Early risk prediction method and system for acute aortic dissection based on blood test

Abstract

The present invention belongs to the field of medical data processing technology, and relates to an early risk prediction method and system for acute aortic dissection based on blood tests. The method includes: constructing several types of single models; using historical patients' blood test index data and acute aortic dissection label data to train the single models respectively; inputting the blood test index data into the trained single models to obtain the prediction results of each single model; constructing an integrated model based on a neural network model to train an acute aortic dissection prediction model; inputting the prediction results of the single model into the acute aortic dissection prediction model to obtain the prediction results of the acute aortic dissection label data of the current patient. The present invention solves the problem that primary medical institutions cannot perform early prediction of acute aortic dissection through imaging means and misdiagnosis and missed diagnosis occurs, solves the problem that the prediction of acute aortic dissection depends on high-quality data, and improves the accuracy and robustness of the integrated model prediction.

Description

Technical Field

[0001] The present invention belongs to the technical field, and in particular, relates to a method and system for early risk prediction of acute aortic dissection based on blood test. Background Art

[0002] Acute aortic dissection (AAD) is a cardiovascular disease, and early diagnosis and timely intervention are extremely important. The traditional method is mainly to predict through computed tomography angiography, but it requires expensive equipment and high-level diagnostic doctors. Since hospitals do not have the conditions for CT angiography (CTA), this is often difficult to achieve in primary medical institutions.

[0003] Existing AAD prediction models based on imaging or electrocardiograms rely on high-quality data, but the availability of these data is limited in primary care institutions. Existing AAD prediction models perform poorly in early prediction and lack robustness, and are unable to make effective predictions before patients develop obvious symptoms.

[0004] Therefore, how to use the blood test indicators obtained by the existing better blood test conditions in primary medical institutions to build an artificial intelligence model to predict the patient's AAD risk early and intervene early has become a problem that needs to be solved urgently. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a method and system for early risk prediction of acute aortic dissection based on blood test.

[0006] In a first aspect, the present invention provides a method for early risk prediction of acute aortic dissection based on blood test, comprising:

[0007] Collect historical patient blood test index data and patient acute aortic dissection tag data from the hospital's electronic medical record system;

[0008] Construct several types of single models;

[0009] Use blood test index data and corresponding acute aortic dissection label data to train each type of single model to obtain a trained single model;

[0010] Input the blood test index data into the trained single model to obtain the prediction results of each single model;

[0011] Construct an integrated model based on a neural network model, use the prediction results of each single model as the input of the integrated model, use the patient's acute aortic dissection label data as the output of the integrated model, update the parameters of the neural network integrated model, train the integrated model, and obtain an acute aortic dissection prediction model;

[0012] Obtain the blood test index data of the current patient;

[0013] The blood test index data of the current patient is input into the trained single model to obtain the prediction results of each single model corresponding to the blood test index data of the current patient;

[0014] The prediction results of each single model corresponding to the current patient's blood test index data are input into the acute aortic dissection prediction model to obtain the prediction results of the current patient's acute aortic dissection label data.

[0015] In a second aspect, the present invention provides an early risk prediction system for acute aortic dissection based on blood test, comprising a data collection unit, a model building unit, a single model training unit, a first input unit, an integrated model building unit, an acquisition unit, a second input unit and an output unit;

[0016] A data collection unit for collecting historical patient blood test index data and patient acute aortic dissection tag data from the hospital's electronic medical record system;

[0017] A model building unit, used to build several types of single models;

[0018] A single model training unit, used to train each type of single model using blood test index data and corresponding acute aortic dissection label data to obtain a trained single model;

[0019] A first input unit, used to input blood test index data into the trained single model to obtain prediction results of each single model;

[0020] An integrated model building unit is used to build an integrated model based on a neural network model, use the prediction results of each single model as the input of the integrated model, use the patient's acute aortic dissection label data as the output of the integrated model, update the parameters of the neural network integrated model, train the integrated model, and obtain an acute aortic dissection prediction model;

[0021] An acquisition unit, used to acquire blood test index data of the current patient;

[0022] The second input unit is used to input the blood test index data of the current patient into the trained single model to obtain the prediction results of each single model corresponding to the blood test index data of the current patient;

[0023] The output unit is used to input the prediction results of each single model corresponding to the current patient's blood test index data into the acute aortic dissection prediction model to obtain the prediction results of the current patient's acute aortic dissection label data.

[0024] Based on the above technical solution, the present invention can also be improved as follows.

[0025] Furthermore, the single models include logistic regression model, random forest model, support vector machine model, gradient boosting model and decision tree model.

[0026] Furthermore, the logistic regression model adopts a binary classification model, the loss function is a logarithmic loss function, and the regularization parameter is 1.0; the random forest model adopts a five-fold cross validation, the number of trees is 100, and the depth of each tree is 5; the support vector machine model adopts a linear kernel function, the penalty coefficient is 1.0, and the number of iterations is 100; in the gradient boosting model, the learning rate is set to 0.1, and the number of iterations is 100; in the decision tree model, the maximum depth is set to 5 and the minimum number of sample splits is set to 5.

[0027] Further, after collecting the historical patient's blood test index data and the patient's acute aortic dissection label data, the blood test index data is classified according to the time of the test result to obtain the first blood test index data and the second blood test index data;

[0028] Construct several types of first single models and second single models; use the first blood test index data as the input of the first single model, use the second blood test index data as the input of the second single model, use the patient's acute aortic dissection label data as the output of the first single model and the second single model, and train each first single model and each second single model respectively.

[0029] Further, when constructing an integrated model based on a neural network model, a first integrated sub-model and a second integrated sub-model are constructed;

[0030] The prediction results of each first single model are used as the input of the first integrated sub-model, the patient's acute aortic dissection label data is used as the output of the first integrated sub-model, the first integrated sub-model is trained, and the first acute aortic dissection prediction model is obtained;

[0031] The prediction results of each first single model are used as the input of the first integrated sub-model, and the patient's acute aortic dissection label data is used as the output of the second integrated sub-model. The second integrated sub-model is trained to obtain a second acute aortic dissection prediction model.

[0032] Furthermore, the first blood test index data includes complete blood cell test index, brain natriuretic peptide test index and myocardial injury marker test index; the second blood test index data includes blood test electrolyte test index and DIC test index.

[0033] Furthermore, the complete blood cell examination indicators include at least one of the following: white blood cell count, neutrophil count, lymphocyte count, monocyte count, eosinophil count, basophil count, neutrophil rate, lymphocyte rate, monocyte rate, eosinophil rate, basophil rate, red blood cell count, hemoglobin, hematocrit, mean corpuscular volume, mean corpuscular hemoglobin, mean corpuscular hemoglobin concentration, standard deviation of corpuscular volume distribution width, coefficient of variation of corpuscular volume distribution width, platelet count, mean platelet volume, hematocrit, platelet volume distribution width and large platelet ratio; the myocardial injury marker examination indicators include at least one of high-sensitivity troponin, myoglobin and creatine kinase isoenzyme; the blood test electrolyte examination indicators include alanine aminotransferase, aspartate aminotransferase, aspartate aminotransferase, The test parameters include the ratio of amino acid aminotransferase to alanine aminotransferase, total protein, albumin, globulin, the ratio of albumin to globulin, total bilirubin, direct bilirubin, indirect bilirubin, total bile acid, r-glutamyl transferase, alkaline phosphatase, prealbumin, urea, uric acid, creatinine, total cholesterol, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, apolipoprotein A1, apolipoprotein B, blood glucose, fructosamine, potassium, sodium, chloride, calcium, carbon dioxide, retinol-binding protein, glomerular filtration rate, anion gap and prothrombin time; the DIC examination indicators include at least one of the international normalized ratio, prothrombin time ratio, prothrombin time activity, activated partial thromboplastin time, fibrinogen, D-dimer, antithromboplastin-III, fibrinogen degradation products and thrombin time.

[0034] The beneficial effects of the present invention are:

[0035] (1) The present invention realizes the application of an artificial intelligence model construction method based on blood tests in the early prediction of acute aortic dissection, solving the problem that primary medical institutions are unable to perform early prediction of acute aortic dissection through imaging methods such as CTA due to poor equipment and insufficient diagnostic capabilities, as well as the frequent occurrence of misdiagnosis and missed diagnosis;

[0036] (2) The present invention solves the problem that the existing prediction model for acute aortic dissection based on imaging or electrocardiogram relies on high-quality data when imaging or electrocardiogram data is difficult to obtain in primary medical institutions;

[0037] (3) The present invention solves the problem that the existing prediction model for acute aortic dissection performs poorly in processing early predictions, has insufficient robustness, and cannot effectively predict before the patient shows obvious symptoms. By combining a single model with an integrated model, the accuracy and robustness of the integrated model prediction are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A schematic diagram of the early risk prediction method for acute aortic dissection based on blood test provided in Example 1 of the present invention;

[0039] Figure 2 This is a schematic diagram of the early risk prediction system for acute aortic dissection based on blood test provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0041] Example 1

[0042] As an example, Figure 1 As shown, in order to solve the above technical problems, this embodiment provides an early risk prediction method for acute aortic dissection based on blood test, including:

[0043] Collect historical patient blood test index data and patient acute aortic dissection tag data from the hospital's electronic medical record system;

[0044] Construct several types of single models;

[0045] Use blood test index data and corresponding acute aortic dissection label data to train each type of single model to obtain a trained single model;

[0046] Input the blood test index data into the trained single model to obtain the prediction results of each single model;

[0047] Construct an integrated model based on a neural network model, use the prediction results of each single model as the input of the integrated model, use the patient's acute aortic dissection label data as the output of the integrated model, update the parameters of the neural network integrated model, train the integrated model, and obtain an acute aortic dissection prediction model;

[0048] Obtain the blood test index data of the current patient;

[0049] The blood test index data of the current patient is input into the trained single model to obtain the prediction results of each single model corresponding to the blood test index data of the current patient;

[0050] The prediction results of each single model corresponding to the current patient's blood test index data are input into the acute aortic dissection prediction model to obtain the prediction results of the current patient's acute aortic dissection label data.

[0051] The acute aortic dissection label data is used to characterize whether the patient has acute aortic dissection.

[0052] Optionally, single models include logistic regression model, random forest model, support vector machine model, gradient boosting model and decision tree model.

[0053] Optionally, the logistic regression model uses a binary classification model, the loss function is a logarithmic loss function, and the regularization parameter is 1.0; the random forest model uses a five-fold cross validation, the number of trees is 100, and the depth of each tree is 5; the support vector machine model uses a linear kernel function, the penalty coefficient is 1.0, and the number of iterations is 100; in the gradient boosting model, the learning rate is set to 0.1 and the number of iterations is 100; in the decision tree model, the maximum depth is set to 5 and the minimum number of sample splits is set to 5.

[0054] Optionally, after collecting the historical patient's blood test index data and the patient's acute aortic dissection label data, the blood test index data are classified according to the time taken for the test results to obtain the first blood test index data and the second blood test index data;

[0055] Construct several types of first single models and second single models; use the first blood test index data as the input of the first single model, use the second blood test index data as the input of the second single model, use the patient's acute aortic dissection label data as the output of the first single model and the second single model, and train each first single model and each second single model respectively.

[0056] Optionally, when constructing an integrated model based on a neural network model, a first integrated sub-model and a second integrated sub-model are constructed;

[0057] The prediction results of each first single model are used as the input of the first integrated sub-model, the patient's acute aortic dissection label data is used as the output of the first integrated sub-model, the first integrated sub-model is trained, and the first acute aortic dissection prediction model is obtained;

[0058] The prediction results of each first single model are used as the input of the first integrated sub-model, and the patient's acute aortic dissection label data is used as the output of the second integrated sub-model. The second integrated sub-model is trained to obtain a second acute aortic dissection prediction model.

[0059] Optionally, the first blood test index data include complete blood cell test indexes, brain natriuretic peptide test indexes and myocardial injury marker test indexes; the second blood test index data include blood test electrolyte test indexes and DIC (disseminated intravascular coagulation) test indexes.

[0060] Optionally, the complete blood cell examination indicators include at least one of white blood cell count, neutrophil count, lymphocyte count, monocyte count, eosinophil count, basophil count, neutrophil rate, lymphocyte rate, monocyte rate, eosinophil rate, basophil rate, red blood cell count, hemoglobin, hematocrit, mean corpuscular volume, mean corpuscular hemoglobin, mean corpuscular hemoglobin concentration, standard deviation of red blood cell volume distribution width, coefficient of variation of red blood cell volume distribution width, platelet count, mean platelet volume, hematocrit, platelet volume distribution width and large platelet ratio; the myocardial injury marker examination indicators include at least one of high-sensitivity troponin, myoglobin and creatine kinase isoenzyme; the blood test electrolyte examination indicators include alanine aminotransferase, aspartate aminotransferase, aspartate aminotransferase, The test parameters include the ratio of amino acid aminotransferase to alanine aminotransferase, total protein, albumin, globulin, the ratio of albumin to globulin, total bilirubin, direct bilirubin, indirect bilirubin, total bile acid, r-glutamyl transferase, alkaline phosphatase, prealbumin, urea, uric acid, creatinine, total cholesterol, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, apolipoprotein A1, apolipoprotein B, blood glucose, fructosamine, potassium, sodium, chloride, calcium, carbon dioxide, retinol-binding protein, glomerular filtration rate, anion gap and prothrombin time; the DIC examination indicators include at least one of the international normalized ratio, prothrombin time ratio, prothrombin time activity, activated partial thromboplastin time, fibrinogen, D-dimer, antithromboplastin-III, fibrinogen degradation products and thrombin time.

[0061] The blood test index data is classified according to the time it takes to get the test results, such as dividing the blood test index data into the first blood test index data with results within 15 minutes and the second blood test index data with results greater than 15 minutes and less than 2 hours. The first blood test index data with results within 15 minutes is as follows: Assume that the variables in the first blood test index data are set to X1, X2, X3, ... X 28, X1 is the number of white blood cells, X2 is the number of neutrophils, X3 is the number of lymphocytes, X4 is the number of monocytes, X5 is the number of eosinophils, X6 is the number of basophils, X7 is the neutrophil rate, X8 is the lymphocyte rate, X9 is the monocyte rate, X 10 is the eosinophil rate, X 11 is the basophil rate, X 12 is the number of red blood cells, X 13 is hemoglobin, X 14 is the hematocrit, X 15 is the average volume of red blood cells, X 16 is the average hemoglobin content of red blood cells, X 17 is the mean hemoglobin concentration of red blood cells, X 18 is the standard deviation of red blood cell volume distribution width, X 19 is the coefficient of variation of red blood cell volume distribution width, X 20 is the platelet count, X 21 is the average platelet volume, X 22 is the platelet volume, X 23 is the platelet volume distribution width, X 24 is the large platelet ratio, X 25 For brain natriuretic peptide, X 26 High-sensitivity troponin T, X 27 is myoglobin, X 28 is creatine kinase isoenzyme; the second blood test index data with a result greater than 15 minutes and less than 2 hours is as follows: the variables in the second blood test index data are set to X 29 , X 30 , X 31 ,……X 71 , X 29 Alanine aminotransferase, X 30 Aspartate aminotransferase, X 31 is the ratio of aspartate aminotransferase to aspartate aminotransferase, X 32 is total protein, X 33 is albumin, X 34 is globulin, X 35 is albumin / globulin, X 36 is total bilirubin, X 37 is direct bilirubin, X 38 is indirect bilirubin, X 39 is total bile acid, X 40 is r-glutamyl transferase, X 41 is alkaline phosphatase, X 42 is prealbumin, X 43 is urea, X 44 For uric acid, X 45is creatinine, X 46 is total cholesterol, X 47 is triglyceride, X 48 is high-density lipoprotein cholesterol, X 49 is low-density lipoprotein cholesterol, X 50 Apolipoprotein A1, X 51 Apolipoprotein B, X 52 is blood sugar, X 53 is fructosamine, X 54 is potassium, X 55 is sodium, X 56 is chlorine, X 57 For calcium, X 58 is carbon dioxide, X 59 Retinol binding protein, X 60 is the glomerular filtration rate, X 61 is the anion gap, X 62 is the prothrombin time, X 63 is the international normalized ratio, X 64 is the prothrombin time ratio, X 65 is the prothrombin time activity, X 66 is the activated partial thromboplastin time, X 67 is fibrinogen, X 68 is D-dimer, X 69 Antithrombin-III, X 70 X is a fibrinogen degradation product. 71 The blood test index data classification table is shown in Table 1.

[0062] Table 1 Classification table of blood test index data

[0063]

[0064] In actual application, let the set of input variables of the single model be X, and the set of output variables be Y. Let the nth blood test index data be represented by X n , the acute aortic dissection of the zth single model is denoted as Y z .

[0065] Assume that the variables in the first blood test index data are set to X1, X2, X3, ... X 28 , then the set of input variables of each single model is X (X1, X2, X3, ... X 28 ), the output variable of the single model is the patient's acute aortic dissection and is represented by Y z(AAD, non-AAD), the input variable of the integrated model is represented by Y(Y1,Y2,Y3,Y4,Y5) and the output variable of the integrated model is represented by Y0(1,0), 1 indicates that the patient meets the criteria of acute aortic dissection, and 0 indicates that the patient does not meet the criteria of acute aortic dissection.

[0066] The single model takes the logistic regression model, random forest model, support vector machine model, gradient boosting model and decision tree model as examples. When the input variable of the single model is the first blood test value index data, the corresponding relationship between the input variable and the output variable of the integrated model is shown in Table 2. When the input variable of the single model is the second blood test index data, the corresponding relationship between the input variable and the output variable of the integrated model is shown in Table 3.

[0067] Table 2 Correspondence between input variables and output variables of the integrated model when the single model input is the first blood test index data

[0068]

[0069] Table 3 Correspondence between input variables and output variables of the integrated model when the single model input is the second blood test index data

[0070]

[0071] An integrated model is constructed based on a neural network model. When the single model input is the first blood test index data and the second blood test index data, the prediction results of the above five single models (Y1, Y2, Y3, Y4, Y5) are integrated and learned respectively, the parameters of the neural network integrated model are updated, and the prediction results of each single model are comprehensively considered, thereby improving the accuracy and robustness of the integrated model prediction. When the probability of aortic dissection predicted by the integrated model is greater than the set threshold, the doctor is prompted to take medical measures such as imaging examinations or surgery to obtain the best time window for timely treatment of patients.

[0072] The optimized single model and integrated model are deployed to the hospital's electronic medical record system to obtain historical patient blood test index data in real time and output the risk prediction results of the integrated model for acute aortic dissection. Doctors can intervene and treat patients early based on the prediction results of the integrated model, thereby reducing the mortality rate of AAD.

[0073] The present invention realizes the application of the artificial intelligence model construction method based on blood test in the early prediction of acute aortic dissection, solves the problem that primary medical institutions are unable to perform early prediction of acute aortic dissection through imaging methods such as CTA due to poor equipment and insufficient diagnostic capabilities, and often suffer from misdiagnosis and missed diagnosis, thereby improving the accuracy of early prediction of acute aortic dissection.

[0074] In primary medical institutions, it is difficult to obtain imaging or electrocardiogram data. The present invention solves the problem that the existing prediction model for acute aortic dissection based on imaging or electrocardiogram relies on high-quality data.

[0075] The present invention solves the problem that the existing prediction model for acute aortic dissection performs poorly in processing early predictions, has deficiencies in robustness, and cannot make effective predictions before patients show obvious symptoms. By combining a single model with an integrated model, the accuracy and robustness of the integrated model prediction are improved.

[0076] Example 2

[0077] Based on the same principle as the method shown in Example 1 of the present invention, as shown in the attached Figure 2 As shown, an embodiment of the present invention also provides an early risk prediction system for acute aortic dissection based on blood test, including a data collection unit, a model building unit, a single model training unit, a first input unit, an integrated model building unit, an acquisition unit, a second input unit and an output unit;

[0078] A data collection unit for collecting historical patient blood test index data and patient acute aortic dissection tag data from the hospital's electronic medical record system;

[0079] A model building unit, used to build several types of single models;

[0080] A single model training unit, used to train each type of single model using blood test index data and corresponding acute aortic dissection label data to obtain a trained single model;

[0081] A first input unit, used to input blood test index data into the trained single model to obtain prediction results of each single model;

[0082] An integrated model building unit is used to build an integrated model based on a neural network model, use the prediction results of each single model as the input of the integrated model, use the patient's acute aortic dissection label data as the output of the integrated model, update the parameters of the neural network integrated model, train the integrated model, and obtain an acute aortic dissection prediction model;

[0083] An acquisition unit, used to acquire blood test index data of the current patient;

[0084] The second input unit is used to input the blood test index data of the current patient into the trained single model to obtain the prediction results of each single model corresponding to the blood test index data of the current patient;

[0085] The output unit is used to input the prediction results of each single model corresponding to the current patient's blood test index data into the acute aortic dissection prediction model to obtain the prediction results of the current patient's acute aortic dissection label data.

[0086] Optionally, single models include logistic regression model, random forest model, support vector machine model, gradient boosting model and decision tree model.

[0087] Optionally, the logistic regression model uses a binary classification model, the loss function is a logarithmic loss function, and the regularization parameter is 1.0; the random forest model uses a five-fold cross validation, the number of trees is 100, and the depth of each tree is 5; the support vector machine model uses a linear kernel function, the penalty coefficient is 1.0, and the number of iterations is 100; in the gradient boosting model, the learning rate is set to 0.1 and the number of iterations is 100; in the decision tree model, the maximum depth is set to 5 and the minimum number of sample splits is set to 5.

[0088] Optionally, after collecting the historical patient's blood test index data and the patient's acute aortic dissection label data, the blood test index data are classified according to the time taken for the test results to obtain the first blood test index data and the second blood test index data;

[0089] Construct several types of first single models and second single models; use the first blood test index data as the input of the first single model, use the second blood test index data as the input of the second single model, use the patient's acute aortic dissection label data as the output of the first single model and the second single model, and train each first single model and each second single model respectively.

[0090] Optionally, when constructing an integrated model based on a neural network model, a first integrated sub-model and a second integrated sub-model are constructed;

[0091] The prediction results of each first single model are used as the input of the first integrated sub-model, the patient's acute aortic dissection label data is used as the output of the first integrated sub-model, the first integrated sub-model is trained, and the first acute aortic dissection prediction model is obtained;

[0092] The prediction results of each first single model are used as the input of the first integrated sub-model, and the patient's acute aortic dissection label data is used as the output of the second integrated sub-model. The second integrated sub-model is trained to obtain a second acute aortic dissection prediction model.

[0093] Optionally, the first blood test index data includes complete blood cell test indexes, brain natriuretic peptide test indexes and myocardial injury marker test indexes; the second blood test index data includes blood test electrolyte test indexes and DIC test indexes.

[0094] Optionally, the complete blood cell examination indicators include at least one of white blood cell count, neutrophil count, lymphocyte count, monocyte count, eosinophil count, basophil count, neutrophil rate, lymphocyte rate, monocyte rate, eosinophil rate, basophil rate, red blood cell count, hemoglobin, hematocrit, mean corpuscular volume, mean corpuscular hemoglobin, mean corpuscular hemoglobin concentration, standard deviation of red blood cell volume distribution width, coefficient of variation of red blood cell volume distribution width, platelet count, mean platelet volume, hematocrit, platelet volume distribution width and large platelet ratio; the myocardial injury marker examination indicators include at least one of high-sensitivity troponin, myoglobin and creatine kinase isoenzyme; the blood test electrolyte examination indicators include alanine aminotransferase, aspartate aminotransferase, aspartate aminotransferase, The test parameters include the ratio of amino acid aminotransferase to alanine aminotransferase, total protein, albumin, globulin, the ratio of albumin to globulin, total bilirubin, direct bilirubin, indirect bilirubin, total bile acid, r-glutamyl transferase, alkaline phosphatase, prealbumin, urea, uric acid, creatinine, total cholesterol, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, apolipoprotein A1, apolipoprotein B, blood glucose, fructosamine, potassium, sodium, chloride, calcium, carbon dioxide, retinol-binding protein, glomerular filtration rate, anion gap and prothrombin time; the DIC examination indicators include at least one of the international normalized ratio, prothrombin time ratio, prothrombin time activity, activated partial thromboplastin time, fibrinogen, D-dimer, antithromboplastin-III, fibrinogen degradation products and thrombin time.

[0095] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An early risk prediction system for acute aortic dissection based on blood test, characterized in that: It includes a data collection unit, a model building unit, a single model training unit, a first input unit, an integrated model building unit, an acquisition unit, a second input unit and an output unit; The data collection unit is used to collect blood test index data of historical patients and acute aortic dissection label data of patients from the electronic medical record system of the hospital; after collecting the blood test index data of historical patients and acute aortic dissection label data of patients, the blood test index data are classified according to the time of the examination result to obtain the first blood test index data and the second blood test index data; the first blood test index data includes the whole blood cell examination index, the brain natriuretic peptide examination index and the myocardial injury marker examination index; the second blood test index data includes the blood test Electrolyte examination indicators and DIC examination indicators; complete blood cell examination indicators include white blood cell count, neutrophil count, lymphocyte count, monocyte count, eosinophil count, basophil count, neutrophil rate, lymphocyte rate, monocyte rate, eosinophil rate, basophil rate, red blood cell count, hemoglobin, hematocrit, mean corpuscular volume, mean corpuscular hemoglobin, mean corpuscular hemoglobin concentration, standard deviation of red blood cell volume distribution width, coefficient of variation of red blood cell volume distribution width, platelet count, mean platelet volume, blood platelet count ... At least one of platelet volume, platelet volume distribution width and large platelet ratio; myocardial injury markers include at least one of high-sensitivity troponin, myoglobin and creatine kinase isoenzyme; blood electrolyte test indicators include alanine aminotransferase, aspartate aminotransferase, aspartate aminotransferase to alanine aminotransferase ratio, total protein, albumin, globulin, albumin to globulin ratio, total bilirubin, direct bilirubin, indirect bilirubin, total bile acid, r-glutamyl transferase, alkaline phosphatase, prealbumin, urea, urine At least one of the following: acid, creatinine, total cholesterol, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, apolipoprotein A1, apolipoprotein B, blood glucose, fructosamine, potassium, sodium, chloride, calcium, carbon dioxide, retinol-binding protein, glomerular filtration rate, anion gap and prothrombin time; DIC examination indicators include at least one of the following: international normalized ratio, prothrombin time ratio, prothrombin time activity, activated partial thromboplastin time, fibrinogen, D-dimer, antithromboplastin-III, fibrinogen degradation products and thrombin time; The model building unit is used to build several types of first single models and second single models; the single models include logistic regression model, random forest model, support vector machine model, gradient boosting model and decision tree model; the logistic regression model adopts a binary classification model, the loss function is a logarithmic loss function, and the regularization parameter is 1.0; the random forest model adopts a five-fold cross validation, the number of trees is 100, and the depth of each tree is 5; the support vector machine model adopts a linear kernel function, the penalty coefficient is 1.0, and the number of iterations is 100; in the gradient boosting model, the learning rate is set to 0.1 and the number of iterations is 100; in the decision tree model, the maximum depth is set to 5 and the minimum number of sample splits is set to 5; The single model training unit is used to respectively use the blood test index data and the corresponding acute aortic dissection label data to train each type of single model to obtain a trained single model; the first blood test index data is used as the input of the first single model, the second blood test index data is used as the input of the second single model, and the patient's acute aortic dissection label data is used as the output of the first single model and the second single model, and each first single model and each second single model are trained respectively; A first input unit, used to input blood test index data into the trained single model to obtain prediction results of each single model; An integrated model construction unit, used to construct an integrated model based on a neural network model to obtain a first integrated sub-model and a second integrated sub-model; The prediction results of each first single model are used as the input of the first integrated sub-model, the patient's acute aortic dissection label data is used as the output of the first integrated sub-model, the first integrated sub-model is trained, and the first acute aortic dissection prediction model is obtained; The prediction results of each second single model are used as the input of the second integrated sub-model, and the patient's acute aortic dissection label data is used as the output of the second integrated sub-model, and the second integrated sub-model is trained to obtain a second acute aortic dissection prediction model; when training the integrated model, a cross-validation method is used to evaluate the integrated model, and the integrated model parameters are adjusted to obtain an acute aortic dissection prediction model; An acquisition unit, used to acquire blood test index data of the current patient; The second input unit is used to input the blood test index data of the current patient into the trained single model to obtain the prediction results of each single model corresponding to the blood test index data of the current patient; The output unit is used to input the prediction results of each single model corresponding to the current patient's blood test index data into the acute aortic dissection prediction model to obtain the prediction results of the current patient's acute aortic dissection label data.

Images