Method, system, storage medium and device for predicting hematoma change probability

Abstract

The present invention relates to a prediction method, system, storage medium and device for the probability of hematoma change, which obtains brain image data and the diagnosis and treatment history data of the corresponding patient; extracts the volume, position, shape and distribution information of hematoma and edema based on the image data, associates it with the diagnosis and treatment history data, forms a data set and preprocesses it; the preprocessed data set forms different training subsets by extracting different features, predicts whether different types of hematoma change events occur, and obtains the probability of hematoma change in the input brain image data through clustering processing.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a method, system, storage medium and device for predicting the probability of hematoma change. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] In the medical field, stroke is a disease that causes damage to brain tissue, including ischemic and hemorrhagic types. Among them, hemorrhagic stroke has no history of trauma, but instead causes bleeding in the skull and brain parenchyma to form hematomas that damage the patient's brain nerves.

[0004] Existing technology uses brain image data combined with information such as blood pressure, coagulation function, bleeding volume, and patient age to predict the probability of expansion of a patient's brain hematoma and changes in the morphology of edema around the hematoma based on the doctor's subjective experience. This is inefficient and has low accuracy. Although some computer programs can carry trained prediction models to help doctors capture biomarkers or imaging features related to the development of hematomas, thereby outputting the probability of hematoma changes, the accuracy of the model is not ideal due to the quality and scale of the data set, as well as possible problems such as inaccurate labeling and unbalanced samples. As a result, the model exhibits poor generalization ability when processing data from different sources, different devices, or different populations. This means that the model may perform well on a specific data set, but its performance in actual applications is greatly reduced. Summary of the invention

[0005] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a method, system, storage medium and device for predicting the probability of hematoma change, which predicts the probability of hematoma change based on brain images, corresponding patient information and treatment plans, combined with prognosis data.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A first aspect of the present invention provides a method for predicting the probability of hematoma change, comprising the following steps:

[0008] Obtain brain image data and corresponding patient's diagnosis and treatment history data;

[0009] Extract the volume, location, shape and distribution information of hemorrhage and edema based on image data, associate it with diagnosis and treatment history data, form a data set and preprocess it;

[0010] The preprocessed data set is used to form different training subsets by extracting different features to predict whether different types of hematoma change events will occur. The probability of hematoma changes in the input brain image data is obtained through clustering processing.

[0011] A second aspect of the present invention provides a prediction system for hematoma change probability, comprising:

[0012] The raw data acquisition module is configured to: acquire brain image data and diagnosis and treatment history data of the corresponding patient;

[0013] The data set construction module is configured to: extract the volume, location, shape and distribution information of hemorrhage and edema based on the image data, associate it with the diagnosis and treatment history data, form a data set and pre-process it;

[0014] The prediction model module is configured as follows: the preprocessed data set is formed into different training subsets by extracting different features, and different types of hematoma change events are predicted to occur respectively, and the probability of hematoma change in the input brain image data is obtained through clustering processing.

[0015] Furthermore, the patient's medical history data is obtained by extracting text data from the patient's medical record to obtain the patient's personal information, medical history, and medical history.

[0016] Furthermore, the preprocessing transforms the feature data into a distribution with a mean of 0 and a standard deviation of 1 by subtracting the mean and dividing by the standard deviation.

[0017] Furthermore, each training subset corresponds to a decision tree, and multiple decision trees form an integrated prediction model.

[0018] Furthermore, by determining the maximum tree depth, the number of trees, and the class weight parameters of the integrated prediction model, the optimal hyperparameters of the integrated prediction model are obtained through grid search.

[0019] Furthermore, based on the set factors in the patient's medical history data, the patients were divided into different groups, and the curves of the edema volume changing over time for each group were obtained. After mean processing, the eigenvector of the covariance matrix was obtained, and the diagonal matrix that the matrix satisfies was found. The diagonal values ​​were arranged in descending order, and the number of principal components was selected to construct a matrix as the component matrix.

[0020] Furthermore, in the obtained component matrix, k points are randomly initialized as cluster centroids, each point in the sample set is assigned to a cluster, and the distance between each point and the centroid is calculated, and it is assigned to the cluster corresponding to the nearest centroid, and the centroid of the cluster is continued to be updated, and the centroid is updated to the average value of all points in the current cluster, and it is iterated repeatedly until the set number of iterations is met to determine the probability of a change in the hematoma in the input brain image data.

[0021] The third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-mentioned method for predicting the probability of hematoma change.

[0022] The fourth aspect of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the program, the steps in the above-mentioned method for predicting the probability of hematoma change are implemented.

[0023] Compared with the prior art, one or more of the above technical solutions have the following beneficial effects:

[0024] Based on brain images, corresponding patient information and treatment plans, the probability of hematoma changes is predicted, and different training subsets are used to predict whether the hematoma will undergo corresponding changes. During the splitting process of each node, a part of the features are randomly selected for evaluation instead of considering all features. The randomness helps to improve the diversity and generalization ability of the model, and the integration helps to reduce the risk of overfitting and improve the stability and accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0026] Figure 1 is a K-Means clustering flow chart for predicting the period during which a hematoma changes, provided by one or more embodiments of the present invention;

[0027] Figure 2 is a fully connected neural network diagram for predicting a period during which a hematoma changes, provided by one or more embodiments of the present invention;

[0028] Figure 3 is a schematic diagram of the F1 scores and accuracy of the random forest algorithm and the XGBoost algorithm provided by one or more embodiments of the present invention;

[0029] Figure 4 A schematic diagram of lithotripsy during a period of predicted hematoma changes provided by one or more embodiments of the present invention;

[0030] Figure 5 is a schematic diagram of the change of edema volume over time of a normalized category 1 data group provided by one or more embodiments of the present invention;

[0031] Figure 6 is a schematic diagram of the change of edema volume over time of a category 2 data group provided by one or more embodiments of the present invention;

[0032] Figure 7 is a schematic diagram of the change of edema volume over time of a category 3 data group provided by one or more embodiments of the present invention;

[0033] Figure 8 It is a schematic diagram of model performance indicators for predicting the probability of hematoma occurrence provided by one or more embodiments of the present invention. DETAILED DESCRIPTION

[0034] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0035] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0036] As introduced in the background technology, computer models in the existing technology are affected by the quality and scale of the data set, as well as possible problems such as inaccurate labeling and unbalanced samples. The accuracy of the model is not ideal, resulting in the model showing poor generalization ability when processing data from different sources, different devices or different populations. This means that the model may perform well on a specific data set, but its performance in actual applications is greatly reduced.

[0037] Therefore, the following embodiments provide a method, system, storage medium and device for predicting the probability of hematoma change, which predicts the probability of hematoma change based on brain images, corresponding patient information and treatment plans, combined with prognosis data.

[0038] Embodiment 1:

[0039] like Figure 1-Figure 8 As shown, the prediction system of the probability of hematoma change includes:

[0040] The raw data acquisition module is configured to: acquire brain image data and diagnosis and treatment history data of the corresponding patient;

[0041] The data set construction module is configured to: extract the volume, location, shape and distribution information of hemorrhage and edema based on the image data, associate it with the diagnosis and treatment history data, form a data set and pre-process it;

[0042] The prediction model module is configured as follows: the preprocessed data set is formed into different training subsets by extracting different features, and different types of hematoma change events are predicted to occur respectively, and the probability of hematoma change in the input brain image data is obtained through clustering processing.

[0043] In order to solve the problem of data imbalance and build a model to predict the changes of hematoma in patients, the oversampling method in the imblearn library (a Python library for processing unbalanced data sets) is considered to synthesize new minority class samples in the feature space. The personal history, disease history, onset-related information, and imaging examination results of hematoma and edema volume and location, as well as the shape and grayscale distribution of hematoma and edema are integrated and processed to form a comprehensive data set, which is divided into features (x) and targets (y). The features include personal history, disease history, onset-related information, hematoma and edema volume, location and shape, grayscale distribution, etc. The target is the probability of hematoma changes. The oversampling method in the imblearn library is used to handle data imbalance, and the data set is reconstructed by random oversampling method.

[0044] The feature data sets of personal history, disease history, onset correlation, hematoma and edema volume, location and shape, and grayscale distribution were standardized using StandardScaler in the scikit-learn library (data standardization tool in the scikit-learn library). The feature data was converted to a distribution with a mean of 0 and a standard deviation of 1 by subtracting the mean and dividing by the standard deviation. The dimensional differences between different features can be eliminated so that they have the same scale, which is helpful for the subsequent algorithm training and optimization process. This can be achieved through the following formula:

[0045] z=(x-μ) / σ;

[0046] Where z is the data after standardization of each column of patient data, x is the original data for different columns, μ is the mean of the original data for different columns, and σ is the standard deviation of the original data.

[0047] In order to obtain the best model performance, this embodiment performs comprehensive hyperparameter tuning on two target classification models, Random Forest and XGBoost.

[0048] Hyperparameter tuning is a key step to improve model performance. Its purpose is to find a set of hyperparameter combinations that can achieve the best model performance indicators by adjusting the model's hyperparameters on a given data set. By selecting the best parameter configuration, the model can achieve the best effect on a given data set, thus providing a basis for classification tasks.

[0049] (1) Use the random forest algorithm for regression tasks, build multiple decision trees to form an integrated model, and use the Random Forest Classifier class to model the classification task to predict whether the hematoma expansion event occurs.

[0050] When dealing with standardized training sets of different features of different patients, the random forest algorithm generates multiple different training subsets by random sampling, and each subset is used to build an independent decision tree model. In the construction process of each decision tree, in the splitting process of each node, not all features are considered, but a part of the features are randomly selected for evaluation. This randomness increases the diversity of the model, reduces the risk of overfitting, and improves the generalization ability of the model. Therefore, between "different training subsets" and each "node", each training subset is used to build a separate decision tree, and each node randomly selects a part of the features for evaluation during the splitting process to form a different decision tree. Finally, by combining the prediction results of these decision trees, the overall prediction result of the random forest model is obtained.

[0051] This embodiment uses the randomness of the random forest algorithm to help improve the diversity and generalization ability of the model, and the integration helps reduce the risk of overfitting and improve the stability and accuracy of the model. The model is trained on the standardized patient feature training set, predicted on the standardized test set, and the performance of the model is evaluated. The accuracy can be calculated and output by the accuracy_score() function.

[0052] For the random forest model, we mainly adjust the maximum tree depth max_depth, the number of trees n_estimators, and the class weight parameter class_weight. After grid search, we get the best hyperparameters for the random forest.

[0053] The optimization parameters in this embodiment include the number of decision trees, the maximum tree depth, the minimum number of samples required for internal node re-division, the minimum number of samples for leaf nodes, and the maximum number of features, etc. This method exhaustively searches for pre-defined parameter combinations and uses cross-validation to evaluate model performance, and finally selects the best performing parameter combination as the optimal hyperparameter to improve the generalization ability and performance of the model.

[0054] Consider the type of problem to be solved and use the weighted average method for regression problems. The calculation formula is as follows:

[0055]

[0056] In the formula, h i (x)∈R; w i is the weight of different indicators for patients, w i >0 and Obtained from the training data; T is h i The number of .

[0057] (2) Use XGBoost for regression tasks. For the XGBoost model, the maximum tree depth max_depth and the number of trees n_estimators are mainly adjusted. Finally, the optimal hyperparameters are determined. After adjusting the parameters, a shallow maximum tree depth and an appropriate number of trees can obtain the best generalization error. By combining the results of multiple weak learners as the final output, a powerful integrated model is formed. Each weak learner will learn from the errors of the previous learner to gradually improve the performance of the overall model. For K trained trees, the final prediction value for the i-th sample is:

[0058]

[0059] Among them, x i represents the sample characteristics, f k (x i ) represents the kth tree for sample x i The prediction results are added together to get the final result

[0060] Its objective function consists of two parts: loss function and regularization term. The loss function measures the prediction error of the model, and the calculation formula is as follows:

[0061]

[0062] In order to reduce the complexity of the model, the formula is used:

[0063]

[0064] In the formula, y i Represents the true result label.

[0065] The regularization term is used to control the complexity of the model and prevent overfitting. The optimization process of the objective function is to find the model parameters that minimize the objective function. After adding regularization, the calculation formula of the loss function accumulated to the current k trees is as follows:

[0066]

[0067] In the above formula, Ω(f k ) is the control model complexity among k weak learners; γ is the adjustment function; L m is the number of indicators to evaluate the importance of patient characteristics; ω is used to consider the complexity of the model according to its modulus.

[0068] This embodiment aims to obtain the correlation between hematoma expansion and perihematoma edema and the prognosis of hemorrhagic stroke. Since there is edema around the hematoma, the effect of hematoma on stroke and the effect of edema volume on stroke are obtained respectively.

[0069] Use grouping or classification to construct curves of edema volume progression over time for different populations. For example, patients can be divided into different groups based on specific factors in their personal history (such as age, gender, disease type, etc.), and curves of edema volume change over time for each group can be drawn separately. Use the obtained curves to establish corresponding polynomial regression models to better predict the curves of edema volume change over time for different populations, so that medical staff can better propose targeted treatment plans for patients' hospitalization.

[0070] On the basis of data processing, each data is fitted with a curve to obtain the parameters of the fitting curve. SPSS is used to perform principal component analysis on the relevant data. The principal component analysis algorithm is as follows:

[0071] (1) The data is converted to zero mean. The corresponding mean is subtracted from each column of data to obtain a zero mean matrix.

[0072]

[0073] In the above formula, X0 is the original data of the patient's edema volume, is the corresponding mean, and X1 is the patient edema volume data obtained after processing.

[0074] (2) Calculate the covariance matrix. The processed matrix X is as follows:

[0075]

[0076] The variance and covariance are:

[0077]

[0078]

[0079] According to the above data processing, we can get:

[0080]

[0081] (3) Calculate the eigenvectors of the covariance matrix, find the diagonal matrix that the matrix satisfies, and arrange the diagonal values ​​in descending order. Select the number of principal components to construct a matrix as the component matrix, as shown in the following formula:

[0082] Diag=P·Cov·P T ;

[0083] (4) Data dimensionality reduction: The original data is reduced in dimension according to the component matrix obtained in the previous step, as shown in the following formula:

[0084] T = P·X0;

[0085] When performing principal component analysis, KMO and Bartlett's test are obtained, and the total variance explained and scree plot indicate significant statistical differences. After obtaining the corresponding principal components, the corresponding principal components are extracted into the corresponding table, and then K-means clustering in the unsupervised learning algorithm is performed to minimize the intra-class gap and maximize the inter-class gap. According to the input sample set X = {x1, x2, ..., x m}, initialize the number of clusters k, and get the output cluster C = {C1, C2, ..., C k By randomly initializing k points as cluster centroids, each point in the sample set is assigned to a cluster, and the distance between each point and the centroid is calculated, and it is assigned to the cluster corresponding to the nearest centroid. The centroid of the cluster is updated continuously, and the centroid is updated to the average value of all points in the current cluster, and it is repeated until the given number of iterations is met. The algorithm flow is as follows Figure 1 shown.

[0086] According to the clustering effect evaluation index SSE, the formula is as follows:

[0087] in

[0088] This embodiment uses different methods to perform time prediction. By given data, according to prediction and regression analysis, the F1_score and Accuracy of the random forest algorithm and the XGBoost algorithm can be obtained, such as Figure 3 As shown in Tables 1 and 2, the confusion matrix of XGBoost predicting the data and the confusion matrix of random forest predicting the data show that the random forest model has a better effect.

[0089] Table 1 Confusion matrix of XGBoost predicting data

[0090]

[0091]

[0092] Table 2 Confusion matrix of random forest prediction of data

[0093]

[0094] By observing the confusion matrices of the two models in Table 1 and Table 2, we can have a deeper understanding of their respective classification performance. It can be seen that Random Forest has a more balanced judgment on positive and negative samples, while XGBoost is better at judging negative samples than positive samples, which also leads to the fact that the various evaluation indicators of XGBoost are not as balanced as those of Random Forest.

[0095] Finally, the performance of the optimized random forest model and XGBoost model was evaluated on the entire dataset. The accuracy of the random forest model reached 0.935 and the F1 score reached 0.923, showing good classification performance.

[0096] From the classification results, we can see that the prediction effect for both categories is good, with high average precision and recall. The accuracy of the XGBoost model is slightly lower than that of the random forest, which is 0.774. The F1 score is also relatively low, which is 0.774. From the classification results, XGBoost predicts the negative class better than the positive class, so the recall rate of the positive class is only 0.63, resulting in a low F1 score. Comparing the two models, the random forest performs more balanced and excellent. Its accuracy and F1 score are significantly higher than the XGBoost model. The classification report also shows that the prediction performance of the random forest on different categories is more balanced. Therefore, it can be considered that the random forest model has better generalization ability for the current task and is a better choice of classification model.

[0097] About using the random forest model to improve generalization ability. First, random forest is trained by randomly selecting features and samples. During the training process of each decision tree, random forest will extract a part of samples and features from the original data set with replacement, so that each decision tree is trained based on different data subsets and feature subsets. This randomness can reduce the risk of overfitting and make random forest have better generalization ability. Secondly, random forest makes predictions by integrating multiple decision trees. Each decision tree is trained independently, and the final prediction result is obtained by voting or averaging. This integrated learning method can reduce the error of a single decision tree and improve the accuracy and generalization ability of the overall model.

[0098] According to the processed data, KMO and Bartlett test are obtained, as shown in Table 1.

[0099] Table 1 KMO and Bartlett's test

[0100]

[0101] Through the total variance explanation and scree plot, 18 principal components can be extracted. The scree plot (such as Figure 4 ). Through K-means clustering, the (first 100) patients were classified and labeled according to different clustering effects. By comparing the K-means clustering results, it can be seen that when k=3, the clustering effect is the best, and the data is divided into equal proportions, as shown in Table 2.

[0102] Table 2 Number of cases in each cluster category

[0103]

[0104]

[0105] According to the different number of groups divided, time was used as the independent variable and edema volume as the dependent variable. The time variable and edema volume variable were normalized, and then the edema volume progression curves of different populations (divided into 3 subgroups) were drawn over time. Figure 5 , Figure 6 , Figure 7 shown.

[0106] Each subgroup has common characteristics. The patients' personal history, medical history, and onset-related factors are counted, and a different curve is fitted for each group. A model of the progression of edema volume over time is established for each of the three subgroups, so that the development of edema in each subgroup can be obtained.

[0107] Embodiment 2:

[0108] The method for predicting the probability of hematoma change by the system described in the first embodiment includes the following steps:

[0109] Obtain brain image data and corresponding patient's diagnosis and treatment history data;

[0110] Extract the volume, location, shape and distribution information of hemorrhage and edema based on image data, associate it with diagnosis and treatment history data, form a data set and preprocess it;

[0111] The preprocessed data set is used to form different training subsets by extracting different features to predict whether different types of hematoma change events will occur. The probability of hematoma changes in the input brain image data is obtained through clustering processing.

[0112] Embodiment three:

[0113] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps in the method for predicting the probability of hematoma change as described in the above-mentioned embodiment 2 are implemented.

[0114] Embodiment 4:

[0115] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the method for predicting the probability of hematoma change as described in the above-mentioned embodiment 2 are implemented.

[0116] The steps involved in the above embodiments 2 to 4 correspond to those in embodiment 1. For the specific implementation, please refer to the relevant description of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0117] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for predicting the probability of hematoma change, characterized in that: The following steps are involved: Obtain brain image data and corresponding patient's diagnosis and treatment history data; The volume, location, shape and distribution information of hemorrhage and edema are extracted based on the image data, associated with the diagnosis and treatment history data, and a data set is formed and preprocessed; the preprocessing includes: The oversampling method in the imblearn library is used to handle data imbalance, and the data set is reconstructed by the random oversampling method; the feature data sets of personal history, disease history, disease correlation, hematoma and edema volume, location and shape, and grayscale distribution are standardized, and the feature data are converted into a distribution with a mean of 0 and a standard deviation of 1 by subtracting the mean and dividing by the standard deviation. The specific formula is: z = (x-μ) / σ, where z is the data after each column of patient data is standardized, x is the original data for different columns, μ is the mean of the original data for different columns, and σ is the standard deviation of the original data; the preprocessed data set is formed into different training subsets by extracting different features, and different types of hematoma change events are predicted respectively. The probability of hematoma change in the input brain image data is obtained by clustering processing; The probability of a change in a hematoma in the input brain image data is obtained through clustering processing, specifically: based on the set factors in the patient's medical history data, the patients are divided into different groups, and the curves of the edema volume change over time of each group are obtained respectively, each data is curve-fitted to obtain the parameters of the fitting curve, and the relevant data are subjected to principal component analysis using SPSS. The SPSS principal component analysis method is: after mean processing, the eigenvector of the covariance matrix is ​​obtained, the diagonal matrix satisfied by the matrix is ​​found, and the diagonal values ​​are arranged in descending order, and the number of principal components is selected to construct a matrix as a component matrix; clustering processing is performed on the obtained component matrix.

2. A prediction system for the probability of hematoma change, characterized in that: include: The raw data acquisition module is configured to: acquire brain image data and diagnosis and treatment history data of the corresponding patient; The data set construction module is configured to: extract the volume, position, shape and distribution information of hematoma and edema based on image data, associate it with the diagnosis and treatment history data, form a data set and preprocess it; the preprocessing includes: using the oversampling method in the imblearn library to deal with data imbalance, and reconstructing the data set through a random oversampling method; performing data standardization on the feature data sets of personal history, disease history, onset correlation, hematoma and edema volume, position and shape, and grayscale distribution, and converting the feature data into a distribution with a mean of 0 and a standard deviation of 1 by subtracting the mean and dividing by the standard deviation. The specific formula is: z = (x-μ) / σ, where z is the data after each column of patient data is standardized, x is the original data for different columns, μ is the mean of the original data for different columns, and σ is the standard deviation of the original data; The prediction model module is configured as follows: the preprocessed data set is formed into different training subsets by extracting different features, respectively predicting whether different types of hematoma change events occur, and obtaining the probability of hematoma changes in the input brain image data through clustering processing; The probability of a change in a hematoma in the input brain image data is obtained through clustering processing, specifically: based on the set factors in the patient's medical history data, the patients are divided into different groups, and the curves of the edema volume change over time of each group are obtained respectively, each data is curve-fitted to obtain the parameters of the fitting curve, and the relevant data are subjected to principal component analysis using SPSS. The SPSS principal component analysis method is: after mean processing, the eigenvector of the covariance matrix is ​​obtained, the diagonal matrix satisfied by the matrix is ​​found, and the diagonal values ​​are arranged in descending order, and the number of principal components is selected to construct a matrix as a component matrix; clustering processing is performed on the obtained component matrix.

3. The prediction system for hematoma change probability according to claim 2, characterized in that: The patient's medical history data is obtained by extracting text data from the patient's medical record to obtain the patient's personal information, medical history, and medical history.

4. The prediction system for hematoma change probability according to claim 2, characterized in that: Each training subset corresponds to a decision tree, and multiple decision trees form an integrated prediction model.

5. The prediction system for hematoma change probability according to claim 4, characterized in that: By determining the maximum tree depth, the number of trees, and the class weight parameters of the integrated prediction model, the optimal hyperparameters of the integrated prediction model are obtained through grid search.

6. The prediction system for hematoma change probability according to claim 1, characterized in that: In the obtained component matrix, k points are randomly initialized as cluster centroids, each point in the sample set is assigned to a cluster, and the distance between each point and the centroid is calculated, and it is assigned to the cluster corresponding to the nearest centroid. The centroid of the cluster is continued to be updated, and the centroid is updated to the average value of all points in the current cluster. It is iterated repeatedly until the set number of iterations is met to determine the probability of a change in the hematoma in the input brain image data.

7. A computer-readable storage medium, characterized in that: A computer program is stored thereon, which, when executed by a processor, implements the steps in the method for predicting the probability of hematoma change as described in claim 1.

8. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the method for predicting the probability of hematoma change as claimed in claim 1 are implemented.

Images