with surrounding tissues from the standardized clinical data set, and generating a tumor spatial structure model using three-dimensional reconstruction technology; The embodiment of the present invention extracts the patient's imaging data from a standardized clinical data set, and extracts the size, position, shape of the tumor and its relationship with surrounding tissues through an automated image processing algorithm. The specific operation steps include segmenting the imaging data, extracting the edge and morphological features of the tumor through a convolutional neural network (CNN), and accurately locating the position of the tumor relative to key areas of the brain in combination with anatomical knowledge. Using three-dimensional reconstruction technology, the structure of the tumor is constructed from a two-dimensional slice image into a three-dimensional model, showing its spatial morphology and relative position to surrounding tissues, and finally generating a tumor spatial structure model for analysis.

[0037] Step S22: numerically processing the tumor spatial structure model and extracting quantitative indicators of the tumor and key brain tissues, thereby obtaining tumor spatial characteristic data; The embodiment of the present invention uses numerical calculation tools to digitize the generated three-dimensional tumor model. The spatial characteristics of the tumor are further extracted by calculating quantitative indicators such as the volume, surface area, and minimum distance from key brain tissues (such as cortex, brainstem, and nerve fiber bundles). The distance between the tumor and these key structures is automatically calculated using a spatial geometry algorithm to ensure accurate spatial positioning. These quantitative indicators can reflect the complexity of the tumor's shape, its potential impact on surrounding tissues, and the difficulty of surgical operations, and ultimately generate detailed tumor spatial feature data.

[0038] Step S23: extracting the patient's basic disease information and the patient's preoperative status information from the standardized clinical data set, thereby obtaining basic disease data and preoperative status data; The embodiment of the present invention extracts the patient's basic disease information and preoperative status data from the standardized clinical data set. Basic disease information includes past medical history such as hypertension, diabetes, and heart disease, while preoperative status information covers physiological indicators such as blood pressure, heart rate, and blood oxygen saturation of the patient. In this process, a specific rule matching algorithm is used to extract relevant fields from the electronic medical record to ensure the accuracy and completeness of the extracted data, and finally obtain basic disease data and preoperative status data to provide basic data support for subsequent analysis.

[0039] Step S24: performing feature engineering processing based on feature selection, feature combination and feature coding on the basic disease data and preoperative status data, thereby obtaining preliminary physiological feature data; The embodiment of the present invention performs feature engineering on the extracted basic disease data and preoperative status data. First, feature selection is performed, and the variance threshold is used to screen out features that have a significant impact on surgical risks, such as the patient's blood pressure control status; then feature combination is performed to construct interactive features such as "coexistence of hypertension and diabetes"; finally, feature encoding is performed on all features, and categorical data is converted into numerical data, and one-hot encoding is used to process categorical variables. Through these steps, preliminary physiological feature data is obtained, providing high-quality feature input for subsequent processing of the model.

[0040] Step S25: Perform dimensionality reduction processing on the preliminary physiological characteristic data, and perform feature fusion according to the tumor spatial characteristic data, so as to obtain physiological characteristic data.

[0041] In order to reduce the data dimension and improve the operational efficiency of the model, the embodiment of the present invention performs dimensionality reduction processing on the preliminary physiological characteristic data. In the specific operation, a dimensionality reduction algorithm such as principal component analysis (PCA) is used to retain the main features that can explain the data variance, while removing redundant information. Then, the physiological characteristic data after dimensionality reduction is fused with the tumor spatial feature data, and the spatial features and physiological features of the tumor are integrated by the feature weighting method to provide complete patient data for surgical risk assessment. The physiological characteristic data finally generated contains both the spatial information of the tumor and the patient's underlying disease and preoperative status data.

[0042] The three-dimensional reconstruction technology of the present invention can intuitively display the actual morphology of the tumor and its spatial relationship with the surrounding tissues, help doctors to more accurately understand the size, location and shape of the tumor, and help improve the accuracy of diagnosis and treatment plan formulation; through the three-dimensional model, doctors can more accurately evaluate the potential impact of different treatment plans on patients, and support the implementation of personalized and precision medicine; the three-dimensional model provides surgeons with more detailed anatomical information, which helps to optimize surgical path planning, reduce surgical risks and improve surgical success rates. Converting the three-dimensional structure into numerical data allows complex tumor characteristics to be accurately quantified and analyzed, which is helpful for subsequent statistical analysis and machine learning model construction; through numerical processing, subjective judgment is converted into objective data, reducing the impact of human factors on the evaluation results, and improving the consistency and repeatability of the results; quantitative tumor spatial feature data can be used to analyze the relationship between the tumor and the surrounding tissues and their impact on surgical risks, providing data support for risk assessment and prediction. The basic disease and preoperative status data cover the overall health status of the patient and provide important background information for surgical risk assessment; the basic diseases and preoperative status of different patients will affect the surgical risk. By extracting these data, the specific risks of each patient can be more accurately assessed; collecting comprehensive patient information, so that subsequent data analysis can be based on more comprehensive and accurate patient status data. Feature engineering helps improve the training effect of machine learning models by selecting and encoding the most representative features, reducing redundant information and noise, and ensuring that the data set is more representative; feature selection and feature combination can reduce the dimension of data, reduce computational complexity and storage requirements, and improve algorithm execution efficiency; feature engineering helps to select the physiological features that have the greatest impact on surgical risks, improve the model's ability to identify key factors, and enhance the model's interpretability. Dimensionality reduction can remove low-correlation or redundant features, simplify the model, avoid the "dimensionality disaster" problem, and improve the model's training efficiency and generalization ability; by integrating with tumor spatial feature data, a more comprehensive physiological feature data set is formed, allowing the model to take into account both tumor characteristics and patient physiological status, enhancing the accuracy of risk assessment; the optimized feature set enables the model to better capture the relationship between complex features and improve the performance and accuracy of predicting surgical risks.

[0043] Preferably, step S21 includes the following steps: Step S211: performing image segmentation processing on the imaging data in the standardized clinical data set to obtain tumor region segmentation data; The embodiment of the present invention preprocesses the imaging data in the standardized clinical data set, including denoising and smoothing operations. Then, a deep learning segmentation model based on a convolutional neural network (CNN) is used to automatically segment the image and accurately identify the tumor area. The segmentation model has been pre-trained with a large amount of annotated medical imaging data and can quickly and accurately segment the boundary area of ​​the tumor. The final tumor area segmentation data is a binary image containing the tumor outline, which clearly distinguishes the boundary between the tumor and the surrounding normal tissue.

[0044] Step S212: using a deep learning algorithm to perform feature extraction on the tumor region segmentation data, thereby obtaining tumor feature data, wherein the feature extraction specifically includes information on the size, location, and shape of the tumor; The embodiment of the present invention uses a deep learning algorithm to extract features based on the tumor region segmentation data. In the specific operation, based on the segmented tumor region, the key features of the tumor, such as volume, position, shape, etc., are extracted. A three-dimensional convolutional neural network (3D CNN) is used to further analyze the features of the tumor at different slice levels, and the volume, center position coordinates, and shape features of the tumor are automatically generated. These feature data can be used to build subsequent models and provide a quantitative analysis basis for surgical risk assessment.

[0045] Step S213: performing three-dimensional reconstruction processing on the tumor characteristic data to obtain a preliminary three-dimensional model; The embodiment of the present invention uses three-dimensional reconstruction technology to reconstruct tumor feature data. By stacking the tumor segmentation data of each slice in a spatial sequence and combining the volume and shape information of the tumor, a preliminary three-dimensional tumor model is generated. The model can reflect the actual size, morphology and spatial distribution of the tumor in the patient's brain. The preliminary three-dimensional model provides a complete structural image of the tumor, preparing for the subsequent spatial registration with normal brain tissue.

[0046] Step S214: using image registration technology to register the preliminary three-dimensional model with the normal brain tissue image in the standardized clinical data set, thereby obtaining the relationship data between the tumor and the surrounding tissue; In order to more accurately evaluate the relationship between the tumor and the surrounding tissue, the embodiment of the present invention uses image registration technology to register the preliminary three-dimensional tumor model with the normal brain tissue image obtained from the standardized clinical data set. The registration technology uses an algorithm based on feature point matching to gradually adjust the spatial position of the tumor model by identifying the common feature points of the tumor model and normal brain tissue in space, so that it is accurately aligned with the surrounding tissue image. After registration, the relative relationship between the spatial position of the tumor and key brain regions, such as the cortex, brainstem and other structures, is accurately described, and the relationship data between the tumor and the surrounding tissue is generated.

[0047] Step S215: Perform three-dimensional visualization processing using computer graphics technology based on the tumor and surrounding tissue relationship data to obtain a tumor spatial structure model.

[0048] The embodiment of the present invention uses computer graphics technology to perform three-dimensional visualization processing based on the data of the relationship between the tumor and the surrounding tissue. The specific steps include visualizing the model of the tumor and its surrounding tissues through three-dimensional graphics rendering tools such as OpenGL or VTK to construct a three-dimensional structure. The visualization process can clearly show the shape, position, and spatial relationship of the tumor with the surrounding brain tissue, providing doctors with an intuitive reference for surgical planning. The resulting tumor spatial structure model is used to support surgical decisions and can be used for further risk assessment and preoperative preparation analysis.

[0049] The image segmentation of the present invention can accurately identify the area where the tumor is located from the imaging data, effectively distinguish the tumor from normal brain tissue, and ensure the accuracy of subsequent analysis; the automated image segmentation technology reduces the subjective bias of relying on human interpretation of images, and improves the consistency and reliability of data processing; the automated segmentation method can quickly process a large amount of image data, significantly improve the data processing speed, and save time for clinical and scientific researchers. The deep learning algorithm can automatically extract complex tumor features from the tumor area, reduce the workload of manual analysis, and improve the efficiency and accuracy of feature extraction; the deep learning technology can identify complex patterns and features in the image that are difficult to extract by traditional methods, and provide rich input data for subsequent modeling; by extracting a variety of tumor features (such as size, location, shape), it provides a more comprehensive information basis for the surgical risk assessment model and improves the predictive performance of the model. Three-dimensional reconstruction can generate a three-dimensional structural view of the tumor from a two-dimensional image, provide intuitive three-dimensional information, and help doctors better understand the spatial distribution of the tumor; the three-dimensional model provides doctors with a more intuitive reference, which helps to accurately plan the surgical path, reduce intraoperative risks, and improve the success rate of surgery; the three-dimensional model can better display the spatial characteristics of the tumor, support personalized surgery and treatment plan design, and optimize the patient's treatment effect. Image registration technology can accurately align tumor models and normal brain tissue images, provide the relative position relationship between tumors and surrounding tissues, and help evaluate the tumor invasion range and surgical risks; registration technology can integrate data from different image sources to provide a unified reference framework for subsequent 3D visualization and analysis; through registration, information from different types of images (such as MRI and CT) can be used simultaneously to enhance the overall analysis ability of the relationship between tumors and surrounding tissues. 3D visualization provides a three-dimensional display of the relationship between tumors and surrounding tissues, allowing doctors to more clearly understand the specific location and shape of tumors, and improve the intuitiveness and accuracy of diagnosis; 3D visualization images not only help doctors understand cases, but can also be used to explain the condition and surgical plan to patients and their families, and improve the communication effect between doctors and patients; 3D visualization provides an intuitive reference for the surgical navigation system, which helps to accurately guide the surgical process, reduce intraoperative accidents, and improve the safety and effectiveness of surgery.

[0050] Preferably, step S3 comprises the following steps: Step S31: integrating the tumor spatial structure model and physiological characteristic data, and performing preliminary analysis using descriptive statistical methods to obtain characteristic statistical description data, wherein the preliminary analysis specifically includes calculation of the mean, standard deviation and quantile of each characteristic; The embodiment of the present invention integrates the tumor spatial structure model with the patient's physiological characteristic data to construct a comprehensive data set containing tumor morphological characteristics and physiological characteristics. Then, the data set is preliminarily analyzed using descriptive statistical methods to calculate statistical indicators such as the mean, standard deviation, minimum, maximum, and 25%, 50%, and 75% quantiles of each feature. The statistical descriptive data of these features can provide a global perspective to help understand the distribution of each variable in the patient population and lay the foundation for subsequent risk analysis.

[0051] Step S32: performing a preliminary contribution evaluation based on a chi-square test on the correlation between each feature and surgical risk according to the feature statistical description data, thereby obtaining a univariate analysis result; After obtaining the characteristic statistical description data, the embodiment of the present invention uses the chi-square test method to evaluate the correlation between each feature and the surgical risk. In the specific operation, an independent chi-square test is performed on each tumor feature and physiological feature to calculate the contribution of each feature to the surgical risk. The significance level used in this process is set to 0.05. If the p value is less than the threshold, it is considered that the feature is significantly correlated with the surgical risk. Through this univariate analysis, a preliminary contribution assessment of each feature to the surgical risk is obtained to form a univariate analysis result.

[0052] Step S33: Based on the single factor analysis results, an interaction analysis based on the construction and evaluation of pairwise feature interaction items is performed to obtain interaction feature data; Based on the univariate analysis, the embodiment of the present invention constructs pairwise feature interaction terms to evaluate the impact of the interaction between features on surgical risk. In the specific steps, the generalized linear model (GLM) is used to construct the interaction term model, and the regression coefficient and its significance level of each interaction term are calculated. Interaction analysis can reveal the synergistic effect between two features and help understand the complex factors affecting surgical risk. The interaction feature data finally obtained will be used for further model optimization and risk assessment.

[0053] Step S34: performing significance tests on each factor on the single factor analysis results and the interactive feature data, thereby obtaining significance test result data, wherein the significance test includes a Wald test and a likelihood ratio test; The embodiment of the present invention performs a significance test on the univariate analysis results and the interactive feature data. The significance of each feature and interaction term is evaluated by the Wald test and the likelihood ratio test to test whether it has a significant impact on the surgical risk. The Wald test evaluates its impact on the surgical risk by calculating the standard error of the feature regression coefficient, while the likelihood ratio test compares the difference in goodness of fit between the model containing the feature and the model without the feature. The results of the significance test can screen out the features that have a greater impact on the surgical risk.

[0054] Step S35: screening risk factors according to the significance test result data, and performing weight calculation on the screened risk factor set, thereby obtaining preliminary risk factor weight data; According to the results of the significance test, the embodiment of the present invention further screens the risk factors after screening, and calculates the weights of the screened risk factor set. In the specific operation, the regression coefficient of each risk factor is calculated by multivariate regression analysis as the preliminary weight of the risk factor. The size of the weight reflects the relative contribution of each risk factor to the surgical risk, and the preliminary risk factor weight data obtained is used for the construction of the subsequent model.

[0055] Step S36: Perform stability analysis on the preliminary risk factor weight data based on bootstrap resampling and cross-validation, and adjust the weights to obtain the risk factor weight data.

[0056] The embodiment of the present invention performs stability analysis and weight adjustment on the preliminary risk factor weight data based on the bootstrap resampling method and cross-validation. Through multiple resampling, the weight fluctuation of each risk factor on different data sets is evaluated to ensure the robustness of the model in different patient groups. Cross-validation is used to verify the predictive performance of the model and select the best risk factor weight combination. After adjustment, the final risk factor weight data will be used to construct a surgical risk assessment model to provide support for clinical decision-making.

[0057] The present invention integrates multiple data sources to ensure that a complete and consistent data set is obtained before analysis, thereby increasing the reliability of the analysis; descriptive statistics provide a preliminary understanding of the basic characteristics of the data, help identify potential patterns and anomalies in the data, and lay the foundation for subsequent in-depth analysis; by understanding the distribution of each feature, it is possible to better select suitable data conversion or preprocessing methods and optimize the quality of the input data of the model. Univariate analysis can preliminarily screen out important features that may affect surgical risk, help narrow the scope of analysis, and improve the efficiency of subsequent analysis; the chi-square test can detect the significant relationship between features and risks, providing a basis for subsequent multivariate analysis; the results of univariate analysis can guide which features should be further studied, thereby optimizing the process of feature selection and improving the performance of the model. By analyzing the interaction between features, more complex relationships that cannot be revealed by single feature analysis can be found, providing richer input information for the accuracy of the model; interactive feature data can significantly improve the predictive performance of the model, especially when there are nonlinear or complex relationships between features; interactive analysis can help understand how features jointly affect surgical risk, thereby providing more basis for clinical decision-making. The significance test helps to confirm which features have a statistically significant effect on surgical risk, thereby improving the credibility of the model; through the significance test, insignificant or redundant features can be removed, the model structure can be simplified, and the computational complexity and overfitting risk can be reduced; the significance test results provide clear guidance for further model tuning, which helps to improve the stability and performance of the model. By screening significant risk factors, the analysis model can be streamlined, noise and unnecessary computational burden can be reduced; weight calculation provides the relative contribution of each risk factor to surgical risk, which is convenient for quantifying and explaining the impact of risk factors; the weight of risk factors can be used for customized risk assessment, which helps to provide personalized surgical risk analysis and recommendations for different patients. Using resampling and cross-validation methods to perform stability analysis on weight data can reduce sample bias and improve the consistency of model performance on different data sets; through weight adjustment, the actual impact of each risk factor can be more accurately reflected, and the reliability of the model for surgical risk assessment can be enhanced; through multiple verification and adjustment of risk factor weights, the generalization ability of the model can be improved, making its performance on new data more robust.

[0058] Preferably, step S33 includes the following steps: Step S331: classify the continuous variables in the univariate analysis results to obtain classified risk factor data; The embodiment of the present invention performs classification processing on the continuous variables in the univariate analysis results. In the specific operation, the continuous variable is converted into multiple categorical variables based on the quantile method. For example, for the continuous variable of tumor size, it is divided into three categories: small, medium, and large according to the 25%, 50%, and 75% quantiles. The goal of the classification processing is to convert the continuous variable into a discrete variable that is easy to analyze, thereby obtaining classified risk factor data. These classified data are convenient for the construction and analysis of subsequent interaction items.

[0059] Step S332: construct all possible pairwise feature interaction items based on the classified risk factor data, thereby obtaining a candidate set of interaction items; After obtaining the classified risk factor data, the embodiment of the present invention constructs all possible pairwise feature interaction terms based on the data. In the specific operation, an exhaustive method is used to generate the combination of each two classified risk factors, and their interaction is used as a new feature. For example, the interaction between tumor size and patient age is used as a new interaction term. In this way, by systematically constructing interaction terms, a candidate set of interaction terms is formed, which includes all possible pairwise interaction terms. This process enables subsequent analysis to capture the synergistic effect between features.

[0060] Step S333: performing a Gini index evaluation on the importance of each interaction item in the interaction item candidate set, thereby obtaining an importance ranking of the interaction items; The embodiment of the present invention uses the Gini index to evaluate the importance of each interaction item in the interaction item candidate set. The Gini index is an indicator commonly used in decision tree models, which can evaluate the importance of each interaction item to the surgical risk classification. The Gini index of each interaction item is calculated, and the interaction items are sorted according to the index value. By sorting the Gini index, the interaction items with the greatest influence on the surgical risk prediction can be identified, forming the interaction item importance sorting result.

[0061] Step S334: sorting the importance of the interaction items according to a preset threshold to filter the interaction items with higher importance, thereby obtaining a filtered interaction item set; The embodiment of the present invention screens the interaction item importance ranking results according to the preset Gini index threshold, and selects the interaction items with higher importance. In the specific operation, assuming that the preset Gini index threshold is 0.01, all interaction items with a Gini index greater than 0.01 will be retained, and interaction items below the threshold will be eliminated. Through this screening process, a set of screened interaction items with significant impact on surgical risk is obtained, providing more streamlined and efficient feature data for subsequent model construction.

[0062] Step S335: The screened interaction item set is combined with the univariate analysis result to construct an extended feature set, thereby obtaining interaction feature data.

[0063] The embodiment of the present invention combines the screened interaction item set with the univariate analysis results to construct an extended feature set. The extended feature set not only includes the significant risk factors in the univariate analysis, but also includes the screened important interaction items. Through this extended feature set, the influence of a single factor and its interaction can be comprehensively considered to obtain more comprehensive interaction feature data. These data will be further used for training and optimization of the surgical risk assessment model to improve the predictive performance of the model.

[0064] The present invention converts continuous variables into categorical variables, which can simplify the processing and analysis of variables and help to more intuitively understand the distribution characteristics and risk patterns of data; categorical variable processing reduces the instability caused by extreme values ​​or noise, thereby improving the robustness and robustness of the model; categorical variables are easy to interpret, especially in clinical scenarios, which can make risk assessment results more intuitive and operational. Pairwise interaction terms can reveal more complex relationships between features and capture patterns and trends that cannot be explained by a single variable; by constructing interaction terms, the feature space can be significantly expanded, providing the model with richer feature inputs and improving predictive ability; in clinical decision-making, understanding how different risk factors interact can better support personalized treatment and intervention plans. The Gini index can effectively measure the contribution of each interaction term to the model output (surgical risk), which helps to identify which interaction terms have higher discriminatory power; by sorting the importance of interaction terms, computing resources and model complexity can be concentrated on the most valuable interaction terms to optimize model performance; screening important interaction terms can reduce the interference of noise features on the model, thereby reducing the risk of overfitting. By screening important interaction terms, the feature dimension can be reduced, thereby simplifying the model structure, improving computational efficiency and interpretability; removing low-importance interaction terms can help improve the generalization ability and predictive performance of the model, especially when the sample size is limited; screening out the most clinically significant interaction features can help doctors better understand and use risk assessment results. The extended feature set includes single-factor features and important interaction terms, providing more comprehensive risk factor information and helping to improve the model's predictive accuracy; by combining single-factor and interaction features, it can more deeply explain the impact of each risk factor and its interaction on surgical risk, providing a more detailed basis for clinical decision-making; comprehensively considering single-factor and interaction features enables the model to more accurately perform individualized risk assessment and support the formulation of personalized medical plans.

[0065] Preferably, step S35 includes the following steps: Step S351: setting a significance threshold according to the significance test result data to screen each risk factor, thereby obtaining a preliminary risk factor set; The embodiment of the present invention sets a significance threshold according to the significance test result data and screens each risk factor. For example, a p value less than 0.05 is set as the significance threshold, and all risk factors that meet this condition are included in the preliminary risk factor set. This screening process ensures that only factors that have a significant impact on surgical risk are included, thereby obtaining a more accurate preliminary risk factor set, removing noise data, and enhancing the prediction ability of the model.

[0066] Step S352: classify the continuous variables in the preliminary risk factor set and convert them into categorical variables, thereby obtaining discretized risk factor data; determine the segmentation points of the discretized risk factor data and perform risk stratification, thereby obtaining risk stratification data; The embodiment of the present invention classifies the continuous variables in the preliminary risk factor set and converts them into categorical variables. For example, for the continuous variable of patient age, it is divided into three categories: less than 40 years old, 40 to 60 years old, and more than 60 years old. By determining a reasonable split point, risk stratification can be performed, and a corresponding risk level is given for each stratification. In this way, the discretized risk factor data is easier to perform further statistical analysis and stratified evaluation.

[0067] Step S353: Estimate the risk coefficient of each risk factor according to the risk stratification data, and convert it into a relative risk value, thereby obtaining relative risk data; The embodiment of the present invention estimates the risk coefficient of each risk factor based on the risk stratification data. In the specific operation, the Cox proportional risk model is used to calculate the risk coefficient of each risk factor, and these coefficients are converted into relative risk values. For example, for patients in a certain age group, the risk coefficient is 1.5, then its relative risk value indicates that the surgical risk of patients in this age group is 50% higher than that of other patients. This relative risk value can more intuitively reflect the differences in surgical risks under different stratifications.

[0068] Step S354: normalize the relative risk data and use the min-max standardization method to assign weights with a sum of weights equal to 1, thereby obtaining preliminary risk factor weight data.

[0069] The embodiment of the present invention normalizes the relative risk data. In the specific operation, the min-max normalization method is used to map the relative risk value to the interval [0, 1], and ensure that the sum of the weights of all risk factors is 1. Through this normalization step, the impact of the scale differences of different factors is eliminated, making the weight calculation more reasonable. For example, assuming that the relative risk value of a risk factor is 0.8, after standardization, its weight is 0.2. This processing ensures the reasonable distribution of the weights of risk factors in the model, thereby obtaining preliminary risk factor weight data, which provides a basis for the optimization of subsequent risk assessment models.

[0070] The present invention reduces the dimension of features, simplifies the computational complexity of the model, and improves the training efficiency by eliminating insignificant risk factors; only retains factors significantly related to surgical risk, so that the output of the model is more clinically meaningful and interpretable, and is easy for clinicians to understand and use; eliminating insignificant factors helps to reduce the noise components in the data, reduce the overfitting risk of the model, and enhance the generalization ability of the model. Converting continuous variables into discrete variables can simplify the data analysis process, making the model easier to understand and explain; classification processing reduces the impact of abnormal values ​​or extreme values ​​in continuous data on the model, and enhances the robustness of the model; risk stratification enables the model to be grouped and evaluated according to different risk levels, which helps clinicians to more clearly understand the risk level of the patient when formulating a treatment plan. By calculating the relative risk value, the degree of influence of different risk factors on surgical risk can be quantified, making the evaluation more accurate; the relative risk value can provide individualized risk assessment information for each patient, supporting more personalized clinical decisions; using the relative risk value can more accurately measure the effect of risk factors, thereby improving the predictive ability of the risk assessment model. Normalization makes the relative weights of all risk factors at the same level, avoiding the uneven impact of certain risk factors on the model due to scale differences; through standardization and weight normalization, errors caused by different dimensions in the data can be eliminated, and the stability and consistency of the model can be enhanced; the distribution method with a total weight of 1 ensures that the contribution of risk factors in the model is appropriately adjusted, thereby improving the accuracy and reliability of risk assessment.

[0071] Preferably, step S4 comprises the following steps: Step S41: constructing an initial neural network structure including an input layer, a hidden layer, and an output layer according to the risk factor weight data, and training it using a standardized clinical data set to obtain a surgical risk assessment model; The embodiment of the present invention uses risk factor weight data to construct an initial neural network structure. In specific operations, the number of nodes in the input layer is set equal to the number of risk factors, the number of nodes in the hidden layer is initially set to twice the number of risk factors, and the output layer contains a node for outputting the surgical risk score. The neural network model is preliminarily trained using a standardized clinical data set, and trained using the Adam optimizer with an initial learning rate of 0.01 until the model error on the training data decreases and tends to stabilize, and finally an initial surgical risk assessment model is obtained.

[0072] Step S42: Divide the standardized clinical data set into K subsets, wherein K-1 subsets are used for training and the remaining 1 subset is used for validation, and perform K-fold cross validation on the surgical risk assessment model to obtain cross validation result data; The embodiment of the present invention divides the standardized clinical data set into 5 subsets, 4 of which are used for model training and the remaining 1 subset is used for model validation, and 5-fold cross validation is performed. After each training is completed, the prediction results on the validation set are recorded, and the average accuracy, sensitivity and specificity on the validation set are calculated, and finally the average results of 5 cross validations are obtained to form the cross validation result data. Cross validation ensures the generalization performance of the model on different data sets and reduces the risk of overfitting.

[0073] Step S43: performing a model evaluation on the surgical risk assessment model based on average accuracy, sensitivity and specificity according to the cross-validation result data, thereby obtaining model performance evaluation data; The embodiment of the present invention performs a performance evaluation on the surgical risk assessment model based on the cross-validation result data. The overall performance of the model is evaluated by calculating the average accuracy, sensitivity, specificity and other indicators. For example, assuming that the average accuracy of the model is 85%, the sensitivity is 82%, and the specificity is 88%, these evaluation data will serve as preliminary reference indicators of the model performance and provide a basis for subsequent optimization.

[0074] Step S44: optimizing the hyperparameters of the surgical risk assessment model based on the learning rate, the number of hidden layers, and the number of neurons in each layer according to the model performance evaluation data by a grid search method, and performing optimization training, thereby obtaining an optimized surgical risk assessment model; The embodiment of the present invention optimizes the hyperparameters of the surgical risk assessment model by a grid search method based on the model performance evaluation data. For example, a search is performed between learning rates of 0.001, 0.01, and 0.1, while adjusting the number of hidden layers (e.g., 1 to 3 layers) and the number of neurons in each layer (e.g., 64, 128, 256). Each set of hyperparameter combinations is trained and verified, and the optimal hyperparameter combination is recorded, and finally an optimized surgical risk assessment model is obtained.

[0075] Step S45: performing overfitting control based on the early stopping method on the surgical risk assessment optimization model through L1 regularization and L2 regularization, thereby obtaining an anti-overfitting model; In order to avoid overfitting of the model, the embodiment of the present invention uses L1 and L2 regularization methods to control overfitting, and combines the early stopping method to monitor the error change on the validation set to prevent the model from continuing to train when the error on the validation set increases. For example, the L1 regularization coefficient is set to 0.01 and the L2 regularization coefficient is set to 0.001. If the error on the validation set does not decrease for 5 consecutive rounds, the early stopping mechanism is triggered. Finally, an anti-overfitting model that has been overfitting controlled is obtained.

[0076] Step S46: pre-setting a portion of data as an independent data set according to the standardized clinical data set, and performing a performance test on the anti-overfitting model according to the independent data set, thereby obtaining model test result data; calculating the accuracy, precision, recall rate and F1 score of the anti-overfitting model according to the model test result data, thereby obtaining model evaluation index data; The embodiment of the present invention sets an independent data set from the standardized clinical data set in advance, and uses the data set to perform performance testing on the anti-overfitting model. The actual prediction ability of the model is measured by calculating the accuracy, precision, recall and F1 score on the independent data set. For example, the accuracy of the model is 87%, the precision is 85%, the recall is 80%, and the F1 score is 0.825, forming the model test result data.

[0077] Step S47: Draw the ROC curve and the PR curve according to the model evaluation index data, and calculate the AUC value, so as to obtain the model performance curve data; use the Bootstrap method to estimate the confidence interval of the model performance curve data, so as to obtain the model performance confidence interval data; The embodiment of the present invention draws the ROC curve and PR curve according to the model test result data, and calculates the AUC value. For example, the AUC value of the model is 0.91, indicating that the performance of the model under different thresholds is relatively good. Then, the Bootstrap method is used to estimate the confidence interval of the performance curve data of the model, for example, the 95% confidence interval is calculated to be [0.88, 0.94], and finally the model performance confidence interval data is obtained.

[0078] Step S48: Optimize the prediction performance of the anti-overfitting model according to the model performance confidence interval data, so as to obtain a surgical risk prediction model.

[0079] The embodiment of the present invention further optimizes the prediction performance of the anti-overfitting model based on the model performance confidence interval data. By adjusting the prediction threshold or re-optimizing the regularization coefficient, the prediction ability of the model in different scenarios is enhanced, and finally a surgical risk prediction model is obtained, which can provide stable and accurate surgical risk prediction results.

[0080] The neural network of the present invention can handle complex nonlinear relationships and automatically learn deep features from risk factor data, thereby improving the accuracy of risk assessment; the design of the initial neural network structure provides flexibility, and the appropriate number of layers and neurons can be selected according to the characteristics of the data to meet the needs of different data sets; through initial model training, a baseline prediction capability is established to lay the foundation for subsequent model optimization. Cross-validation can make full use of limited data and reduce model errors caused by uneven data division through multiple training-validation cycles, thereby reducing the risk of overfitting; by evaluating on different training-validation set combinations, it is ensured that the model has stronger generalization ability and can be better applied to new data; the cross-validation result data provides richer information for the comprehensive evaluation of model performance, including performance indicators such as average accuracy and standard deviation. Accuracy, sensitivity and specificity are evaluated at the same time to ensure the balanced performance of the model in different evaluation dimensions and avoid the bias of a single indicator; it is clear in which aspects of the model need to be improved (such as improving sensitivity or specificity) to provide a basis for the next step of model optimization. Hyperparameter optimization can significantly improve the performance of the model, making the model better fit the data while avoiding underfitting; through reasonable parameter selection, the training time and computing resource consumption can be reduced; the grid search method automatically adjusts the parameters, reduces human intervention, and improves the efficiency and effect of model parameter adjustment. Regularization technology effectively reduces the complexity of the model and avoids overfitting caused by excessive weights; the early stopping method automatically terminates the training process, allowing the model to stop before reaching the best performance, avoiding overtraining and enhancing the prediction ability of new data; L1 regularization helps to sparse the model, automatically remove unimportant features, and improve the simplicity and interpretability of the model. The performance on the independent test set can truly reflect the generalization ability of the model and ensure its applicability to unknown data; the comprehensive use of multiple indicators such as accuracy, precision, recall and F1 score can more comprehensively evaluate the performance of the model and help identify the advantages and disadvantages of the model; independent data set testing avoids the risk of exposing test data during training and improves the fairness of the evaluation. The ROC curve and AUC value directly reflect the model's ability to distinguish under different thresholds and are important tools for evaluating the quality of binary classification models. The Bootstrap method provides confidence intervals for the model's performance indicators, quantifies the statistical uncertainty of the model's performance, and makes the evaluation more scientific and reliable. The curve graph shows the performance of the model under different conditions, making the evaluation results more intuitive and easy to understand. The use of confidence interval optimization makes the model's performance on different samples more stable, reducing dependence on the results of a single data set; further optimizing the model parameters and structure makes the model have higher prediction accuracy and reliability in practical applications; the optimized surgical risk prediction model has better performance, which can more effectively assist doctors in making decisions in clinical environments and improve patient treatment outcomes.

[0081] Preferably, step S5 comprises the following steps: Step S51: Designing an API interface based on data input, model call and result output functions according to the surgical risk prediction model, thereby obtaining model API interface data; According to the surgical risk prediction model, the embodiment of the present invention first designs an API interface to support data input, model call and result output functions. In the specific operation, the API is designed using the RESTful architecture, and the input format is defined as JSON, which contains the patient's basic information, imaging data and physiological characteristics data. The model call sends data to the server through a POST request and returns the prediction result. To ensure data security, OAuth 2.0 is used for identity authentication, and the format verification of the input data is implemented to ensure the robustness and security of the interface. Finally, the model API interface data is obtained to ensure seamless integration with other systems.

[0082] Step S52: docking the surgical risk prediction model with the medical system according to the model API interface data, and performing functional testing to obtain integrated model data; The embodiment of the present invention connects the surgical risk prediction model with the hospital's medical system based on the model API interface data. During the connection process, the medical system development team cooperates with the model development team to use the interface document for functional testing, including simulating input data, calling the API and recording the return results. During the test, multiple scenarios are set, such as normal cases, complex cases, and abnormal cases, and it is confirmed that the prediction results returned by the model meet expectations. Finally, the integrated model data is formed to ensure that the surgical risk prediction model can be correctly embedded in the hospital's workflow.

[0083] Step S53: Integrate the integrated model data with the hospital's existing electronic medical record system and image archiving and communication system to assist surgical decision-making.

[0084] The embodiment of the present invention deeply integrates the integrated model data with the hospital's existing electronic medical record system (EMR) and picture archiving and communication system (PACS). In the specific implementation, a new field is first added to the EMR to store the surgical risk prediction results, and a corresponding user interface is designed to facilitate doctors to view and use the prediction data. At the same time, the API call is integrated into the PACS system, and doctors can directly retrieve the risk prediction data through image data. Multiple joint tests are carried out to ensure smooth data flow in a real operating environment, ultimately assisting surgical decision-making and improving patient treatment effects and safety.

[0085] The API interface designed by the present invention enables the surgical risk prediction model to be integrated with different systems and applications. Through the standardized API interface, the model can be easily embedded in various medical systems without rewriting or adjusting the code. The API interface provides a standardized data input and output method for the model, ensuring that the model can be smoothly connected with the hospital information system (such as the electronic medical record system) and other related systems. By defining a clear interface function, the calling process of the model can be simplified, so that the user or system can efficiently and accurately obtain the prediction results of the model; the API interface design can include data encryption and security authentication mechanisms to ensure the security of the data transmission process and protect the privacy of patients. By connecting the model API interface with the medical system, the integration effect of the model in the actual system can be verified to ensure that its function is normal and stable; functional testing can check the compatibility of the model with the medical system to ensure that operations such as data transmission, model calling and result output are carried out smoothly; during the integration process, functional testing can help discover potential technical problems or functional defects, so as to repair them in time to avoid affecting the normal use of the system; through sufficient testing, the stability and reliability of the integrated system can be improved to ensure that the performance in actual use is consistent with expectations. After integrating the prediction model with the hospital's existing system, the patient's surgical risk prediction results can be obtained in real time, assisting doctors in making more scientific and data-driven decisions during the surgical decision-making process. The integrated system can automatically process and display surgical risk assessment results, reduce manual operation and data input time, and improve doctors' work efficiency. By providing real-time risk prediction and decision support, potential surgical risks can be identified in advance, so that preventive measures can be taken to improve patient safety and treatment effects. Integrating model data into existing systems can better utilize medical resources, optimize surgical arrangements, reduce unnecessary surgical risks, and thus reduce medical costs.

[0086] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0087] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for constructing a risk assessment model for minimally invasive surgery of craniocerebral tumors, characterized in that: The following steps are involved: Step S1: Obtaining clinical data of minimally invasive surgery for craniocerebral tumors, including patient imaging data, surgical record data, postoperative complication data, and patient basic data; cleaning the surgical clinical data and performing standardization processing to unify the data format, thereby obtaining a standardized clinical data set; Step S2: extracting the size, location, shape and relationship of the tumor with surrounding tissues from the standardized clinical data set, and generating a tumor spatial structure model using three-dimensional reconstruction technology; extracting physiological characteristics based on the underlying disease and preoperative status from the standardized clinical data set, thereby obtaining physiological characteristic data; Step S3: Identify the main factors affecting surgical risk based on the tumor spatial structure model and physiological characteristic data by applying statistical analysis methods, thereby generating risk factor weight data, wherein the main factor identification includes the contribution analysis of each feature to surgical risk, the interaction between features and their impact on surgical risk, the significance test of each feature, and the screening of risk factors; Step S4: Generate a surgical risk assessment model based on a neural network model according to the risk factor weight data; optimize the risk assessment model according to the standardized clinical data set through cross-validation, and use an independent data set in the pre-set standardized clinical data set to evaluate the prediction performance and accuracy, thereby obtaining a surgical risk prediction model; Step S5: The surgical risk prediction model is embedded into the medical system through the API interface, and integrated with the hospital's existing electronic medical record system and image archiving and communication system to assist surgical decision-making.

2. The method for constructing a risk assessment model for minimally invasive surgery for craniocerebral tumors according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: extracting basic data of patients with minimally invasive brain tumors through the electronic medical record system, including age, gender, weight, height, and medical history; and obtaining patient imaging data including MRI and CT through the image archiving and communication system; Step S12: collecting the operation time, operation method and operation difficulty rating information through the operation information management system to obtain the operation record data; obtaining the postoperative complication data from the postoperative follow-up system, including the complication type, severity and occurrence time; Step S13: combining the patient's imaging data, surgical record data, postoperative complication data, and basic patient data into clinical data of minimally invasive surgery for craniocerebral tumors; Step S14: Check the data integrity of the surgical clinical data and use the multiple interpolation method to handle missing data, so as to obtain complete clinical data; Step S15: performing anomaly detection and outlier processing based on the interquartile range on the complete clinical data, and converting the numerical data therein into the interval [0, 1], thereby obtaining a preprocessed clinical data set; Step S16: performing data consistency check and repeatability check on the preprocessed clinical data set to obtain validity verification data; deleting or merging duplicate records and sorting the preprocessed clinical data set based on the validity verification data to obtain surgical clinical optimization data; Step S17: The surgical clinical optimization data is converted into a unified format based on date format, numerical precision and unit, and the same concepts are mapped to unified standard terms, thereby standardizing the clinical data set.

3. The method for constructing a risk assessment model for minimally invasive surgery of craniocerebral tumors according to claim 2, characterized in that: Step S17 includes the following steps: Step S171: converting the surgical clinical optimization data into a format based on a unified date format, numerical precision, and unit, and mapping the same concepts to unified standard terms, thereby obtaining a formatted standard data set; Step S172: performing desensitization processing on the formatted standard data set to obtain a desensitized clinical data set, wherein the desensitization processing specifically includes identifying and removing or encrypting the patient's personal identification information, removing image elements containing patient information from the imaging data in the formatted standard data set, and generating a unique research ID; Step S173: establishing a mapping relationship according to the desensitized clinical data set and the formatted standard data set, and storing them securely, thereby obtaining desensitized mapping relationship data; Step S174: Implement data version control processing on the desensitized mapping relationship data to obtain a standardized clinical data set with version control, wherein the implementation of the data version control processing specifically includes assigning a unique version number to the data set at each processing stage, recording the data processing operations, timestamps and operator information of each version, implementing a data rollback mechanism and establishing a data processing log.

4. The method for constructing a risk assessment model for minimally invasive surgery for craniocerebral tumors according to claim 3, characterized in that: Step S2 includes the following steps: Step S21: extracting the size, location, shape of the tumor and its relationship with surrounding tissues from the standardized clinical data set, and generating a tumor spatial structure model using three-dimensional reconstruction technology; Step S22: numerically processing the tumor spatial structure model and extracting quantitative indicators of the tumor and key brain tissues, thereby obtaining tumor spatial characteristic data; Step S23: extracting the patient's basic disease information and the patient's preoperative status information from the standardized clinical data set, thereby obtaining basic disease data and preoperative status data; Step S24: performing feature engineering processing based on feature selection, feature combination and feature coding on the basic disease data and preoperative status data, thereby obtaining preliminary physiological feature data; Step S25: Perform dimensionality reduction processing on the preliminary physiological characteristic data, and perform feature fusion according to the tumor spatial characteristic data, so as to obtain physiological characteristic data.

5. The method for constructing a risk assessment model for minimally invasive surgery for craniocerebral tumors according to claim 4, characterized in that: Step S21 includes the following steps: Step S211: performing image segmentation processing on the imaging data in the standardized clinical data set to obtain tumor region segmentation data; Step S212: using a deep learning algorithm to perform feature extraction on the tumor region segmentation data, thereby obtaining tumor feature data, wherein the feature extraction specifically includes information on the size, location, and shape of the tumor; Step S213: performing three-dimensional reconstruction processing on the tumor characteristic data to obtain a preliminary three-dimensional model; Step S214: using image registration technology to register the preliminary three-dimensional model with the normal brain tissue image in the standardized clinical data set, thereby obtaining the relationship data between the tumor and the surrounding tissue; Step S215: Perform three-dimensional visualization processing using computer graphics technology based on the tumor and surrounding tissue relationship data to obtain a tumor spatial structure model.

6. The method for constructing a risk assessment model for minimally invasive surgery for craniocerebral tumors according to claim 5, characterized in that: Step S3 includes the following steps: Step S31: integrating the tumor spatial structure model and physiological characteristic data, and performing preliminary analysis using descriptive statistical methods to obtain characteristic statistical description data, wherein the preliminary analysis specifically includes calculation of the mean, standard deviation and quantile of each characteristic; Step S32: performing a preliminary contribution evaluation based on a chi-square test on the correlation between each feature and surgical risk according to the feature statistical description data, thereby obtaining a univariate analysis result; Step S33: Based on the single factor analysis results, an interaction analysis based on the construction and evaluation of pairwise feature interaction items is performed to obtain interaction feature data; Step S34: performing significance tests on each factor on the single factor analysis results and the interactive feature data, thereby obtaining significance test result data, wherein the significance test includes a Wald test and a likelihood ratio test; Step S35: screening risk factors according to the significance test result data, and performing weight calculation on the screened risk factor set, thereby obtaining preliminary risk factor weight data; Step S36: Perform stability analysis on the preliminary risk factor weight data based on bootstrap resampling and cross-validation, and adjust the weights to obtain the risk factor weight data.

7. The method for constructing a risk assessment model for minimally invasive surgery of craniocerebral tumors according to claim 6, characterized in that: Step S33 includes the following steps: Step S331: classify the continuous variables in the univariate analysis results to obtain classified risk factor data; Step S332: construct all possible pairwise feature interaction items based on the classified risk factor data, thereby obtaining a candidate set of interaction items; Step S333: performing a Gini index evaluation on the importance of each interaction item in the interaction item candidate set, thereby obtaining an importance ranking of the interaction items; Step S334: sorting the importance of the interaction items according to a preset threshold to filter the interaction items with higher importance, thereby obtaining a filtered interaction item set; Step S335: The screened interaction item set is combined with the univariate analysis result to construct an extended feature set, thereby obtaining interaction feature data.

8. The method for constructing a risk assessment model for minimally invasive surgery for craniocerebral tumors according to claim 7, characterized in that: Step S35 includes the following steps: Step S351: setting a significance threshold according to the significance test result data to screen each risk factor, thereby obtaining a preliminary risk factor set; Step S352: classify the continuous variables in the preliminary risk factor set and convert them into categorical variables, thereby obtaining discretized risk factor data; determine the segmentation points of the discretized risk factor data and perform risk stratification, thereby obtaining risk stratification data; Step S353: Estimate the risk coefficient of each risk factor according to the risk stratification data, and convert it into a relative risk value, thereby obtaining relative risk data; Step S354: normalize the relative risk data and use the min-max standardization method to assign weights with a sum of weights equal to 1, thereby obtaining preliminary risk factor weight data.

9. The method for constructing a risk assessment model for minimally invasive surgery of craniocerebral tumors according to claim 8, characterized in that: Step S4 includes the following steps: Step S41: constructing an initial neural network structure including an input layer, a hidden layer, and an output layer according to the risk factor weight data, and training it using a standardized clinical data set to obtain a surgical risk assessment model; Step S42: Divide the standardized clinical data set into K subsets, wherein K-1 subsets are used for training and the remaining 1 subset is used for validation, and perform K-fold cross validation on the surgical risk assessment model to obtain cross validation result data; Step S43: performing a model evaluation on the surgical risk assessment model based on average accuracy, sensitivity and specificity according to the cross-validation result data, thereby obtaining model performance evaluation data; Step S44: optimizing the hyperparameters of the surgical risk assessment model based on the learning rate, the number of hidden layers, and the number of neurons in each layer according to the model performance evaluation data by a grid search method, and performing optimization training, thereby obtaining an optimized surgical risk assessment model; Step S45: performing overfitting control based on the early stopping method on the surgical risk assessment optimization model through L1 regularization and L2 regularization, thereby obtaining an anti-overfitting model; Step S46: pre-setting a portion of data as an independent data set according to the standardized clinical data set, and performing a performance test on the anti-overfitting model according to the independent data set, thereby obtaining model test result data; calculating the accuracy, precision, recall rate and F1 score of the anti-overfitting model according to the model test result data, thereby obtaining model evaluation index data; Step S47: Draw the ROC curve and the PR curve according to the model evaluation index data, and calculate the AUC value, so as to obtain the model performance curve data; use the Bootstrap method to estimate the confidence interval of the model performance curve data, so as to obtain the model performance confidence interval data; Step S48: Optimize the prediction performance of the anti-overfitting model according to the model performance confidence interval data, so as to obtain a surgical risk prediction model.

10. The method for constructing a risk assessment model for minimally invasive surgery of craniocerebral tumors according to claim 9, characterized in that: Step S5 includes the following steps: Step S51: Designing an API interface based on data input, model call and result output functions according to the surgical risk prediction model, thereby obtaining model API interface data; Step S52: docking the surgical risk prediction model with the medical system according to the model API interface data, and performing functional testing to obtain integrated model data; Step S53: Integrate the integrated model data with the hospital's existing electronic medical record system and image archiving and communication system to assist surgical decision-making.

Images