Method and system for predicting hematoma expansion event of stroke patient

Abstract

Disclosed in the present invention are a method and system for predicting a hematoma expansion event of a stroke patient. The method comprises: collecting patient data information, and performing data preprocessing; using an XG-Boost model to calculate the SHAP average value and the contribution degrees of different types of data, and extracting a feature quantity with the highest contribution degree from each type of data; performing multivariate linear regression analysis to obtain the weight of the effect of each type of data on hematoma expansion, and weighting all the feature quantities on the basis of the weight to obtain weighted feature quantities; using an extreme learning machine to predict a hematoma expansion probability; and using a northern goshawk optimization algorithm to search predicted values for the boundary of occurrence of a hematoma expansion event, and verifying the prediction accuracy. The present invention can accurately analyze and predict the expansion risk of hematoma after hemorrhagic stroke, which is crucial to early diagnosis and timely treatment.

Description

A method and system for predicting hematoma expansion events in stroke patients Technical Field

[0001] The present invention relates to the field of clinical intelligent diagnosis and treatment technology, specifically a method and system for predicting hematoma expansion events in stroke patients. Background Art

[0002] Hemorrhagic stroke, defined as cerebral hemorrhage caused by non-traumatic intraparenchymal vascular rupture, is an acute cerebrovascular disease characterized by rapid onset and progression, often accompanied by severe neurological impairment. Due to its high mortality rate, hemorrhagic stroke imposes a significant health and economic burden on patients and their families. Following a hemorrhagic stroke, hematoma expansion is considered a key factor in poor prognosis. The expansion of the hematoma may lead to increased intracranial pressure, further deteriorating neurological function, and even be life-threatening. Furthermore, edema surrounding the hematoma, a hallmark of secondary injury after cerebral hemorrhage, has a significant impact on the patient's neurological impairment.

[0003] Advances in medical imaging technology have provided effective, noninvasive means of dynamically monitoring brain tissue damage and changes after hemorrhagic stroke. Furthermore, the rapid development of artificial intelligence in the medical field has provided new opportunities for processing and analyzing large amounts of medical data. This has made it feasible to accurately predict the prognosis of hemorrhagic stroke patients by combining imaging features with patient clinical information and treatment plans.

[0004] In response to the occurrence and development of hematoma expansion and perihematomal edema after hemorrhagic stroke, the present invention proposes a data-driven method for early identification and prediction to improve the patient's prognosis and quality of life. The present invention preprocesses and analyzes the patient's clinical information, the volume, location, shape and grayscale information of the hematoma and edema, and applies methods such as the XG-Boost model, SHAP value, multivariate linear regression and extreme learning machine (ELM) to establish a mathematical model for the impact of hematoma and edema information on the patient's blood vessel expansion. The model can effectively predict the risk of hematoma expansion and is of great significance for clinical treatment decisions and patient management. In addition, the model combines the Northern Goshawk Optimization Algorithm (NGO) to improve the prediction accuracy, further enhancing the practicality and accuracy of the model.

[0005] Summary of the Invention

[0006] In view of the above-mentioned problems, the present invention is proposed.

[0007] Therefore, the technical problem addressed by this invention is the early identification and accurate prediction of hematoma expansion and perihematomal edema in patients with hemorrhagic stroke. Specifically, the technical problem involves effectively integrating and analyzing patient clinical information, as well as hematoma and edema volume, location, shape, and grayscale information, to develop a reliable model for predicting the risk of hematoma expansion after hemorrhagic stroke.

[0008] To solve the above technical problems, the present invention provides the following technical solutions: a method for predicting hematoma expansion events in stroke patients, comprising: collecting patient data information and performing data preprocessing;

[0009] Use the XG-Boost model to calculate the SHAP average and contribution of different types of data, and extract the feature with the highest contribution in each type of data;

[0010] Perform multiple linear regression analysis to obtain the weight of each type of data's influence on hematoma expansion, and then weight all feature quantities according to the weight to obtain a feature quantity weighted by the SHAP average contribution.

[0011] Use extreme learning machine to predict the probability of hematoma expansion;

[0012] The Northern Goshawk optimization algorithm was used to find the boundary of hematoma expansion events in the predicted values ​​and verify the accuracy of the prediction.

[0013] As a preferred embodiment of the method for predicting hematoma expansion events in stroke patients described in the present invention, the data preprocessing includes merging various types of information corresponding to the serial number of the patient's first examination, calculating the time interval from the patient's onset to the first hospitalization imaging examination, encoding the blood pressure and gender, and re-extracting index sorting and normalization.

[0014] As a preferred embodiment of the method for predicting hematoma expansion events in stroke patients according to the present invention, the XG-Boost model includes:

[0015] The XG-Boost model is applied to the training dataset. The model iteratively adds trees, each time adding a tree to fit the residual of the previous round of prediction. The specific process follows the objective function:

[0016] in, It represents the accumulation of loss functions, where n represents the number of samples. Represents the loss function of a single sample, used to measure the model prediction value With the actual value y i the differences between; represents the accumulation of regularization terms, K represents the number of trees in the model; γ is a parameter used to control the complexity of the model, which is achieved by penalizing the number of leaf nodes T; represents the L2 regularization term of the leaf node weight; λ is the regularization strength parameter, and w represents the weight of each leaf node in the tree.

[0017] Update the model's predictions in each round using the formula:

[0018] in, represents the predicted value of the previous round, f t (x i ) represents the tree pair sample x newly added to the model in this round i The predicted value of .

[0019] As a preferred embodiment of the method for predicting hematoma expansion events in stroke patients according to the present invention, the calculation of the SHAP average and contribution of different types of data includes:

[0020] For each feature x in the model j , calculate its SHAP value φ j To reflect the average contribution of the feature quantity to the model prediction results, the specific formula is as follows:

[0021] Among them, φ j (νal) represents the characteristic x j SHAP value of Consider all the features that do not contain x j The feature subset S consists of all the features in the model except x j All other features except ; |S| represents the number of features in the feature subset S; p represents the total number of features in the model;

[0022] in, Represents the weight factor, which is used to balance the contribution of feature subsets of different sizes in the calculation, taking into account the feature x j The impact of adding or not adding subset S on the prediction results; val(S∪{x j})-val(S) represents the feature x j The influence of the existence or non-existence on the prediction results of the model; val(S∪{x j}) indicates that the feature x is included j In the case of , the predicted value of feature subset S; val(S) represents the predicted value of feature subset S when feature x is not included. j In the case of , the predicted value of feature subset S.

[0023] As a preferred embodiment of the method for predicting hematoma expansion events in stroke patients according to the present invention, the multivariate linear regression analysis includes using a multivariate linear regression model to analyze the most influential feature quantities extracted and calculate the weight of their influence on hematoma expansion. The analysis model follows the form: Y = β0 + β1X1 + β2X2 + ... + β p X p +ε

[0024] Among them, Y represents the dependent variable, x1, x2, ..., x p represents the independent variable, β0,β1,β2,...,β p represents the regression coefficient, β0 represents the intercept, and ε represents the error term.

[0025] As a preferred embodiment of the method for predicting hematoma expansion events in stroke patients according to the present invention, the method of predicting the probability of hematoma expansion using an extreme learning machine includes the following calculation formula:

[0026] Where L represents the number of hidden units; N represents the number of training samples; β i represents the weight vector between the i-th hidden layer and the output; w represents the weight vector between the input and output; g represents the activation function; b represents the bias vector; x represents the input vector;

[0027] Using the ELM feedforward neural network structure, the weighted feature quantity is used as input to construct the hidden layer output matrix H, set the target output matrix T and the weight matrix β from the hidden layer to the output layer. The specific form is:

[0028] Among them, H represents the hidden layer output matrix, the size of which is N×L, where N is the number of training samples and L is the number of hidden layer units; g(w i ·x j +b i ) represents the activation function of the hidden layer, for each hidden unit i (i = 1, ..., L) and each training sample j (j = 1, ..., N), where w i is the weight from the input layer to the i-th hidden unit, b i is the bias term; β represents the weight matrix from the hidden layer to the output layer, with a size of L×m, where m is the dimension of the output layer; T represents the label matrix of the training set, with a size of N×m; represents the target label of the jth training sample.

[0029] As a preferred embodiment of the method for predicting hematoma expansion events in stroke patients described in the present invention, the method uses the Northern Goshawk optimization algorithm to find the boundary of hematoma expansion events in the predicted value, and verifies the accuracy of the prediction, including:

[0030] Initialize the northern goshawk population X:

[0031] Among them, X i represents the i-th proposed solution, x i,j represents the value of the jth variable specified by the i-th proposed solution, N represents the number of population members, and m represents the number of problem variables;

[0032] In the first stage of the northern goshawk optimization algorithm, the position of each goshawk is updated according to the following formula: P i =X k ,i=1,2,...,N,k=1,2,...,i-1,i+1,...,N

[0033] Among them, P i represents the prey location of the i-th northern goshawk, F Pi Represents the objective function value, k represents a random natural number in the interval [1, N], represents the new state of the i-th proposed solution, represents its j-th dimension, It represents the objective function value based on the first stage of NGO; r represents a random number in the interval [0,1], and I represents a random number;

[0034] In the second stage of the northern goshawk optimization algorithm, the escape behavior of the prey after being attacked and the hawk's rapid pursuit action are simulated according to the following formula:

[0035] Where t represents the iteration counter, T represents the maximum number of iterations, represents the new state of the i-th proposed solution, represents its j-th dimension, It represents the objective function value based on the second stage of NGO.

[0036] A system for predicting hematoma expansion events in stroke patients, characterized by comprising:

[0037] Data preprocessing module: performs data preprocessing on the patient's personal history, medical history, onset-related information, hematoma volume and location, hematoma shape and grayscale information;

[0038] SHAP value calculation module: uses the XG-Boost model to calculate the SHAP average and contribution of different types of data, and extracts the feature with the highest contribution in each type of data;

[0039] Multiple linear regression analysis module: Perform multiple linear regression analysis to obtain the weight of the impact of each type of data on hematoma expansion, weight all feature quantities according to the weight, and obtain a feature quantity weighted by the SHAP average contribution;

[0040] Extreme learning machine prediction module: uses extreme learning machine to predict the probability of hematoma expansion;

[0041] Northern Goshawk Optimization Algorithm Module: Use the Northern Goshawk Optimization Algorithm to find the boundary of hematoma expansion events in the predicted values ​​and verify the accuracy of the prediction.

[0042] A computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0043] A computer-readable storage medium stores a computer program, which implements the steps of the method described above when executed by a processor.

[0044] Beneficial effects of the present invention: The present invention can accurately analyze and predict the risk of hematoma expansion after hemorrhagic stroke, which is crucial for early diagnosis and timely treatment. The prediction model provided by the present invention can perform personalized risk assessment based on the patient's specific situation, which helps to formulate a more accurate treatment plan. By identifying high-risk patients early and taking appropriate intervention measures, serious complications caused by hematoma expansion can be effectively reduced, thereby reducing long-term medical costs. Accurate risk prediction and timely therapeutic intervention help reduce further damage to brain tissue, thereby improving the patient's survival rate and quality of life. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort. Among them:

[0046] FIG1 is an overall flow chart of a method for predicting hematoma expansion events in stroke patients provided by a first embodiment of the present invention;

[0047] FIG2 is a diagram showing the Shapley mean and contribution mean of the characteristic values ​​in Table 1 of a method for predicting hematoma expansion events in stroke patients provided by a second embodiment of the present invention;

[0048] FIG3 is a diagram showing the Shapley mean and contribution mean of the characteristic values ​​in Table 2 of a method for predicting hematoma expansion events in stroke patients provided by a second embodiment of the present invention;

[0049] FIG4 is a diagram showing the Shapley mean and contribution mean of the characteristic values ​​in Table 3 of a method for predicting hematoma expansion events in stroke patients provided by a second embodiment of the present invention;

[0050] FIG5 is a multivariate linear regression process diagram of a method for predicting hematoma expansion events in stroke patients provided by a second embodiment of the present invention;

[0051] FIG6 is a diagram showing the weights of different types of data in a method for predicting hematoma expansion events in stroke patients provided by a second embodiment of the present invention;

[0052] FIG7 is a comparison diagram of actual values ​​and predicted values ​​of a training set of a method for predicting hematoma expansion events in stroke patients provided by a second embodiment of the present invention;

[0053] FIG8 is a confusion matrix diagram of sub001 to sub100 training sets of a method for predicting hematoma expansion events in stroke patients provided by the second embodiment of the present invention. DETAILED DESCRIPTION

[0054] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0055] Example 1

[0056] 1 , which shows an embodiment of the present invention, provides a method for predicting hematoma expansion events in stroke patients, comprising:

[0057] S1: Collect patient data information and perform data preprocessing.

[0058] Data preprocessing is performed on the patient's personal history, disease history, onset relatedness, hematoma volume and location, hematoma shape and grayscale information.

[0059] The various types of information corresponding to the serial number of the patient's first examination were merged, the time interval from the patient's onset to the first hospital imaging examination was calculated, the blood pressure and gender were coded, and the index was re-extracted, sorted, and normalized.

[0060] S2: Use the XG-Boost model to calculate the SHAP average and contribution of different types of data, and extract the feature with the highest contribution in each type of data.

[0061] XG-Boost is a type of boosting algorithm. The idea behind boosting algorithms is to ensemble many weak classifiers into a single strong classifier. Because XGBoost is a boosted tree model, it ensembles many tree models into a single, strong classifier. The tree model used is the CART regression tree model.

[0062] The CART regression tree assumes that the tree is a binary tree and continuously splits the features. For example, the current tree node is split based on the jth eigenvalue. Let the samples with the eigenvalue less than s be divided into the left subtree, and the samples with the eigenvalue greater than s be divided into the right subtree. R1(j,s)={x|x (j) ≤s}andR2(j,s)={x|x (j) >s} (5-1)

[0063] R1 and R2 can be regarded as the sets of data points contained in the two subtrees split by the threshold s on feature j.

[0064] The CART regression tree essentially divides the sample space in this feature dimension, and the optimization of this space division is an NP-hard problem. Therefore, a heuristic method is used to solve it in the decision tree model. The objective function generated by a typical CART regression tree is:

[0065] y i is the value of the target quantity actually observed, f(x i ) is the value predicted by the model, x i is a data point in the feature space, R m It represents a set of data points divided according to the eigenvalues ​​at a node in the decision tree, that is, the area corresponding to a node.

[0066] In order to solve the optimal segmentation feature j and the optimal segmentation point s, it is transformed into solving an objective function:

[0067] It means to minimize the expression with respect to j and s, where j is the index of the feature and s is the value of the split point.

[0068] is the sum of squared errors of the left subtree; where x i is the sample point that falls on the left subtree R1 after segmentation, y i is the true value of the sample point, c1 is the average of the target values ​​of all sample points in the left subtree; It means that we need to find a value c1 that minimizes the sum of square errors of the left subtree. It belongs to the right subtree.

[0069] By traversing all the split points of all features, we can find the optimal split features and split points, and finally get a regression tree.

[0070] The XG-Boost algorithm works by continuously adding trees and performing feature splitting to grow a tree. Each time a tree is added, a new function is learned to fit the residual of the previous prediction. After training, k trees are obtained. Each tree will fall into a leaf node, and each leaf node is assigned a score. Finally, the predicted value for the sample is simply calculated by adding up the scores of each tree.

[0071] in, It represents the estimation of the target variable output by the model; φ(x i ) is the prediction function for the i-th sample, which is the weighted sum of a series of basis functions; K represents the total number of trees in the model; f k (x i ) represents the prediction result of the k-th tree for the i-th sample. It means that the prediction results of all K trees are added together to get the final prediction value. F is the function space of the model, that is, the set of all possible prediction trees. f(x) = w q(x) For the input feature x, the decision tree uses the function q(x) to determine which leaf node it falls on, and uses w to index the weight corresponding to this leaf node. q:R m →T is a mapping function that maps the m-dimensional feature space to the leaf nodes of the tree. T is the set of leaf nodes of the tree. w∈R T Each leaf node T is assigned a real-valued weight.

[0072] The XG-Boost objective function is defined as:

[0073] is the training loss, is the complexity of the tree.

[0074] The objective function consists of two parts: the first measures the difference between the predicted score and the true score, and the second is a regularization term. T represents the number of leaf nodes, and w represents the score of a leaf node. γ controls the number of leaf nodes, while λ prevents overfitting by limiting the leaf node score.

[0075] The newly generated tree is to fit the residual of the last prediction, and the prediction score can be written as:

[0076] The objective function can be rewritten as:

[0077] The idea of ​​XG-Boost is to use its t = 0. So the objective function is approximately:

[0078] g i is the first-order derivative, h i is the second-order derivative.

[0079] Since the residual of the prediction scores of the first t-1 trees and y has no effect on the optimization of the objective function, the objective function can be simplified as:

[0080] Reorganize all samples of the same leaf node, and we have:

[0081] By rewriting formula (5-11), the optimal w and objective function formulas are:

[0082] XG-Boost uses many strategies to prevent overfitting; the objective function optimization utilizes the second-order derivative of the loss function with respect to the function to be solved; nodes at the same level can be run in parallel; for a certain node, the best split point is selected within the node, and the gains of the candidate split points are calculated in parallel using multiple threads, which makes the training speed fast.

[0083] For each feature x in the model j , calculate its SHAP value φ j To reflect the average contribution of the feature quantity to the model prediction results.

[0084] The main idea behind SHAP values ​​is Shapley values, a method from coalitional game theory. The Shapley value of each eigenvalue is the contribution of that eigenvalue to the prediction, obtained by weighting and summing all possible combinations of eigenvalues:

[0085] Where S is the subset of features used in the model, x is the vector of feature values ​​of the instance to be explained, and p is the number of features;

[0086] in, is the weight of subset S, which is related to the order.

[0087] val(S) is the prediction for subset S.

[0088] S3: Perform multiple linear regression analysis to obtain the weight of the influence of each type of data on hematoma expansion. All feature quantities are weighted according to the weight to obtain a feature quantity weighted by the SHAP average contribution.

[0089] A multivariate linear regression model was used to analyze the most influential features extracted and calculate their weights on hematoma expansion. The analysis model followed the form: Y = β0 + β1X1 + β2X2 + ... + β p X p +ε (5-14)

[0090] Where Y is the dependent variable, x1, x2, ..., x p are independent variables, β0,β1,β2,...,β p is the regression coefficient, β0 is the intercept, and ε is the error term.

[0091] It should be noted that multiple linear regression (MLR), also known as multiple regression, is a statistical technique that uses multiple explanatory variables to predict the outcome of a response variable. The goal of multiple linear regression is to model the linear relationship between the explanatory (independent) variables and the response (dependent) variable. Essentially, multiple regression is an extension of ordinary least squares (OLS) regression because it involves multiple explanatory variables.

[0092] S4: Use extreme learning machine to predict the probability of hematoma expansion.

[0093] Extreme Learning Machines (ELM) are feedforward neural networks with good generalization performance and extremely fast learning capabilities. ELM does not require gradient-based backpropagation to adjust weights, but instead uses Moore-Penrose generalized inverse to set weights. Specifically, the ELM calculation process is as follows:

[0094] Where: L is the number of hidden units; N is the number of training samples; β is the weight vector between the i-th hidden layer and the output; ω is the weight vector between the input and output; g is the activation function; b is the bias vector; x is the input vector.

[0095] The computational process of the extreme learning machine is very similar to that of a standard back-propagation neural network, but the weight matrix between the hidden layer and the output is a pseudo-inverse matrix.

[0096] The above formula can be simplified as: T=Hβ (5-24)

[0097] Where: m is the number of outputs; H is the hidden layer output matrix; T is the training set target matrix.

[0098] S5: Use the Northern Goshawk optimization algorithm to find the boundary of hematoma expansion events in the predicted values ​​and verify the accuracy of the prediction.

[0099] At the beginning of the algorithm, the population members are randomly initialized in the search space. The population matrix in the NGO algorithm is determined using (5-15).

[0100] Where X is the population of northern goshawks, X i is the i-th proposed solution, x i,j is the value of the jth variable specified by the i-th proposed solution, N is the number of population members, and m is the number of problem variables.

[0101] These values ​​obtained for the objective function can be represented as a vector using

[0102] In the first phase of hunting, a northern goshawk randomly selects a prey item and quickly attacks it. This phase increases the NGO's exploration power by randomly selecting prey items within the search space. This phase results in a global search of the search space, aiming to identify the optimal region. The concepts expressed in the first phase are mathematically modeled using (5-17) to (5-19). P i =X k ,i=1,2,...,N,k=1,2,...,i-1,i+1,...,N (5-19)

[0103] P i is the prey position of the i-th northern goshawk, is the objective function value, k is a random natural number in the interval [1, N], is the new state of the i-th proposed solution, is its j-th dimension, is the objective function value based on the first stage of NGO; r is a random number in the interval [0,1], and I is a random number that can be 1 or 2.

[0104] In the second phase of a northern goshawk's hunting, prey, after being attacked, will attempt to escape. Therefore, the northern goshawk continues to pursue its prey during the tailing and pursuit phases. Due to its high speed, the northern goshawk can pursue and ultimately capture its prey in almost any situation. Simulating this behavior enhances the algorithm's ability to exploit local search space. In the proposed NGO algorithm, the pursuit process between a northern goshawk and its prey is assumed to be enclosed within an attack position with a radius of R. Figure 3 shows the chase process between the northern goshawk and its prey. The concepts expressed in the second phase are mathematically modeled using (5-20) to (5-22).

[0105] t is the iteration counter, T is the maximum number of iterations, is the new state of the i-th proposed solution, is its j-th dimension, It is the objective function value based on the second stage of NGO.

[0106] It should be noted that the Northern Goshawk Optimization Algorithm (NGOA) is a heuristic algorithm inspired by the goshawk's hunting ability. It simulates the goshawk's hunting behavior to find the optimal solution. NGOA is a population-based algorithm, with the northern goshawk as the searcher. In NGOA, each population member represents a proposed solution to the problem and determines the value of the variables. From a mathematical perspective, each population member is a vector, and these vectors together constitute the algorithm's population matrix.

[0107] The above embodiments also include a system for predicting hematoma expansion events in stroke patients, specifically:

[0108] Data preprocessing module: performs data preprocessing on the patient's personal history, medical history, onset-related information, hematoma volume and location, hematoma shape and grayscale information;

[0109] SHAP value calculation module: uses the XG-Boost model to calculate the SHAP average and contribution of different types of data, and extracts the feature with the highest contribution in each type of data;

[0110] Multiple linear regression analysis module: Perform multiple linear regression analysis to obtain the weight of the impact of each type of data on hematoma expansion, weight all feature quantities according to the weight, and obtain a feature quantity weighted by the SHAP average contribution;

[0111] Extreme learning machine prediction module: uses extreme learning machine to predict the probability of hematoma expansion;

[0112] Northern Goshawk Optimization Algorithm Module: Use the Northern Goshawk Optimization Algorithm to find the boundary of hematoma expansion events in the predicted values ​​and verify the accuracy of the prediction.

[0113] The computer device may be a server. The computer device includes a processor, a memory, an input / output interface (I / O) and a communication interface. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data cluster data of the power monitoring system. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for predicting hematoma expansion events in stroke patients is implemented.

[0114] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in the various embodiments provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processor involved in the various embodiments provided in this application may be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computing, etc., but are not limited to these.

[0115] Example 2

[0116] 2 to 8 , which illustrate an embodiment of the present invention, a method for predicting hematoma expansion events in stroke patients is provided. To verify the beneficial effects of the present invention, scientific demonstration is performed through economic benefit calculations and simulation / comparative experiments.

[0117] Gender and blood pressure were first encoded (1 for male and 0 for female, 1 for blood pressure outside the normal range and 0 for blood pressure within the normal range). When building the model, different types of data were taken into consideration. In order to more accurately predict the probability of hemodialysis in patients, hemodialysis was first used as the target quantity. Different types of feature quantities were calculated using the XG-Boost model as the carrier to calculate the SHAP average of different types of feature quantities, and the contribution of each feature quantity was calculated based on this average.

[0118] It should be noted that since the collected data content is relatively complex, Table 1 shown in the present invention is only a partial content.

[0119] Table 1 List of some patients and clinical information

[0120] The calculation results are shown in Figures 2, 3, and 4, respectively. Figure 2 shows that the interval from onset to first imaging examination and a history of diabetes contribute significantly to blood vessel expansion. Figure 3 shows that HM_MCA_R_Ratio and HM_volume contribute most significantly to blood vessel expansion. Figure 4 shows that NCCT_original_firstorder_Entropy and NCCT_original_firstorder_InterquartileRange contribute significantly to blood vessel expansion. Based on these results, the top two contributing features in each table were extracted for feature dimensionality reduction. Multiple linear regression analysis was then used to analyze the weight of each feature type's impact on the target variable, blood vessel expansion. The analysis process is shown in Figure 5, and the results are shown in Figure 6. This weight was added to the previous SHAP average contribution to obtain the contribution of all features. This contribution was then multiplied across all features, and all data were normalized.

[0121] Then, all the feature data and blood expansion data of patients from sub001 to sub100 were used as the training set and put into the extreme learning machine (ELM) for learning, obtaining a series of blood expansion probabilities from sub001 to sub100. Then, we used the obtained probabilities and the actual label quantity as input, and used the NGO fitness function with the minimum variance of the obtained multiple iteration accuracy to optimize the setting value of the classification boundary. We found that the optimal setting value was 0.47. Then, we established a confusion matrix of the actual values ​​and predicted values ​​as shown in Figures 5-8 below. Finally, we used the trained model to predict the probability of blood expansion from sub100 to sub160.

[0122] As can be seen from Figure 7, the overall prediction accuracy of the model for blood vessel dilation reached 87%, and the accuracy of blood vessel dilation (label quantity is 1) was 80.65%.

[0123] From the confusion matrix in Figure 8, we can see that when the true value is 0, the prediction accuracy is as high as 91.2%. When the true value is 1, the prediction accuracy is also close to 80%. The overall classification results show a main diagonal distribution, and the model has a high prediction accuracy.

[0124] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for predicting hematoma expansion events in stroke patients, characterized in that: include: Collect patient data information and perform data preprocessing; Use the XG-Boost model to calculate the SHAP average and contribution of different types of data, and extract the feature with the highest contribution in each type of data; Perform multiple linear regression analysis to obtain the weight of the impact of each type of data on hematoma expansion, and weight all feature quantities according to the weight to obtain a feature quantity weighted by the SHAP average contribution; Use extreme learning machine to predict the probability of hematoma expansion; The Northern Goshawk optimization algorithm was used to find the boundary of hematoma expansion events in the predicted values ​​and verify the accuracy of the prediction.

2. The method for predicting hematoma expansion events in stroke patients according to claim 1, characterized in that: The data preprocessing includes merging various types of information corresponding to the serial number of the patient's first examination, calculating the time interval from the patient's onset to the first hospital imaging examination, encoding blood pressure and gender, and re-extracting index sorting and normalization.

3. The method for predicting hematoma expansion events in stroke patients according to claim 2, characterized in that: The XG-Boost model includes, The XG-Boost model is applied to train the training data set. The model adds trees iteratively, adding one tree each time to fit the residual of the previous round of prediction. The specific process follows the objective function Obj as follows: in, It represents the accumulation of loss functions; n represents the number of samples; Represents the loss function of a single sample, used to measure the model prediction value With the actual value y i The difference between represents the accumulation of regularization terms; K represents the number of trees in the model; γ represents a parameter used to control the complexity of the model, which is achieved by penalizing the number of leaf nodes T; represents the L2 regularization term of the leaf node weight; λ represents the regularization strength parameter, and w represents the weight of each leaf node in the tree. In each round, the model's prediction value is updated, expressed as: in, represents the predicted value of the previous round, f t (x i ) represents the tree pair sample x newly added to the model in this round i The predicted value of .

4. The method for predicting hematoma expansion events in stroke patients according to claim 3, characterized in that: The calculation of SHAP average values ​​and contributions of different types of data includes: For each feature x in the model j , calculate its SHAP value φ j To reflect the average contribution of the feature quantity to the model prediction results, the specific formula is as follows: Among them, φ j (νal) represents the characteristic x j SHAP value of Consider all the features that do not contain x j The feature subset S consists of all features in the model except x j All other features except ; |S| represents the number of features in the feature subset S; p represents the total number of features in the model; in, Represents the weight factor, which is used to balance the contribution of feature subsets of different sizes in the calculation, taking into account the feature x j The impact of adding or not adding subset S on the prediction results; val(S∪{x j })-val(S) represents the feature x j The influence of the existence or non-existence on the prediction results of the model; val(S∪{x j }) indicates that in the case of containing feature x j In the case of , the predicted value of feature subset S; val(S) represents the predicted value of feature subset S without feature x j In the case of , the predicted value of feature subset S.

5. The method for predicting hematoma expansion events in stroke patients according to claim 4, characterized in that: The multivariate linear regression analysis includes using a multivariate linear regression model to analyze the most influential feature quantity extracted and calculate the weight of its influence on hematoma expansion. The analysis model follows the form: Y = β0 + β1X1 + β2X2 + ... + β p X p +ε Among them, Y represents the dependent variable, x1, x2, ..., x p represents the independent variable, β0,β1,β2,...,β p represents the regression coefficient, β0 represents the intercept, and ε represents the error term.

6. The method for predicting hematoma expansion events in stroke patients according to claim 5, characterized in that: The prediction of the probability of hematoma expansion using an extreme learning machine includes the following calculation formula: Where L represents the number of hidden units; N represents the number of training samples; β i represents the weight vector between the i-th hidden layer and the output; w represents the weight vector between the input and the output; g represents the activation function; b represents the bias vector; x represents the input vector; Using the ELM feedforward neural network structure, the weighted feature quantity is used as input, the hidden layer output matrix H is constructed, and the target output matrix T and the weight matrix β from the hidden layer to the output layer are set. The specific expression is: Among them, H represents the hidden layer output matrix, the size is N×L, where N is the number of training samples and L is the number of hidden layer units; g(w i ·x j +b i ) represents the activation function of the hidden layer. For each hidden unit i (i = 1, ..., L) and each training sample j (j = 1, ..., N), where w i is the weight from the input layer to the i-th hidden unit, b i is the bias term; β represents the weight matrix from the hidden layer to the output layer, with a size of L×m, where m is the dimension of the output layer; T represents the label matrix of the training set, with a size of N×m; represents the target label of the jth training sample.

7. The method for predicting hematoma expansion events in stroke patients according to claim 6, characterized in that: The use of the Northern Goshawk optimization algorithm includes: Initialize the northern goshawk population X: Among them, X i represents the i-th proposed solution, x i,j represents the value of the jth variable specified by the i-th proposed solution, N represents the number of population members, and m represents the number of problem variables; In the first stage of the northern goshawk optimization algorithm, the position of each goshawk is updated according to the following formula: P i =X k ,i=1,2,…,N,k=1,2,…,i-1,i+1,…,N Among them, P i represents the prey position of the i-th northern goshawk, represents the objective function value, k represents a random natural number in the interval [1, N], represents the new state of the i-th proposed solution, represents its j-th dimension, represents the objective function value based on the first stage of NGO; r represents a random number in the interval [0,1], and I represents a random number; In the second stage of the northern goshawk optimization algorithm, the escape behavior of the prey after being attacked and the rapid pursuit action of the goshawk are simulated according to the following formula: Where t represents the iteration counter, T represents the maximum number of iterations, represents the new state of the i-th proposed solution, represents its j-th dimension, It represents the objective function value based on the second stage of NGO.

8. A system for predicting hematoma expansion events in stroke patients using the method according to any one of claims 1 to 7, characterized in that: Data preprocessing module: preprocess the patient's personal history, medical history, disease-related information, hematoma volume and location, hematoma shape and grayscale information; SHAP value calculation module: Use the XG-Boost model to calculate the SHAP average and contribution of different types of data, and extract the feature with the highest contribution in each type of data; Multiple linear regression analysis module: Perform multiple linear regression analysis to obtain the weight of the impact of each type of data on hematoma expansion, weight all feature quantities according to the weight, and obtain a feature quantity weighted by the SHAP average contribution; Extreme learning machine prediction module: Use extreme learning machine to predict the probability of hematoma expansion; Northern Goshawk Optimization Algorithm Module: Use the Northern Goshawk Optimization Algorithm to find the boundary of hematoma expansion events in the predicted values ​​and verify the accuracy of the prediction.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images