Acute abdominal disease pre-examination triage progressive prediction method, system and device

Abstract

The invention relates to an acute abdominal disease pre-examination triage progressive prediction method, system and device, and the method comprises the steps: collecting the data of an acute abdominal disease patient, and carrying out the data preprocessing, and obtaining an acute abdominal disease data set; inputting the acute abdominal disease data set into a plurality of basic learners for training based on a stacked ensemble learning method to obtain meta-features; the meta-features and the acute abdominal disease data are gathered and input into a meta-model learning device for prediction, and a pre-examination triage prediction result is obtained. The method is a triage scheme of multi-step prediction and progressive correction, according to the scheme, according to the time progress of the emergency examination process of an emergency patient as the sequence, a stacked integrated learning method is applied to four acute abdominal disease data sets, an emergency classification result is output, and when the triage process is carried out in the clockwise direction, the triage classification result is obtained. The classification prediction result of the triage prediction system is dynamically updated; the method conforms to the actual triage process, omission and misjudgment of a traditional single-time fixed manual triage method can be avoided, and the accuracy of acute abdominal triage is improved.

Description

Technical Field

[0001] The present invention relates to the field of medical triage, and particularly to a progressive prediction method, system and device for emergency abdominal pain pre-triage and triage. Background Art

[0002] Emergency abdominal pain is a common and complex disease diagnosed in the emergency department. It refers to an acute pathological change that occurs in the abdominal cavity, pelvic cavity, and retroperitoneal tissues and organs, resulting in clinical syndromes with abdominal symptoms and signs as the main manifestations, accompanied by systemic reactions. Common emergency abdominal pains include: acute appendicitis, acute perforation of peptic ulcer, acute intestinal obstruction, acute biliary tract infection and cholelithiasis, acute pancreatitis, aortic dissection, urinary tract stones, abdominal trauma, ectopic pregnancy, uterine rupture, and so on. Such diseases cover a wide range of causes, have diverse manifestations, are complex in condition, and vary in criticality. They are a type of emergency with a high risk, rapid disease progression, poor patient comfort, and require emergency medical attention and concern.

[0003] Emergency triage is a basic intervention measure for the treatment of emergency patients and the key to the treatment of critically ill patients. Scientific and safe management of patient triage to ensure that patients in need of immediate medical treatment receive emergency treatment has always been a challenge and focus of research in emergency triage. At present, the emergency triage of patients with emergency abdominal pain mainly relies on the patient's chief complaint. Triage nurses classify patients based on personal experience and grade the patient's condition according to the pain score (1-10 points). The triage mode based on experience is prone to increasing the triage error rate of patients with emergency abdominal pain, prolonging the waiting time, and thus delaying the best treatment opportunity. Summary of the Invention

[0004] The present invention provides a progressive prediction method, system and device for emergency abdominal pain pre-triage and triage to solve at least one of the above technical problems.

[0005] The technical solution of the present invention to solve the above technical problems is as follows: A progressive prediction method for emergency abdominal pain pre-triage and triage includes:

[0006] S1, collecting first quadrant data of emergency abdominal pain patients based on objective assessment, and performing data preprocessing to obtain a first quadrant emergency abdominal pain data set; inputting the first quadrant emergency abdominal pain data set into a plurality of basic learners for training based on the stacked ensemble learning method to obtain first quadrant meta-features; combining the first quadrant meta-features and the first quadrant emergency abdominal pain data set and inputting them into a meta-model learner for prediction to obtain a first quadrant pre-triage and triage prediction result;

[0007] S2. Based on the pre-triage prediction results in the first quadrant, collect the data of the second quadrant of the acute abdomen patient based on manual assessment, perform data preprocessing, and merge it with the acute abdomen dataset in the first quadrant to obtain the acute abdomen dataset in the second quadrant. Based on the stacked ensemble learning method, input the acute abdomen dataset in the second quadrant into the multiple base learners for training to obtain the meta-features in the second quadrant. Merge the meta-features in the second quadrant and the acute abdomen dataset in the second quadrant and input them into the meta-model learner for prediction to obtain the pre-triage prediction results in the second quadrant;

[0008] S3. Based on the pre-triage prediction results in the second quadrant, collect the data of the third quadrant of the acute abdomen patient based on the chief complaint, perform data preprocessing, and merge it with the acute abdomen dataset in the second quadrant to obtain the acute abdomen dataset in the third quadrant. Based on the stacked ensemble learning method, input the acute abdomen dataset in the third quadrant into the multiple base learners for training to obtain the meta-features in the third quadrant. Merge the meta-features in the third quadrant and the acute abdomen dataset in the third quadrant and input them into the meta-model learner for prediction to obtain the pre-triage prediction results in the third quadrant;

[0009] S4. Based on the pre-triage prediction results in the third quadrant, collect the data of the fourth quadrant of the acute abdomen patient based on auxiliary examinations, perform data preprocessing, and merge it with the acute abdomen dataset in the third quadrant to obtain the acute abdomen dataset in the fourth quadrant. Based on the stacked ensemble learning method, input the acute abdomen dataset in the fourth quadrant into the multiple base learners for training to obtain the meta-features in the fourth quadrant. Merge the meta-features in the fourth quadrant and the acute abdomen dataset in the fourth quadrant and input them into the meta-model learner for prediction to obtain the final pre-triage prediction results.

[0010] Based on the above technical solutions, the present invention can be further improved as follows.

[0011] Further, the sample features of the acute abdomen dataset in the first quadrant include: gender, age, height, weight, way of coming for diagnosis, body temperature, pulse, respiration, blood pressure, and blood oxygen saturation.

[0012] Further, the sample features of the acute abdomen dataset in the second quadrant include: gender, age, height, weight, way of coming for diagnosis, body temperature, pulse, respiration, blood pressure, blood oxygen saturation, airway, respiration, circulation, and consciousness.

[0013] Further, the sample features of the acute abdomen dataset in the third quadrant include: gender, age, height, weight, way of coming for diagnosis, body temperature, pulse, respiration, blood pressure, blood oxygen saturation, airway, respiration, circulation, consciousness, chief complaint, symptoms, pain score, and medical history.

[0014] Further, the sample characteristics of the dataset of acute abdominal pain in the fourth quadrant include: gender, age, height, weight, way of coming for diagnosis, body temperature, pulse, respiration, blood pressure, blood oxygen saturation, airway, respiration, circulation, consciousness, chief complaint, symptoms, pain score, medical history, red blood cells, white blood cells, arterial oxygen partial pressure, lactic acid, serum creatinine, blood urea nitrogen, blood glucose, transaminase, direct bilirubin, albumin, PCT, CRP, electrocardiogram, and imaging diagnosis results.

[0015] Further, the multiple base learners include: support vector machine, random forest, and multi-layer perceptron neural network.

[0016] Further, the data preprocessing includes at least one of feature extraction processing, data merging processing, data cleaning processing, One-hot encoding processing, and data normalization processing.

[0017] Based on the above progressive prediction method for pre-triage of acute abdominal pain, the present invention also provides a progressive prediction system for pre-triage of acute abdominal pain.

[0018] A progressive prediction system for pre-triage of acute abdominal pain, comprising:

[0019] A first quadrant pre-triage prediction module, which is used to collect first quadrant data of acute abdominal pain patients based on objective assessment, and perform data preprocessing to obtain a first quadrant acute abdominal pain dataset; input the first quadrant acute abdominal pain dataset into multiple base learners for training to obtain first quadrant meta-features; merge and input the first quadrant meta-features and the first quadrant acute abdominal pain dataset into a meta-model learner for prediction to obtain a first quadrant pre-triage prediction result;

[0020] A second quadrant pre-triage prediction module, which is used to, based on the first quadrant pre-triage prediction result, collect second quadrant data of the acute abdominal pain patients based on artificial assessment and perform data preprocessing, and merge it with the first quadrant acute abdominal pain dataset to obtain a second quadrant acute abdominal pain dataset; input the second quadrant acute abdominal pain dataset into the multiple base learners for training to obtain second quadrant meta-features; merge and input the second quadrant meta-features and the second quadrant acute abdominal pain dataset into the meta-model learner for prediction to obtain a second quadrant pre-triage prediction result;

[0021] The third quadrant pre-triage prediction module is used to collect the third quadrant data of the acute abdomen patient based on the chief complaint on the basis of the second quadrant pre-triage prediction result, perform data preprocessing, and merge it with the second quadrant acute abdomen data set to obtain the third quadrant acute abdomen data set; input the third quadrant acute abdomen data set into the multiple basic learners for training to obtain the third quadrant meta-features; input the merged third quadrant meta-features and the third quadrant acute abdomen data set into the meta-model learner for prediction to obtain the third quadrant pre-triage prediction result;

[0022] The fourth quadrant pre-triage prediction module is used to collect the fourth quadrant data of the acute abdomen patient based on the auxiliary examination on the basis of the third quadrant pre-triage prediction result, perform data preprocessing, and merge it with the third quadrant acute abdomen data set to obtain the fourth quadrant acute abdomen data set; input the fourth quadrant acute abdomen data set into the multiple basic learners for training to obtain the fourth quadrant meta-features; input the merged fourth quadrant meta-features and the fourth quadrant acute abdomen data set into the meta-model learner for prediction to obtain the final pre-triage prediction result.

[0023] Based on the above-mentioned progressive prediction method for acute abdomen pre-triage, the present invention also provides a progressive prediction device for acute abdomen pre-triage.

[0024] A progressive prediction device for acute abdomen pre-triage includes a processor, a memory, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the progressive prediction method for acute abdomen pre-triage as described above.

[0025] The beneficial effects of the present invention are as follows: The progressive prediction method, system and device for acute abdomen pre-triage of the present invention improve the traditional single fixed manual triage method into a multi-step prediction and progressive correction triage scheme. This scheme is in the order of the time progress of the emergency examination process (objective evaluation, manual assessment, chief complaint, auxiliary examination) of emergency patients, corresponding to four acute abdomen data sets for the four steps of the emergency examination. The stacked ensemble learning method is applied to each acute abdomen data set to output the emergency classification result. When the triage process proceeds in the four quadrants in the clockwise direction, as the pre-triage data becomes gradually complete, especially the superposition of laboratory data and imaging data, the classification prediction result of the triage prediction system is also dynamically updated; this method conforms to the actual triage process, can avoid the omissions and misjudgments of the traditional single fixed manual triage method, and improve the accuracy of acute abdomen triage; in addition, the present invention adopts the machine learning method of stacked ensemble learning, which is applied to the ring triage process node data set, integrating the advantages of multiple basic learners, and its prediction accuracy is higher than that of a single machine learning. Description of the Drawings

[0026] Figure 1 This is a flowchart of a progressive prediction method for pre - triage of acute abdomen in the present invention;

[0027] Figure 2 This is a schematic diagram of a four - quadrant data structure;

[0028] Figure 3 This is a schematic diagram of an acute abdomen data set;

[0029] Figure 4 This is a schematic diagram of a stacking ensemble learning process;

[0030] Figure 5 This is a schematic diagram of data row - column conversion;

[0031] Figure 6 This is a schematic diagram of one - hot encoding;

[0032] Figure 7 This is a structural block diagram of a progressive prediction system for pre - triage of acute abdomen in the present invention. Detailed implementation manners

[0033] The principles and features of the present invention will be described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0034] As Figure 1 shown, a progressive prediction method for pre - triage of acute abdomen includes:

[0035] S1. Collect the first - quadrant data of acute abdomen patients based on objective evaluation, and perform data pre - processing to obtain the first - quadrant acute abdomen data set; based on the stacking ensemble learning method, input the first - quadrant acute abdomen data set into multiple base learners for training to obtain the first - quadrant meta - features; combine the first - quadrant meta - features and the first - quadrant acute abdomen data set and input them into the meta - model learner for prediction to obtain the first - quadrant pre - triage prediction result;

[0036] S2. On the basis of the first - quadrant pre - triage prediction result, collect the second - quadrant data of the acute abdomen patients based on manual assessment and perform data pre - processing, and combine it with the first - quadrant acute abdomen data set to obtain the second - quadrant acute abdomen data set; based on the stacking ensemble learning method, input the second - quadrant acute abdomen data set into the multiple base learners for training to obtain the second - quadrant meta - features; combine the second - quadrant meta - features and the second - quadrant acute abdomen data set and input them into the meta - model learner for prediction to obtain the second - quadrant pre - triage prediction result;

[0037] S3. Based on the pre-triage prediction results in the second quadrant, collect the data of the third quadrant of the acute abdomen patient based on the chief complaint and perform data preprocessing, and merge it with the acute abdomen dataset in the second quadrant to obtain the acute abdomen dataset in the third quadrant; based on the stacked ensemble learning method, input the acute abdomen dataset in the third quadrant into the multiple base learners for training to obtain the meta-features in the third quadrant; merge the meta-features in the third quadrant and the acute abdomen dataset in the third quadrant and input them into the meta-model learner for prediction to obtain the pre-triage prediction results in the third quadrant;

[0038] S4. Based on the pre-triage prediction results in the third quadrant, collect the data of the fourth quadrant of the acute abdomen patient based on the auxiliary examination and perform data preprocessing, and merge it with the acute abdomen dataset in the third quadrant to obtain the acute abdomen dataset in the fourth quadrant; based on the stacked ensemble learning method, input the acute abdomen dataset in the fourth quadrant into the multiple base learners for training to obtain the meta-features in the fourth quadrant; merge the meta-features in the fourth quadrant and the acute abdomen dataset in the fourth quadrant and input them into the meta-model learner for prediction to obtain the final pre-triage prediction results.

[0039] The stacked ensemble learning method of the present invention is a machine learning method, in which different learning algorithms are trained and then combined to obtain the final prediction. The ensemble algorithm trains different models on the same dataset and then combines their results to obtain a more accurate prediction. Its basic principle is that when multiple base learners are combined together, they can form a predictor with higher quality and reliability. Compared with relying on any single model, the ensemble method trains multiple learning models to adapt to the weaknesses and biases of other models. Ensemble learning can produce more predictions than a single machine learning method and can improve the overall accuracy and robustness of the system. Ensemble learning helps to solve some inherent problems in machine learning models, such as overfitting, underfitting, overvariance, and sensitivity to noise or anomalies. The key assumption of ensemble learning is that different base learners will produce uncorrelated errors. When the predictions from multiple base learners are intelligently aggregated, the errors are cancelled out and the correct predictions are strengthened. Ensemble learning is used in many different machine learning tasks where prediction accuracy is important. The ensemble is mainly responsible for improving the performance of the classification model. It can capture non-linear decision boundaries and complex interaction effects for classification problems. Healthcare example: Healthcare professionals can diagnose diseases more accurately by combining the outputs of individual models trained on various medical datasets (such as imaging data, patient health records).

[0040] Meanwhile, the present invention improves the traditionally single - time fixed artificial triage method into a multi - step prediction and progressive correction triage scheme. This scheme is in the order of the time progress of the emergency examination process (objective assessment, artificial evaluation, chief complaint, auxiliary examination) of emergency patients, corresponding to four acute abdomen datasets for the four steps of emergency examination. The stacked ensemble learning method is applied to each acute abdomen dataset to output the emergency classification result. When the triage process proceeds in a clockwise direction, as the pre - examination data gradually becomes complete, especially with the superposition of laboratory data and imaging diagnosis data, the classification prediction of the triage prediction system is also dynamically updated. This method conforms to the actual triage process, can avoid the omissions and misjudgments of the traditional single - time fixed artificial triage method, and improve the accuracy of acute abdomen triage.

[0041] In this specific embodiment:

[0042] Based on the subjective and objective grading criteria and combined with the important clinical parameters of acute abdomen patients, the present invention establishes a big data platform for acute abdomen patients that is scientific and standardized. Taking the objective assessment parameters and artificial evaluation indicators as the four - quadrant data, with the visit time as the axis, the clinical data of acute abdomen patients are collected clockwise. The four - quadrant dataset modules are Firstevaluation (corresponding to the first - quadrant data), Second evaluation (corresponding to the second - quadrant data), Thirdevaluation (corresponding to the third - quadrant data), and Fourth evaluation (corresponding to the fourth - quadrant data). The four - quadrant data structure is as Figure 2 shown. First evaluation includes the basic information and vital signs of the patient. Second evaluation includes A (airway), B (breathing), C (circulation), D (consciousness). Third evaluation includes the chief complaint symptoms and medical history of the patient. Fourth evaluation includes the electrocardiogram, laboratory indicators, imaging data, etc. of the patient. These are the indicators that can best reflect the severity of the patient's condition.

[0043] In this specific embodiment:

[0044] The sample features of the first - quadrant acute abdomen dataset include: gender, age, height, weight, way of coming to the hospital, body temperature, pulse, respiration, blood pressure, and blood oxygen saturation.

[0045] The sample features of the second - quadrant acute abdomen dataset include: gender, age, height, weight, way of coming to the hospital, body temperature, pulse, respiration, blood pressure, blood oxygen saturation, airway, breathing, circulation, and consciousness.

[0046] The sample features of the third quadrant acute abdomen dataset include: gender, age, height, weight, way of coming for diagnosis, body temperature, pulse, respiration, blood pressure, blood oxygen saturation, airway, breathing, circulation, consciousness, chief complaint, symptoms, pain score, and medical history.

[0047] The sample features of the fourth quadrant acute abdomen dataset include: gender, age, height, weight, way of coming for diagnosis, body temperature, pulse, respiration, blood pressure, blood oxygen saturation, airway, breathing, circulation, consciousness, chief complaint, symptoms, pain score, medical history, red blood cells, white blood cells, arterial oxygen partial pressure, lactic acid, serum creatinine, urea nitrogen, blood glucose, transaminase, direct bilirubin, albumin, PCT, CRP, electrocardiogram, and imaging diagnosis results.

[0048] As can be seen from the above, the first quadrant acute abdomen dataset, the second quadrant acute abdomen dataset, the third quadrant acute abdomen dataset, and the fourth quadrant acute abdomen dataset in the present invention are in an inclusive relationship in sequence, that is: Figure 3 is a schematic diagram of the acute abdomen dataset; among them, AAD - Base represents the first quadrant acute abdomen dataset, AAD - Manual represents the second quadrant acute abdomen dataset, AAD - Complain represents the third quadrant acute abdomen dataset, and AAD - Laboratory represents the fourth quadrant acute abdomen dataset.

[0049] In this specific embodiment:

[0050] The multiple base learners include: support vector machine, random forest, and multi - layer perceptron neural network.

[0051] The present invention adopts a multi - dataset stacking (Stacking) ensemble learning method. The stacking ensemble learning method combines the predictions of multiple models to create a potentially more accurate final prediction.

[0052] In the present invention, the stacking ensemble learning method is independently used to train the model for the acute abdomen dataset. In each model, the stacking ensemble learning process is as Figure 4 shown, and it is specifically divided into two stages:

[0053] Stage 1: Train the base learners. First, a group of different base learners are trained according to the acute abdomen dataset. Three different machine learning algorithms are used in this stage of training, and the three different machine learning algorithms are specifically support vector machine (SVM), random forest (RF), and multi - layer perceptron neural network (MLP).

[0054] Phase 2: Generate meta-features and construct the final model. The predictions made by each base learner (SVM, RF, MLP) in Phase 1 will be used to create a new set of features called meta-features. These meta-features capture the unique attributes of each data point. These meta-features are input into the final meta-model learner. The meta-model learner combines the acute abdomen dataset with the corresponding meta-features, and after training, weighs and combines the predictions from the base learners to make the final prediction.

[0055] In this specific embodiment:

[0056] The data preprocessing includes at least one of feature extraction processing, data merging processing, data cleaning processing, One-hot encoding processing, and data normalization processing.

[0057] Feature extraction processing:

[0058] Through the emergency specialist medical consortium platform, select the emergency data of acute abdomen patients. Data selection criteria: ① Patients who register and seek medical treatment in the emergency department; ② Patients over 14 years old; ③ Patients with the main complaint of chest and abdominal pain; ④ Informed consent. Exclusion criteria: ① Patients who do not pay or receive treatment on time according to the emergency department visit process; ② Patients in mass incidents; ③ Patients or their families who voluntarily terminate treatment or refuse rescue treatment; ④ Patients with incomplete case data.

[0059] Data desensitization processing. According to the inclusion and exclusion criteria of the research object, and in accordance with the patient reception and diagnosis and treatment process, delete / hide personal information such as the patient's name, visit number, and identity information in the clinical data.

[0060] Data merging processing:

[0061] Data merging is a preprocessing step for the original data. That is, the original data collected for constructing the four acute abdomen datasets is collected from the triage, outpatient, and laboratory databases, and the attribute values need to be merged according to the patient's visit ID number. It involves two steps:

[0062] (1) Data row-column conversion. That is, convert the row data of the test into column data, and use the Unstack method of the Pandas data processing tool to convert the stacked data into a flat layout along the column direction; among them, an example of data row-column conversion is Figure 5 as shown.

[0063] (2) Inner join of tables. The inner join associates the data scattered in multiple tables through the visit number ID. Query the data from two or more related tables through the inner join statement and return the records with equal fields in the two tables. In the present invention, through the inner join operation, all the attribute data of patients with the same visit number are queried and merged and output to one table.

[0064] Data cleaning processing:

[0065] Data cleaning refers to processing the extracted and merged data, including deleting duplicate or redundant data, performing logical verification on variable values, handling outliers, and handling missing data, etc. The data cleaning of the present invention involves spelling check, missing value handling, outlier detection and handling.

[0066] Spelling check. Check whether there are spelling mistakes in the text type attributes of each dataset, and adopt the Levenshtein Distance algorithm. The edit distance is a method to measure the difference between two strings. The smaller the edit distance between two strings, the more similar they are. After being processed by the edit distance algorithm, the text type attributes of the dataset are aligned and consistent.

[0067] Outlier handling. Outliers usually refer to data points that deviate from the normal range and do not conform to the expected pattern. The present invention adopts the interquartile range (IQR) outlier detection method to find outliers of numerical attributes. For abnormal data that significantly deviates from the upper and lower bounds of the empirical values, according to the type of abnormality, it is corrected by replacing with the upper and lower bound values or the median of the empirical values.

[0068] Missing value handling. First, check the missing situation of each sample data attribute row by row. For records where the proportion of the number of missing value attributes in all attribute numbers exceeds 60%, the strategy of deletion processing is adopted; after deletion processing, check the missing values of the remaining samples column by column. For the missing situation of categorical types, directly delete the record; for the missing situation of numerical types, since the data processed in the present invention is the laboratory test data of human biochemical indicators, which has a significant normal value range and data upper and lower bounds, the median of the empirical values is used for filling. After handling the missing values, through statistical, visualization and other methods, check whether the data distribution, correlation, etc. are reasonable and whether it conforms to the common sense of emergency business. This processing step adopts multiple iterative processes until the data meets the requirements.

[0069] One-hot encoding processing:

[0070] The role of one-hot encoding is to convert discrete categorical features into continuous features with one valid encoding. Using one-hot encoding for discrete features will make the distance calculation between features more reasonable. In the present invention, categorical attributes such as gender, way of coming for diagnosis, chief complaint, medical history, etc. are converted using one-hot encoding. For example, after performing one-hot encoding conversion on the way of coming for diagnosis, the categorical attribute is converted into 001, 010, and 100. Among them, the one-hot encoding example is as Figure 6 shown.

[0071] Data normalization processing:

[0072] The role of data normalization is to eliminate the influence of the dimension between attributes and scale the attribute values to the same comparable range. The present invention adopts a linear normalization method, also known as Min-max normalization. The linear normalization method performs a linear transformation on the original data and maps the data values to the range of [0,1]. It is expressed by the formula:

[0073] x′ = x - min(x) / (max(x) - min(x)).

[0074] Based on the above-mentioned progressive prediction method for acute abdomen pre-triage, the present invention also provides a progressive prediction system for acute abdomen pre-triage.

[0075] As Figure 7 shown, a progressive prediction system for acute abdomen pre-triage includes:

[0076] The first quadrant pre-triage prediction module is used to collect the first quadrant data of acute abdomen patients based on objective evaluation, perform data preprocessing to obtain the first quadrant acute abdomen data set; input the first quadrant acute abdomen data set into multiple basic learners for training to obtain the first quadrant meta-features; input the first quadrant meta-features and the first quadrant acute abdomen data set into the meta-model learner for prediction to obtain the first quadrant pre-triage prediction result;

[0077] The second quadrant pre-triage prediction module is used to, based on the first quadrant pre-triage prediction result, collect the second quadrant data of the acute abdomen patients based on manual evaluation and perform data preprocessing, and merge it with the first quadrant acute abdomen data set to obtain the second quadrant acute abdomen data set; input the second quadrant acute abdomen data set into the multiple basic learners for training to obtain the second quadrant meta-features; input the second quadrant meta-features and the second quadrant acute abdomen data set into the meta-model learner for prediction to obtain the second quadrant pre-triage prediction result;

[0078] The third quadrant pre-triage prediction module is used to, based on the second quadrant pre-triage prediction result, collect the third quadrant data of the acute abdomen patients based on the chief complaint and perform data preprocessing, and merge it with the second quadrant acute abdomen data set to obtain the third quadrant acute abdomen data set; input the third quadrant acute abdomen data set into the multiple basic learners for training to obtain the third quadrant meta-features; input the third quadrant meta-features and the third quadrant acute abdomen data set into the meta-model learner for prediction to obtain the third quadrant pre-triage prediction result;

[0079] The fourth quadrant pre - triage prediction module is used to, based on the pre - triage prediction result of the third quadrant, collect the fourth quadrant data of the acute abdomen patient based on auxiliary examinations and perform data pre - processing, and merge it with the third quadrant acute abdomen data set to obtain the fourth quadrant acute abdomen data set; input the fourth quadrant acute abdomen data set into the multiple base learners for training to obtain the fourth quadrant meta - features; merge the fourth quadrant meta - features and the fourth quadrant acute abdomen data set and input them into the meta - model learner for prediction to obtain the final pre - triage prediction result.

[0080] Based on the above - mentioned progressive prediction method for acute abdomen pre - triage, the present invention also provides an acute abdomen pre - triage four - quadrant progressive prediction device.

[0081] An acute abdomen pre - triage four - quadrant progressive prediction device includes a processor, a memory, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the acute abdomen pre - triage progressive prediction method as described above.

[0082] In summary, in the acute abdomen pre - triage progressive prediction method, system and device of the present invention: (1) An innovative progressive triage dynamic prediction method is applied to acute abdomen triage. The present invention develops a triage prediction method that conforms to the actual situation of triage, improving the traditional single - time fixed manual triage method to a multi - step prediction and progressive correction triage method. This method is in the order of the time progress of the emergency examination process (objective assessment, manual evaluation, chief complaint, auxiliary examination) of emergency patients, corresponding to four acute abdomen data sets for the four steps of the emergency examination. The stacked ensemble learning method is applied to each data set to output the emergency classification result. When the triage process proceeds in a clockwise direction, as the pre - triage data gradually becomes complete, especially with the superposition of laboratory data and imaging data, the classification prediction result of the triage prediction system is also dynamically updated. When the model output of the last step (auxiliary examination) is completed, the system outputs the final prediction result of the triage level of the patient. This method conforms to the actual triage process, can avoid the omissions and misjudgments of the traditional single - time fixed manual triage method, and improve the accuracy of acute abdomen triage. (2) A machine learning method of stacked ensemble learning is applied to the acute abdomen triage prediction. This method is applied to the ring - shaped triage process node data set, integrating the advantages of support vector machines, random forests, and multi - layer perceptron classifiers, and outputting the final prediction through the meta - learner. The experimental evaluation shows that the prediction accuracy of the ensemble learning model is higher than that of the single - mode triage prediction model.

[0083] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A progressive prediction method for pre-diagnosis and triage of acute abdomen, characterized in that: include: S1, collect the first quadrant data of acute abdomen patients based on objective evaluation, and perform data preprocessing to obtain the first quadrant acute abdomen data set; Based on the stacked ensemble learning method, the first quadrant acute abdomen data set is input into multiple basic learners for training to obtain the first quadrant meta-features; The first quadrant meta-features and the first quadrant acute abdomen data set are combined and input into the meta-model learner for prediction to obtain the first quadrant pre-examination triage prediction result; S2, based on the prediction results of the pre-examination and triage in the first quadrant, collect the second quadrant data of the acute abdomen patients based on manual evaluation and perform data preprocessing, and merge it with the first quadrant acute abdomen data set to obtain the second quadrant acute abdomen data set; based on the stacked ensemble learning method, input the second quadrant acute abdomen data set into the multiple basic learners for training to obtain the second quadrant meta-features; The second quadrant meta-features and the second quadrant acute abdomen data set are combined and input into the meta-model learner for prediction to obtain the second quadrant pre-examination triage prediction result; S3, based on the prediction results of the pre-examination and triage in the second quadrant, collecting the third quadrant data of the acute abdomen patient based on the chief complaint and performing data preprocessing, and merging it with the second quadrant acute abdomen data set to obtain the third quadrant acute abdomen data set; inputting the third quadrant acute abdomen data set into the multiple basic learners for training based on the stacked ensemble learning method to obtain the third quadrant meta-features; The third quadrant meta-features and the third quadrant acute abdomen data set are combined and input into the meta-model learner for prediction to obtain the third quadrant pre-examination triage prediction result; S4, based on the prediction results of the pre-examination and triage in the third quadrant, collecting the fourth quadrant data of the acute abdomen patient based on the auxiliary examination and performing data preprocessing, and merging it with the third quadrant acute abdomen data set to obtain a fourth quadrant acute abdomen data set; inputting the fourth quadrant acute abdomen data set into the multiple basic learners for training based on the stacked ensemble learning method to obtain the fourth quadrant element features; The fourth quadrant meta-features and the fourth quadrant acute abdomen data set are combined and input into the meta-model learner for prediction to obtain the final pre-examination and triage prediction results.

2. The method for predicting acute abdomen pre-diagnosis and triage according to claim 1, characterized in that: The sample characteristics of the first quadrant acute abdomen data set include: gender, age, height, weight, diagnosis method, body temperature, pulse, respiration, blood pressure and blood oxygen saturation.

3. The method for predicting acute abdomen pre-diagnosis and triage according to claim 2, characterized in that: The sample characteristics of the second quadrant acute abdomen dataset include: gender, age, height, weight, diagnosis method, body temperature, pulse, respiration, blood pressure, blood oxygen saturation, airway, respiration, circulation and consciousness.

4. The method for predicting acute abdomen pre-diagnosis and triage according to claim 3, characterized in that: The sample characteristics of the third quadrant acute abdomen dataset include: gender, age, height, weight, consultation method, body temperature, pulse, respiration, blood pressure, blood oxygen saturation, airway, respiration, circulation, consciousness, chief complaint, symptoms, pain score and medical history.

5. The method for progressive prediction of acute abdomen pre-diagnosis and triage according to claim 4, characterized in that: The sample characteristics of the fourth quadrant acute abdomen dataset include: gender, age, height, weight, consultation method, body temperature, pulse, respiration, blood pressure, blood oxygen saturation, airway, breathing, circulation, consciousness, chief complaint, symptoms, pain score, medical history, red blood cells, white blood cells, arterial oxygen partial pressure, lactic acid, blood creatinine, urea nitrogen, blood sugar, transaminase, direct bilirubin, albumin, PCT, CRP, electrocardiogram and imaging diagnosis results.

6. The method for progressive prediction of acute abdomen pre-diagnosis and triage according to claim 1, characterized in that: The multiple basic learners include: support vector machine, random forest and multi-layer perceptron neural network.

7. The method for progressive prediction of acute abdomen pre-diagnosis and triage according to claim 1, characterized in that: The data preprocessing includes at least one of feature extraction processing, data merging processing, data cleaning processing, one-hot encoding processing and data normalization processing.

8. An acute abdomen pre-diagnosis and triage progressive prediction system, characterized in that: include: The first quadrant pre-examination triage prediction module is used to collect the first quadrant data of acute abdomen patients based on objective evaluation, and perform data preprocessing to obtain the first quadrant acute abdomen data set; the first quadrant acute abdomen data set is input into multiple basic learners for training to obtain the first quadrant meta-features; The first quadrant meta-features and the first quadrant acute abdomen data set are combined and input into the meta-model learner for prediction to obtain the first quadrant pre-examination triage prediction result; A second quadrant pre-examination and triage prediction module, which is used to collect the second quadrant data of the acute abdomen patient based on manual assessment and perform data preprocessing on the basis of the first quadrant pre-examination and triage prediction result, and merge it with the first quadrant acute abdomen data set to obtain the second quadrant acute abdomen data set; Inputting the second quadrant acute abdomen data set into the multiple basic learners for training to obtain second quadrant meta-features; The second quadrant meta-features and the second quadrant acute abdomen data set are combined and input into the meta-model learner for prediction to obtain the second quadrant pre-examination triage prediction result; A third quadrant pre-examination and triage prediction module is used to collect the third quadrant data of the acute abdomen patient based on the chief complaint based on the second quadrant pre-examination and triage prediction result, perform data pre-processing, and merge it with the second quadrant acute abdomen data set to obtain the third quadrant acute abdomen data set; Inputting the third quadrant acute abdomen data set into the multiple basic learners for training to obtain third quadrant element features; The third quadrant meta-features and the third quadrant acute abdomen data set are combined and input into the meta-model learner for prediction to obtain the third quadrant pre-examination triage prediction result; A fourth quadrant pre-examination and triage prediction module is used to collect the fourth quadrant data of the acute abdomen patient based on auxiliary examinations and perform data preprocessing on the basis of the third quadrant pre-examination and triage prediction results, and merge it with the third quadrant acute abdomen data set to obtain a fourth quadrant acute abdomen data set; Inputting the fourth quadrant acute abdomen data set into the multiple basic learners for training to obtain fourth quadrant element features; The fourth quadrant meta-features and the fourth quadrant acute abdomen data set are combined and input into the meta-model learner for prediction to obtain the final pre-examination and triage prediction results.

9. A progressive prediction device for pre-diagnosis and triage of acute abdomen, characterized in that: The method comprises a processor, a memory and a computer program stored in the memory, wherein when the computer program is executed by the processor, the method for pre-examination and triage progressive prediction of acute abdomen as claimed in any one of claims 1 to 7 is implemented.

Images