Neurosurgical disease diagnosis and prognosis prediction system and method based on machine learning

Abstract

The invention relates to the technical field of medical informatization, in particular to a neurosurgical disease diagnosis and prognosis prediction system and method based on machine learning, and the system comprises a data collection module, a processing module, a disease diagnosis module, a prognosis prediction module, a model evaluation module and a result output module. The system fuses clinical, image and survival data, and after preprocessing, screening and feature extraction, disease classification and prognosis prediction are performed by using a support vector machine and a random forest algorithm. Diagnosis and prognosis prediction are obtained by comparing algorithm results, and the model evaluation module dynamically adjusts the algorithm weight and optimizes the prediction accuracy. And finally, the system integrates diagnosis, prediction and evaluation results to generate a visual diagnosis report. According to the innovative design, multi-source data fusion, multi-model collaboration and dynamic optimization are achieved, the diagnosis accuracy and prediction reliability are remarkably improved, powerful support is provided for clinical decision making, and the method is expected to play an important role in improving the diagnosis and treatment level, improving patient prognosis and optimizing medical resource allocation.

Description

Technical Field

[0001] The present invention relates to the technical field of medical informatization, and particularly to a system and method for diagnosing and predicting the prognosis of neurosurgical diseases based on machine learning. Background Art

[0002] With the aging of the population and the change of lifestyle, the incidence of neurosurgical diseases shows an upward trend, causing great pressure on the quality of life of patients and social medical resources. Accurate disease diagnosis and prognosis prediction are crucial for formulating reasonable treatment plans and improving the prognosis of patients. Traditional diagnosis and prognosis prediction of neurosurgical diseases mainly rely on doctors' clinical experience and limited examination methods, with problems such as strong subjectivity, insufficient accuracy, and low efficiency.

[0003] In recent years, with the rapid development of medical imaging technology and artificial intelligence, auxiliary diagnosis systems based on machine learning have been widely used in the medical field. However, existing neurosurgical disease diagnosis and prognosis prediction systems still have many deficiencies. First, most systems only focus on a single data source, such as only using imaging data or clinical data, ignoring the information gain brought by multi-source data fusion. Second, existing systems usually adopt a single machine learning model, which is difficult to adapt to the complexity and diversity of neurosurgical diseases. Third, many systems lack a dynamic optimization mechanism and cannot adjust the model in a timely manner according to new data and clinical feedback, resulting in a decline in prediction performance over time. In addition, existing systems often directly present the prediction results to doctors, lacking the intervention of expert knowledge and an auxiliary decision-making mechanism, which limits the application value of the system in actual clinical practice. Summary of the Invention

[0004] In view of the above problems, there is an urgent need to develop a neurosurgical disease diagnosis and prognosis prediction system that can integrate multi-source data, integrate multiple machine learning models, have the ability of dynamic optimization, and can be organically combined with expert knowledge. The present invention is an innovative solution proposed in response to this need.

[0005] The present invention provides a system and method for diagnosing and predicting the prognosis of neurosurgical diseases based on machine learning, including:

[0006] A data acquisition module, configured to:

[0007] Collect neurosurgical clinical data, image data, and patient survival data;

[0008] A data processing module, communicatively connected to the data acquisition module, configured to:

[0009] Receive the neurosurgical clinical data, image data, and patient survival data sent by the data acquisition module;

[0010] Preprocess, screen, and extract features from the neurosurgical clinical data and patient survival data;

[0011] A disease diagnosis module, communicatively connected to the data processing module, for:

[0012] Based on the data processed by the data processing module, perform classification prediction through the support vector machine algorithm and the random forest algorithm;

[0013] Compare the classification prediction results of the support vector machine algorithm and the random forest algorithm to obtain the disease diagnosis result;

[0014] A prognosis prediction module, communicatively connected to the data processing module and the disease diagnosis module, for:

[0015] Based on the patient survival data processed by the data processing module, perform classification prediction through the support vector machine algorithm and the random forest algorithm;

[0016] Compare the classification prediction results of the support vector machine algorithm and the random forest algorithm to obtain the prognosis prediction result;

[0017] A model evaluation module, communicatively connected to the disease diagnosis module and the prognosis prediction module, for:

[0018] Evaluate the prediction results of the disease diagnosis module and the prognosis prediction module;

[0019] Dynamically adjust the weights of the support vector machine algorithm and the random forest algorithm based on the evaluation results;

[0020] A result output module, communicatively connected to the disease diagnosis module, the prognosis prediction module, and the model evaluation module, for:

[0021] Receive and integrate the disease diagnosis result, the prognosis prediction result, and the model evaluation result;

[0022] Generate a diagnostic report and a visual display.

[0023] Preferably, the data processing module includes:

[0024] A data cleaning unit for performing outlier processing and missing value processing on the neurosurgical clinical data and patient survival data;

[0025] A feature selection unit, communicatively connected to the data cleaning unit, for selecting highly relevant features from the neurosurgical clinical data and patient survival data;

[0026] A feature extraction unit, communicatively connected to the feature selection unit, for:

[0027] Extract high-order clinical features from the neurosurgical clinical data;

[0028] Extract image features, statistical features, and texture features from the image data, where the texture features include GLCM and Gabor texture features.

[0029] Preferably, the disease diagnosis module includes:

[0030] A support vector machine model training unit for training the support vector machine model, learning the support vector machine model through training data, and adjusting the model parameters;

[0031] A random forest model training unit for training the random forest model, learning the random forest model through training data, and adjusting the model parameters;

[0032] A model fusion unit, communicatively connected to the support vector machine model training unit and the random forest model training unit, for:

[0033] Receiving the prediction results of the support vector machine model and the random forest model;

[0034] Based on a predefined weight strategy, fusing the prediction results to obtain a final disease diagnosis result.

[0035] Preferably, the prognosis prediction module includes:

[0036] A survival data processing unit for extracting features from the patient survival data, extracting features including operation time, postoperative complications, postoperative high fever, and physical condition at discharge;

[0037] A prognosis model training unit, communicatively connected to the survival data processing unit, for:

[0038] Based on the features extracted by the survival data processing unit, training support vector machine models and random forest models;

[0039] Optimizing the parameters of the trained models;

[0040] A prognosis prediction unit, communicatively connected to the prognosis model training unit, for:

[0041] Using the trained support vector machine models and random forest models to perform prognosis prediction on new patient data;

[0042] Combining the prediction results of the two models to obtain a final prognosis prediction result.

[0043] Preferably, the model evaluation module includes:

[0044] An accuracy evaluation unit for:

[0045] Divide the data collected by the data acquisition module into training data and test data sets;

[0046] Evaluate the accuracy of the classification prediction results through the test data set;

[0047] The model tuning unit, communicatively connected to the accuracy evaluation unit, is configured to:

[0048] When the accuracy of the test data set is lower than the set threshold, trigger the retraining of the model;

[0049] When the accuracy of the test data set is higher than the set threshold, confirm that the current model is available;

[0050] The weight adjustment unit, communicatively connected to the model tuning unit, is configured to:

[0051] Dynamically adjust the weights of the support vector machine model and the random forest model in the result fusion based on the accuracy evaluation results.

[0052] Preferably, it further includes:

[0053] The data visualization module, communicatively connected to the result output module, is configured to:

[0054] Receive the diagnostic report generated by the result output module;

[0055] Convert the diagnostic report into visual charts, including drawing visual charts, graphs, heat maps, and artifact maps.

[0056] Preferably, the data acquisition module includes:

[0057] The clinical data acquisition unit is used to collect clinical information such as the gender, age, course of disease, initial symptoms, preoperative complications, preoperative test results, pathological results, postoperative test results, postoperative complications, and disease prognosis of the patient;

[0058] The imaging data acquisition unit is used to collect the cranial CT and MRI results of the patient and convert them into the NIFTI format as a neuroimaging analysis atlas;

[0059] The survival data acquisition unit is used to collect survival-related data such as the operation time of the patient, whether severe complications occur after the operation, whether high fever occurs after the operation, and the physical condition at discharge.

[0060] Preferably, it further includes:

[0061] The expert-assisted decision-making module, communicatively connected to the result output module, is configured to:

[0062] Receive the diagnostic report and visual display generated by the result output module;

[0063] Provide an interface for doctors to manually review and adjust the prediction results;

[0064] Based on the doctor's feedback, record and analyze the manual adjustment situation to provide a basis for subsequent model optimization.

[0065] Preferably, it further includes:

[0066] Cloud deployment module, used for:

[0067] Deploy the system on a cloud server;

[0068] Provide different levels of data access and management permissions for administrators and physicians;

[0069] Realize remote access and multi-device synchronization of the system.

[0070] A method for diagnosing and predicting the prognosis of neurosurgical diseases based on machine learning, using the above system, includes the following steps:

[0071] S1. Collect patient data, including neurosurgical clinical data, image data, and patient survival data;

[0072] S2. Preprocess the neurosurgical clinical data and patient survival data, including:

[0073] Use the K-MEANS clustering algorithm for outlier and missing value processing;

[0074] Match the preprocessed clinical data with the image data;

[0075] S3. Extract features from the preprocessed data, including:

[0076] Extract high-order clinical features from the clinical data;

[0077] Extract image features, statistical features, and texture features from the image data;

[0078] S4. Use the support vector machine algorithm and the random forest algorithm to train the extracted features to obtain a disease diagnosis model;

[0079] S5. Use the disease diagnosis model to predict the disease diagnosis of new patient data;

[0080] S6. Based on the patient survival data, use the support vector machine algorithm and the random forest algorithm to train a prognosis prediction model;

[0081] S7. Use the prognosis prediction model to predict the prognosis of new patient data;

[0082] S8. Comprehensively evaluate the disease diagnosis results and prognosis prediction results;

[0083] S9. Dynamically adjust the weights of the support vector machine algorithm and the random forest algorithm in the result fusion based on the evaluation results.

[0084] S10. Generate a diagnostic report and a visual display, and provide an expert-assisted decision-making interface.

[0085] The system and method for diagnosing and predicting the prognosis of neurosurgical diseases based on machine learning provided by the present invention realizes the full-process intelligence from data collection, processing, analysis to result output through the collaborative work of multiple modules. The innovation and beneficial effects of this system are mainly reflected in the following aspects:

[0086] First of all, the present invention realizes the deep fusion of multi-source heterogeneous data. By simultaneously collecting and processing clinical data, imaging data and patient survival data, the system can comprehensively capture the health status and disease characteristics of patients. This multi-dimensional data integration not only improves the accuracy of diagnosis and prediction, but also provides the possibility to discover potential disease-related factors.

[0087] Secondly, the present invention adopts a dual-model collaborative prediction strategy. By combining two machine learning models with different characteristics, namely the support vector machine and the random forest, the system can better adapt to the complexity and diversity of neurosurgical diseases. This integrated learning method not only improves the stability and robustness of prediction, but also can overcome the limitations of a single model to a certain extent.

[0088] Furthermore, the present invention introduces a dynamic model optimization mechanism. The system can evaluate the model performance in real time according to the newly added data and the prediction effect, and trigger model retraining when necessary. This adaptive learning ability ensures the continuous optimization of the system performance, enabling it to adapt to the dynamic changes of disease patterns and the progress of medical technologies.

[0089] In addition, the present invention designs an expert-assisted decision-making module, which organically combines the prediction results of artificial intelligence with the professional judgment of doctors. This "AI + expert experience" decision-making mode not only improves the interpretability and credibility of the prediction results, but also provides more comprehensive decision-making support for clinicians.

[0090] Finally, the cloud deployment design of the present invention greatly improves the accessibility and scalability of the system. Through the microservices architecture and the load balancing strategy, the system can efficiently process a large number of concurrent requests, laying a foundation for multi-center and large-sample clinical applications. At the same time, strict access control and data encryption mechanisms ensure patient privacy and data security.

[0091] Generally speaking, through innovative designs such as multi-source data fusion, multi-model collaboration, dynamic optimization, and expert-assisted decision-making, the present invention constructs a comprehensive, accurate, and reliable diagnosis and prognosis prediction system for neurosurgical diseases. This system can not only significantly improve the diagnostic accuracy and prediction reliability but also provide strong support for clinical decision-making, and is expected to play an important role in improving the level of neurosurgical diagnosis and treatment, improving the prognosis of patients, and optimizing the allocation of medical resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] Figure 1 is the overall logic block diagram of the system of the present invention;

[0093] Figure 2 is the logic block diagram of the data acquisition module of the present invention;

[0094] Figure 3 is the logic block diagram of the data processing module of the present invention;

[0095] Figure 4 is the logic block diagram of the disease diagnosis module of the present invention;

[0096] Figure 5 is the logic block diagram of the prognosis prediction module of the present invention;

[0097] Figure 6 is the logic block diagram of the model evaluation module of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0098] Please refer to the attached Figures 1-6 , the present invention provides a system and method for diagnosing and predicting the prognosis of neurosurgical diseases based on machine learning. The system includes multiple functional modules, and through data acquisition, processing, analysis, and prediction, it realizes the intelligent diagnosis and prognosis evaluation of neurosurgical diseases. The following will describe the present invention in detail in combination with specific embodiments.

[0099] The system of the present invention includes a data acquisition module 1, a data processing module 2, a disease diagnosis module 3, a prognosis prediction module 4, a model evaluation module 5, and a result output module 6. These modules are closely connected through data streams and control streams, and jointly complete the entire process from data input to result output.

[0100] The data acquisition module 1 is used to collect neurosurgical clinical data, image data, and patient survival data. Preferably, the clinical data includes the basic information of the patient (such as age, gender, etc.), symptoms, signs, laboratory test results, etc. The image data mainly refers to the CT and MRI scan results. The patient survival data includes information such as surgery, treatment, and follow-up. This module can obtain data through multiple channels such as the hospital information system (HIS) and the radiology information system (RIS), and perform preliminary format unification and storage.

[0101] The data processing module 2 is communicatively connected to the data acquisition module 1, and is used for receiving the collected raw data and performing preprocessing, screening, and feature extraction. Specifically, this module first cleans the data, including handling missing values, outliers, etc. The present invention uses an improved K-means clustering algorithm for outlier detection, and its core idea is to identify samples far from the cluster center after clustering the data. The algorithm is as follows:

[0102]

[0103] J is the objective function, k is the number of clusters, and n is the number of samples. is the i-th sample in the j-th class, and c j is the cluster center of the j-th class.

[0104] For missing values, the present invention uses mean / median filling or multiple imputation methods according to the data type. For example, for continuous variables, the missing values can be estimated using the following formula:

[0105]

[0106] where, is the estimated missing value, and are the means of the known x and y respectively, and O i is the set of known observations paired with the i-th observation.

[0107] After data cleaning, the data processing module 2 performs feature extraction. For clinical data, high-order features such as symptom duration and abnormal degree of laboratory indicators are mainly extracted. For image data, in addition to conventional shape and texture features, the present invention also introduces deep learning methods to extract high-level semantic features. For example, a pre-trained convolutional neural network is used to extract features, and the feature map of the l-th layer can be expressed as:

[0108] h l = f(W l * h l-1 + b l ),

[0109] h l is the feature map of the l-th layer, W l and b l are the weights and biases of this layer respectively, f is the activation function, and * represents the convolution operation.

[0110] After feature extraction is completed, the data processing module 2 also performs feature selection to reduce the data dimension and improve the model efficiency. The present invention uses Lasso regression based on L1 regularization for feature selection, and its objective function is:

[0111]

[0112] Among them, y is the target variable, X is the feature matrix, β is the regression coefficient, and λ is the regularization parameter. By adjusting λ, the strictness of feature selection can be controlled.

[0113] The disease diagnosis module 3 is communicatively connected to the data processing module 2, receives the processed data, and performs classification prediction using the support vector machine (SVM) and random forest (RF) algorithms. The core of SVM is to find the maximum margin hyperplane, and its decision function is:

[0114]

[0115] Among them, α i is the Lagrange multiplier, y i is the class label, K(x i , x) is the kernel function, and b is the bias term. The present invention adopts the radial basis function (RBF) kernel:

[0116] K(x i , x j ) = exp(-γ||x i -x j || 2 ),

[0117] Among them, γ is the kernel parameter, which controls the curvature of the decision boundary.

[0118] Random forest realizes classification by constructing multiple decision trees and voting. In the construction process of each tree, the criterion for selecting the best splitting feature at node m is to maximize the information gain:

[0119]

[0120] Among them, H is the entropy function, D m is the data set at node m, f is the feature, is the subset when the feature f takes the value v.

[0121] The disease diagnosis module 3 will compare the prediction results of SVM and RF, and adopt the weighted voting method to obtain the final diagnosis result. The weights can be dynamically adjusted according to the performance of each model on the validation set.

[0122] Through innovations such as multi-source data fusion, multi-model integration, and dynamic weight adjustment, the system of the present invention effectively improves the accuracy and reliability of neurosurgical disease diagnosis. The modular design of the system also enables each part to be independently optimized and upgraded, with good scalability. In practical applications, the system can be used as an intelligent assistant tool for doctors, providing objective and comprehensive diagnostic suggestions, which helps to improve the diagnostic efficiency and accuracy.

[0123] The prognosis prediction module 4 of the present invention. This module is communicatively connected to the data processing module 2 and the disease diagnosis module 3, and is mainly responsible for predicting the prognosis of patients. Prognosis prediction is an important part of neurosurgical clinical practice and can provide important references for doctors to formulate treatment plans and patient management.

[0124] The prognosis prediction module 4 of the present invention adopts a machine learning method based on patient survival data. Specifically, this module first extracts features from patient survival data, focusing on key indicators such as operation time, postoperative complications, postoperative high fever, and physical condition at discharge. These indicators are usually closely related to the prognosis of neurosurgical patients. For example, too long operation time may increase the risk of postoperative complications; postoperative high fever may indicate potential infections or inflammatory reactions; the physical condition at discharge directly reflects the patient's recovery.

[0125] Based on the feature extraction, the prognosis prediction module 4 uses two algorithms, support vector machine (SVM) and random forest (RF), to construct a prediction model. These two algorithms each have their own advantages in the prognosis prediction task: SVM is good at dealing with high-dimensional data and non-linear relationships, while RF can effectively handle the interaction between features and provide a ranking of feature importance.

[0126] For the SVM model, the present invention adopts the kernel trick to handle non-linear relationships. The decision function for prognosis prediction can be expressed as:

[0127]

[0128] x is the input feature vector, x i is the support vector, y i is the corresponding label, α i is the Lagrange multiplier, K(x i , x) is the kernel function, which is used to handle non-linear relationships, and b is the bias term. The present invention preferably adopts the radial basis function (RBF) kernel:

[0129] K(x i , x j ) = exp(-γ||x i -x j || 2 ),

[0130] where γ is the kernel parameter, which controls the complexity of the decision boundary. In practical applications, the value of γ usually ranges from 0.01 to 1, and can be optimized by methods such as cross-validation.

[0131] For the random forest model, the present invention constructs multiple decision trees, and each tree is trained on a randomly selected subset of features. When splitting nodes, the Gini index is used as the impurity measure:

[0132]

[0133] Among them, D is the data set, K is the number of categories, and p k is the proportion of samples in the k-th category. Preferably, the random forest model of the present invention includes 100 - 500 decision trees, and the maximum depth of each tree is set to the square root of the number of data features to balance the complexity and generalization ability of the model.

[0134] The prognosis prediction module 4 will comprehensively consider the prediction results of the SVM and RF models. The present invention uses weighted averaging to fuse the output probabilities of the two models:

[0135] P final = w SVM ·P SVM + w RF ·P RF ,

[0136] Among them, P final is the final prediction probability, P SVM and P RF are the prediction probabilities of the SVM and RF models respectively, and w SVM and w RF are the corresponding weights. Initially, the two weights can be set to 0.5, and can be dynamically adjusted later according to the performance of the models on the validation set.

[0137] The model evaluation module 5 of the present invention. This module is communicatively connected to the disease diagnosis module 3 and the prognosis prediction module 4, and is mainly responsible for evaluating the prediction results and dynamically adjusting the model performance. Model evaluation is a key link to ensure the prediction accuracy and reliability of the system.

[0138] The model evaluation module 5 of the present invention first divides the data collected by the data collection module 1 into training data and test data sets. Preferably, the stratified sampling method is used for data division to ensure that the test set can represent the overall data distribution. The proportion of the test set is usually set to 20% - 30%, and can be adjusted specifically according to the size of the data volume.

[0139] For classification tasks (such as disease diagnosis), the model evaluation module 5 mainly uses indicators such as accuracy, precision, recall, and F1-score. The calculation formulas for these indicators are as follows:

[0140] Accuracy:

[0141] Precision:

[0142] Recall:

[0143] F1-score:

[0144] Among them, TP is the true positive, TN is the true negative, FP is the false positive, and FN is the false negative.

[0145] For tasks involving probability estimation such as prognosis prediction, the model evaluation module 5 also calculates metrics such as log loss and AUC-ROC (area under the receiver operating characteristic curve). The calculation formula for log loss is:

[0146]

[0147] Among them, N is the number of samples, y i is the true label, and p i is the predicted probability.

[0148] When the accuracy of the test data set is lower than the set threshold, the model evaluation module 5 will trigger the model to be retrained. The present invention sets different thresholds for different tasks: for disease diagnosis, the accuracy threshold is set to 0.85; for prognosis prediction, the AUC-ROC threshold is set to 0.75. These threshold settings are based on the requirements for prediction accuracy in clinical practice and also consider the feasibility of the model.

[0149] The model evaluation module 5 is also responsible for dynamically adjusting the weights of the support vector machine model and the random forest model in the result fusion. The adjustment strategy is based on the performance of each model in the recent period of time and can be expressed as:

[0150]

[0151] Among them, w i is the weight of model i, performance i is the performance metric of model i (such as accuracy or AUC-ROC), and T is the temperature parameter that controls the smoothness of the weight distribution. A smaller T value will make the weight distribution more "sharp", that is, more inclined to the model with the best performance. In the preferred embodiment of the present invention, the T value is set to 0.1.

[0152] The present invention also includes a data visualization module, which is communicatively connected to the result output module 6. This module can convert the diagnostic report into intuitive visualization charts, including but not limited to bar charts, line charts, heat maps, and pseudocolor maps, etc. For example, for the prognosis prediction of cancer patients, the Kaplan-Meier survival curve can be used to intuitively show the change of the survival probability of different risk groups over time. The estimation formula for the survival function is:

[0153]

[0154] Among them, t iObserved survival time, d i Time t i Number of events, n i Time t i Size of the risk set

[0155] Through these visualization means, the system of the present invention can more intuitively display the diagnostic and prediction results to doctors, helping doctors quickly understand the key information and improving the efficiency of clinical decision-making.

[0156] The present invention also includes an expert-assisted decision-making module, which is communicatively connected to the result output module 6 and aims to combine the prediction results of artificial intelligence with the professional judgment of doctors to improve the accuracy and reliability of diagnosis and prognosis prediction.

[0157] The expert-assisted decision-making module first receives the diagnostic report and visualization display generated by the result output module 6. This information includes disease diagnosis results, prognosis prediction results, and relevant statistical data and charts. The module provides an interactive interface that allows doctors to review and adjust the prediction results.

[0158] In practical applications, doctors can view the prediction results of the system through this interface and confirm, modify, or supplement the results based on their clinical experience and understanding of the specific situation of the patient. For example, if the system predicts that a patient has an 80% probability of having a certain neurological disease, but the doctor believes based on other clinical clues that this probability may be overestimated, the doctor can adjust this prediction probability through the interface.

[0159] Preferably, the expert-assisted decision-making module of the present invention also includes a feedback mechanism. When doctors adjust the prediction results, the system records these adjustments and analyzes them. This analysis can help identify possible biases or deficiencies in the model and provide valuable information for subsequent model optimization. For example, if it is found that the prediction accuracy of the system in a certain group of patients is significantly lower than the overall level, this may indicate that more training data needs to be collected for this group of patients or the model structure needs to be adjusted.

[0160] In one embodiment of the present invention, the expert-assisted decision-making module also integrates a case library. This case library contains a large number of historical cases, and each case includes the patient's clinical data, imaging materials, diagnosis results, treatment plans, and prognosis. When the system diagnoses and predicts the prognosis of a new patient, the expert-assisted decision-making module will automatically retrieve historical cases similar to the current patient from the case library and present them to the doctor for reference.

[0161] The calculation of case similarity uses the weighted Euclidean distance:

[0162]

[0163] Among them, x and y respectively represent the feature vectors of the current patient and historical cases, and w i is the weight of the i-th feature, and n is the total number of features. The feature weights can be set by expert experience or determined by machine learning methods such as the feature importance of random forests.

[0164] By referring to similar cases, doctors can more comprehensively evaluate the situation of the current patient and make more accurate diagnoses and prognostic judgments. This comprehensive decision-making model of "AI + expert experience + historical cases" gives full play to the computing power of artificial intelligence and the clinical experience of doctors, and is expected to significantly improve the diagnosis and treatment level of neurosurgical diseases.

[0165] The cloud deployment module of the present invention. The design of this module aims to improve the accessibility, scalability, and data security of the system. In the preferred embodiment of the present invention, the entire neurosurgical disease diagnosis and prognosis prediction system is deployed on a cloud server.

[0166] The cloud deployment module is first responsible for deploying each functional module of the system to the cloud server. This includes the data acquisition module 1, the data processing module 2, the disease diagnosis module 3, the prognosis prediction module 4, the model evaluation module 5, and the result output module 6, etc. By using containerization technology (such as Docker), each module can be encapsulated into an independent container, which is convenient for deployment and management.

[0167] Preferably, the present invention adopts a microservices architecture, splitting different functions of the system into multiple independent services. This architecture has the following advantages:

[0168] 1. High scalability: Each service can be independently scaled according to the load situation.

[0169] 2. Fault isolation: The failure of a single service will not affect the operation of the entire system.

[0170] 3. Technical stack flexibility: Different services can be implemented using the most suitable technical stack.

[0171] The cloud deployment module is also responsible for implementing the load balancing of the system. When the system faces high-concurrency requests, the load balancer can distribute the requests to multiple service instances, improving the response speed and stability of the system. The present invention adopts a load balancing strategy that combines round robin and least connections. The round robin strategy can be expressed as:

[0172] ServerIndex = (CurentIndex + 1) mod N,

[0173] where N is the total number of servers, and CurrentIndex is the current server index.

[0174] The least connection strategy selects the server with the fewest current active connections:

[0175] SelectedServer = argmin i Connections(Server i )

[0176] where Connections(Server i ) represents the current active connection count of server i.

[0177] To ensure data security and privacy, the cloud deployment module implements a strict access control mechanism. The system provides different levels of data access and management permissions for different roles (such as administrators, doctors, researchers, etc.). Permission control adopts the Role-Based Access Control (RBAC) model. In the RBAC model, permission assignment can be expressed as:

[0178] Permissions(USer) = ∪ Role∈Roles (User)Permissions(Role),

[0179] where Roles(User) represents the set of roles owned by the user, and Permissions(Role) represents the set of permissions owned by a certain role.

[0180] In addition, the cloud deployment module also implements encrypted data storage and transmission. All sensitive data is encrypted during storage and the SSL / TLS protocol is used for encryption during network transmission. This ensures that even in the event of a data breach, unauthorized personnel cannot access patients' private information.

[0181] Finally, the present invention also provides a method for diagnosing and predicting the prognosis of neurosurgical diseases based on machine learning. This method closely corresponds to the aforementioned system and includes steps such as data collection, data preprocessing, feature extraction, model training, disease diagnosis, prognosis prediction, result evaluation, and visualization.

[0182] In the data collection step, this method simultaneously obtains the patient's clinical data, imaging data, and survival data. Clinical data includes the patient's basic information, symptoms, signs, laboratory test results, etc.; imaging data is mainly the CT and MRI scan results; survival data includes surgical conditions, complications, follow-up results, etc.

[0183] The data preprocessing step uses the K - MEANS clustering algorithm for outlier handling and missing value imputation. The objective function of the K - MEANS algorithm is:

[0184]

[0185] where k is the number of clusters, n is the number of samples, is the i - th sample in the j - th class, and c j is the cluster center of the j - th class.

[0186] The feature extraction step uses different methods for different types of data. For clinical data, mainly high - order statistical features are extracted; for imaging data, deep learning models are used to extract high - level semantic features. For example, a pre - trained convolutional neural network is used to extract features, and the feature map of the l - th layer can be expressed as:

[0187] h l = f(W l * h l-1 + b l ),

[0188] where h l is the feature map of the l - th layer, W l and b l are the weights and biases of this layer respectively, f is the activation function, and * represents the convolution operation.

[0189] In the model training step, this method trains both the Support Vector Machine (SVM) and Random Forest (RF) models simultaneously. The decision function of the SVM is:

[0190]

[0191] where α i is the Lagrange multiplier, y i is the class label, K(x i , x) is the kernel function, and b is the bias term.

[0192] The random forest realizes classification by constructing multiple decision trees and voting. The information gain is used as the evaluation criterion when splitting nodes:

[0193]

[0194] where H is the entropy function, D m is the data set at node m, f is the feature, is the subset when the feature f takes the value v.

[0195] In the steps of disease diagnosis and prognosis prediction, this method comprehensively considers the prediction results of SVM and RF, and uses the method of weighted voting to obtain the final result. The weights can be dynamically adjusted according to the performance of each model on the validation set.

[0196] In the result evaluation step, multiple evaluation metrics are used, including accuracy, precision, recall, F1-score, etc. At the same time, this method also implements a dynamic update mechanism for the model. When the model performance drops below the preset threshold, the retraining process will be automatically triggered.

[0197] Finally, this method presents the prediction results in an intuitive visual way, including various statistical charts and pseudo-color enhancement displays of medical images, etc. These visual results are combined with the expert system to provide comprehensive and accurate reference information for doctors' clinical decisions.

[0198] Through the above steps, the method of the present invention realizes the intelligent diagnosis and accurate prognosis prediction of neurosurgical diseases, providing strong technical support for improving the clinical diagnosis and treatment level of neurosurgery.

[0199] It should be noted that the above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A neurosurgical disease diagnosis and prognosis prediction system based on machine learning, characterized in that: include: Data acquisition module for: Collect neurosurgery clinical data, image data, and patient survival data; A data processing module is connected to the data acquisition module for: Receiving neurosurgery clinical data, image data and patient survival data sent by the data acquisition module; Preprocessing, screening and feature extraction of the neurosurgery clinical data and patient survival data; A disease diagnosis module, which is in communication with the data processing module, is used to: Based on the data processed by the data processing module, classification prediction is performed using a support vector machine algorithm and a random forest algorithm; Comparing the classification prediction results of the support vector machine algorithm and the random forest algorithm to obtain a disease diagnosis result; The prognosis prediction module is in communication with the data processing module and the disease diagnosis module and is used to: Based on the patient survival data processed by the data processing module, classification prediction is performed using a support vector machine algorithm and a random forest algorithm; Comparing the classification prediction results of the support vector machine algorithm and the random forest algorithm to obtain a prognosis prediction result; The model evaluation module is in communication with the disease diagnosis module and the prognosis prediction module and is used to: Evaluating the prediction results of the disease diagnosis module and the prognosis prediction module; Dynamically adjust the weights of the support vector machine algorithm and the random forest algorithm based on the evaluation results; The result output module is connected to the disease diagnosis module, the prognosis prediction module and the model evaluation module for: Receiving and integrating the disease diagnosis results, prognosis prediction results and model evaluation results; Generate diagnostic reports and visualizations.

2. The system according to claim 1, characterized in that The data processing module comprises: A data cleaning unit, used for performing outlier processing and missing value processing on the neurosurgery clinical data and patient survival data; a feature selection unit, in communication with the data cleaning unit, for selecting highly correlated features from the neurosurgery clinical data and the patient survival data; A feature extraction unit is communicatively connected to the feature selection unit and is used to: extracting high-level clinical features from the neurosurgery clinical data; Image features, statistical features and texture features are extracted from the image data, wherein the texture features include GLCM and Gabor texture features.

3. The system according to claim 1, characterized in that The disease diagnosis module comprises: A support vector machine model training unit is used to train the support vector machine model, learn the support vector machine model through training data, and adjust the model parameters; A random forest model training unit is used to train the random forest model, learn the random forest model through training data, and adjust the model parameters; A model fusion unit is communicatively connected with the support vector machine model training unit and the random forest model training unit, and is used for: Receiving prediction results of the support vector machine model and the random forest model; Based on a predefined weight strategy, the prediction results are fused to obtain the final disease diagnosis result.

4. The system according to claim 1, characterized in that The prognosis prediction module comprises: A survival data processing unit, used for extracting features from the patient's survival data, including operation time, postoperative complications, postoperative high fever, and physical condition at discharge; The prognosis model training unit is communicatively connected to the survival data processing unit and is used to: Based on the features extracted by the survival data processing unit, training a support vector machine model and a random forest model; Optimize the parameters of the trained model; A prognosis prediction unit, which is in communication with the prognosis model training unit, is used to: Use the trained support vector machine model and random forest model to predict the prognosis of new patient data; The prediction results of the two models are combined to obtain the final prognosis prediction result.

5. The system according to claim 1, characterized in that The model evaluation module includes: Accuracy evaluation unit, used to: Divide the data collected by the data collection module into training data and test data sets; Evaluate the accuracy of classification prediction results through the test data set; A model tuning unit is communicatively connected to the accuracy evaluation unit and is used to: When the accuracy of the test data set is lower than the set threshold, the model is triggered to retrain; When the accuracy of the test data set is higher than the set threshold, it is confirmed that the current model is applicable; A weight adjustment unit is connected to the model optimization unit for: Based on the accuracy evaluation results, the weights of the support vector machine model and the random forest model in the result fusion are dynamically adjusted.

6. The system according to claim 1, characterized in that Also includes: A data visualization module is connected to the result output module for: Receiving a diagnostic report generated by the result output module; The diagnostic report is converted into a visual chart, including drawing a visual chart, a graph, a heat map and an artifact map.

7. The system according to claim 1, characterized in that The data acquisition module comprises: Clinical data collection unit, used to collect clinical information such as gender, age, course of disease, first symptom, preoperative complications, preoperative laboratory test results, pathological results, postoperative laboratory test results, postoperative complications and disease outcomes of patients; The imaging data acquisition unit is used to collect the patient's cranial CT and MRI results and convert them into N IFTI format as a neuroimaging analysis atlas; The survival data collection unit is used to collect survival-related data such as the patient's operation time, whether serious complications occurred after the operation, whether high fever occurred after the operation, and physical condition at discharge.

8. The system according to claim 1, characterized in that Also includes: The expert decision-making support module is connected to the result output module for: Receiving the diagnostic report and visual display generated by the result output module; Provide an interface for doctors to manually review and adjust prediction results; Based on the doctors’ feedback, manual adjustments are recorded and analyzed to provide a basis for subsequent model optimization.

9. The system according to claim 1, characterized in that Also includes: Cloud deployment module for: Deploy the system on a cloud server; Provide different levels of data access and management permissions for administrators and physicians; Realize remote access and multi-device synchronization of the system.

10. A method for diagnosing and predicting prognosis of neurosurgical diseases based on machine learning, using the system according to any one of claims 1 to 9, characterized in that: The following steps are involved: S1. Collect patient data, including neurosurgery clinical data, image data, and patient survival data; S2. Preprocessing the neurosurgery clinical data and patient survival data, including: Use K-MEANS clustering algorithm to handle outliers and missing values; Matching the preprocessed clinical data with the image data; S3. Extract features from the preprocessed data, including: Extract high-level clinical features from clinical data; Extract image features, statistical features and texture features from image data; S4. Use support vector machine algorithm and random forest algorithm to train the extracted features to obtain a disease diagnosis model; S5. Use the disease diagnosis model to predict disease diagnosis for new patient data; S6. Based on patient survival data, the prognosis prediction model was trained using the support vector machine algorithm and the random forest algorithm; S7. Using the prognosis prediction model to perform prognosis prediction on new patient data; S8. Comprehensively evaluate the disease diagnosis results and prognosis prediction results; S9. Based on the evaluation results, dynamically adjust the weights of the support vector machine algorithm and the random forest algorithm in the result fusion; S10. Generate diagnostic reports and visual displays, and provide expert-assisted decision-making interfaces.

Images