Stroke prognosis prediction method and device, medium and equipment

Abstract

The invention discloses a stroke prognosis prediction method and device, a medium and equipment. The method comprises the following steps: acquiring resting-state electroencephalogram data of an acute-stage acute ischemic stroke patient, calculating a power spectrum and a micro-state index of the resting-state electroencephalogram data, combining baseline information and clinical examination data of the patient to construct a stroke prognosis prediction feature set, inputting the stroke prognosis prediction feature set into a model obtained by training through Lasso regression and a plurality of machine learning algorithms, and obtaining the stroke prognosis prediction feature set. And outputting a functional prognosis result of the patient. In the process, the advantage that electroencephalogram data reflect brain function changes is fully utilized, the defect that the brain function value is neglected in the prior art is overcome, interference of human factors is reduced through machine learning, and the objectivity and accuracy of prognosis evaluation are improved.

Description

Technical Field

[0001] The present invention relates to the field of stroke prognosis prediction, and in particular to a stroke prognosis prediction method, device, medium and equipment. Background Art

[0002] Acute ischemic stroke (AIS) is a neurological disease with high morbidity, disability, and recurrence rates. According to the latest statistics, the in-hospital mortality rate for patients with acute ischemic stroke in my country is 0.5%, and the complication rate is 12.8%. The three-month mortality rate and the one-year mortality rate range from 1.5% to 3.2% and 3.4% to 6.0% respectively. The three-month disability rate and the one-year disability rate range from 14.6% to 23.1% respectively. Therefore, for neurologists, accurately assessing the long-term functional prognosis of patients with acute ischemic stroke and early identification of patients with poor functional prognosis are crucial for developing treatment and rehabilitation plans.

[0003] Currently, the main methods for evaluating the long-term prognosis of patients with acute ischemic stroke include the following:

[0004] Patient baseline conditions: including the patient's age, general health status, onset and progression, past medical history (such as hypertension, diabetes, heart disease, etc.), infarction location and area, etc.

[0005] Imaging assessment: Imaging is an important tool for assessing the prognosis of acute ischemic stroke. CT scans can be used to rapidly screen patients suspected of having a stroke, while CT angiography (CTA) and CT perfusion imaging (CTP) can assess intracranial vascular status and ischemic perfusion in brain tissue. Magnetic resonance imaging (MRI-DWI) can more clearly demonstrate the location and extent of acute ischemic stroke lesions and is crucial for guiding revascularization therapy.

[0006] Scale assessment: including the National Institutes of Health Stroke Scale (NIHSS), modified Rankin Score (mRS), activities of daily living (ADL) assessment, quality of life index (QLI), etc. These assessments require professional physicians to score according to the patient's current condition.

[0007] However, existing evaluation methods have the following shortcomings:

[0008] Ignoring the value of brain function: Existing methods mainly focus on the predictive effect of brain structural damage (such as infarct location, infarct area, etc.) on prognosis, but ignore the impact of brain function changes on prognosis.

[0009] Limitations of imaging examinations: Imaging examinations are time-consuming, not suitable for bedside assessment, expensive, and require high standards of equipment and operators.

[0010] Subjectivity of scale assessment: Clinical assessment scales rely on the patient's cooperation and the clinician's subjective judgment. Different doctors may have different observations and assessments of the patient's symptoms and signs, resulting in inconsistent scoring.

[0011] These deficiencies result in the inability of existing technologies to accurately assess the long-term functional prognosis of patients with acute ischemic stroke. Summary of the Invention

[0012] The present invention provides a stroke prognosis prediction method, apparatus, medium and equipment to solve the problem in the prior art that the long-term functional prognosis of acute ischemic stroke patients cannot be accurately assessed.

[0013] In a first aspect, the present application provides a method for predicting stroke prognosis, comprising:

[0014] Obtaining resting-state electroencephalogram data of patients with acute ischemic stroke to be predicted during the acute phase;

[0015] Calculating the power spectrum and microstate indicators of the resting-state electroencephalogram data based on the resting-state electroencephalogram data;

[0016] Constructing a stroke prognosis prediction feature set based on the power spectrum, microstate indicators, preset patient baseline information, and preset clinical test data;

[0017] Inputting the stroke prognosis prediction feature set into a preset machine learning model so that the machine learning model outputs a functional prediction result of the patient;

[0018] Among them, the preset machine learning model is obtained by screening variables based on the Lasso regression algorithm, and then training the initial machine learning model and historical resting-state EEG data using multiple machine learning algorithms.

[0019] This application obtains resting-state EEG data from patients in the acute phase of acute ischemic stroke. The resting-state EEG examination has the characteristics of being non-invasive, high temporal resolution, convenient, fast, and low-cost, making it more suitable for widespread clinical application; and calculates its power spectrum and microstate indicators, and constructs a feature set based on the patient's baseline information and clinical test data, and then inputs it into a model trained based on Lasso regression and multiple machine learning algorithms to achieve accurate prediction of the patient's functional prognosis. This method not only makes up for the shortcomings of the existing technology that only relies on brain structural damage assessment, but also reduces interference from human factors through machine learning, thereby improving the objectivity and accuracy of prognosis assessment. This application effectively solves the problem that the existing technology cannot accurately assess the long-term functional prognosis of patients with acute ischemic stroke.

[0020] As a preferred embodiment of the first aspect, the preset machine learning model is obtained by performing variable screening based on the Lasso regression algorithm, and then training the initial machine learning model and the preset historical stroke prognosis prediction feature set using multiple machine learning algorithms, specifically:

[0021] Obtain the preset historical stroke prognosis prediction feature set;

[0022] Inputting the historical stroke prognosis prediction feature set into an initial machine learning model for training, so that the initial machine learning model performs feature screening on the power spectrum, microstate indicators, baseline information, and clinical test data of the historical stroke prognosis prediction feature set according to a Lasso regression algorithm, and outputs each feature with the highest correlation with functional prognosis;

[0023] According to the naive Bayes algorithm, random forest algorithm, support vector machine algorithm or preset machine learning algorithm, the features with the highest correlation with functional prognosis are trained to obtain the preset machine learning model.

[0024] In this preferred embodiment, the present application uses the Lasso regression algorithm to perform feature screening on the historical stroke prognosis prediction feature set, which can accurately identify the features with the highest correlation with functional prognosis, thereby effectively reducing the interference of redundant features on model performance and improving the accuracy and reliability of the model. On this basis, a variety of machine learning algorithms such as naive Bayes, random forest or support vector machine are used to train the screened features to further optimize the predictive ability of the model. This method of combining feature screening with multiple algorithm training not only improves the accuracy of the model in predicting the functional prognosis of patients with acute ischemic stroke, but also enhances the generalization ability of the model, so that it can be more stably applied to the prognosis assessment of different patient groups.

[0025] As a preferred embodiment of the first aspect, the initial machine learning model performs feature screening on the power spectrum, microstate indicators, baseline information, and clinical test data of the historical stroke prognosis prediction feature set according to the Lasso regression algorithm, and outputs the features with the highest correlation with functional prognosis, specifically:

[0026] The characteristics most closely associated with functional prognosis included the patient's age, admission NIHSS score, whether they had undergone tirofiban-enhanced antiplatelet therapy, fibrinogen, erythrocyte sedimentation rate, whether the carotid artery was stenotic, whether the subclavian artery was stenotic, Delta band frontal lobe power spectral density, DTABR, and microstate A coverage.

[0027] In this preferred embodiment, the present application uses the Lasso regression algorithm to perform feature screening on the historical stroke prognosis prediction feature set, and accurately identifies the key features with the highest correlation with functional prognosis, including the patient's age, NIHSS score on admission, whether tirofiban-enhanced anti-platelet therapy has been performed, fibrinogen, erythrocyte sedimentation rate, whether the carotid artery is stenotic, whether the subclavian artery is stenotic, Delta band frontal lobe power spectral density, DTABR and microstate A coverage. The selection of these features is based on the analysis of a large amount of historical data, ensuring their importance and effectiveness in predicting the functional prognosis of patients. By focusing on these key features, the model can more accurately assess the patient's prognosis, reduce the noise and interference caused by irrelevant features, and thus improve the accuracy and reliability of the prediction. This method not only improves the performance of the model, but also provides clinicians with a more targeted and practical prognosis assessment tool, which helps to formulate more accurate treatment and rehabilitation plans.

[0028] As a preferred embodiment of the first aspect, the step of obtaining a preset historical stroke prognosis prediction feature set is specifically:

[0029] According to the preset modified Rankin scoring rules, the recovery status of each patient who has suffered acute ischemic stroke was scored using mRS;

[0030] According to the mRS score results, the patients were divided into a good functional prognosis group and a poor functional prognosis group;

[0031] Matching variable scores were performed on the patients in the good functional prognosis group and the poor functional prognosis group; wherein the matching variables included gender, age, history of hypertension, diabetes, heart disease, ischemic stroke, and NIHSS score at admission;

[0032] A preset historical stroke prognosis prediction feature set is constructed based on the matching variable scoring results and the resting-state electroencephalogram data of each patient who has suffered from acute ischemic stroke.

[0033] In this preferred embodiment, the present application first scores the patient's recovery according to the mRS scoring rules, and divides the patients into a good functional prognosis group and a non-good functional prognosis group accordingly. This classification provides a clear goal and direction for subsequent analysis. Then, the two groups of patients are scored for matching variables, taking into account important variables such as gender and age, ensuring the comparability of the two groups in these key factors, thereby reducing the interference of confounding factors on the research results. Finally, a feature set is constructed by combining the matching variable scoring results and the patient's resting-state EEG data. This feature set not only contains rich clinical information, but also ensures the balance and representativeness of the data through the matching process. This rigorous data processing method provides a solid foundation for subsequent model training and prediction, so that the constructed model can more accurately reflect the patient's functional prognosis, thereby improving the model's predictive performance and clinical application value.

[0034] In a second aspect, the present application provides a stroke prognosis prediction device. The stroke prognosis prediction device includes an acquisition module, a calculation module, a construction module, and a prediction module;

[0035] The acquisition module is used to obtain resting-state electroencephalogram data of patients with acute ischemic stroke to be predicted during the acute phase;

[0036] The calculation module is used to calculate the power spectrum and microstate indicators of the resting-state electroencephalogram data based on the resting-state electroencephalogram data;

[0037] The construction module is used to construct a stroke prognosis prediction feature set based on the power spectrum, microstate indicators, preset patient baseline information and preset clinical test data;

[0038] The prediction module is used to input the stroke prognosis prediction feature set into a preset machine learning model so that the machine learning model outputs the patient's functional prediction result;

[0039] Among them, the preset machine learning model is obtained by screening variables based on the Lasso regression algorithm, and then training the initial machine learning model and historical resting-state EEG data using multiple machine learning algorithms.

[0040] This device uses four modules to divide the work and coordinate work to more accurately predict the prognosis of stroke. This application obtains resting-state EEG data of patients with acute ischemic stroke in the acute phase. The resting-state EEG examination has the characteristics of non-invasiveness, high temporal resolution, convenience, speed, and low cost, making it more suitable for widespread clinical application; and calculates its power spectrum and microstate indicators, and combines the patient's baseline information with clinical test data to construct a feature set, which is then input into a model trained based on Lasso regression and multiple machine learning algorithms to achieve accurate prediction of the patient's functional prognosis. This method not only makes up for the shortcomings of the existing technology that only relies on brain structural damage assessment, but also reduces human interference through machine learning, and improves the objectivity and accuracy of prognosis assessment. This application effectively solves the problem that the existing technology cannot accurately assess the long-term functional prognosis of patients with acute ischemic stroke.

[0041] As a preferred embodiment of the second aspect, the preset machine learning model is obtained by performing variable screening based on the Lasso regression algorithm, and then training the initial machine learning model and the preset historical stroke prognosis prediction feature set using multiple machine learning algorithms, specifically:

[0042] Obtain the preset historical stroke prognosis prediction feature set;

[0043] Inputting the historical stroke prognosis prediction feature set into an initial machine learning model for training, so that the initial machine learning model performs feature screening on the power spectrum, microstate indicators, baseline information, and clinical test data of the historical stroke prognosis prediction feature set according to a Lasso regression algorithm, and outputs each feature with the highest correlation with functional prognosis;

[0044] According to the naive Bayes algorithm, random forest algorithm, support vector machine algorithm or preset machine learning algorithm, the features with the highest correlation with functional prognosis are trained to obtain the preset machine learning model.

[0045] In this preferred embodiment, the present application uses the Lasso regression algorithm to perform feature screening on the historical stroke prognosis prediction feature set, which can accurately identify the features with the highest correlation with functional prognosis, thereby effectively reducing the interference of redundant features on model performance and improving the accuracy and reliability of the model. On this basis, a variety of machine learning algorithms such as naive Bayes, random forest or support vector machine are used to train the screened features to further optimize the predictive ability of the model. This method of combining feature screening with multiple algorithm training not only improves the accuracy of the model in predicting the functional prognosis of patients with acute ischemic stroke, but also enhances the generalization ability of the model, so that it can be more stably applied to the prognosis assessment of different patient groups.

[0046] As a preferred embodiment of the second aspect, the initial machine learning model performs feature screening on the power spectrum, microstate indicators, baseline information, and clinical test data of the historical stroke prognosis prediction feature set according to the Lasso regression algorithm, and outputs the features with the highest correlation with functional prognosis, specifically:

[0047] The characteristics most closely associated with functional prognosis included the patient's age, admission NIHSS score, whether they had undergone tirofiban-enhanced antiplatelet therapy, fibrinogen, erythrocyte sedimentation rate, whether the carotid artery was stenotic, whether the subclavian artery was stenotic, Delta band frontal lobe power spectral density, DTABR, and microstate A coverage.

[0048] In this preferred embodiment, the present application uses the Lasso regression algorithm to perform feature screening on the historical stroke prognosis prediction feature set, and accurately identifies the key features with the highest correlation with functional prognosis, including the patient's age, NIHSS score on admission, whether tirofiban-enhanced anti-platelet therapy has been performed, fibrinogen, erythrocyte sedimentation rate, whether the carotid artery is stenotic, whether the subclavian artery is stenotic, Delta band frontal lobe power spectral density, DTABR and microstate A coverage. The selection of these features is based on the analysis of a large amount of historical data, ensuring their importance and effectiveness in predicting the functional prognosis of patients. By focusing on these key features, the model can more accurately assess the patient's prognosis, reduce the noise and interference caused by irrelevant features, and thus improve the accuracy and reliability of the prediction. This method not only improves the performance of the model, but also provides clinicians with a more targeted and practical prognosis assessment tool, which helps to formulate more accurate treatment and rehabilitation plans.

[0049] As a preferred embodiment of the second aspect, the obtaining of a preset historical stroke prognosis prediction feature set is specifically:

[0050] According to the preset modified Rankin scoring rules, the recovery status of each patient who has suffered acute ischemic stroke was scored using mRS;

[0051] According to the mRS score results, the patients were divided into a good functional prognosis group and a poor functional prognosis group;

[0052] Matching variable scores were performed on the patients in the good functional prognosis group and the poor functional prognosis group; wherein the matching variables included gender, age, history of hypertension, diabetes, heart disease, ischemic stroke, and NIHSS score at admission;

[0053] A preset historical stroke prognosis prediction feature set is constructed based on the matching variable scoring results and the resting-state electroencephalogram data of each patient who has suffered from acute ischemic stroke.

[0054] In this preferred embodiment, the present application first scores the patient's recovery according to the preset Rankin scoring rules, and divides the patients into a good functional prognosis group and a poor functional prognosis group accordingly. This classification provides a clear goal and direction for subsequent analysis. Then, the two groups of patients are scored for matching variables, taking into account important variables such as gender and age, ensuring the comparability of the two groups in these key factors, thereby reducing the interference of confounding factors on the research results. Finally, a feature set is constructed by combining the matching variable scoring results and the patient's resting-state EEG data. This feature set not only contains rich clinical information, but also ensures the balance and representativeness of the data through the matching process. This rigorous data processing method provides a solid foundation for subsequent model training and prediction, so that the constructed model can more accurately reflect the patient's functional prognosis, thereby improving the model's predictive performance and clinical application value.

[0055] In a third aspect, the present application provides a computer-readable storage medium, comprising a stored computer program. When the computer program is executed, the computer-readable storage medium controls a device containing the computer-readable storage medium to execute a stroke prognosis prediction method as described above. The beneficial effects of the method are the same as those of the stroke prognosis prediction method provided in the first aspect of the present application.

[0056] In a fourth aspect, the present application provides a terminal device comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements any one of the stroke prognosis prediction methods described in the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 : A flow chart of an embodiment of a method for predicting stroke prognosis provided in this application;

[0058] Figure 2 : A structural schematic diagram of an embodiment of the present application for comparing power spectra of different frequency bands between a good functional prognosis group and a poor functional prognosis group at 90 days;

[0059] Figure 3 : A structural diagram of an embodiment of the comparison of microstate indicators (frequency, duration, and time coverage) between a good functional prognosis group and a poor functional prognosis group provided in the present application at 90 days;

[0060] Figure 4 : A schematic diagram of an embodiment of the algorithm performance (AUC value) results provided for this application;

[0061] Figure 5: A structural schematic diagram of an embodiment of a stroke prognosis prediction device provided in this application. DETAILED DESCRIPTION

[0062] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0063] Example 1

[0064] Please refer to Figure 1 , which is a stroke prognosis prediction method provided by an embodiment of the present invention.

[0065] In this embodiment, the process of the stroke prognosis prediction method in this application is described in detail through steps S01-S04.

[0066] This application uses the acute resting-state EEG data of hospitalized patients with acute ischemic stroke, combined with the baseline information of the patients at admission, to construct a machine learning model to predict the patients' independent functional prognosis 90 days after onset (based on the modified Rankin score), and verifies it in external validation data, thereby forming a clinical prognosis machine learning model for acute ischemic stroke that combines resting-state EEG data with clinical data.

[0067] S01: Obtain resting-state electroencephalogram data of patients with acute ischemic stroke to be predicted during the acute phase.

[0068] As a preferred embodiment of the first embodiment, the step of obtaining resting-state electroencephalogram data of a patient to be predicted for acute ischemic stroke in the acute phase is specifically as follows:

[0069] This application includes a total of 165 patients who were admitted to the Department of Neurology of our hospital due to acute ischemic stroke from 2020 to 2022 and completed resting-state EEG in the acute phase (within five days after admission). Resting-state EEG data of ≥5 minutes with eyes closed were extracted, and the data were preprocessed and the patients' resting-state EEG power spectrum and microstate indicators were calculated.

[0070] S02: Calculating the power spectrum and microstate indicators of the resting-state EEG data based on the resting-state EEG data.

[0071] S03: Constructing a stroke prognosis prediction feature set based on the power spectrum, microstate indicators, preset patient baseline information and preset clinical test data.

[0072] S04: inputting the stroke prognosis prediction feature set into a preset machine learning model, so that the machine learning model outputs a functional prediction result of the patient;

[0073] Among them, the preset machine learning model is obtained by screening variables based on the Lasso regression algorithm, and then training the initial machine learning model and historical resting-state EEG data using multiple machine learning algorithms.

[0074] As a preferred embodiment of the first embodiment, the preset machine learning model is obtained by performing variable screening based on the Lasso regression algorithm, and then training the initial machine learning model and historical resting-state EEG data using multiple machine learning algorithms, specifically:

[0075] Obtain the preset historical stroke prognosis prediction feature set;

[0076] Inputting the historical stroke prognosis prediction feature set into an initial machine learning model for training, so that the initial machine learning model performs feature screening on the power spectrum, microstate indicators, baseline information, and clinical test data of the historical stroke prognosis prediction feature set according to a Lasso regression algorithm, and outputs each feature with the highest correlation with functional prognosis;

[0077] According to other machine learning algorithms such as Naive Bayes, Random Forest or Support Vector Machine algorithm, the features with the highest correlation with functional prognosis are trained to obtain the preset machine learning model.

[0078] In this preferred embodiment, the present application uses the Lasso regression algorithm to perform feature screening on the historical stroke prognosis prediction feature set, which can accurately identify the features with the highest correlation with functional prognosis, thereby effectively reducing the interference of redundant features on model performance and improving the accuracy and reliability of the model. On this basis, a variety of machine learning algorithms such as naive Bayes, random forest or support vector machine are used to train the screened features to further optimize the predictive ability of the model. This method of combining feature screening with multiple algorithm training not only improves the accuracy of the model in predicting the functional prognosis of patients with acute ischemic stroke, but also enhances the generalization ability of the model, so that it can be more stably applied to the prognosis assessment of different patient groups.

[0079] Furthermore, the characteristics that are most correlated with functional prognosis include the age of each patient, NIHSS score upon admission, whether or not tirofiban-enhanced anti-platelet therapy has been performed, fibrinogen, erythrocyte sedimentation rate, whether or not the carotid artery is stenotic, whether or not the subclavian artery is stenotic, Delta frequency band frontal lobe power spectral density, DTABR (a relative power ratio parameter, DTABR = (δ+θ) / (α+β)) and microstate A coverage.

[0080] This application uses the Lasso regression algorithm to perform feature screening on the historical stroke prognosis prediction feature set, and accurately identifies the key features with the highest correlation with functional prognosis, including the age of each patient, NIHSS score on admission, whether or not tirofiban-enhanced anti-platelet therapy has been performed, fibrinogen, erythrocyte sedimentation rate, whether the carotid artery is stenotic, whether or not the subclavian artery is stenotic, Delta band frontal lobe power spectral density, DTABR, and microstate A coverage. The selection of these features is based on the analysis of a large amount of historical data, ensuring their importance and effectiveness in predicting the functional prognosis of patients. By focusing on these key features, the model can more accurately assess the patient's prognosis, reduce the noise and interference caused by irrelevant features, and thus improve the accuracy and reliability of the prediction. This method not only improves the performance of the model, but also provides clinicians with a more targeted and practical prognosis assessment tool, which helps to formulate more accurate treatment and rehabilitation plans.

[0081] Furthermore, the preset historical stroke prognosis prediction feature set is obtained, specifically:

[0082] According to the preset modified Rankin scoring rule, the recovery status of each patient who has suffered from acute ischemic stroke is scored by mRS, wherein the preset modified Rankin scoring rule is mRS scoring; that is, the mRS score after 90 days is obtained, and the patients are divided into a good functional prognosis group and a poor functional prognosis group according to the 90-day mRS score of 0-2 or 3-6 points; Figure 2 and Figure 3 As shown, Figure 2 Comparison of power spectra of different frequency bands between the good and poor functional prognosis groups at 90 days. Figure 3 The microstate indicators (frequency, duration, and time coverage) were compared between the good and poor functional prognosis groups at 90 days.

[0083] A 1:2 propensity matching was performed based on gender, age, history of hypertension, diabetes, heart disease, ischemic stroke, and admission NIHSS score. The resting-state EEG indicators of the two matched groups (28 patients in the functionally dependent group and 50 patients in the functionally independent group) were compared.

[0084] Matching variable scores were performed on the patients in the good functional prognosis group and the poor functional prognosis group; wherein the matching variables included gender, age, history of hypertension, diabetes, heart disease, ischemic stroke, and NIHSS score at admission;

[0085] A preset historical stroke prognosis prediction feature set is constructed based on the matching variable scoring results and the resting-state electroencephalogram data of each patient who has suffered from acute ischemic stroke.

[0086] In this preferred embodiment, the present application first scores the patient's recovery according to the preset modified Rankin scoring rules, and divides the patients into a good functional prognosis group and a poor functional prognosis group accordingly. This classification provides a clear goal and direction for subsequent analysis. Then, the two groups of patients are scored for matching variables, taking into account important variables such as gender and age, ensuring the comparability of the two groups in these key factors, thereby reducing the interference of confounding factors on the research results. Finally, a feature set is constructed by combining the matching variable scoring results and the patient's resting-state EEG data. This feature set not only contains rich clinical information, but also ensures the balance and representativeness of the data through the matching process. This rigorous data processing method provides a solid foundation for subsequent model training and prediction, so that the constructed model can more accurately reflect the patient's functional prognosis, thereby improving the model's predictive performance and clinical application value.

[0087] As a preferred embodiment of the first embodiment, after inputting the stroke prognosis prediction feature set into a preset machine learning model so that the machine learning model outputs a functional prediction result of the patient, the method further includes:

[0088] The prediction results showed the best performance (AUC value reached 0.970 and above) in the Random Forest algorithm and CatBoost algorithm. Figure 4 As shown, Figure 4 The performance (AUC values) of different algorithms for machine learning models to predict whether the 90-day functional prognosis is good or not.

[0089] This application obtains resting-state EEG data from patients in the acute phase of acute ischemic stroke. The resting-state EEG examination has the characteristics of being non-invasive, high temporal resolution, convenient, fast, and low-cost, making it more suitable for widespread clinical application; and calculates its power spectrum and microstate indicators, and constructs a feature set based on the patient's baseline information and clinical test data, and then inputs it into a model trained based on Lasso regression and multiple machine learning algorithms to achieve accurate prediction of the patient's functional prognosis. This method not only makes up for the shortcomings of the existing technology that only relies on brain structural damage assessment, but also reduces interference from human factors through machine learning, thereby improving the objectivity and accuracy of prognosis assessment. This application effectively solves the problem that the existing technology cannot accurately assess the long-term functional prognosis of patients with acute ischemic stroke.

[0090] Example 2

[0091] Please refer to Figure 5 , which is a stroke prognosis prediction device provided in an embodiment of the present application.

[0092] In this embodiment, the stroke prognosis prediction device includes an acquisition module 10 , a calculation module 20 , a construction module 30 and a prediction module 40 .

[0093] This application uses the acute resting-state EEG data of hospitalized patients with acute ischemic stroke, combined with the baseline information of the patients at admission, to construct a machine learning model to predict the patients' independent functional prognosis 90 days after onset (based on the modified Rankin score), and verifies it in external validation data, thereby forming a clinical prognosis machine learning model for acute ischemic stroke that combines resting-state EEG data with clinical data.

[0094] The acquisition module 10 is used to obtain resting-state electroencephalogram data of a patient to be predicted for acute ischemic stroke during the acute phase.

[0095] As a preferred embodiment of the second embodiment, the step of obtaining resting-state electroencephalogram data of a patient to be predicted for acute ischemic stroke in the acute phase is specifically as follows:

[0096] This application includes a total of 165 patients who were admitted to the Department of Neurology of our hospital due to acute ischemic stroke from 2020 to 2022 and completed resting-state EEG in the acute phase (within five days after admission). Resting-state EEG data of ≥5 minutes with eyes closed were extracted, and the data were preprocessed and the patients' resting-state EEG power spectrum and microstate indicators were calculated.

[0097] The calculation module 20 is used to calculate the power spectrum and microstate indicators of the resting-state EEG data based on the resting-state EEG data.

[0098] The construction module 30 is used to construct a stroke prognosis prediction feature set based on the power spectrum, microstate indicators, preset patient baseline information and preset clinical test data.

[0099] The prediction module 40 is used to input the stroke prognosis prediction feature set into a preset machine learning model so that the machine learning model outputs the patient's functional prediction result;

[0100] Among them, the preset machine learning model is obtained by training the initial machine learning model and historical resting-state EEG data based on the Lasso regression algorithm and multiple machine learning algorithms.

[0101] As a preferred embodiment of the second embodiment, the preset machine learning model is obtained by training the initial machine learning model and historical resting-state EEG data according to the Lasso regression algorithm and multiple machine learning algorithms, specifically:

[0102] Obtain the preset historical stroke prognosis prediction feature set;

[0103] Inputting the historical stroke prognosis prediction feature set into an initial machine learning model for training, so that the initial machine learning model performs feature screening on the power spectrum, microstate indicators, baseline information, and clinical test data of the historical stroke prognosis prediction feature set according to a Lasso regression algorithm, and outputs each feature with the highest correlation with functional prognosis;

[0104] According to other machine learning algorithms such as Naive Bayes, Random Forest or Support Vector Machine algorithm, the features with the highest correlation with functional prognosis are trained to obtain the preset machine learning model.

[0105] In this preferred embodiment, the present application uses the Lasso regression algorithm to perform feature screening on the historical stroke prognosis prediction feature set, which can accurately identify the features with the highest correlation with functional prognosis, thereby effectively reducing the interference of redundant features on model performance and improving the accuracy and reliability of the model. On this basis, a variety of machine learning algorithms such as naive Bayes, random forest or support vector machine are used to train the screened features to further optimize the predictive ability of the model. This method of combining feature screening with multiple algorithm training not only improves the accuracy of the model in predicting the functional prognosis of patients with acute ischemic stroke, but also enhances the generalization ability of the model, so that it can be more stably applied to the prognosis assessment of different patient groups.

[0106] Furthermore, the features that are most correlated with functional prognosis include the age of each patient, the NIHSS score upon admission, whether or not tirofiban-enhanced anti-platelet therapy has been performed, fibrinogen, erythrocyte sedimentation rate, whether or not the carotid artery is stenotic, whether or not the subclavian artery is stenotic, the power spectral density of the Delta band frontal lobe area, DTABR (a type of relative power ratio parameter, DTABR = (δ+θ) / (α+β)) and microstate A coverage.

[0107] This application uses the Lasso regression algorithm to perform feature screening on the historical stroke prognosis prediction feature set, and accurately identifies the key features with the highest correlation with functional prognosis, including the age of each patient, NIHSS score on admission, whether or not tirofiban-enhanced anti-platelet therapy has been performed, fibrinogen, erythrocyte sedimentation rate, whether the carotid artery is stenotic, whether or not the subclavian artery is stenotic, Delta band frontal lobe power spectral density, DTABR, and microstate A coverage. The selection of these features is based on the analysis of a large amount of historical data, ensuring their importance and effectiveness in predicting the functional prognosis of patients. By focusing on these key features, the model can more accurately assess the patient's prognosis, reduce the noise and interference caused by irrelevant features, and thus improve the accuracy and reliability of the prediction. This method not only improves the performance of the model, but also provides clinicians with a more targeted and practical prognosis assessment tool, which helps to formulate more accurate treatment and rehabilitation plans.

[0108] Furthermore, the preset historical stroke prognosis prediction feature set is obtained, specifically:

[0109] According to the preset modified Rankin scoring rule, the recovery status of each patient who has suffered from acute ischemic stroke is scored by mRS, wherein the preset modified Rankin scoring rule is mRS scoring; that is, the mRS score after 90 days is obtained, and the patients are divided into a good functional prognosis group and a poor functional prognosis group according to the 90-day mRS score of 0-2 or 3-6 points; Figure 2 and Figure 3 As shown, Figure 2 Comparison of power spectra of different frequency bands between the good and poor functional prognosis groups at 90 days. Figure 3 The microstate indicators (frequency, duration, and time coverage) were compared between the good and poor functional prognosis groups at 90 days.

[0110] A 1:2 propensity matching was performed based on gender, age, history of hypertension, diabetes, heart disease, ischemic stroke, and admission NIHSS score. The resting-state EEG indicators of the two matched groups (28 patients in the functionally dependent group and 50 patients in the functionally independent group) were compared.

[0111] Matching variable scores were performed on the patients in the good functional prognosis group and the poor functional prognosis group; wherein the matching variables included gender, age, history of hypertension, diabetes, heart disease, ischemic stroke, and NIHSS score at admission;

[0112] A preset historical stroke prognosis prediction feature set is constructed based on the matching variable scoring results and the resting-state electroencephalogram data of each patient who has suffered from acute ischemic stroke.

[0113] In this preferred embodiment, the present application first scores the patient's recovery according to the preset modified Rankin scoring rules, and divides the patients into a good functional prognosis group and a poor functional prognosis group accordingly. This classification provides a clear goal and direction for subsequent analysis. Then, the two groups of patients are scored for matching variables, taking into account important variables such as gender and age, ensuring the comparability of the two groups in these key factors, thereby reducing the interference of confounding factors on the research results. Finally, a feature set is constructed by combining the matching variable scoring results and the patient's resting-state EEG data. This feature set not only contains rich clinical information, but also ensures the balance and representativeness of the data through the matching process. This rigorous data processing method provides a solid foundation for subsequent model training and prediction, so that the constructed model can more accurately reflect the patient's functional prognosis, improving the model's predictive performance and clinical application value.

[0114] As a preferred embodiment of the second embodiment, after inputting the stroke prognosis prediction feature set into a preset machine learning model so that the machine learning model outputs a functional prediction result of the patient, the method further includes:

[0115] The prediction results showed the best performance (AUC value reached 0.970 and above) in the Random Forest algorithm and CatBoost algorithm. Figure 4 As shown, Figure 4 The performance (AUC values) of different algorithms for machine learning models to predict whether the 90-day functional prognosis is good or not.

[0116] This device uses four modules to divide the work and coordinate work to more accurately predict the prognosis of stroke. This application obtains resting-state EEG data of patients with acute ischemic stroke in the acute phase. The resting-state EEG examination has the characteristics of non-invasiveness, high temporal resolution, convenience, speed, and low cost, making it more suitable for widespread clinical application; and calculates its power spectrum and microstate indicators, and combines the patient's baseline information with clinical test data to construct a feature set, which is then input into a model trained based on Lasso regression and multiple machine learning algorithms to achieve accurate prediction of the patient's functional prognosis. This method not only makes up for the shortcomings of the existing technology that only relies on brain structural damage assessment, but also reduces human interference through machine learning, and improves the objectivity and accuracy of prognosis assessment. This application effectively solves the problem that the existing technology cannot accurately assess the long-term functional prognosis of patients with acute ischemic stroke.

[0117] Example 3:

[0118] An embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the stroke prognosis prediction method;

[0119] Wherein, the stroke prognosis prediction method, if implemented in the form of a software functional unit and used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by the processor, it can implement the steps of the above-mentioned various method embodiments. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium can include: any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0120] Example 4

[0121] The present application provides a terminal device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, any one of the stroke prognosis prediction methods described in Example 1 is implemented.

[0122] The specific embodiments described above further illustrate the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for predicting stroke prognosis, characterized in that: include: Obtaining resting-state electroencephalogram data of patients with acute ischemic stroke to be predicted during the acute phase; Calculating the power spectrum and microstate indicators of the resting-state electroencephalogram data based on the resting-state electroencephalogram data; Constructing a stroke prognosis prediction feature set based on the power spectrum, microstate indicators, preset patient baseline information, and preset clinical test data; Inputting the stroke prognosis prediction feature set into a preset machine learning model so that the machine learning model outputs a functional prediction result of the patient; Among them, the preset machine learning model is obtained by training the initial machine learning model and historical resting-state EEG data based on the Lasso regression algorithm and multiple machine learning algorithms.

2. The stroke prognosis prediction method according to claim 1, characterized in that: The preset machine learning model is obtained by training the initial machine learning model and the preset historical stroke prognosis prediction feature set based on the Lasso regression algorithm and multiple machine learning algorithms, specifically: Obtain the preset historical stroke prognosis prediction feature set; Inputting the historical stroke prognosis prediction feature set into an initial machine learning model for training, so that the initial machine learning model performs feature screening on the power spectrum, microstate indicators, baseline information, and clinical test data of the historical stroke prognosis prediction feature set according to a Lasso regression algorithm, and outputs each feature with the highest correlation with functional prognosis; According to the naive Bayes algorithm, random forest algorithm, support vector machine algorithm or preset machine learning algorithm, the features with the highest correlation with functional prognosis are trained to obtain the preset machine learning model.

3. The method for predicting stroke prognosis according to claim 2, wherein: The initial machine learning model is configured to perform feature screening on the power spectrum, microstate indicators, baseline information, and clinical test data of the historical stroke prognosis prediction feature set according to the Lasso regression algorithm, and output the features with the highest correlation with functional prognosis, specifically: The characteristics most closely associated with functional prognosis included the patient's age, admission NIHSS score, whether they had undergone tirofiban-enhanced antiplatelet therapy, fibrinogen, erythrocyte sedimentation rate, whether the carotid artery was stenotic, whether the subclavian artery was stenotic, Delta band frontal lobe power spectral density, DTABR, and microstate A coverage.

4. The method for predicting stroke prognosis according to claim 2, wherein: The method of obtaining a preset historical stroke prognosis prediction feature set is specifically: According to the preset modified Rankin scoring rules, the recovery status of each patient who has suffered acute ischemic stroke was scored using mRS; According to the mRS score results, the patients were divided into a good functional prognosis group and a poor functional prognosis group; Matching variable scores were performed on the patients in the good functional prognosis group and the poor functional prognosis group; wherein the matching variables included gender, age, history of hypertension, diabetes, heart disease, ischemic stroke, and NIHSS score at admission; A preset historical stroke prognosis prediction feature set is constructed based on the matching variable scoring results and the resting-state electroencephalogram data of each patient who has suffered from acute ischemic stroke.

5. A stroke prognosis prediction device, characterized in that: include: Acquisition module, calculation module, construction module and prediction module; The acquisition module is used to obtain resting-state electroencephalogram data of patients with acute ischemic stroke to be predicted during the acute phase; The calculation module is used to calculate the power spectrum and microstate indicators of the resting-state electroencephalogram data based on the resting-state electroencephalogram data; The construction module is used to construct a stroke prognosis prediction feature set based on the power spectrum, microstate indicators, preset patient baseline information and preset clinical test data; The prediction module is used to input the stroke prognosis prediction feature set into a preset machine learning model so that the machine learning model outputs the patient's functional prediction result; Among them, the preset machine learning model is obtained by screening variables based on the Lasso regression algorithm, and then training the initial machine learning model and historical resting-state EEG data using multiple machine learning algorithms.

6. The stroke prognosis prediction device according to claim 5, characterized in that: The preset machine learning model is obtained by performing variable screening based on the Lasso regression algorithm, and then training the initial machine learning model and the preset historical stroke prognosis prediction feature set using multiple machine learning algorithms, specifically: Obtain the preset historical stroke prognosis prediction feature set; Inputting the historical stroke prognosis prediction feature set into an initial machine learning model for training, so that the initial machine learning model performs feature screening on the power spectrum, microstate indicators, baseline information, and clinical test data of the historical stroke prognosis prediction feature set according to a Lasso regression algorithm, and outputs each feature with the highest correlation with functional prognosis; According to the naive Bayes algorithm, random forest algorithm, support vector machine algorithm or preset machine learning algorithm, the features with the highest correlation with functional prognosis are trained to obtain the preset machine learning model.

7. The stroke prognosis prediction device according to claim 6, characterized in that: The initial machine learning model is configured to perform feature screening on the power spectrum, microstate indicators, baseline information, and clinical test data of the historical stroke prognosis prediction feature set according to the Lasso regression algorithm, and output the features with the highest correlation with functional prognosis, specifically: The characteristics most closely associated with functional prognosis included the patient's age, admission NIHSS score, whether they had undergone tirofiban-enhanced antiplatelet therapy, fibrinogen, erythrocyte sedimentation rate, whether the carotid artery was stenotic, whether the subclavian artery was stenotic, Delta band frontal lobe power spectral density, DTABR, and microstate A coverage.

8. The stroke prognosis prediction device according to claim 6, characterized in that: The method of obtaining a preset historical stroke prognosis prediction feature set is specifically: According to the preset modified Rankin scoring rules, the recovery status of each patient who has suffered acute ischemic stroke was scored using mRS; According to the mRS score results, the patients were divided into a good functional prognosis group and a poor functional prognosis group; Matching variable scores were performed on the patients in the good functional prognosis group and the poor functional prognosis group; wherein the matching variables included gender, age, history of hypertension, diabetes, heart disease, ischemic stroke, and NIHSS score at admission; A preset historical stroke prognosis prediction feature set is constructed based on the matching variable scoring results and the resting-state electroencephalogram data of each patient who has suffered from acute ischemic stroke.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the stroke prognosis prediction method according to any one of claims 1 to 4.

10. A terminal device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the stroke prognosis prediction method according to any one of claims 1 to 4 when executing the computer program.

Images