Method, device and storage medium for obtaining stroke prediction value

Abstract

The present invention provides a method, device and storage medium for obtaining a stroke prediction value, including: dividing a plurality of evaluation areas from a plain scan computer tomography image obtained from a brain scan; extracting texture features for each evaluation area to obtain a texture feature sample set; training the texture feature sample set, and performing five-fold cross-validation of the training results during the training process; and calculating a prediction value of a stroke risk based on a trained random forest model. The technical solution provided by the present invention can automatically evaluate and predict stroke using feature engineering and random forest technology, and accurately and reliably determine the ASPECTS score.

Description

Technical Field

[0001] The present application relates to the field of machine learning, and more specifically, to a method, device and storage medium for obtaining stroke prediction values. Background Art

[0002] In 2010, approximately 17 million people worldwide suffered from stroke, and 330 million people had a history of stroke and were still alive. In 2013, stroke was the second most common cause of death in the world, second only to coronary artery disease, accounting for about 12% of the total number of deaths, of which about 3.3 million people died of ischemic stroke. In my country, the situation is becoming more and more serious. Stroke is the leading cause of death and disability among adults in my country, and at least 1 in every 5 deaths is caused by stroke. With the acceleration of social aging and urbanization, unhealthy lifestyles among residents are prevalent, and cardiovascular risk factors are widely exposed. The burden of stroke in my country has an explosive growth trend, and it shows a rapid growth in low-income groups, obvious gender and regional differences, and a younger trend. Ischemic stroke refers to brain tissue dysfunction and necrosis caused by insufficient blood supply to the brain. There are four causes of ischemic stroke: thrombosis (blocking blood clots in the brain), embolism (embolism from elsewhere), systemic insufficiency (general systemic insufficiency of blood flow, such as shock), and cerebral venous sinus thrombosis. About half of people who have a stroke die within a year. Therefore, the long-term prognosis of ischemic stroke patients will have a significant impact on treatment decisions.

[0003] When evaluating the onset of ischemic stroke in patients, the current gold standard recognized by the clinical medical community is the Alberta stroke program early CT score (ASPECTS) based on non-contrast CT (NCCT). Machine learning technology is becoming increasingly suitable for use in the medical field due to its high accuracy. In recent years, increasing evidence has shown that the automatic ASPECTS scoring method based on machine learning is comparable to the expert reading of ASPECTS. ASPECTS is a simple method, but scoring early ischemic changes based on various indicators remains a challenge, especially for less experienced users. Various technical factors, such as X-ray energy peak image processing and display procedures; patient factors, such as old cerebral infarction, brain atrophy and leukomalacia; and factors such as experience, training and expertise may affect the parameters of ASPECTS.

[0004] Therefore, there is an urgent need for a scoring method that is simple to operate and can improve the precision of the ASPECTS score to further help and support doctors in interpreting brain scans of patients with acute ischemic stroke. Summary of the invention

[0005] The embodiment of the present invention provides a method, device and storage medium for obtaining a stroke prediction value, so as to solve the problem that the ASPECTS scoring in the prior art is complicated in operation and low in precision.

[0006] A first aspect of an embodiment of the present invention provides a method for obtaining a stroke prediction value, comprising:

[0007] Delineating multiple assessment regions from a plain-contrast computed tomography image obtained for a brain scan;

[0008] Extract texture features for each evaluation area to obtain a texture feature sample set;

[0009] Training the texture feature sample set, and performing five-fold cross-validation on the training results during the training process;

[0010] The predicted value of stroke risk was calculated based on the trained random forest model.

[0011] Optionally, in a possible implementation manner of the first aspect, before extracting texture features for each evaluation area to obtain a texture feature sample set, the method further includes:

[0012] A median filter process is performed on the regional image of each evaluation region in the plain scan computer tomography image.

[0013] Optionally, in a possible implementation manner of the first aspect, the training of the texture feature sample set and five-fold cross-validation of the training result during the training process include:

[0014] The texture features are divided into five test sets, and data not included in the respective test sets are used for training and validation;

[0015] Different variables were selected for the five random forest classifiers obtained after training and validation;

[0016] The average accuracy reduction values ​​of different variables were ranked from small to large, and the variable with the highest ranking was selected as the predictor variable.

[0017] Optionally, in a possible implementation manner of the first aspect, extracting texture features for each evaluation area to obtain a texture feature sample set includes:

[0018] The ischemic hemisphere was determined based on imaging and clinical parameters;

[0019] Regional feature difference values ​​were obtained by subtracting the regional values ​​of the ischemic hemisphere from those of the normal hemisphere.

[0020] Optionally, in a possible implementation manner of the first aspect, the training of the texture feature sample set and five-fold cross-validation of the training result during the training process include:

[0021] sorting the texture features based on linear discriminant analysis;

[0022] The sorted texture features are input into the random forest model;

[0023] Selecting training parameters using five-fold cross validation, the training parameters including: the number of trees in the forest of each region, the maximum depth of the trees, and the number of texture features within a preset ranking;

[0024] Set different training weights according to the ratio of abnormal data to normal data in the training data;

[0025] The training parameters are trained by the random forest model.

[0026] Optionally, in a possible implementation of the first aspect, the calculation of the predicted value of stroke risk based on the trained random forest model includes: using the trained random forest model to calculate the ASPECTS score of each assessment area, and weighting the ASPECTS scores of all assessment areas to obtain a final predicted value, wherein the weight value is pre-set.

[0027] A second aspect of an embodiment of the present invention provides a device for obtaining a stroke prediction value, comprising:

[0028] A segmentation module for segmenting a plurality of evaluation regions from a plain scan computed tomography image obtained for a brain scan;

[0029] A feature extraction module is used to extract texture features for each evaluation area to obtain a texture feature sample set;

[0030] A model training module, used for training the texture feature sample set and performing five-fold cross-validation on the training results during the training process;

[0031] The prediction module is used to calculate the predicted value of stroke risk based on the trained random forest model.

[0032] Optionally, in a possible implementation manner of the second aspect, the feature extraction module is further used to perform the following steps, including:

[0033] A median filter process is performed on the regional image of each evaluation region in the plain scan computer tomography image.

[0034] Optionally, in a possible implementation of the second aspect, the model training module is also used to perform the following steps, including: dividing the texture features into five test sets, and using data not included in each test set for training and verification; selecting different variables for the five random forest classifiers obtained after training and verification; sorting the average accuracy reduction values ​​of different variables from small to large, and selecting the variable with the highest ranking as the prediction variable.

[0035] Optionally, in a possible implementation manner of the second aspect, a difference value calculation module is further included, and the difference value calculation module is used to perform the following steps, including:

[0036] The ischemic hemisphere was determined based on imaging and clinical parameters; regional feature difference values ​​were obtained by subtracting the regional values ​​of the ischemic hemisphere from those of the normal hemisphere.

[0037] Optionally, in a possible implementation manner of the second aspect, the model training module is further used to perform the following steps, including:

[0038] sorting the texture features based on linear discriminant analysis;

[0039] The sorted texture features are input into the random forest model;

[0040] Selecting training parameters using five-fold cross validation, the training parameters including: the number of trees in the forest of each region, the maximum depth of the trees, and the number of texture features within a preset ranking;

[0041] Set different training weights according to the ratio of abnormal data to normal data in the training data;

[0042] The training parameters are trained by the random forest model.

[0043] Optionally, in a possible implementation manner of the second aspect, the prediction module is further configured to perform the following steps, including:

[0044] The trained random forest model is used to calculate the ASPECTS score of each evaluation area, and the ASPECTS scores of all evaluation areas are weighted to obtain the final prediction value, and the weight value is preset.

[0045] According to a third aspect of an embodiment of the present invention, a readable storage medium is provided, in which a computer program is stored. When the computer program is executed by a processor, it is used to implement the method of the first aspect of the present invention and various possible designs of the first aspect.

[0046] The stroke prediction value acquisition method, device and storage medium provided by the present invention can predict the risk of stroke using computed tomography technology, automatically evaluate and predict stroke using feature engineering and random forest technology, and accurately and reliably determine the ASPECTS score, further helping and supporting doctors to interpret brain scans of patients with acute ischemic stroke. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A flowchart of a first implementation method of a method for obtaining a stroke prediction value;

[0048] Figure 2 A flowchart of a first implementation method of a method for obtaining a stroke prediction value;

[0049] Figure 3 Label the ASPECTS area map;

[0050] Figure 4 Export comparison chart for ASPECTS

[0051] Figure 5 This is a structural diagram of a first embodiment of a device for acquiring a stroke prediction value. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in sequences other than those illustrated or described herein.

[0054] It should be understood that in various embodiments of the present invention, the size of the sequence number of each process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0055] It should be understood that in the present invention, "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0056] It should be understood that in the present invention, "plurality" refers to two or more than two. "And / or" is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "Contains A, B and C", "Contains A, B, C" means that A, B, and C are all included, "Contains A, B or C" means that one of A, B, and C is included, and "Contains A, B and / or C" means that any one, any two, or any three of A, B, and C are included.

[0057] It should be understood that in the present invention, "B corresponding to A", "B corresponding to A", "A corresponds to B" or "B corresponds to A" means that B is associated with A and B can be determined based on A. Determining B based on A does not mean determining B based only on A, but B can also be determined based on A and / or other information. A and B match when the similarity between A and B is greater than or equal to a preset threshold.

[0058] Depending on the context, "if" as used herein may be interpreted as "when" or "when" or "in response to determining" or "in response to detecting."

[0059] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0060] The ASPECTS scoring method of the present invention is developed based on feature engineering and random forests. As one of the latest and most popular enhancement methods, random forests have demonstrated classification performance for difficult problems in many medical image analysis applications compared to other classifiers. Random forest is an integrated learning method that combines multiple weak classifiers (decision trees) and lets these decision trees vote for the most popular categories. Each tree in the forest relies on an independently sampled random vector, and all trees in the forest have the same distribution. The growth of the tree is controlled by the random vector. A randomness measure is introduced in the training to prevent the training classifier from falling into the local minimum, thereby improving accuracy and reducing the chance of overfitting.

[0061] The present invention provides a method for obtaining a stroke prediction value, such as Figure 1 and Figure 2 The flow chart shown includes:

[0062] Step S110, a plurality of evaluation regions are divided from a plain scan computed tomography image obtained from a brain scan. The plain scan computed tomography image is also called a plain scan CT, NCCT, which is a useful screening tool in the acute stroke workflow; NCCT is also used to measure the extent of early ischemic changes (EIC) in the ischemic brain. The Alberta Stroke Project Early CT Score (ASPECTS) is a simple and systematic method for evaluating EIC within the middle cerebral artery region of NCCT. In this step, if more than 20% of a region in the NCCT image is occupied by diffusion, it is considered to be an affected region.

[0063] Step S120: extract texture features from each evaluation area to obtain a texture feature sample set.

[0064] In step S120, the affected area needs to be subjected to median filtering before feature extraction. After median filtering, 376 texture features can be extracted from each ASPECTS area of ​​each patient image, such as high-order statistics and image texture features, image texture coarseness, density and other features. Common texture features include gray-level co-occurrence matrix, autoregressive texture model, Tamura texture features, wavelet transform, etc. Feature extraction and classification of each ASPECTS area are performed in 3D.

[0065] Step S130, training the texture feature sample set, and five-fold cross-validation of the training results during the training process. In this step, cross-validation is to group the original data in a certain sense, with one part as the training set and the other as the validation set. First, the classifier is trained with the training set, and then the trained model is tested with the validation set, which is used as a performance indicator for evaluating the classifier. In the present invention, five-fold cross-validation is used, that is, the initial data set is divided into five parts, and four of them are used for training and one for validation in turn. The average of the five results is used as an estimate of the accuracy of the algorithm. Generally, multiple five-fold cross-validation is required to calculate the average.

[0066] Step S140: Calculate the predicted value of stroke risk based on the trained random forest model.

[0067] In one embodiment, before extracting texture features for each evaluation area and obtaining a texture feature sample set, it also includes: performing median filtering on the regional image of each evaluation area in the plain scan computer tomography image. In this step, the median filtering process adopted is a filtering method that averages the neighborhood values ​​of the pixel points of the image; the extracted texture signs are the characteristics of the image itself, such as the average, maximum, and median values ​​of the CT values ​​of the area; the frequency distribution of different CT values; high-order features such as the image after being processed by a certain matrix operator, and then extracting the above information, and the extraction method is to perform mathematical and matrix operations on the segmented image.

[0068] In one embodiment, the texture feature sample set is trained and the training results are cross-validated five-fold during the training process, including: dividing the texture features into five test sets, and using data not included in each test set for training and validation; selecting different variables for the five random forest classifiers obtained after training and validation; sorting the average accuracy reduction values ​​of different variables from small to large, and selecting the variable with the highest ranking as the predictor variable. In the step, the importance of the variables is evaluated by sorting the average accuracy reduction of the variables and then selecting the most important predictor variable for outcome prediction.

[0069] In one embodiment, extracting texture features for each evaluation region to obtain a texture feature sample set includes: determining the ischemic hemisphere based on imaging and clinical parameters; and obtaining a regional feature difference value by subtracting the regional value of the ischemic hemisphere from the regional value of the normal hemisphere.

[0070] In one embodiment, the texture feature sample set is trained, and the training results are cross-validated five-fold during the training process, including: sorting the texture features based on linear discriminant analysis; inputting the sorted texture features into a random forest model; using five-fold cross-validation to select training parameters, the training parameters include: the number of trees in the forest of each area, the maximum depth of the tree, and the number of texture features within the preset ranking of the sorting; setting different training weights according to the ratio of abnormal data to normal data in the training data; training parameters are trained through a random forest model. In this step, after the texture features are calculated, linear discriminant analysis (LDA) is used to sort the texture features, and the sorted features are input into the random forest model for training. In addition, five-fold cross-validation is used to train parameters, wherein the training parameters also include using the ASPECTS scores made by experts on nuclear magnetic DWI images as labels, which are used together as training data.

[0071] In one embodiment, the prediction value of the stroke risk based on the trained random forest model includes: using the trained random forest model to calculate the ASPECTS score of each evaluation area, and weighting the ASPECTS scores of all evaluation areas to obtain the final prediction value, and the weight value is preset. In this step, the random forest model obtained by training the texture feature sample set is used to calculate the ASPECTS score of each evaluation area, and the evaluation area includes: lenticular nucleus L, insula I, caudate nucleus C, internal capsule IC, middle cerebral artery M, wherein the middle cerebral artery is divided into: M1, M2, M3, M4, M5, M6; a total of 10 evaluation areas; then the ASPECTS score calculated for each evaluation area is set to a corresponding weight value, and the initial score of each area is multiplied by the respective weight value; then the ASPECTS scores of the 10 evaluation areas are summed to obtain the final score. Considering that different evaluation areas have different effects on the risk of stroke, corresponding weight values ​​are set according to the size of the impact of different evaluation areas on the risk of stroke, so that the score value calculated by weighted summation is more accurate.

[0072] In addition, the ASPECTS scoring method: the full score is 10 points; among the 10 segmented areas, if one area is affected (more than 20% is occupied by diffusion), 1 point is deducted. The lower the score, the larger the early ischemic lesion. When the score is less than or equal to 4, the risk of ischemic stroke is considered high.

[0073] The technical solution provided by the present invention is developed based on machine learning to detect early signs of ischemic damage on CT. It uses ASPECTS values ​​as output. The present invention can select different parameters or algorithm configurations, namely the operating point (OP), to determine the probability of ischemic stroke in patients. The indicators obtained by the invention are simple and reliable, and are powerful artificial intelligence predictive indicators for functional outcomes and symptomatic cerebral hemorrhage after thrombolytic therapy. Existing ASPECTS automated methods generally use first-order image features, such as Hounsfield Unit (HU), or density and HU differences between the ischemic side and the contralateral side as features of their algorithms, but these first-order image features have limitations in patients with smaller ischemic changes and patients with lower signal-to-noise ratios and more obvious motion artifacts in the image; in the present automatic ASPECTS method, multiple higher-order computational texture features are used as part of the machine learning algorithm, which can help improve the effectiveness of the technology, such as Figure 3 As shown, an example of automatically labeling each ASPECTS region using the present invention, L represents the lentiform nucleus, I represents the insula, C represents the caudate nucleus, IC represents the internal capsule, and M represents the middle cerebral artery.

[0074] The present invention simulates the behavior of different physicians by running with different configurations. Studies have shown that the present invention can be adjusted to a suitable working point so that its performance (including sensitivity and score) in correctly identifying a single ischemic ASPECTS area is significantly better than all trainees and some experts. The difference in specificity is small (<5%), but it is important because most of the data are unaffected areas, which dominate when calculating specificity. The difference in sensitivity is large (the mean difference is 23.6%), which illustrates the different behaviors of trainees and experts, that is, the present invention has a relatively large sensitivity improvement at the expense of a small reduction in specificity relative to experts. The present technology uses two more advanced measurement methods to describe the present invention, and the consistency of ASPECTS with DWI ground truth data is the best compared with experts and trainees. Compared with manual measurement, the Matthews correlation coefficient (MCC) of the present invention is 0.44, which also shows the value closest to 1. The consistency of these results obtained using different methodological methods further enhances the robustness of the findings of the present technology.

[0075] Example verification:

[0076] For verification, the present invention measured 100 patients, and the intraclass correlation coefficient between the automatic ASPECTS measurement method and the DWI ASPECTS expert reading score was 0.76 (95% confidence interval, 0.67-0.83), and the average ASPECTS difference value of the consistency evaluation diagram was 0.3 (consistency limit, -3.3, 2.6). A separate ASPECTS regional level analysis showed that for 100×10 (each patient was measured 10 times separately) ASPECTS areas, the present method obtained a linear weighted κ=0.60, a sensitivity of 66.2%, a specificity of 91.8%, and an area under the curve of 0.79. In addition, when ASPECTS was divided into two groups according to a given threshold (>4 or ≤4), κ=0.78, and the method had a sensitivity of 97.8%, a specificity of 80%, and an area under the curve of 0.89 compared with the expert reading of DWI. The ICC (intraclass correlation coefficient) of total ASPECTS between expert-rated NCCT and automatic CT ASPECTS was 0.61 (95% confidence interval, 0.47-0.72). The conflict between expert-rated CT ASPECTS and automatic CT ASPECTS was moderate (κ = 0.48; 95% confidence interval, 0.28–0.68), and the ASPECTS threshold for dichotomy was >4≤. Application examples include Figure 4As shown, the left picture shows the automatic export of ASPECTS using the present invention, and the right picture shows the manual export of ASPECTS by imaging experts.

[0077] Therefore, the parameters derived by the automatic ASPECTS method of the present invention are very consistent with the DWIASPECTS within the region read by the expert and the total ASPECTS. In addition, the good agreement between the ASPECTS cut-off dichotomy threshold (>4 pairs or ≤4) of the DWI ASPECTS read by the automatic method and the expert helps to assess whether there is a large area of ​​infarction at baseline in the patient. These results also show that the automatic ASPECTS method is not inferior to the ASPECTS read by the expert on NCCT.

[0078] The present invention showed better performance using ASPECTS in CT scoring of patients with acute ischemic stroke. These findings were supported by several different analyses that evaluated the agreement between the present invention and physicians using ground truth DWI data. The present invention is a promising tool for clinical practice and clinical decision making, especially in the upcoming situation of triaging patients due to limited resources for endovascular stroke treatment. According to the test results, patients with ASPECTS>5 on CT can be successfully selected for thrombectomy. The purpose of this software is not to replace the expert evaluation of the scan, but to assist the physician in clinical decision making and research. The advantage lies in the potential time efficiency improvement using a relatively unbiased and standardized method to triage neuroradiologists or experienced stroke clinicians. It can be concluded that the present technology is already comparable to the manual reading parameters derived by ASPECTS experts.

[0079] The embodiment of the present invention also provides a device for obtaining a stroke prediction value, such as Figure 5 As shown, including:

[0080] A segmentation module for segmenting a plurality of evaluation regions from a plain scan computed tomography image obtained for a brain scan;

[0081] A feature extraction module is used to extract texture features for each evaluation area to obtain a texture feature sample set;

[0082] A model training module, used for training the texture feature sample set and performing five-fold cross-validation on the training results during the training process;

[0083] The prediction module is used to calculate the predicted value of stroke risk based on the trained random forest model.

[0084] In one embodiment, the feature extraction module is further used to perform the following steps, including: performing median filtering on the regional image of each evaluation area in the plain scan computed tomography image.

[0085] In one embodiment, it also includes a difference value calculation module, which is used to perform the following steps, including: determining the ischemic cerebral hemisphere based on imaging and clinical parameters; obtaining the regional feature difference value by subtracting the regional value of the ischemic cerebral hemisphere from the regional value of the normal cerebral hemisphere.

[0086] In one embodiment, the model training module is also used to perform the following steps, including: sorting the texture features based on linear discriminant analysis; inputting the sorted texture features into a random forest model; selecting training parameters using five-fold cross validation, the training parameters including: the number of trees in the forest in each area, the maximum depth of the tree, and the number of texture features within a preset ranking in the sorting; setting different training weights according to the ratio of abnormal data to normal data in the training data; and training the training parameters through a random forest model.

[0087] In one embodiment, the prediction module is further used to perform the following steps, including: using the trained random forest model to calculate the ASPECTS score of each evaluation area, and weighting the ASPECTS scores of all evaluation areas to obtain a final prediction value, where the weight value is pre-set and the weight values ​​of different evaluation areas are different.

[0088] Among them, the readable storage medium can be a computer storage medium or a communication medium. The communication medium includes any medium that facilitates the transmission of a computer program from one place to another. The computer storage medium can be any available medium that can be accessed by a general or special-purpose computer. For example, a readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an application-specific integrated circuit (ASIC). In addition, the ASIC can be located in a user device. Of course, the processor and the readable storage medium can also exist in a communication device as discrete components. The readable storage medium can be a read-only memory (ROM), a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device, etc.

[0089] The present invention also provides a program product, which includes an execution instruction, which is stored in a readable storage medium. At least one processor of a device can read the execution instruction from the readable storage medium, and at least one processor executes the execution instruction so that the device implements the methods provided in the above various embodiments.

[0090] In the above-mentioned terminal or server embodiment, it should be understood that the processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the present invention can be directly embodied as being executed by a hardware processor, or can be executed by a combination of hardware and software modules in the processor.

[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for obtaining a stroke prediction value, using computed tomography to predict the risk of stroke, characterized in that: include: Delineating multiple assessment regions from a plain-contrast computed tomography image obtained for a brain scan; Extract texture features for each evaluation area to obtain a texture feature sample set; The random forest model is trained according to the texture feature sample set, and the training results are cross-validated five times during the training process; Calculate the predicted value of stroke risk based on the trained random forest model; The step of extracting texture features from each evaluation area to obtain a texture feature sample set includes: The ischemic hemisphere was determined based on imaging and clinical parameters; Regional feature difference values ​​were obtained by subtracting the regional values ​​of the ischemic hemisphere from those of the normal hemisphere; The random forest model is trained according to the texture feature sample set, and the training result is five-fold cross-validated during the training process, including: sorting the texture features based on linear discriminant analysis; The sorted texture features are input into the random forest model; Selecting training parameters using five-fold cross validation, the training parameters including: the number of trees in the forest of each region, the maximum depth of the trees, and the number of texture features within a preset ranking; Set different training weights according to the ratio of abnormal data to normal data in the training data; The training parameters are trained through the random forest model; The predicted value of stroke risk is calculated based on the trained random forest model, including: The random forest model trained by the texture feature sample set was used to calculate the ASPECTS score for each evaluation area. The evaluation areas included the lentiform nucleus L, insula I, caudate nucleus C, internal capsule IC, and middle cerebral artery M, where the middle cerebral artery was divided into M1, M2, M3, M4, M5, and M6. There were 10 evaluation areas in total. The ASPECTS score calculated for each evaluation area was then set to a corresponding weight value, and the initial score of each area was multiplied by its own weight value. The ASPECTS scores of the 10 evaluation areas were then summed to obtain the final score.

2. The method for obtaining a stroke prediction value according to claim 1, characterized in that: Before extracting texture features from each evaluation area to obtain a texture feature sample set, the method further includes: A median filter process is performed on the regional image of each evaluation region in the plain scan computer tomography image.

3. The method for obtaining a stroke prediction value according to claim 1, characterized in that: The random forest model is trained according to the texture feature sample set, and the training result is five-fold cross-validated during the training process, including: The texture features are divided into five test sets, and data not included in the respective test sets are used for training and validation; Different variables were selected for the five random forest classifiers obtained after training and validation; The average accuracy reduction values ​​of different variables were ranked from small to large, and the variable with the highest ranking was selected as the predictor variable.

4. The method for obtaining stroke prediction value according to claim 1, characterized in that: The method of calculating the predicted value of stroke risk based on the trained random forest model includes: The trained random forest model is used to calculate the ASPECTS score of each evaluation area, and the ASPECTS scores of all evaluation areas are weighted to obtain the final prediction value. The weight value is pre-set and the weight value of different evaluation areas is different.

5. A device for obtaining a stroke prediction value, characterized in that: include: A segmentation module for segmenting a plurality of evaluation regions from a plain scan computed tomography image obtained for a brain scan; A feature extraction module is used to extract texture features for each evaluation area to obtain a texture feature sample set; A model training module, used for training a random forest model according to the texture feature sample set, and performing five-fold cross-validation on the training results during the training process; A prediction module is used to calculate the predicted value of stroke risk based on the trained random forest model; The step of extracting texture features from each evaluation area to obtain a texture feature sample set includes: The ischemic hemisphere was determined based on imaging and clinical parameters; Regional feature difference values ​​were obtained by subtracting the regional values ​​of the ischemic hemisphere from those of the normal hemisphere; The random forest model is trained according to the texture feature sample set, and the training result is five-fold cross-validated during the training process, including: sorting the texture features based on linear discriminant analysis; The sorted texture features are input into the random forest model; Selecting training parameters using five-fold cross validation, the training parameters including: the number of trees in the forest of each region, the maximum depth of the trees, and the number of texture features within a preset ranking; Set different training weights according to the ratio of abnormal data to normal data in the training data; The training parameters are trained through the random forest model; The predicted value of stroke risk is calculated based on the trained random forest model, including: The random forest model trained by the texture feature sample set was used to calculate the ASPECTS score for each evaluation area. The evaluation areas included the lentiform nucleus L, insula I, caudate nucleus C, internal capsule IC, and middle cerebral artery M, where the middle cerebral artery was divided into M1, M2, M3, M4, M5, and M6. There were 10 evaluation areas in total. The ASPECTS score calculated for each evaluation area was then set to a corresponding weight value, and the initial score of each area was multiplied by its own weight value. The ASPECTS scores of the 10 evaluation areas were then summed to obtain the final score.

6. The stroke prediction value acquisition device according to claim 5, characterized in that: The system further comprises a filtering processing module, wherein the filtering processing module is used to perform the following steps, including: A median filter process is performed on the regional image of each evaluation region in the plain scan computer tomography image.

7. A readable storage medium, characterized in that: The readable storage medium stores a computer program, and the computer program is used to implement any one of the methods of claims 1 to 4 when executed by a processor.

Images