A stroke prognosis prediction method, system, computer device and storage medium

Abstract

The application provides a stroke prognosis prediction method, system, computer equipment and storage medium, the method comprises: obtaining the to-be-analyzed feature data of a stroke patient; the to-be-analyzed feature data comprises age, gender, native place, mRS score at discharge, stroke-related medical history characteristics and admission and discharge NIHSS score change characteristics; the to-be-analyzed data is input into a pre-constructed stroke prognosis prediction model for prognosis prediction, and a corresponding prognosis prediction result is obtained; the stroke prognosis prediction model is trained based on a data set having the same characteristics as the to-be-analyzed feature data. The application trains and constructs a stroke prognosis prediction model by screening key data characteristics which are highly available and can be used for precise prediction of different post-stroke outcomes at the same time, can effectively improve the efficiency and accuracy of post-stroke outcome prediction, guarantees the generalization and ease of use of the model, and further provides reliable technical support for health management of stroke patients.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of stroke prognosis prediction, in particular to a stroke prognosis prediction method, system, computer device and storage medium. BACKGROUND

[0002] Stroke has the characteristics of high incidence, high disability, high mortality, high recurrence and limited treatment time window, and is the most important cause of death and disability in the world. Its treatment cost also causes a huge economic burden to the country, society and family. Timely and reliable stroke prognosis prediction is crucial for the health management of stroke patients.

[0003] The existing stroke prognosis prediction methods mainly include establishing a stroke outcome prediction model for predicting the outcome of stroke patients based on traditional regression analysis methods and machine learning methods. However, the existing stroke outcome prediction model has great application limitations and is difficult to effectively promote the use: 1) the construction of the model mainly depends on the imaging data or laboratory examination results of stroke patients, and such indicators involve patient privacy, which is less available in actual application; 2) different data features need to be used to construct different stroke outcome prediction models for different stroke outcome prediction, which cannot meet the convenience of predicting different stroke outcomes with the same features; 3) the Area Under Curve (ACU) of the model prediction is poor, which cannot meet the demand of clinical application. Therefore, it is urgent to provide a stroke prognosis prediction method that does not completely rely on imaging data or laboratory examination results and meets the common features of different stroke outcome prediction with high reliability. SUMMARY

[0004] The purpose of the present application is to provide a stroke prognosis prediction method, which screens out key data features that do not depend on imaging data or laboratory examination results, have strong data availability and can be used for precise prediction of different stroke outcomes at the same time through analysis of stroke research data sets, and trains and constructs an efficient, precise and high-risk identification stroke prognosis prediction model based on the key data features for predicting the prognosis of stroke patients. The use can effectively improve the efficiency and accuracy of stroke outcome prediction, while ensuring the generalization and ease of use of the prediction model, facilitating wide application and use, and providing reliable technical support for the health management of stroke patients.

[0005] In order to achieve the above purpose, it is necessary to provide a stroke prognosis prediction method, system, computer device and storage medium to solve the above technical problems.

[0006] In a first aspect, the present application provides a stroke prognosis prediction method, which comprises the following steps:

[0007] obtaining stroke patient data to be analyzed; the data to be analyzed includes age, gender, hometown, mRS score at discharge, stroke-related medical history characteristics, and admission and discharge NIHSS score change characteristics;

[0008] inputting the data to be analyzed into a pre-constructed stroke prognosis prediction model for prognosis prediction, to obtain corresponding prognosis prediction results; the stroke prognosis prediction model is trained based on a data set having the same characteristics as the data to be analyzed.

[0009] Further, the stroke-related medical history characteristics include whether there is a history of transient ischemic attack, a history of diabetes, and a history of transient ischemic attack; the admission and discharge NIHSS score change characteristics include the total NIHSS score change value and the corresponding several admission and discharge NIHSS single score change values.

[0010] Further, the construction steps of the stroke prognosis prediction model include:

[0011] obtaining a stroke research data set; the stroke research data set includes a first stroke patient data set and a second stroke patient data set; the first stroke patient data set and the second stroke patient data set both include age, gender, hometown, mRS score at discharge, stroke-related medical history characteristics, admission NIHSS total score and corresponding several admission NIHSS single scores, discharge NIHSS total score and corresponding several discharge NIHSS single scores;

[0012] After preprocessing the stroke research data set, a first training set, a first external validation set, and a second external validation set are obtained;

[0013] According to the first training set and the XGBoost model, stroke prognosis prediction key features are extracted from the data characteristics of the first training set;

[0014] According to the stroke prognosis prediction key features and the corresponding prognosis outcomes in the first training set, the first external validation set, and the second external validation set, respectively, corresponding second training sets, first validation sets, and second validation sets are obtained;

[0015] According to the second training set, the XGBoost model is trained, and through K-fold cross-validation, an optimal parameter model is obtained, to obtain the stroke prognosis prediction model, and according to the first validation set and the second validation set, the stroke prognosis prediction model is verified by combining homologous data and non-homologous data, to obtain a prediction model verification result.

[0016] Further, after preprocessing the stroke patient data in the stroke research data set, the first training set, the first external validation set, and the second external validation set are obtained.

[0017] differences between the total NIHSS score at discharge and the total NIHSS score at admission in each piece of stroke patient data in the stroke research dataset, and differences between the individual NIHSS score at discharge and the individual NIHSS score at admission, to obtain corresponding NIHSS score changes at admission and discharge;

[0018] add the NIHSS score changes at admission and discharge and the corresponding NIHSS individual score changes at admission and discharge in each piece of stroke patient data in the stroke research dataset to the stroke research dataset, to obtain an updated stroke research dataset;

[0019] divide the first stroke patient data set part in the updated stroke research dataset into a first training set and a first external validation set, and randomly select a preset number of patient data from the second stroke patient data set part in the updated stroke research dataset to obtain a second external validation set.

[0020] Further, the step of extracting stroke prognosis prediction key features from the data features of the first training set according to the first training set and the XGBoost model comprises:

[0021] training the XGBoost model according to the first training set, and obtaining a corresponding feature importance ranking table after training is completed;

[0022] obtaining first stroke prognosis prediction important features according to the feature importance ranking table; the first stroke prognosis prediction important features include NIHSS score changes at admission and discharge and mRS score at discharge;

[0023] based on the availability of features in clinical analysis, obtaining second stroke prognosis prediction important features from the data features of the first training set; the second stroke prognosis prediction important features include age, gender, place of origin and stroke-related medical history features;

[0024] combining the first stroke prognosis prediction important features and the second stroke prognosis prediction important features to obtain the stroke prognosis prediction key features.

[0025] Further, the step of training the XGBoost model according to the second training set, obtaining an optimal parameter model through K-fold cross-validation to obtain the stroke prognosis prediction model, and combining homologous data and non-homologous data to verify the stroke prognosis prediction model according to the first validation set and the second validation set to obtain the prediction model verification result comprises:

[0026] According to the training set and the internal and external K-fold cross-validation method, the XGBoost model is trained, and the optimal parameters of the XGBoost model are obtained through grid search, so as to obtain a stroke prognosis prediction model corresponding to the optimal parameters; the optimal parameters include a learning rate and a maximum depth;

[0027] According to the first validation set, the stroke prognosis prediction model is verified by homologous data, and a corresponding model homologous data verification result is obtained;

[0028] According to the second validation set, the stroke prognosis prediction model is verified by non-homologous data, and a corresponding model heterologous data verification result is obtained.

[0029] Further, the method further comprises:

[0030] According to the feature importance ranking, corresponding rehabilitation exercise guidance suggestions are generated.

[0031] In a second aspect, an embodiment of the present application provides a stroke prognosis prediction system, and the system comprises:

[0032] A data acquisition module is configured to acquire feature data of a stroke patient to be analyzed; the feature data to be analyzed comprises age, gender, place of origin, mRS score at discharge, stroke-related medical history features, and admission and discharge NIHSS score change features.

[0033] A prognosis prediction module is configured to input the data to be analyzed into a pre-constructed stroke prognosis prediction model to perform prognosis prediction, so as to obtain a corresponding prognosis prediction result; the stroke prognosis prediction model is trained based on a data set having the same features as the feature data to be analyzed.

[0034] In a third aspect, an embodiment of the present application further provides a computer device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the above method when executing the computer program.

[0035] In a fourth aspect, an embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of the above method.

[0036] The present invention provides a method, system, computer device, and storage medium for predicting stroke prognosis. The method acquires characteristic data of stroke patients, including age, gender, place of origin, mRS score at discharge, stroke-related medical history characteristics, and changes in NIHSS scores from admission to discharge. This data is then input into a stroke prognosis prediction model pre-trained on a dataset with the same characteristics as the target data for prognosis prediction, yielding the corresponding prognosis prediction result. Compared with existing technologies, this stroke prognosis prediction method, through research and analysis of stroke research datasets, selects key data features that are independent of imaging data or laboratory test results, have high data availability, and can be applied simultaneously to accurately predict different post-stroke outcomes. Based on these key data features, an efficient, accurate, and highly risk-discriminating stroke prognosis prediction model is trained for use in predicting the prognosis of stroke patients. This not only effectively improves the efficiency and accuracy of stroke outcome prediction but also ensures the generalizability and ease of use of the prediction model, facilitating widespread adoption and providing reliable technical support for the health management of stroke patients. Attached Figure Description

[0037] Figure 1 This is a flowchart illustrating the stroke prognosis prediction method in an embodiment of the present invention.

[0038] Figure 2 This is a schematic diagram of the feature data to be analyzed in an embodiment of the present invention;

[0039] Figure 3 This is an embodiment of the present invention representing the intentional ranking of the importance of features for screening important features for predicting the prognosis of the first stroke.

[0040] Figure 4 This is a schematic diagram showing the comparison of prediction results between the DISCO model and the existing clinically commonly used RRE model in this embodiment of the invention.

[0041] Figure 5 This is a schematic diagram illustrating the effectiveness of the DISCO model in stratifying relapse risk over a period of 3 months on the second training set, the first validation set, and the second validation set in an embodiment of the present invention.

[0042] Figure 6 This is another flowchart illustrating the stroke prognosis prediction method in this embodiment of the invention;

[0043] Figure 7 This is a schematic diagram of the feature importance ranking results of the DISCO model obtained by SHAP attribution analysis in an embodiment of the present invention;

[0044] Figure 8 This is a schematic diagram of the stroke prognosis prediction system in an embodiment of the present invention;

[0045] Figure 9 is another structural schematic diagram of a stroke prognosis prediction system in an embodiment of the present application;

[0046] Figure 10 is an internal structure diagram of a computer device in an embodiment of the present application. DETAILED DESCRIPTION

[0047] In order to make the objectives, technical solutions and beneficial effects of the present application clearer, further detailed description will be made to the present application in combination with the accompanying drawings and embodiments. Obviously, the following described embodiments are only a part of the embodiments of the present application, and are only used to explain the present application, but not to limit the scope of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0048] In one embodiment, as shown in Figure 1 , a stroke prognosis prediction method is provided, comprising the following steps:

[0049] S11, obtaining the to-be-analyzed feature data of a stroke patient; wherein the to-be-analyzed feature data can be understood as the discharge patient data indexes that can be used for analyzing different prognosis outcomes, which are obtained based on the relevant report information of the stroke patient at the time of discharge and admission, and are based on the stroke prognosis prediction key features obtained by analyzing the existing stroke patient research data set; in this embodiment, the to-be-analyzed feature data preferably includes age, gender, native place, mRS (modified Rankin Score) score at discharge, stroke-related medical history features and admission and discharge NIHSS score change features. Wherein the stroke-related medical history features can be understood as the relevant medical history data that may affect the development of stroke, specifically, the stroke-related medical history features include whether there is a history of transient ischemic attack, whether there is a history of diabetes and whether there is a history of transient ischemic attack, in actual application, the values of such indexes are 0 or 1, 0 represents no, 1 represents yes, which can be obtained based on the patient's complaint or registration information.

[0050] The admission and discharge NIHSS (National Institutes of Health Stroke Scale) score change features in this embodiment can be understood as the data change information obtained by comparing and analyzing the NIHSS score table (U.S. National Institutes of Health Stroke Scale) detected at the time of discharge and admission of the patient, that is, by calculating the difference between the data at the time of discharge and the data at the time of admission; specifically, the admission and discharge NIHSS score change features include the admission and discharge NIHSS total score change value and a plurality of corresponding admission and discharge NIHSS single score change values, and the number of admission and discharge NIHSS single score change values can be selected according to actual application requirements, in order to ensure the comprehensiveness of the analysis, for example,Figure 2 As shown, the single item score change value can include 15 aspects in the NIHSS scale, such as consciousness, speech, visual field, facial movement, upper limb movement, lower limb movement, coordinated movement, sensation, language understanding, and reading.

[0051] S12, input the to-be-analyzed data into a pre-constructed stroke prognosis prediction model for prognosis prediction, to obtain a corresponding prognosis prediction result; the construction principle of the stroke prognosis prediction model can be obtained by using an existing machine learning model construction method capable of realizing prognosis prediction analysis, but in order to ensure the efficiency, accuracy and generalization of the prediction analysis, the embodiment preferably uses the XGBoost algorithm to analyze and screen the existing multiple classic stroke research data sets for stroke prognosis prediction key features, and then trains and verifies the XGBoost model based on the prognosis prediction key features of different data sets to obtain the final required stroke prognosis prediction model. Specifically, the construction steps of the stroke prognosis prediction model include:

[0052] obtain a stroke research data set; the stroke research data set includes a first stroke patient data set and a second stroke patient data set; the first stroke patient data set and the second stroke patient data set both include age, gender, hometown, mRS score at discharge, stroke-related medical history characteristics, total NIHSS score at admission and corresponding several NIHSS single item scores at admission, total NIHSS score at discharge and corresponding several NIHSS single item scores at discharge. In order to ensure the reliability of the research data, the first stroke patient data set in the embodiment is the ischemic stroke patient data after excluding the data of patients with transient ischemic attack (TIA) from the China National Stroke Registry-III (CNSR-III, a total of 15166 patients were included, which is a nationwide stroke or TIA registry, involving 201 hospitals in China, and the registration time is from August 2015 to March 22, 2018), and the second stroke patient data set is the ischemic stroke patient data after excluding the data of patients with TIA from the CHANCE-2 (Clopidogrel plus Aspirin versus Aspirin Alone in Patients With Acute Ischemic Stroke or Transient Ischemic Attack) data. It should be noted that the admission NIHSS single item scores and the discharge NIHSS single item scores in the first stroke patient data set and the second stroke patient data set are consistent with the NIHSS single item scores in the aforementioned to-be-analyzed feature data.

[0053] After preprocessing the stroke research data set, a first training set, a first external validation set and a second external validation set are obtained; wherein the preprocessing mainly includes comparing and analyzing the admission NIHSS total score and the corresponding several admission NIHSS single item scores, the discharge NIHSS total score and the corresponding several discharge NIHSS single item scores of each patient, obtaining the admission-discharge NIHSS total score change value and the corresponding several admission-discharge NIHSS single item score change values, which are used to add to the current data set to obtain a new data set; specifically, the step of obtaining the training set, the first external validation set and the second external validation set after preprocessing the stroke patient data in the stroke research data set includes:

[0054] The discharge NIHSS total score and the admission NIHSS total score in each stroke patient data in the stroke research data set are subtracted to obtain the corresponding admission-discharge NIHSS total score change value, and the corresponding discharge NIHSS single item score and the admission NIHSS single item score are subtracted to obtain the corresponding admission-discharge NIHSS single item score change value;

[0055] The admission-discharge NIHSS total score change value and the corresponding admission-discharge NIHSS single item score change value of each stroke patient data in the stroke research data set are added to the stroke research data set to obtain an updated stroke research data set, wherein the updated stroke research data set can be understood as adding the admission-discharge NIHSS total score change value and the admission-discharge NIHSS single item score change value obtained by the above analysis to the original stroke research data set.

[0056] The first stroke patient data set in the updated stroke research data set is divided into a first training set and a first external validation set, and a second external validation set is obtained by randomly selecting a preset number of patient data from a second stroke patient data set in the updated stroke research data set; wherein the division ratio of the first training set and the first external validation set of the first stroke patient data set can be set according to actual application requirements, for example, if the first stroke patient data set is the ischemic stroke patient data set after removing TIA (transient ischemic attack) patient data from the CNSR-III data as described above, then 11813 patients from 169 hospitals in the CNSR-III cohort can be randomly selected and assigned to the first training set, and 2127 patients from another 40 independent hospitals can be assigned to the first external validation set. It should be noted that the preset number used to construct the second external validation set can be determined according to actual application requirements, which is not limited here, for example, if the second stroke patient data set is the ischemic stroke patient data set after removing TIA patient data from the CHANCE-2 cohort data as described above, then 5158 ischemic stroke patients can be selected as research objects.

[0057] According to the first training set and the XGBoost model, stroke prognosis prediction key features are extracted from the data features of the first training set; wherein the stroke prognosis prediction key features can be understood as data features that have a greater impact on the outcome after stroke; specifically, the step of extracting stroke prognosis prediction key features from the data features of the first training set according to the first training set and the XGBoost model comprises:

[0058] The XGBoost model is trained according to the first training set, and after training is completed, a corresponding feature importance ranking table is obtained; wherein the feature importance ranking table can be obtained by using the feature importance analysis and ranking function provided by the XGBoost algorithm after the XGBoost model is trained based on the first training set, or by analyzing the prediction contribution of each data feature in the first training set to the XGBoost model based on the SHAP attribution analysis method, or by comprehensively analyzing the analysis results of the two methods, the indicators with higher importance ranking in the results obtained by the two methods are selected, which are not limited here.

[0059] According to the feature importance ranking table, a first stroke prognosis prediction important feature is obtained; wherein the first stroke prognosis prediction important feature can be understood as a data feature with a relatively high ranking in the feature importance ranking table, and the specific number of features can be determined according to actual application requirements. In order to improve the convenience and efficiency of data analysis on the basis of ensuring the reliability of subsequent model training, the embodiment preferably screens two data features with particularly prominent importance, such as Figure 3 As shown in FIG. 13, the admission-discharge NIHSS score change feature and the discharge mRS score are both in the top two in the feature importance ranking table obtained by the feature importance analysis and sorting function of the XGBoost algorithm and the SHAP attribution analysis method, and the corresponding scores are very high, so the first stroke prognosis prediction important feature can be determined to include the admission-discharge NIHSS score change feature and the discharge mRS score.

[0060] Based on the clinical analysis of the availability of features, a second stroke prognosis prediction important feature is obtained from the data features of the first training set; wherein the clinical analysis of the availability of features can be understood as the research and consultation results of relevant clinical experts on the influence indicators of the post-stroke outcome. Similarly, considering the convenience and efficiency of data acquisition, the embodiment preferably sets the second stroke prognosis prediction important feature to include age, gender, place of origin (geographical location of the patient in China), and stroke-related medical history features, and the stroke-related medical history features include whether there is a history of transient ischemic attack (TOAST classification of the patient), whether there is a history of diabetes, and whether there is a history of transient ischemic attack (TIA history), etc.

[0061] The first stroke prognosis prediction important feature and the second stroke prognosis prediction important feature are combined to obtain the stroke prognosis prediction key features; wherein the stroke prognosis prediction key features preferably include age, gender, place of origin, discharge mRS (modified Rankin Score) score, whether there is a history of transient ischemic attack (TOAST classification of the patient), whether there is a history of diabetes, whether there is a history of transient ischemic attack (TIA history), admission-discharge NIHSS total score change value, and corresponding 15 admission-discharge NIHSS single score change values.

[0062] According to the stroke prognosis prediction key features and the corresponding prognosis outcomes in the first training set, the first external validation set, and the second external validation set, the corresponding second training set, the first validation set, and the second validation set are obtained; that is, the second training set, the first validation set, and the second validation set are data sets constructed by screening the stroke prognosis prediction key features and the corresponding prognosis outcomes from the first training set, the first external validation set, and the second external validation set.

[0063] According to the second training set, the XGBoost model is trained, the optimal parameter model is obtained through K-fold cross validation, the stroke prognosis prediction model is obtained, and the stroke prognosis prediction model is verified according to the combination of homologous data and non-homologous data according to the first verification set and the second verification set, and the prediction model verification result is obtained. Specifically, the step of training the XGBoost model according to the second training set, obtaining the optimal parameter model through K-fold cross validation, obtaining the stroke prognosis prediction model, and verifying the stroke prognosis prediction model according to the combination of homologous data and non-homologous data according to the first verification set and the second verification set, and obtaining the prediction model verification result comprises:

[0064] According to the second training set and the internal and external K-fold cross validation method, the XGBoost model is trained, and the optimal parameters of the XGBoost model are obtained through grid search, and the stroke prognosis prediction model corresponding to the optimal parameters is obtained. The optimal parameters include learning rate and maximum depth; wherein the training data and verification data division ratio used by the K-fold cross validation method can be set according to actual application requirements, and the embodiment is preferably set to ten-fold cross validation.

[0065] In order to ensure the efficiency and reliability of model training, when using XGB Classifier for model fitting training based on the second training set and K-fold cross validation method, the grid search method is used to adjust the learning rate (eta) and maximum depth (max depth) two hyperparameters, and the precision, recall, F1 score (harmonic mean of precision and recall, F1 score = [(2*precision*recall) / (precision+recall)]), AUC value of ROC curve (area under the curve of receiver operating curve, if the area under the curve is 0.5, it means random judgment, the larger the area under the curve, the better, the ideal value is 1) of the model under different parameter combinations are obtained, and the optimal parameters are selected according to the AUC value of the ROC curve, and the XGBoost model corresponding to the optimal parameters is used as the initial prognosis prediction model.

[0066] According to the first verification set, the stroke prognosis prediction model is verified by homologous data, and the corresponding model homologous data verification result is obtained; wherein the model homologous data verification result can be understood as the index value of the verification model prediction effect obtained based on the first verification set (the same as the training set of the initial prognosis prediction model). The index can select the prediction mean square error (MSE) of the model.

[0067] According to the second verification set, the stroke prognosis prediction model is verified by non-homologous data, and a corresponding model heterologous data verification result is obtained; wherein the model heterologous data verification result can be understood as an index value of the verification model prediction effect obtained based on the second verification set (different from the training set of the initial prognosis prediction model), which is consistent with the index selected in the above model homologous data verification result, and the prediction mean square error (MSE) of the model can also be selected.

[0068] In addition, considering that in actual application, the stroke prognosis prediction model trained based on the second training set may perform poorly on the verification set, in order to improve the generalization ability of the prediction model as much as possible, the embodiment preferably adjusts the parameters of the stroke prognosis prediction model according to the model homologous data verification result and the model heterologous data verification result. Specifically, the process of adjusting the parameters of the stroke prognosis prediction model includes: comparing the obtained model homologous data verification result and model heterologous data verification result with the corresponding homologous verification index threshold and heterologous verification index threshold, respectively; if the model homologous data verification result and the model heterologous data verification result are both less than the corresponding index threshold, it is considered that the generalization ability of the above obtained stroke prognosis prediction model is guaranteed, and it can be directly used as the final stroke prognosis prediction model; otherwise, if any one of the model homologous data verification result and the model heterologous data verification result is greater than the corresponding index threshold, it is considered that the generalization ability of the above obtained stroke prognosis prediction model needs to be further improved, and the model parameters of the stroke prognosis prediction model are optimized and trained based on the second training set and a preset loss function to obtain a final stroke prognosis prediction model; wherein the preset loss function is represented as:

[0069]

[0070] In the formula, L represents the model prediction loss value; M1 and M2 represent the training sample subset and the verification sample subset divided in the K-fold cross-training in the second training set, respectively; n1 and n2 represent the number of samples in the training sample subset and the verification sample subset, respectively; y i，ture and y i，pre respectively represent the true value and the predicted value of the i-th sample in the training sample subset M1; y j，ture and y j，pre respectively represent the true value and the predicted value of the j-th sample in the verification sample subset; λ represents a weight coefficient.

[0071] Through the above method steps, a stroke prognosis prediction model with high accuracy and risk discrimination on different data sources can be obtained. In order to facilitate subsequent description, the obtained stroke prognosis prediction model is named as DISCO (Delta-NIHSS based post-Stroke Composite Outcome) model.

[0072] It should be noted that the above-mentioned stroke prognosis prediction model is constructed in the process of using the post-stroke outcome label, which is different for the prediction needs: when predicting post-stroke recurrence, the corresponding recurrence index value is used as the training label; when predicting post-stroke disability, the corresponding mRS score variation index (mRS>2 is 1, and mRS<=2 is 0) is used as the training label; when predicting post-stroke death, the corresponding death index value is used as the training label. The prediction model of different prognosis outcomes can be obtained by the above-mentioned method steps. If the obtained stroke prognosis prediction model is for predicting stroke recurrence outcome, the corresponding prediction result is the stroke recurrence probability value. If the obtained stroke prognosis prediction model is for predicting stroke death outcome, the corresponding prediction result is the stroke death probability value. If the obtained stroke prognosis prediction model is for predicting post-stroke disability outcome, the corresponding prediction result is the probability value of mRS>2.

[0073] The technical scheme provided by the embodiment of the present application acquires the to-be-analyzed feature data of the stroke patient, including age, gender, hometown, mRS score at discharge, stroke-related medical history characteristics, and admission and discharge NIHSS score change characteristics, inputs the to-be-analyzed data into the stroke prognosis prediction model trained in advance based on the data set having the same characteristics as the to-be-analyzed feature data to perform prognosis prediction, and obtains the corresponding prognosis prediction result. Through the analysis and screening of the stroke research data set, the key data characteristics which are not dependent on the imaging data or laboratory examination results, have strong data availability, and can be simultaneously applied to the precise prediction of different post-stroke outcomes are screened out, and the efficient, precise, and high-risk identification and differentiation stroke prognosis prediction model is trained and constructed based on the key data characteristics for the prognosis outcome prediction of the stroke patient. The present application not only effectively improves the efficiency and precision of the post-stroke outcome prediction, but also guarantees the generalization and ease of use of the prediction model, is convenient for wide range of popularization and use, and further provides reliable technical support for the health management of the stroke patient.

[0074] In order to verify the application effect of the stroke prognosis prediction method provided by the present application, the first training set, the first external validation set, and the second external validation set are obtained based on the first stroke patient data set constructed based on the China National Stroke Registry-III (CNSR-III) data and the second stroke patient data set constructed based on the China CLOpidogrel and Aspirin Sensible AgaiN Study-2 (CHANCE-2) cohort data in the above-mentioned embodiment, taking the 3-month recurrence risk prediction of the 3-month stroke patient as an example, the DISCO model is compared and analyzed with the existing clinical commonly used RRE model, and the analysis results shown in FIG. 2 are obtained, and as shown in FIG. 3, the prediction effect of the DISCO model provided by the present application is better. Figure 4 Figure 4

[0075] ​​In addition, the DISCO model is verified for the prediction effect on the three outcomes of recurrence, death and disability in three months based on the second training set, the first validation set and the second validation set, and the model prediction index results are shown in Tables 1-3; at the same time, the DISCO model is verified for the recurrence risk stratification effect in three months on the second training set, the first validation set and the second validation set, and the results are shown in Table 4. Figure 5

[0076] Table 1 DISCO model prediction index results on the second training set

[0077]

[0078] Table 2 DISCO model prediction index results on the first validation set

[0079]

[0080] Table 3 DISCO model prediction index results on the second validation set

[0081]

[0082]

[0083] Based on the performance of the DISCO model on the second training set in Table 1, the DISCO model has good prediction results for the three-month stroke recurrence event that is most concerned by clinicians, AUC = 0.80, precision = 0.98, recall = 0.93, F1 = 0.95, and the prediction effect on death in three months is better than that on recurrence, AUC > 0.8, precision > 0.99, recall > 0.8, F1 > 0.89, although the prediction on disability mRS in three months is relatively weak than that on recurrence, but can also guarantee the AUC value between 0.7 and 0.8.

[0084] Similarly, based on the performance of the DISCO model on the first validation set in Table 2, the DISCO has AUC = 0.74, precision = 0.96, recall = 0.93, F1 = 0.94 for the three-month stroke recurrence event.

[0085] Based on the performance of the DISCO model on the second validation set in Table 3, the DISCO model has AUC = 0.80, precision = 0.96, recall = 0.93, F1 = 0.95 for the three-month stroke recurrence event. In addition, the prediction effect of the DISCO model on death in three months is also better than that on recurrence in two validation sets, AUC > 0.8, precision > 0.98, recall > 0.80, F1 > 0.88, although the prediction on disability mRS in three months is relatively weak than that on recurrence, but can also guarantee the AUC value around 0.7.​

[0086] As shown in Figure 5 , the DISCO model has a very good ability to identify high-risk groups of stroke recurrence in 3 months.

[0087] In addition, in order to further improve the application value of the above stroke prognosis prediction model, the embodiment of the present application preferably provides personalized rehabilitation exercise guidance suggestions based on the importance ranking of the specific feature data of each patient and the professional suggestions of clinical workers on the basis of the required prognosis outcome prediction of the analyzed feature data provided for the discharged stroke patients. Specifically, as shown in Figure 6 , the method further comprises:

[0088] S13, by SHAP attribution analysis method, the feature importance ranking corresponding to the prognosis prediction result is obtained, and the corresponding rehabilitation exercise guidance suggestion is generated according to the feature importance ranking; wherein the feature importance ranking can be understood as the contribution degree ranking of each analyzed feature data provided by the discharged stroke patient to the prediction outcome. After obtaining the feature importance ranking result of the DISCO model by the SHAP attribution analysis method as shown in Figure 7 , the features with high contribution ranking can be taken as reference basis, and the targeted rehabilitation exercise guidance suggestions can be given combined with the experience of clinical workers, and the rehabilitation exercise guidance suggestions can also be obtained by comparing the actually obtained feature importance ranking with the corresponding guidance suggestion knowledge base constructed based on different feature ranking results. In order to further improve the scientificity and intelligence of the rehabilitation exercise guidance suggestion generation, a reinforcement learning model can also be constructed by taking the feature importance ranking result as the state variable and taking the rehabilitation exercise guidance suggestion as the corresponding action instruction. In actual application, the obtained feature importance ranking is input into the trained reinforcement learning model to obtain the required rehabilitation exercise guidance suggestion.

[0089] The technical scheme of the embodiment of the application can effectively improve the efficiency and accuracy of post-stroke outcome prediction, ensure the generalization and ease of use of the prediction model, facilitate wide range of use, can also give corresponding rehabilitation guidance suggestions based on the prediction results, has high health management guidance significance, and further provides reliable technical support for health management of stroke patients.

[0090] It should be noted that although each step in the above flowchart is displayed in sequence according to the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, the execution of these steps has no strict order limitation, and these steps can be executed in other orders.

[0091] In one embodiment, as shown in Figure 8 a stroke prognosis prediction system is provided, and the system comprises:

[0092] The data acquisition module 1 is configured to acquire the analysis feature data of the stroke patient; the analysis feature data comprises age, gender, hometown, mRS score at discharge, stroke-related medical history characteristics, and change characteristics of NIHSS score at admission and discharge.

[0093] The prognosis prediction module 2 is configured to input the analysis data into a pre-constructed stroke prognosis prediction model for prognosis prediction to obtain a corresponding prognosis prediction result.

[0094] In one embodiment, as shown in Figure 9 a stroke prognosis prediction system is provided, and the system further comprises:

[0095] The suggestion generation module 3 is configured to acquire the feature importance ranking corresponding to the prognosis prediction result by SHAP attribution analysis method, and generate corresponding rehabilitation guidance suggestions according to the feature importance ranking.

[0096] The specific definition of the stroke prognosis prediction system can refer to the definition of the stroke prognosis prediction method in the foregoing, and the corresponding technical effect can also be obtained equally, which will not be described here. Each module in the stroke prognosis prediction system described above can be realized by software, hardware, and a combination thereof in whole or in part. The above-mentioned modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to each module.

[0097] Figure 10 An internal structure diagram of a computer device in an embodiment is shown, which can be a terminal or a server in particular. As shown in the figure, Figure 10 The computer device includes a processor, a memory, a network interface, a display, a camera, and an input device connected through a system bus. 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 and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The network interface of the computer device is used to communicate with external terminals through network connection. The computer program is executed by the processor to implement the stroke prognosis prediction method. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball, or touchpad provided on the shell of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0098] Those skilled in the art can understand, Figure 10 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have the same component arrangement.

[0099] In one embodiment, a computer device is provided, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor implements the steps of the above-mentioned method when executing the computer program.

[0100] In one embodiment, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by the processor to implement the steps of the above-mentioned method.

[0101] In summary, the stroke prognosis prediction method, system, computer device and storage medium provided by the embodiment of the present application realize obtaining the to-be-analyzed feature data of a stroke patient, including age, gender, native place, mRS score at discharge, stroke-related medical history characteristics and admission and discharge NIHSS score change characteristics, inputting the to-be-analyzed data into a stroke prognosis prediction model trained in advance based on a data set having the same characteristics as the to-be-analyzed feature data for prognosis prediction, obtaining the corresponding prognosis prediction result, and obtaining the feature importance ranking corresponding to the prognosis prediction result through SHAP attribution analysis, and generating a corresponding rehabilitation guidance technical scheme according to the feature importance ranking suggestion. The method not only effectively improves the efficiency and accuracy of the post-stroke outcome prediction, but also guarantees the generalization and ease of use of the prediction model, facilitates wide range of use, can also give corresponding rehabilitation guidance suggestions based on the prediction result, has high health management guidance significance, and further provides reliable technical support for health management of stroke patients.

[0102] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts of each embodiment can be referred to each other, and each embodiment mainly describes the difference from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the related parts can be referred to the part of the method embodiment. It should be noted that, the technical features of the above embodiments can be combined arbitrarily, in order to make the description simple, not all possible combinations of the technical features of the above embodiments are described, however, as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0103] The above-described embodiments only express several preferred embodiments of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be noted that, for ordinary skilled in the art, without departing from the technical principles of the present application, a number of improvements and replacements can be made, and these improvements and replacements should be considered as the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the protection scope of the claims.

Claims

1. A method for predicting stroke prognosis, characterized in that, The method, applicable to 3-month stroke prognosis prediction, includes the following steps: Acquire the characteristic data to be analyzed for stroke patients; the characteristic data to be analyzed includes age, gender, place of origin, mRS score at discharge, stroke-related medical history characteristics, and NIHSS score change characteristics upon admission and discharge; the stroke-related medical history characteristics include whether there is a history of transient ischemic attack, whether there is a history of diabetes, and whether there is a history of transient ischemic attack; the NIHSS score change characteristics upon admission and discharge include the total NIHSS score change value upon admission and discharge and the corresponding changes value of several individual NIHSS scores upon admission and discharge. The data to be analyzed is input into a pre-constructed stroke prognosis prediction model for prognosis prediction, and the corresponding prognosis prediction results are obtained; the stroke prognosis prediction model is trained based on a dataset with the same features as the data to be analyzed. The steps for constructing the stroke prognosis prediction model include: Obtain a stroke research dataset; the stroke research dataset includes a first stroke patient dataset and a second stroke patient dataset; the first stroke patient dataset and the second stroke patient dataset are ischemic stroke patient data after removing TIA patient data from the CNSR-III cohort and the CHANCE-2 cohort, respectively; After preprocessing the stroke research dataset, a first training set, a first external validation set, and a second external validation set are obtained. The preprocessing includes comparing and analyzing the total admission NIHSS score and several corresponding admission NIHSS individual scores, as well as the total discharge NIHSS score and several corresponding discharge NIHSS individual scores of each patient to obtain the change value of the total admission and discharge NIHSS score and the change value of several corresponding admission and discharge NIHSS individual scores, which are then added to the current dataset to obtain a new dataset. Based on the first training set and the XGBoost model, key features for stroke prognosis prediction are extracted from the data features of the first training set. Based on the key features for stroke prognosis prediction and the corresponding prognostic outcomes in the first training set, the first external validation set, and the second external validation set, respectively, the corresponding second training set, first validation set, and second validation set are obtained. The XGBoost model is trained using the second training set, and the optimal parameter model is obtained through K-fold cross-validation to obtain the stroke prognosis prediction model. The stroke prognosis prediction model is then validated using a combination of homologous and non-homologous data based on the first and second validation sets to obtain the prediction model validation results. If both the model's validation results from the same source data and the model's validation results from different source data are less than the corresponding indicator thresholds in the prediction model validation results, then the stroke prognosis prediction model will not be fine-tuned; otherwise, the model parameters of the stroke prognosis prediction model will be optimized and trained based on the second training set and a preset loss function; the preset loss function is expressed as: In the formula, This represents the model's predicted loss value; and These represent the training sample subset and validation sample subset, respectively, in the K-fold cross-training of the second training set; and These represent the number of samples in the training sample subset and the validation sample subset, respectively. and They represent the training sample subsets respectively. The first in i The true and predicted values ​​of each sample; and These represent the first and second digits of the validation sample subset, respectively. j The true and predicted values ​​of each sample; This represents the weighting coefficient.

2. The stroke prognosis prediction method as described in claim 1, characterized in that, Both the first stroke patient dataset and the second stroke patient dataset include age, gender, place of origin, mRS score at discharge, stroke-related medical history characteristics, total NIHSS score upon admission and several corresponding individual NIHSS scores upon admission, and total NIHSS score upon discharge and several corresponding individual NIHSS scores upon discharge.

3. The stroke prognosis prediction method as described in claim 2, characterized in that, The step of preprocessing the stroke patient data in the stroke research dataset to obtain the first training set, the first external validation set, and the second external validation set includes: The total discharge NIHSS score in each stroke patient data set in the stroke research dataset is subtracted from the total admission NIHSS score to obtain the corresponding change in the total admission and discharge NIHSS score. The corresponding individual discharge NIHSS score is subtracted from the individual admission NIHSS score to obtain the corresponding change in the individual admission and discharge NIHSS score. The changes in the total NIHSS score upon admission and discharge and the corresponding changes in individual NIHSS scores upon admission and discharge for each stroke patient in the stroke research dataset are added to the stroke research dataset to obtain the updated stroke research dataset. The first stroke patient dataset portion of the updated stroke research dataset is divided into a training set and a first external validation set, and a predetermined number of patient data are randomly selected from the second stroke patient dataset portion of the updated stroke research dataset to obtain a second external validation set.

4. The stroke prognosis prediction method as described in claim 1, characterized in that, The step of extracting key features for stroke prognosis prediction from the data features of the first training set based on the first training set and the XGBoost model includes: The XGBoost model is trained based on the first training set, and after training is completed, the corresponding feature importance ranking table is obtained. Based on the feature importance ranking table, the first important features for predicting stroke prognosis are obtained; the first important features for predicting stroke prognosis include the changes in NIHSS scores upon admission and discharge and the mRS score at discharge. Based on the feature accessibility suggestions of clinical analysis, a second important feature for predicting stroke prognosis is obtained from the data features of the first training set; the second important feature for predicting stroke prognosis includes age, gender, place of origin, and stroke-related medical history. The first important feature for predicting stroke prognosis and the second important feature for predicting stroke prognosis are combined to obtain the key features for predicting stroke prognosis.

5. The stroke prognosis prediction method as described in claim 1, characterized in that, The steps of training the XGBoost model based on the second training set, obtaining the optimal parameter model through K-fold cross-validation to obtain the stroke prognosis prediction model, and validating the stroke prognosis prediction model using a combination of homologous and non-homologous data based on the first and second validation sets to obtain the prediction model validation results include: The XGBoost model is trained using the second training set and the inner and outer K-fold cross-validation method. The optimal parameters of the XGBoost model are obtained through grid search, resulting in a stroke prognosis prediction model corresponding to the optimal parameters. The optimal parameters include the learning rate and the maximum depth. The stroke prognosis prediction model is validated using the first validation set with source data to obtain the corresponding model source data validation results. The stroke prognosis prediction model is validated using non-homogeneous data based on the second validation set to obtain the corresponding model heterogeneous data validation results.

6. The stroke prognosis prediction method as described in claim 1, characterized in that, The method further includes: The SHAP attribution analysis method is used to obtain the feature importance ranking corresponding to the prognostic prediction results, and corresponding rehabilitation exercise guidance suggestions are generated based on the feature importance ranking.

7. A stroke prognosis prediction system, characterized in that, The system, employing the stroke prognosis prediction method as described in claim 1, comprises: The data acquisition module is used to acquire the characteristic data to be analyzed of stroke patients; the characteristic data to be analyzed includes age, gender, place of origin, mRS score at discharge, stroke-related medical history characteristics, and NIHSS score change characteristics from admission to discharge. The prognosis prediction module is used to input the data to be analyzed into a pre-constructed stroke prognosis prediction model to perform prognosis prediction and obtain the corresponding prognosis prediction result; the stroke prognosis prediction model is trained based on a dataset with the same features as the data to be analyzed.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images