CN116403714B

CN116403714B - Cerebral apoplexy END risk prediction model building method and device, END risk prediction system, electronic equipment and medium

Info

Publication number: CN116403714B
Application number: CN202310369098.7A
Authority: CN
Inventors: 张立红; 张策
Original assignee: Dalian central hospital
Current assignee: Dalian central hospital
Priority date: 2023-04-07
Filing date: 2023-04-07
Publication date: 2024-01-26
Anticipated expiration: 2043-04-07
Also published as: CN116403714A

Abstract

The invention relates to the field of data information processing, and aims to solve the technical problem of how to predict and evaluate early nerve function deterioration risks of patients suffering from ischemic cerebral apoplexy. Therefore, the invention provides a method and a device for establishing an early neural function deterioration risk prediction model of a cerebral apoplexy thrombolysis patient, a risk prediction system, electronic equipment and a medium. By collecting personal data of a stroke patient, the stroke patient comprising a patient suffering from early deterioration of neurological function and a patient not suffering from early deterioration of neurological function, wherein the personal data comprises trapezin 1, and/or brain hemoglobin; creating a dataset using the collected personal data; and establishing the early-stage nerve function deterioration risk prediction model by using the data set, and predicting the final recovery effect of the patient by inputting personal data information of the cerebral apoplexy patient into the risk prediction model to generate the output of clinical prediction. The invention can timely, accurately and effectively predict and evaluate the final recovery effect of the ischemic cerebral apoplexy patient, and provide targeted treatment and nursing intervention for the patient, thereby improving the prognosis of the patient.

Description

Cerebral apoplexy END risk prediction model building method and device, END risk prediction system, electronic equipment and medium

Technical Field

The invention belongs to the field of data information processing, and particularly relates to a method and a device for establishing an early nerve function deterioration risk prediction model of a cerebral apoplexy thrombolytic patient, an early nerve function deterioration risk prediction system, electronic equipment and a medium.

Background

The cerebral apoplexy has high morbidity, high disability rate, high mortality rate, high recurrence rate and high economic burden, and is the first cause of death in China. The recanalization treatment is a treatment method for taking out thrombus by adopting medicines to dissolve and/or mechanical methods, and is an effective method for treating the ultra-early cerebral infarction. Acute ischemic stroke is the most common type of stroke, and early diagnosis and early treatment should be emphasized for the treatment of ischemic stroke. The use of recombinant tissue-type plasminogen activator (recombinant tissue plasminogen activator, rt-PA) for intravenous thrombolysis in the ultra-early stage of ischemic stroke (within 4.5 hours of onset) with evidence-based medical evidence currently internationally recognized is currently the most effective treatment for improving the clinical outcome in patients with acute ischemic stroke, but there are still some patients who develop increased neurological deficit, i.e., early neurological deterioration (early neurological deterioration, END) following standard dose of rt-PA thrombolytic therapy. At present, END definitions are not unified, evaluation indexes are different, the occurrence probability of END is 6.7% -29.8% and even up to 40% in the prior art, and END becomes an independent risk factor affecting the prognosis of nerve function, so that the death rate of patients can be directly increased. At present, a unified consensus is not formed on specific pathogenesis, risk factors, intervention measures and the like of the END, so that the intervention measures of the END are not exact.

In the present big data age, the existing medical data are utilized, and the technology can fully play a role in promoting medical progress by combining with an artificial intelligence learning method. The deep learning method in the artificial intelligence field is widely applied in many fields in life nowadays, but the combined research of the artificial intelligence field and the medical service field is still in the research development stage, the research of medical big data research at home and abroad is not widely applied in the aspects of clinical auxiliary diagnosis and monitoring, and the Keras framework and the Tensorflow framework provide brand new tools for big data simulation and operation, but the attempted application of the artificial intelligence tool in clinical research is not widely applied.

The related factors causing the acute ischemic cerebral apoplexy END are very complex, the related factors are reported in domestic and foreign documents in recent years, but the number of cases is small, the indexes are relatively dispersed, and a plurality of compound factor large sample researches based on a data mining method are seldom adopted by a group control research method, so that how to construct a cerebral apoplexy thrombolytic patient early nerve function deterioration risk prediction system based on big data, timely, accurately and effectively predict and evaluate the final recovery effect of the ischemic cerebral apoplexy patient, and provide targeted treatment and nursing intervention for the patient, thereby improving the prognosis of the cerebral apoplexy is a problem to be solved urgently.

Disclosure of Invention

In order to improve the technical problems, the invention provides a method and a device for establishing an early nerve function deterioration risk prediction model of a cerebral apoplexy thrombolytic patient, an early nerve function deterioration risk prediction system, electronic equipment and medium, and the method and the device discover the related compound risk factors by analyzing the related factors of an END patient, and timely, accurately and effectively predict and evaluate the final recovery effect of the ischemic cerebral apoplexy patient so as to play a certain role in preventing cerebral apoplexy progression and improving patient prognosis.

In order to solve the above technical problems, a first aspect of the present invention provides a method for establishing a prediction model of early-stage neural function deterioration risk of a cerebral apoplexy thrombolytic patient, which is characterized in that the method comprises the following steps: collecting personal data of a cerebral apoplexy patient, wherein the cerebral apoplexy patient comprises a patient suffering from early-stage nerve function deterioration and a patient not suffering from early-stage nerve function deterioration, and the personal data comprises trapezin 1 and/or cerebral red protein; the inclusion criteria of the cerebral apoplexy patients are as follows: patients with venous thrombolysis, which are over 18 years old and meet the recommendations of 2018 edition of Chinese acute ischemic cerebral apoplexy diagnosis and treatment guidelines, and patients with early nerve function deterioration after rt-PA thrombolysis intervention treatment process; the exclusion criteria for the stroke patients were: a stroke simulator, a patient with severely compromised efficacy metrics, a patient with special privacy requirements, or a bridging therapy patient; the personal data further includes a plurality of attributes: 1) General demographics selected from gender, age, BMI; 2) An admission vital sign selected from the group consisting of body temperature, pulse, respiration, blood pressure; 3) NIHSS baseline scores before and after treatment; 4) Past history, such as Shi Xuanzi drinking history, smoking history, hypertension history, diabetes history, hyperlipidemia history, atrial fibrillation history, coronary heart disease history, cerebral apoplexy history, transient ischemic attack history, and anti-blood A history of platelet drug administration, a history of statin administration; 5) Laboratory indicators selected from triacylglycerols, total cholesterol, blood glucose, glycosylated hemoglobin, high density lipoproteins, low density lipoproteins, erythrocyte sedimentation rate, fibrinogen, D-dimer, homocysteine, C-reactive protein, leukocytes, lymphocytes, platelets, monocytes, neutrophils, uric acid, serum ferritin, brain natriuretic peptide, aspirin resistance, urine protein/creatinine ratio, serum neuron specific alcoholise, SUA, fib, eGFR, SCr levels, coagulation examination, brain white matter demyelination; 6) The image index is selected from cervical color ultrasound, head and neck CTA or MRA, head and core magnetic scanning, DWI and TOAST parting; 7) A time parameter selected from the group consisting of time of onset to visit, time of arrival of patient at hospital to thrombolysis, and number of days of hospitalization; creating a dataset using the collected personal data; establishing the early neural function deterioration risk prediction model using the data set; the model is a DNN artificial intelligence learning model, the model comprises an input layer, at least one hidden layer and an output layer, any neuron of each layer is connected with any neuron of the next layer, each neuron of the input layer receives patient experiment sample data from a training sample set, the experiment sample data is scaled by respective weights and transmitted to the neurons of the hidden layer to generate intermediate output, each neuron of the hidden layer can further scale the intermediate output through the respective weights, the scaled intermediate output is then forwarded to the output layer, the output layer sums the scaled intermediate output to obtain a prediction output, parameters in the model are updated through back propagation error loss information in the modeling process, error comparison analysis is carried out on a result obtained by each training and an expected result, the weight and the threshold value of each neuron are corrected according to the comparison result, the error loss is converged through multiple training, and the model is enabled to continuously approximate to the expected result, wherein the weights of the neurons of the layers of the first iteration of the model are randomly set by software; the model function expression is: Wi is a weight coefficient of a linear relation, b is a bias quantity, x is a feature domain, and i is a positive integer; for each layer of the model, the parameter param= (the number of neurons of the layer) + (the number of neurons of the layer), wherein (the number of neurons of the layer) represents the input-kernel parameter and the number of neurons of the layer represents the bias parameter; the output layer compares the predicted output with a preset threshold value and outputs information indicating the effect of the patient after treatment; the risk prediction model output information comprises the effects of patients after treatment, NIHSS after thrombolysis and mRS.

A preferred embodiment according to the invention is characterized in that the personal data comprises at least: age, sex, history of coronary disease, mild atherosclerosis and stenosis, history of TIA, NIHSS before and after thrombolysis, history of cerebral apoplexy, history of diabetes, TOAST typing, history of hyperlipidemia, history of fasting abnormal blood glucose, and history of atrial fibrillation.

A preferred embodiment according to the invention is characterized in that the data set is randomly divided into a training sample set and a test sample set; training an early neural function deterioration risk prediction model using a training sample set; and introducing test sample set data to carry out test evaluation on the risk prediction model.

According to a preferred embodiment of the invention, the method is characterized in that the Relu activation function is used in the input layer and the hidden layer, the Sigmoid activation function is used in the output layer, and the result is directly output without the activation function in the face of time such metering data.

In a preferred embodiment of the invention, in the model test evaluation, the gradient descent evaluation of the model is performed using the average absolute error for data whose output variables are the metering data, and the gradient descent evaluation of the model is performed using the cross entropy for data whose output variables are the classification variables.

The second aspect of the present invention provides a device for establishing a model for predicting early-stage nerve function deterioration risk of a cerebral apoplexy thrombolytic patient, comprising: personal data information acquisition module for acquiring cerebral apoplexyPersonal data of a patient, the stroke patient comprising a patient suffering from early deterioration of neurological function and a patient not suffering from early deterioration of neurological function, wherein the personal data comprises trapezin 1, and/or brain hemoglobin; the inclusion criteria of the cerebral apoplexy patients are as follows: patients with venous thrombolysis, which are over 18 years old and meet the recommendations of 2018 edition of Chinese acute ischemic cerebral apoplexy diagnosis and treatment guidelines, and patients with early nerve function deterioration after rt-PA thrombolysis intervention treatment process; the exclusion criteria for the stroke patients were: a stroke simulator, a patient with severely compromised efficacy metrics, a patient with special privacy requirements, or a bridging therapy patient; the personal data further includes a plurality of attributes: 1) General demographics selected from gender, age, BMI; 2) An admission vital sign selected from the group consisting of body temperature, pulse, respiration, blood pressure; 3) NIHSS baseline scores before and after treatment; 4) A past history of Shi Xuanzi drinking history, smoking history, hypertension history, diabetes history, hyperlipidemia history, atrial fibrillation history, coronary heart disease history, cerebral apoplexy history, transient ischemic attack history, antiplatelet medicine taking history and statin medicine taking history; 5) Laboratory indicators selected from triacylglycerols, total cholesterol, blood glucose, glycosylated hemoglobin, high density lipoproteins, low density lipoproteins, erythrocyte sedimentation rate, fibrinogen, D-dimer, homocysteine, C-reactive protein, leukocytes, lymphocytes, platelets, monocytes, neutrophils, uric acid, serum ferritin, brain natriuretic peptide, aspirin resistance, urine protein/creatinine ratio, serum neuron specific alcoholise, SUA, fib, eGFR, SCr levels, coagulation examination, brain white matter demyelination; 6) The image index is selected from cervical color ultrasound, head and neck CTA or MRA, head and core magnetic scanning, DWI and TOAST parting; 7) A time parameter selected from the group consisting of time of onset to visit, time of arrival of patient at hospital to thrombolysis, and number of days of hospitalization; a data set making module for making a data set by using the collected personal data; a model building module for building the early neural function deterioration risk prediction model by using the data set; the model is The DNN artificial intelligence learning model comprises an input layer, at least one hidden layer and an output layer, wherein any neuron of each layer is connected with any neuron of the next layer, each neuron of the input layer receives patient experiment sample data from a training sample set, scales the experiment sample data by respective weights and transmits the experiment sample data to the neurons of the hidden layer to generate intermediate output, each neuron of the hidden layer can further scale the intermediate output by the respective weights, the scaled intermediate output is then forwarded to the output layer, the output layer sums the scaled intermediate output to obtain a predicted output, parameters in the model are updated by back propagation error loss information in the modeling process, error comparison analysis is carried out on a result obtained by each training and an expected result, the weight and the threshold value of each neuron are corrected according to the comparison result, the model is enabled to be approximate to the expected result continuously by multiple times of training and error loss convergence, and the weights of the neurons of the layers of the model are randomly set by software; the model function expression is:wi is a weight coefficient of a linear relation, b is a bias quantity, x is a feature domain, and i is a positive integer; for each layer of the model, the parameter param= (the number of neurons of the layer) + (the number of neurons of the layer), wherein (the number of neurons of the layer) represents the input-kernel parameter and the number of neurons of the layer represents the bias parameter; the output layer compares the predicted output with a preset threshold value and outputs information indicating the effect of the patient after treatment; the risk prediction model output information comprises the effects of patients after treatment, NIHSS after thrombolysis and mRS.

The third aspect of the present invention provides a system for predicting risk of early neurological deterioration of a patient suffering from cerebral apoplexy, wherein the risk prediction is performed by using a prediction model established by the above-mentioned model establishment method, and the risk detection device comprises: the personal data information acquisition module is used for acquiring clinical data information of a cerebral apoplexy patient, wherein the clinical data comprises trapezin 1 and/or cerebral red protein; and the recovery effect prediction module is used for inputting the personal data information of the cerebral apoplexy patient into the risk prediction model, predicting the final recovery effect of the patient and generating the output of clinical prediction.

A fourth aspect of the present invention proposes an electronic device comprising: a processor; and a memory storing computer executable instructions that, when executed, cause the processor to perform the method of the first aspect described above.

A fifth aspect of the present invention proposes a computer readable storage medium storing one or more programs which, when executed by a processor, implement the method of the first aspect described above.

According to the invention, by extracting clinical data of a cerebral apoplexy patient, predicting the risk of early nerve function deterioration, early identifying the early nerve function deterioration and judging the prognosis thereof, and further performing related intervention early, the problem that END is difficult to distinguish at the first time is solved, and the cure rate is improved; secondly, the level of the trapezin 1 and the level of the cerebral red protein are related to the onset risk and prognosis of the END, and the level of the trapezin 1 and the cerebral red protein in cerebral apoplexy patients also have differences, so that the prediction efficiency of the prediction model for the onset risk of the END is obviously improved after the new level of the trapezin 1 and the cerebral red protein are added.

Drawings

In order to make the technical problems solved by the present invention, the technical means adopted and the technical effects achieved more clear, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be noted, however, that the drawings described below are merely illustrative of exemplary embodiments of the present invention and that other embodiments of the drawings may be derived from these drawings by those skilled in the art without undue effort.

Fig. 1 is a schematic flow chart of an embodiment of a method for establishing a model for predicting early risk of neurological deterioration of a cerebral apoplexy thrombolytic patient based on big data according to the present invention.

Fig. 2 is a schematic diagram of one embodiment of a network topology of a DNN-based risk prediction model according to the present invention.

Fig. 3 is a schematic diagram of the final accuracy of a training set according to one embodiment of the present invention.

FIG. 4 is a schematic diagram of a one-time validation set final accuracy in accordance with an embodiment of the invention.

Fig. 5 is a schematic block diagram of an embodiment of an apparatus for establishing a model for predicting early-stage nerve function deterioration risk of a cerebral apoplexy thrombolytic patient based on big data according to the present invention.

Fig. 6 is a schematic block diagram of one embodiment of a system for predicting risk of early neurological deterioration of a stroke thrombolytic patient based on big data in accordance with the present invention.

Fig. 7 is a schematic diagram of a structural framework of an embodiment of a system for predicting risk of early neurological deterioration of a stroke thrombolytic patient based on big data according to the present invention.

Fig. 8 is a block diagram of a more specific embodiment of the system according to the invention.

FIG. 9 is a schematic diagram of one embodiment of a computer readable medium of the present invention.

Detailed Description

Exemplary embodiments of the present invention will now be described more fully hereinafter with reference to the accompanying drawings, in which exemplary embodiments are shown, although the exemplary embodiments may be practiced in various specific ways. Rather, these exemplary embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of the invention to those skilled in the art.

The structures, capabilities, effects, or other features described in a particular embodiment may be incorporated in one or more other embodiments in any suitable manner without departing from the spirit of the present invention.

In describing particular embodiments, specific details of construction, performance, effects, or other features are set forth in order to provide a thorough understanding of the embodiments by those skilled in the art. It is not excluded, however, that one skilled in the art may implement the present invention in a particular situation in a solution that does not include the structures, properties, effects, or other characteristics described above.

The flow diagrams in the figures are merely exemplary flow illustrations and do not represent that all of the elements, operations, and steps in the flow diagrams must be included in the aspects of the present invention, nor that the steps must be performed in the order shown in the figures. For example, some operations/steps in the flowcharts may be decomposed, some operations/steps may be combined or partially combined, etc., and the order of execution shown in the flowcharts may be changed according to actual situations without departing from the gist of the present invention.

The block diagrams in the figures generally represent functional entities and do not necessarily correspond to physically separate entities. That is, the functional entities may be implemented in software, or in one or more hardware modules or integrated circuits, or in different networks and/or processor devices and/or microcontroller devices.

The same reference numerals in the drawings denote the same or similar elements, components or portions, and thus repeated descriptions of the same or similar elements, components or portions may be omitted hereinafter. It will be further understood that, although the terms first, second, third, etc. may be used herein to describe various devices, elements, components or portions, these devices, elements, components or portions should not be limited by these terms. That is, these phrases are merely intended to distinguish one from the other. For example, a first device may also be referred to as a second device without departing from the spirit of the invention. Furthermore, the term "and/or," "and/or" is meant to include all combinations of any one or more of the items listed.

Techniques for performing clinical predictions are disclosed. Clinical predictions may predict a response to treatment of a disease, predict a patient's risk of suffering from the disease, etc. The technical scheme of the invention will be further described in detail below with reference to specific embodiments. It is to be understood that the following examples are illustrative only and are not to be construed as limiting the scope of the invention. All techniques implemented based on the above description of the invention are intended to be included within the scope of the invention.

Example 1

The invention provides a method for establishing an early neural function deterioration risk prediction model of a cerebral apoplexy thrombolytic patient based on big data.

The programming language used in the present invention is Python, and in one embodiment, the modeling software uses Python3.7.4 for analysis. In the neural network model modeling process, the Pandas module and the Numpy module of Python are adopted for data pre-processing (missing values and outliers). Training of the model is completed under a TensorFlow and Keras big data computing framework, and a TensorBoard modeling data visualization panel is adopted for data result visualization.

Those skilled in the art can know that the programming environment in the present invention can be set according to the above data, but the above data is only for ensuring that the early warning method provided in the present application achieves the best early warning effect for END, and other languages and frameworks can be adopted to complete the early warning, which is not limited herein.

The invention selects acute ischemic cerebral Apoplexy (AIS) patients which are not thrombolytic by rt-PA and are accepted by central hospitals of Dalian city from 1 month of 2007 to 7 months of 2022 as study subjects.

Data inclusion criteria: (1) age 18 years and older, men and women are not limited; (2) The intravenous thrombolytic patients which are in accordance with the guidelines for diagnosis and treatment of acute ischemic cerebral apoplexy in 2018 edition are recommended; (3) Patients who underwent early deterioration of neurological function during treatment with rt-PA thrombolytic intervention.

Data exclusion criteria: (1) a stroke simulator; (2) Efficacy indicators, such as NIHSS scoring patients with severe loss; (3) patients with special requirements for privacy security; (4) bridging treatment of the patient.

In a particular embodiment, 1747 patients with acute ischemic stroke alteplase intravenous thrombolysis are enrolled, of which 982 men and 765 women. The patient ages ranged from 52 to 80 years, with an average age of 66.68+ -8.22 years. Patients were classified into END groups (629 cases) and non-END groups (1118 cases) according to whether END occurred, and blood was drawn at different time points to detect trapezin-1 (CAV-1) and brain hemoglobin (Ngb).

Those skilled in the art can know that the inclusion and discharge standard in the present invention can be set according to the above data, but the above data is only for ensuring that the early warning method provided in the present application achieves the best early warning effect on END occurrence, and the inclusion and discharge standard is not limited.

The invention can simultaneously adopt the national institute of health cerebral apoplexy scale (National Institutes of Health Stroke Scale, NIHSS) to score the nerve function. The NIHSS score motor function item is increased by more than or equal to 1 point within 3 days of patient incidence, or the NIHSS total score is increased by more than or equal to 2 points compared with the incidence score, the patient is classified into an early nerve function deterioration group, namely an END group, or the patient is classified into a non-early nerve function deterioration group, namely a non-END group.

The invention carries out the following treatment treatments on patients: patients were given rt-PA thrombolytic therapy within 4.5 hours, conventional anti-thrombolytic, scavenging oxygen free radicals, lipid regulating, anti-platelet aggregation therapy, total dose of 0.9mg/kg, maximum upper limit of 90mg, dissolving drug with the same ml of water for injection as the total dose of rt-PA, adding the dissolved drug to physiological saline to prepare 100ml of liquid, 10% of the 100ml of liquid was administered by intravenous bolus injection within 1 minute, and the remaining 90ml of liquid was intravenous drip for 1 hour.

The present invention collects personal data of a patient. Personal data for each patient includes general characteristics of the patient and clinical treatment conditions including, but not limited to: 1) General demographics such as gender, age, BMI, etc.; 2) Admitted vital signs such as body temperature, pulse, respiration, blood pressure, etc.; 3) NIHSS baseline scores before and after treatment; 4) Past history such as history of drinking, history of smoking, history of hypertension, history of diabetes, history of hyperlipidemia, history of atrial fibrillation, history of coronary heart disease, history of cerebral apoplexy, history of Transient Ischemic Attack (TIA), history of administration of antiplatelet drugs, history of statin drugs, and the like; 5) Laboratory indicators such as trapin 1, cerebral hemoglobin, triacylglycerols, total cholesterol, blood glucose, glycosylated hemoglobin, high density lipoproteins, low density lipoproteins, erythrocyte sedimentation rate, fibrinogen, D-dimer, homocysteine, C-reactive protein, leukocytes, lymphocytes, platelets, monocytes, neutrophils, uric acid, serum ferritin, brain natriuretic peptide, aspirin resistance, urine protein/creatinine ratio, serum neuron specific alcoholises, SUA (serum uric acid), fib (fibrinogen), evfr (estimated glomerular filtration rate), blood creatinine (SCr) levels, coagulation examination, leukodemyelination, and the like; 6) Image indexes such as neck color ultrasound (carotid atherosclerosis stenosis degree), head and neck CTA or MRA, head and core magnetic scanning, DWI and the like; 7) Time parameters such as time of onset to visit, time of patient arrival to hospital to thrombolysis (DNT), and number of days in hospital.

The patients are monitored by routine vital signs after being admitted to the hospital; venous blood was withdrawn and SUA and serum creatinine (SCr) levels were measured using a fully automated biochemical analyzer; measuring the level of Fib by using a fully automatic coagulation analyzer; the eGFR is calculated. Serum Cav-1 levels were detected by ELISA (enzyme-linked immunosorbent assay). The concentration of Ngb was determined by ELISA. CT examination was performed 24 hours after thrombolysis. Patients were followed at 3 months and patient outcomes were assessed using a modified rank scale (Modified Rankin Scale, mRS). The improved rank scale is used for measuring the neurological function recovery condition of a patient after cerebral apoplexy, the result is good by defining mRS score less than or equal to 2, and the result is poor by defining mRS score more than 2.

The difference in sex, age, body Mass Index (BMI), past history (history of smoking, history of drinking, hypertension, diabetes, history of cerebral stroke, atrial fibrillation, coronary heart disease, etc.), pre-thrombolytic systolic pressure, pre-thrombolytic diastolic pressure, pre-thrombolytic National Institutes of Health Stroke Scale (NIHSS) score, hemoglobin, blood glucose, triacylglycerol, total cholesterol, thrombolytic time and responsible infarction, CT low density range ratio and eGFR, SUA, fib, cav-1, NGB levels of two groups of patients with two different prognosis groups of END were observed. Patients were classified into good and bad outcome groups by outcome.

Specifically, the method comprises the following steps:

s101: and acquiring personal data information of the cerebral apoplexy patient, and screening the patient information according to a preset inclusion exclusion standard. Each patient personal data information includes general characteristics of the patient and clinical treatment conditions.

Selecting a plurality of attributes from the general characteristics and clinical treatment conditions of the patient, the plurality of attributes selected from the group consisting of: 1) General demographics such as gender, age, BMI, etc.; 2) Admitted vital signs such as body temperature, pulse, respiration, blood pressure, etc.; 3) NIHSS baseline scores before and after treatment; 4) Past history such as history of drinking, history of smoking, history of hypertension, history of diabetes, history of hyperlipidemia, history of atrial fibrillation, history of coronary heart disease, history of cerebral apoplexy, history of Transient Ischemic Attack (TIA), history of administration of antiplatelet drugs, history of statin drugs, and the like; 5) Laboratory indicators such as trapin 1, cerebral hemoglobin, triacylglycerols, total cholesterol, blood glucose, glycosylated hemoglobin, high density lipoproteins, low density lipoproteins, erythrocyte sedimentation rate, fibrinogen, D-dimer, homocysteine, C-reactive protein, leukocytes, lymphocytes, platelets, monocytes, neutrophils, uric acid, serum ferritin, brain natriuretic peptide, aspirin resistance, urine protein/creatinine ratio, serum neuron specific alcoholises, SUA (serum uric acid), fib (fibrinogen), evfr (estimated glomerular filtration rate), blood creatinine (SCr) levels, coagulation examination, leukodemyelination, and the like; 6) Image indexes such as neck color ultrasound (carotid atherosclerosis stenosis degree), head and neck CTA or MRA, head and core magnetic leveling, DWI, TOAST typing and the like; 7) Time parameters such as time of onset to visit, time of patient arrival to hospital to thrombolysis (DNT), and number of days in hospital.

The above attributes are used as input variables of the model.

Notch 1 (cavol-l, cav-1) is the main structural component of the cell membrane pit (cavol) which is an area formed by inward recession. The cell membrane cellar exists on vascular endothelial cells and fibroblasts in a large quantity and participates in cell life activities such as endocytosis, signal transduction, macromolecule transportation and the like. Cav-1 participates in regulating and controlling free radical formation, blood brain barrier exudation, neuroinflammatory reaction, oxidative stress reaction process, myelin sheath repair, neuronal synapse regeneration and other processes in the central nervous system, thereby playing a central protective role.

The brain hemoglobin is also called as Neurogenin (NGB), is used as a novel high-efficiency oxygen-carrying globulin, has high affinity with oxygen, has the functions of improving the tolerance of brain tissues to ischemia and hypoxia and reducing the damage degree of brain tissues, has important functions on oxygen metabolism and oxygen utilization of tissue cells, and can play the neuroprotection role by increasing the tolerance of tissues to hypoxia, scavenging free radicals, resisting apoptosis and the like. Animal experiments find that when intracranial ischemia and hypoxia, oxidative stress and poison injury occur, the serum NGB expression is increased and is possibly closely related to the severity of ischemic cerebral apoplexy and the disease progression, but related clinical studies are rare, so that the change and evolution rule of the serum NGB level of ischemic cerebral apoplexy patients are clear, and whether the NGB can be used as a specific biological indicator for the occurrence, the development and the prognosis of progressive cerebral apoplexy is known.

Therefore, the trapezin 1 and the cerebral hemoglobin are selected as biological indication markers which indicate independent influence factors before and after thrombolysis and participate in the establishment of a neural network model.

In a particular embodiment, the serum Cav-1 level and the concentration of the brain globin Ngb of the patient are measured and scored on day 1 and day 14, respectively, of the patient's hospital admission therapy.

In a specific embodiment, serum Cav-1 levels are detected using ELISA.

In a particular embodiment, the concentration of Ngb is determined using enzyme-linked immunosorbent assay (ELISA).

In a specific embodiment, the method further comprises the step of preprocessing personal data information of the cerebral apoplexy patient.

S102: and creating a data set according to the collected personal data information.

And screening according to a preset inclusion exclusion standard by adopting a convenient sampling method, and excluding 1747 qualified experimental sample data after missing data are excluded, so as to manufacture an experimental sample data set.

The patient was divided into END groups (i.e., positive samples, for 629 cases) and non-END groups (i.e., negative samples, for 1118 cases) according to whether END occurred. Each sample has a plurality of attributes, selected from the general characteristics of the patient and the clinical treatment conditions described above, the sample data portions of which can be referred to as examples shown in table 1. The post-treatment effect is an experimental label, 0 represents good outcome and 1 represents poor outcome. Specific scoring criteria for mRS are described in table 2.

A digital feature vector may be generated to represent experimental sample data based on a mapping between the experimental sample data and a predetermined code.

Table 1 is a partial data presentation of one example of an experimental sample dataset.

Table 1 experimental sample data set

Specific scoring criteria for mRS are described in table 2.

Patient condition	Scoring criteria
		Completely asymptomatic	0
Although symptomatic, it has no obvious dysfunction and can complete all daily work and life	1
		Mild disability, failure to complete all pre-illness activities, but no help to care for their own daily routine	2
Moderate disability, xu Bufen, but can walk independently	3
		Moderately severe disabilities, can not independently walk, and needs other people to help in daily life	4
Severe disability, lying in bed, urinary incontinence and complete dependence of daily life on others	5
		Death of	6

Table 2 improved Rankine scale

Various weights may be set for each attribute in the experimental sample dataset while participating in the calculation. As shown in table 3, table 3 is one weight example of a partial attribute.

In a specific embodiment, the weights of the neurons of each layer in the first iteration of the model are randomly set by software, and the best effect is achieved through continuous iteration.

Attributes of	Weighting of
		History of coronary heart disease	0.582054
Slight atherosclerosis and restenosis	0.515542
		TIA medical history	0.512427
Sex (sex)	0.485705
		Death during hospitalization 10	0.47655
NIHSS (discharge) (minute)	-0.42113
		History of cerebral apoplexy	0.408869
History of diabetes	0.403163
		NIHSS (Admission time)	-0.32211
TOAST typing	0.321368
		History of hyperlipidemia	-0.31535
NIHSS (aggravation/non-aggravation)	0.291369
		Age of visit (age)	0.275641
Abnormal fasting blood glucose	0.243692
		History of atrial fibrillation	0.132481
Hemorrhage (CT)	0.025152

TABLE 3 Table 3

The invention trains the neural network model by the data set. The data is randomly divided into a training sample set and a test sample set, and the ratio of the training sample set and the test sample set may be predetermined based on the amount of data. The training sample set may be used to train the machine learning model, and the test sample set may be used to evaluate the machine learning model after training.

In one particular embodiment, the training sample set 1397 patients, the test sample set 350 patients, i.e., 80% of the data is the training sample set and 20% is the test sample set.

S103: an END risk prediction model is trained using the training sample set.

The training sample set is input into the model, so that the machine model can be trained, and a trained risk prediction model is obtained. The risk prediction model output data may include post-treatment effects for the patient, post-thrombolytic NIHSS, mRS, etc. scores. The risk prediction model is used for predicting the final recovery effect of the patient before medication by inputting factors before intervention.

In a specific embodiment, the pre-intervention factor is a brain globin.

In a specific embodiment, the pre-intervention factors are trapezium 1 and brain globin.

In a specific embodiment, the pre-intervention factor is trapezium 1.

The patient post-treatment effects include good outcome and poor outcome.

In a specific implementation mode, the invention adopts a TENSORFLOW module of Python to establish a deep neural network (DNN, deep neural network) artificial intelligent model for clinical research, and the basic idea is to establish the relation between input and output through a neural network algorithm and explore disease influence factors.

In a particular embodiment, the model of the present invention uses a training data set for 50 iterations to arrive at a neural network model.

Deep neural network technology originated in the fifth and sixty of the last century, possessing an input layer, an output layer and an hidden layer. These layers consist of neurons, which are the locations where computation occurs, that combine inputs from data with a set of coefficients or weights that either amplify or attenuate the importance of each input to the task that the algorithm is attempting to learn. These input weight products are summed and the sum is then passed through a so-called neuron activation function to determine whether and to what extent the signal further passes through the network to affect the final result. DNN uses a cascade of multiple layers of nonlinear processing units for feature extraction and conversion. Each subsequent layer uses the output of the previous layer as input. Higher level functionality is derived from lower level features, forming a hierarchical representation. The layers following the input layer may be convolutional layers that will generate feature maps that will filter the results of the input and be used by the next convolutional layer.

The perceptron is the basis of a neural network, the deep neural network model is an extension of the multi-layer perceptron, and the DNN comprises a complex fully-connected network structure. DNNs have a fully connected network character, and their layer-to-layer full connection can be interpreted as any, neuron of each layer being connected to all neurons of the next layer. Compared to MLP (Multi-Layer perceptron), DNN presents a complex network structure, while there is not much conflict with the perceptron model from a smaller local model, which is essentially a linear model, and is effectively mapped with an activation function.

In the invention, the feature vector input from the input layer obtains corresponding weight by calculating the hidden layers, and the final change result is transmitted to the output layer to obtain the prediction result.

Fig. 2 is a schematic diagram of one embodiment of the overall architecture of a DNN-based risk prediction model according to the present invention.

The DNN model comprises a 4-layer neural network as a model unit for deep DNN learning and calculation. The deep neural network adopted by the invention is a multi-layer and supervised (label) neural network, the model transmits the output characteristics of the upper layer to the input of the lower layer, continuous characteristic learning is carried out, after the layer-by-layer characteristic mapping, the large data sample characteristics of the existing space are better mapped to the characteristic space of another dimension, and different input characteristics are better expressed through the deep neural network learning. Sample data is better fitted through a plurality of nonlinear mapping transformations. The model adopts a supervised algorithm to perform network pre-training, continuously transmits and maps data characteristics to the next layer through layer-by-layer training, and fine-tunes all layers through a feedback network.

The DNN model effectively overcomes the defects that the traditional neural network is easy to overfit, has low training speed and the like, and improves the prediction accuracy through a high-efficiency training mechanism of the whole neural network.

The neural network is a four-layer neural network structure comprising an input layer, two hidden layers and an output layer, wherein the layers are fully connected, and any neuron of the ith layer is connected with any neuron of the (i+1) th layer. The neural network model is trained, i.e., the magnitudes of the parameters in the neural network model are modified. In the modeling process, the model fitting effect is improved by a method mainly relying on back propagation, error comparison analysis is carried out on the result obtained by each training and the expected result through the back propagation of errors, the weight and the threshold value of each neuron are corrected according to the comparison result, the model is enabled to be close to the expected result continuously through multiple times of training, and the accuracy of the model is improved. Specifically, the input signal is forwarded to the output to generate error loss, and the error loss is converged by the parameters in the neural network model updated by the back propagation error loss information. The back propagation algorithm is a back propagation motion that dominates the error loss, and aims to obtain parameters of the optimal neural network model, such as a weight matrix.

Each neuron of the input layer may receive experimental sample data of the patient and scale the experimental sample data with respective weights. The first weight may represent or be part of a scaling factor and may determine the effect of the experimental sample data on the final predicted output. Each neuron in the first hidden layer may receive scaled sample data from each neuron in the input layer and generate an intermediate output. Each neuron of the second hidden layer may further scale the intermediate output by a respective weight. The scaled intermediate output may then be forwarded to the output layer. The output layer sums the scaled intermediate outputs. And obtaining an intermediate output result by establishing a linear relation between the output and the input.

DNN is a feed-forward neural network in which each neural unit receives a weighted sum of inputs from the anterior layer after the first layer neurons acquire the input data. The DNN calculation and work targets are that an optimal model is obtained through continuous iteration, a certain continuous function is approximated, feedback comparison is carried out by using labeled original data, iteration is carried out, and a batch of optimal neural networks represented by weight values are obtained, so that effective mapping from feature input to a prediction result is obtained.

The left-most input layer and the right-most output layer of the deep neural network are respectively provided, a plurality of hidden layers are designed in the middle, the whole deep neural network model is sequentially calculated from left to right, and after data is input from the left side, a final result is given at the right-most side through a calculation process of a layer-by-layer full-connection layer. In supervised (labeled) learning, if the result of DNN calculation is different from the true value, the model calculates from right to left, calculates the error value of each neuron of each layer, adjusts the weight of each neuron, starts a new round of trial calculation process from left to right after completing data transmission from right to left, and performs continuous iteration until the model converges to a prediction model with highest efficiency, namely a batch of weight values which can be used, and completes the model training process. The weights of the neurons of each layer of the first iteration of the model are randomly set by software, and the best effect is achieved through continuous iteration.

Setting training inputs as:

T＝{(x ₁ ,y ₁ ),……(x _N ,y _N )}

where x is the feature domain, i=1, 2, … … N, y is the tag domain;

the test inputs are set as:

X’＝{x’ ₁ ,……,x’ _N }；

the test output is set as follows:

Y’＝{y’ ₁ ,……,y’ _N }。

the model function expression is as follows:

wherein W is _i The weight coefficient of the linear relation, b is the bias amount bias. And carrying out an operation process of the adder on each neuron, and realizing a data mapping process by activating a function.

As shown in fig. 2, the DNN network includes a full connection layer Dense (Dense) layer, a dense_1 (Dense) layer, a dense_2 (Dense) layer, and a dense_3 (Dense).

Taking the Dense (Dense) layer as an example, two parameters are Output in the Dense (Dense) layer, which are Output shape= (None, 64), param= 200960, respectively.

The first parameter of the Output Shape is Batch size, namely Batch size, when the Batch size is None, the data are put into the model for training at one time; the second is the shape input, and 64 one-dimensional data are input. This layer Param is 2176.

Param for each layer can be calculated as:

param= (number of neurons in the previous layer) × (number of neurons in the present layer) + (number of neurons in the present layer).

The term "input-kernel" is represented by (the number of neurons in the upper layer) × (the number of neurons in the present layer) × (the number of neurons in the bias).

In a specific embodiment, the model is subjected to 50 iteration processes, as shown in fig. 3 and 4, the accuracy of the final training set is 86.24%, the accuracy of the verification set is 87.28%, and the model has high test accuracy, good generalization capability and robustness and no over-fitting phenomenon.

The activation function is a nonlinear activation applied after each fully connected layer, thereby introducing nonlinearities into the model. The input layer and the hidden layer are both activated by Relu (linear rectification function), the output layer is Sigmoid function, and the result is directly output without activating function in the face of time measurement data (such as time from admission to thrombolysis DNT, time from onset to thrombolysis ONT, etc.). Gradient convergence can be accelerated by the ReLU function; the calculation power of the whole DNN neural network is saved through the threshold value.

The Relu activation function is as follows:

f(x))＝max(0,x)

the output layer processes the score with a Sigmoid activation function to generate a predicted output, the Sigmoid function being as follows:

the output layer may compare the predicted output to a preset threshold. If the predicted output meets or exceeds the threshold, the output layer may output an indication that the patient is likely to have a good prognosis, and if the predicted output is below the threshold, the output layer may output an indication that the patient is likely to have a poor prognosis.

S104: the generated model is subjected to test evaluation by inputting test sample set data into the trained model.

In a DNN artificial intelligent network, a model is built through a training set, and the generalization capability of the model is further inspected through a test data set, so that excessive generalization of the model caused by too sparse output is avoided. Specifically, the generated model is subjected to test evaluation by inputting test sample set data into the trained model.

In a specific embodiment, after 50 iterations are performed using the training dataset to obtain a deep neural network model, the test dataset is then validated on the model to determine whether the model has an over-fit or under-fit phenomenon, and the ability and robustness of the model to be generalized are tested.

For example, performance of the machine learning model may be assessed based on predetermined criteria, such as AUC (area under the curve) of a subject operating characteristic (receiver operating characteristic, ROC) curve. Inputting the test sample set into a machine learning model for testing, and obtaining an AUC value of the model; judging whether the AUC value of the model is smaller than a corresponding initial preset AUC value, and continuously classifying the remaining continuous variables when the AUC value of the model is smaller than the preset AUC value; when the AUC value of the model is equal to the preset AUC value, that is, the accuracy of the submodel reaches the maximum, the binning step is not executed. And carrying out multiple binning treatment on the residual continuous variables in the sub-model until the AUC value of the model reaches a preset value (namely a maximum value), so that the scoring of the model is more accurate.

It takes a lot of time to make adjustments after model training is complete to improve performance. And adjusting the model structure according to the reaction of the artificial intelligence model. Under the condition that the evaluation result of the trained neural network model does not meet the preset condition, the related parameters of the neural network model and/or the structure of the neural network model can be further updated until the evaluation result of the updated neural network model meets the preset condition and/or the update times reach the preset times.

In order to improve the accuracy of the patient condition prediction result, the loss function adopted is directly related to the prediction result when the model is trained, and the smaller the loss function is, the more accurate the patient condition prediction result is. The loss function measures the gap between the true value of the target and the model predicted value. Different problems require different loss functions.

In the model evaluation process, the gradient descent evaluation of the model is carried out by adopting an average absolute error according to the data of which the output variable is the metering data, and the gradient descent evaluation of the model is carried out by adopting a cross entropy according to the data of which the output variable is the classification variable (for example, the effect after treatment can be divided into good results or poor results, the results can be divided into different grades according to mRS and the like).

Random gradient descent (SGD) represents the simplest and most commonly used optimizer. Gradient descent algorithms typically calculate gradients (slopes) that are affected by given weight accuracy. If the gradients across the dataset need to be calculated to perform the weight updates, the speed is slower, while random gradient drops will perform the updates for each training image, one at a time. While this may lead to fluctuations in overall target accuracy or loss, it is more generalized than other approaches because it can jump into a new region of the loss parameter landscape and find a new minimum loss function.

Cross entropy loss is a commonly used loss function that tends to perform better than the simple mean square error between the true and predicted values. If the result of the network passes the Softmax layer, the distribution of cross entropy results in better accuracy. This is because it naturally maximizes the likelihood of correctly classifying the input data by not over-weighting the far outliers.

Example 2

As shown in fig. 5, the apparatus includes a clinical data information acquisition module 501, a data set making module 502, a model training module 503, and a model test evaluation module 504.

The clinical data information obtaining module 501 is configured to obtain personal data information of a patient suffering from cerebral apoplexy, and screen the patient information according to a preset inclusion exclusion criterion. Each patient personal data information includes general characteristics of the patient and clinical treatment conditions, and the specific functions are as in the process described in step S101 in example 1. The data set creating module 502 creates a data set using the personal data acquired by the personal data information acquiring module 501, and the specific function is the processing procedure described in step S102 in embodiment 1. The model training module 503 uses the training sample set to train the END risk prediction model, and the specific function is the processing procedure described in step S103 in embodiment 1. And a model test evaluation module 504 for performing test evaluation on the generated model by inputting test sample set data in the trained model, the specific function being as in the processing procedure described in step S104 in embodiment 1.

Since the clinical data information obtaining module 501, the data set making module 502, the model training module 503, and the model test evaluation module 504 correspond to the method steps S101-104 in embodiment 1, respectively, the description thereof will be omitted.

Example 3

The apparatus includes a personal data information acquisition module 601, a model construction module 602, and a restoration effect prediction module 603.

The personal data information acquisition module 602 is configured to acquire personal data information of a patient suffering from cerebral apoplexy, where each patient's personal data information includes general characteristics and clinical treatment conditions of the patient.

The general characteristics and clinical treatment conditions of the patient may include the following attributes: 1) General demographics such as gender, age, BMI, etc.; 2) Admitted vital signs such as body temperature, pulse, respiration, blood pressure, etc.; 3) NIHSS baseline scores before and after treatment; 4) Past history such as history of drinking, history of smoking, history of hypertension, history of diabetes, history of hyperlipidemia, history of atrial fibrillation, history of coronary heart disease, history of cerebral apoplexy, history of Transient Ischemic Attack (TIA), history of administration of antiplatelet drugs, history of statin drugs, and the like; 5) Laboratory indicators such as trapin 1, cerebral hemoglobin, triacylglycerols, total cholesterol, blood glucose, glycosylated hemoglobin, high density lipoproteins, low density lipoproteins, erythrocyte sedimentation rate, fibrinogen, D-dimer, homocysteine, C-reactive protein, leukocytes, lymphocytes, platelets, monocytes, neutrophils, uric acid, serum ferritin, brain natriuretic peptide, aspirin resistance, urine protein/creatinine ratio, serum neuron specific alcoholises, SUA (serum uric acid), fib (fibrinogen), evfr (estimated glomerular filtration rate), blood creatinine (SCr) levels, coagulation examination, leukodemyelination, and the like; 6) Image indexes such as neck color ultrasound (carotid atherosclerosis stenosis degree), head and neck CTA or MRA, head and core magnetic leveling, DWI, TOAST typing and the like; 7) Time parameters such as time of onset to visit, time of patient arrival to hospital to thrombolysis (DNT), and number of days in hospital.

In a specific embodiment, serum Cav-1 levels are detected using ELISA.

The model building module 601 is configured to build and train a risk prediction model for early nerve function deterioration of a cerebral apoplexy thrombolytic patient, wherein the model predicts a final recovery effect of the patient by using personal data information of the cerebral apoplexy patient, and the model building module is specifically implemented by the building device of the early nerve function deterioration risk prediction model of the cerebral apoplexy thrombolytic patient based on big data in embodiment 2, which is not described herein.

The recovery effect prediction module 603 is configured to input the general features of the patient and the clinical treatment condition acquired by the personal data information acquisition module 601 into the trained risk prediction model, predict a final recovery effect of the patient, and generate a clinical prediction output.

The risk prediction model output data may include patient post-treatment effects, NIHSS, mRS, etc. scores. The risk prediction model is used for predicting the final recovery effect of the patient before medication by inputting factors before intervention.

Example 4

FIG. 7 is a schematic block diagram of a risk prediction system for early neurological deterioration of a stroke thrombolytic patient based on big data according to an embodiment of the present invention.

As shown in fig. 7, the system includes a memory for storing a computer executable program (or instructions, etc.), and a data processing device for reading the computer executable program (or instructions, etc.) in the memory to perform the foregoing training method of the risk assessment model, as in the method of embodiment 1. The system may be a local system or a distributed system. The memory of the present invention may be a local memory or a distributed storage system, such as a cloud storage system. The data processor includes at least one device having digital information processing capabilities, such as a CPU, GPU, multiprocessor system, or cloud processor.

Example 5

An example of the structure of a local system is specifically described below with reference to fig. 8. The system may be regarded as an implementation in physical form for the method and apparatus embodiments of the invention described above. The details described in this embodiment of the system of the present invention should be regarded as supplementary to the above-described embodiments of the method or apparatus/system, not as limiting, but merely as an exemplary illustration of a local system condition, and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

As shown in fig. 8, the local system 200 of the exemplary embodiment is in the form of a general purpose data processing device. The components of example 200 of the system may include, but are not limited to: at least one processing unit 210 (i.e., an example of a specific data processing device), at least one storage unit 220 (i.e., an example of a specific memory), a bus 230 connecting the different system components (including the storage unit 220 and the processing unit 210), a display unit 240, and the like.

The storage unit 220 stores therein a computer readable program, which may be a source program or code of a program that is read only. The program may be executed by the processing unit 210 such that the processing unit 210 performs the steps of various embodiments of the present invention. For example, the processing unit 210 may perform the steps of the method of the foregoing embodiment 1.

The memory unit 220 may include readable media in the form of volatile memory units, such as Random Access Memory (RAM) 2201 and/or cache memory 2202, and may further include Read Only Memory (ROM) 2203. The storage unit 220 may also include a program/utility 2204 having a set (at least one) of program modules 2205, such program modules 2205 including, but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of which may include an implementation of a network environment.

Bus 230 may be a bus representing one or more of several types of bus structures including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

The local system 200 can also communicate with one or more external devices 300 (e.g., keyboard, display, network device, bluetooth device, etc.), such that devices can interact with the system 200 via the external devices 300, and/or such that the system 200 can communicate with one or more other data processing devices (e.g., routers, modems, etc.). Such communication may occur through an input/output (I/O) interface 250, and may also occur through a network adapter 260 to one or more networks, such as a Local Area Network (LAN), a Wide Area Network (WAN), and/or a public network, such as the internet. Network adapter 260 may communicate with other modules of electronic device 200 via bus 230. It should be appreciated that although not shown, other hardware and/or software modules may be used in electronic device 200, including, but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, data backup storage systems, and the like.

Example 6

Specifically, a computer readable medium storing one or more programs is also included, wherein the steps of the foregoing embodiment 1, which relate to the method of the present invention, are implemented when the one or more programs are executed by a processor. FIG. 9 is a schematic diagram of one embodiment of a computer readable medium of the present invention. The computer program may be stored on one or more computer readable media, which may be local or distributed, such as cloud storage, etc.

Those skilled in the art will appreciate that all or part of the steps implementing the above-described embodiments are implemented as a program, i.e., a computer program, executed by a data processing apparatus (including a computer). The above-described method provided by the present invention can be implemented when the computer program is executed. Furthermore, the computer program may be stored in a computer readable storage medium, i.e., a computer readable medium, which may be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium would include the following: an electrical connection having one or more wires, a portable disk, a magnetic disk, an optical disk, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or a storage array of any suitable combination of the foregoing, such as a plurality of storage media, for example, a magnetic disk or tape storage array. The computer program, when executed by one or more data processing apparatus, enables the computer readable medium to carry out the above-described methods of the present invention. Further, the storage medium is not limited to the centralized storage, but may be a distributed storage, such as cloud storage based on cloud computing. It should be appreciated that in the above description of exemplary embodiments of the invention, various features of the invention are sometimes described in the context of a single embodiment or with reference to a single figure in order to streamline the invention and aid those skilled in the art in understanding the various aspects of the invention. The present invention should not, however, be construed as including features that are essential to the patent claims in the exemplary embodiments.

Further, those skilled in the art will readily appreciate that the exemplary embodiments described herein may be implemented in software, or may be implemented in software in combination with the necessary hardware. Thus, the technical solution according to the embodiments of the present invention may be embodied in the form of a software product, which may be stored in a computer readable medium (which may be a CD-ROM, a usb disk, a mobile hard disk, etc.) or on a network, comprising several instructions to cause a data processing device (which may be a personal computer, a server, or a network device, etc.) to perform the above-described method according to the present invention. The computer readable medium may include a data signal propagated in baseband or as part of a carrier wave, with readable program code embodied therein. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination of the foregoing. A readable storage medium may also be any readable medium that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a readable storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing. Program code for carrying out operations of the present invention may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device, partly on a remote computing device, or entirely on the remote computing device or server. In the case of remote computing devices, the remote computing device may be connected to the user computing device through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computing device (e.g., connected via the Internet using an Internet service provider).

Thus, the present invention may be embodied in methods, systems, electronic devices, or computer readable media that execute computer programs. Some or all of the functions of the present invention may be implemented in practice using a general purpose data processing device such as a microprocessor or Digital Signal Processor (DSP).

It should be understood that modules, units, components, etc. included in the apparatus of one embodiment of the present invention may be adaptively changed to arrange them in an apparatus different from the embodiment. The different modules, units or components comprised by the apparatus of the embodiments may be combined into one module, unit or component or they may be divided into a plurality of sub-modules, sub-units or sub-components. The modules, units, or components of embodiments of the invention may be implemented in hardware, in software running on one or more processors, or in a combination thereof.

The foregoing description of the embodiments has been provided for the purpose of illustrating the general principles of the invention, and is not meant to limit the invention thereto, but to limit the invention thereto, and any modifications, equivalents, improvements and equivalents thereof may be made without departing from the spirit and principles of the invention.

In summary, the present invention may be implemented in a method, apparatus, system, or computer readable medium that executes a computer program. Some or all of the functions of the present invention may be implemented in practice using a general purpose data processing device such as a microprocessor or Digital Signal Processor (DSP).

The above-described specific embodiments further describe the objects, technical solutions and advantageous effects of the present invention in detail, and it should be understood that the present invention is not inherently related to any particular computer, virtual device or electronic apparatus, and various general-purpose devices may also implement the present invention. The foregoing description of the embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, alternatives, and improvements that fall within the spirit and scope of the invention.

Claims

1. The method for establishing the early nerve function deterioration risk prediction model of the cerebral apoplexy thrombolysis patient is characterized by comprising the following steps:

collecting personal data of a cerebral apoplexy patient, wherein the cerebral apoplexy patient comprises a patient suffering from early-stage nerve function deterioration and a patient not suffering from early-stage nerve function deterioration, and the personal data comprises trapezin 1 and/or cerebral red protein;

The inclusion criteria of the cerebral apoplexy patients are as follows: patients with venous thrombolysis, which are over 18 years old and meet the recommendations of 2018 edition of Chinese acute ischemic cerebral apoplexy diagnosis and treatment guidelines, and patients with early nerve function deterioration after rt-PA thrombolysis intervention treatment process;

the exclusion criteria for the stroke patients were: a stroke simulator, a patient with severely compromised efficacy metrics, a patient with special privacy requirements, or a bridging therapy patient;

the personal data further includes a plurality of attributes:

1) General demographics selected from gender, age, BMI;

2) An admission vital sign selected from the group consisting of body temperature, pulse, respiration, blood pressure;

3) NIHSS baseline scores before and after treatment;

4) A past history of Shi Xuanzi drinking history, smoking history, hypertension history, diabetes history, hyperlipidemia history, atrial fibrillation history, coronary heart disease history, cerebral apoplexy history, transient ischemic attack history, antiplatelet medicine taking history and statin medicine taking history;

5) Laboratory indicators selected from triacylglycerols, total cholesterol, blood glucose, glycosylated hemoglobin, high density lipoproteins, low density lipoproteins, erythrocyte sedimentation rate, fibrinogen, D-dimer, homocysteine, C-reactive protein, leukocytes, lymphocytes, platelets, monocytes, neutrophils, uric acid, serum ferritin, brain natriuretic peptide, aspirin resistance, urine protein/creatinine ratio, serum neuron specific alcoholise, SUA, fib, eGFR, SCr levels, coagulation examination, brain white matter demyelination;

6) The image index is selected from cervical color ultrasound, head and neck CTA or MRA, head and core magnetic scanning, DWI and TOAST parting;

7) A time parameter selected from the group consisting of time of onset to visit, time of arrival of patient at hospital to thrombolysis, and number of days of hospitalization;

creating a dataset using the collected personal data;

establishing the early neural function deterioration risk prediction model using the dataset;

randomly dividing the data set into a training sample set and a test sample set; training an early neural function deterioration risk prediction model using a training sample set; introducing test sample set data to test and evaluate the risk prediction model;

the model is a DNN artificial intelligence learning model, the model comprises an input layer, at least one hidden layer and an output layer, any neuron of each layer is connected with any neuron of the next layer, each neuron of the input layer receives patient experiment sample data from a training sample set, the experiment sample data is scaled by respective weights and transmitted to the neurons of the hidden layer to generate intermediate output, each neuron of the hidden layer further scales the intermediate output by respective weights, the scaled intermediate output is then forwarded to the output layer, the output layer sums the scaled intermediate output to obtain a prediction output, parameters in the model are updated by back propagation error loss information in the modeling process, error comparison analysis is carried out on a result obtained by each training and an expected result, the weight and a threshold value of each neuron are corrected according to the comparison result, the error loss is converged through multiple training, and the model is enabled to continuously approximate to the expected result, wherein the weights of the neurons of the layers of the first iteration of the model are randomly set by software;

The model function expression is:

wherein W is _i The weight coefficient is a linear relation, b is a bias quantity, x is a characteristic domain, and i is a positive integer;

for each layer of the model,

parameter param= (number of neurons in the upper layer) × (number of neurons in the present layer) + (number of neurons in the present layer),

wherein (the number of neurons in the upper layer) × (the number of neurons in the present layer) represents the input-kernel parameter, and the number of neurons in the present layer represents the bias parameter;

the output layer compares the predicted output with a preset threshold value and outputs information indicating the effect of the patient after treatment;

in the model test evaluation process, aiming at the data with the output variable being the metering data, carrying out gradient descent evaluation of the model by adopting an average absolute error, and aiming at the data with the output variable being the classification variable data, carrying out gradient descent evaluation of the model by adopting a cross entropy;

the risk prediction model output information comprises the effects of patients after treatment, NIHSS after thrombolysis and mRS.

2. The method of claim 1, wherein,

the personal data includes at least: age, sex, history of coronary disease, mild atherosclerosis and stenosis, history of TIA, NIHSS before and after thrombolysis, history of cerebral apoplexy, history of diabetes, TOAST typing, history of hyperlipidemia, history of fasting abnormal blood glucose, and history of atrial fibrillation.

3. The method of any one of claim 1 to 2, wherein,

the Relu activation function is adopted in the input layer and the hidden layer, the Sigmoid activation function is adopted in the output layer, and the result is directly output without the activation function for the metering data such as time.

4. An apparatus for establishing an early-stage nerve function deterioration risk prediction model of a cerebral apoplexy thrombolytic patient, comprising:

a personal data information acquisition module that acquires personal data of a stroke patient including a patient suffering from early-stage nerve function deterioration and a patient not suffering from early-stage nerve function deterioration, wherein the personal data includes trapezin 1, and/or brain hemoglobin;

the personal data further includes a plurality of attributes:

1) General demographics selected from gender, age, BMI;

3) NIHSS baseline scores before and after treatment;

a data set making module for making a data set by using the collected personal data;

a model building module that uses the data set to build the early neural function deterioration risk prediction model;

The model function expression is:

for each layer of the model,

5. A system for predicting risk of early nerve function deterioration in a cerebral apoplexy thrombolytic patient, wherein the risk prediction is performed by using a prediction model established by the model establishment method according to any one of claims 1 to 3,

The risk prediction system includes:

the personal data information acquisition module is used for acquiring clinical data information of a cerebral apoplexy patient, wherein the clinical data comprises trapezin 1 and/or cerebral red protein;

and the recovery effect prediction module is used for inputting the personal data information of the cerebral apoplexy patient into the risk prediction model, predicting the final recovery effect of the patient and generating the output of clinical prediction.

6. An electronic device, comprising:

a processor and a memory storing computer-executable instructions;

the computer executable instructions, when executed, cause the processor to perform the method of any of claims 1-3.

7. A computer readable medium storing one or more programs, which when executed by a processor, implement the method of any of claims 1-3.