Preoperative sign evaluation system for cardiovascular medicine department

Abstract

The invention relates to the technical field of preoperative evaluation, in particular to a cardiovascular medicine preoperative sign evaluation system which specifically comprises a data acquisition module, a statistical data calculation module, a data preprocessing module, a data integration module, a data enhancement module and a risk prediction module. In order to solve the problem that in a data enhancement link of cardiovascular medicine operation risk prediction, a sample is easily interfered by noise, unreliable data is generated, and prediction accuracy is reduced, a KNN classifier is used in a data enhancement module to further divide minority classes according to a prediction result so as to better distinguish noise data, and the prediction accuracy is improved. And the density of each minority class of data is calculated by using the improved Gaussian kernel function in the submerged space of the conditional variation auto-encoder, and noise data with relatively low density is filtered, so that the authenticity and reliability of the generated data are ensured, and the accuracy of surgical risk prediction of the cardiovascular medicine department is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of preoperative assessment, and in particular to a preoperative physical sign assessment system for cardiovascular medicine. Background Art

[0002] In the field of cardiovascular surgery, it is particularly important to conduct a comprehensive and accurate physical sign assessment of the patient before surgery. The effective implementation of this link can significantly enhance the safety of the surgery and provide a solid guarantee for the patient's life safety. However, the existing preoperative physical sign assessment system for cardiovascular surgery has the following problems:

[0003] (1) In the data enhancement process of cardiovascular surgery risk prediction, samples are easily disturbed by noise, resulting in unreliable data and reduced prediction accuracy;

[0004] (2) The differences between patient samples in the Department of Cardiology are small, and noisy data are difficult to filter;

[0005] (3) The sample dimensions of patients in the Department of Cardiology are relatively high, and the computational efficiency is low when using kernel functions for density estimation. Summary of the invention

[0006] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a cardiovascular preoperative physical sign assessment system. In view of the problem that in the data enhancement link of cardiovascular surgery risk prediction, samples are easily interfered by noise, thereby generating unreliable data, resulting in a decrease in prediction accuracy, the present invention proposes a data enhancement module, and uses a KNN classifier in the data enhancement module to further divide the minority class according to the prediction results to better distinguish the noise data, and then uses an improved Gaussian kernel function in the latent space of the conditional variational autoencoder to calculate the density of each minority class data, and filters the noise data with less density, thereby ensuring the authenticity and reliability of the generated data and improving the accuracy of cardiovascular surgery risk prediction.

[0007] The technical solution adopted by the present invention is as follows: The present invention provides a cardiovascular medicine preoperative physical sign assessment system, which specifically includes: a data acquisition module, a statistical data calculation module, a data preprocessing module, a data integration module, a data enhancement module and a risk prediction module, wherein the output end of the data acquisition module is respectively connected to the input ends of the statistical data calculation module, the data preprocessing module and the data integration module, the output end of the statistical data calculation module is connected to the input end of the data preprocessing module, the output end of the data preprocessing module is respectively connected to the input ends of the data integration module and the risk prediction module, the output end of the data integration module is connected to the input end of the data enhancement module, and the output end of the data enhancement module is connected to the input end of the risk prediction module;

[0008] The data acquisition module acquires the patient's pathological data, vital sign data and basic information, and specifically includes a pathological data acquisition unit, a vital sign data acquisition unit and a basic information acquisition unit;

[0009] The pathological data acquisition unit acquires pathological data of the patient;

[0010] The vital sign data collection unit collects the vital sign data of the patient;

[0011] The basic information acquisition unit acquires the basic information and survival status of the patient, where the survival status refers to whether the patient survives the cardiovascular surgery, and the survival status includes survival and death;

[0012] The statistical data calculation module calculates statistical data based on the patient's pathological data and physical sign data;

[0013] The data preprocessing module performs standardization processing on the patient's pathological data, physical sign data and statistical data to obtain standardized pathological data, standardized physical sign data and standardized statistical data;

[0014] The data integration module integrates the standardized pathological data, standardized vital sign data, basic information and survival status into integrated data;

[0015] The data enhancement module further divides the category of death into survival status using KNN classifier, filters the noise by conditional variational autoencoder combined with improved Gaussian kernel function, and then constructs new integrated data;

[0016] The risk prediction module uses a random forest model to predict the patient's risk of death from cardiovascular surgery.

[0017] The data preprocessing module performs standardization processing on the patient's pathological data, physical sign data and statistical data to obtain standardized pathological data, standardized physical sign data and standardized statistical data. The specific steps of the standardization processing are as follows:

[0018] Step S1: Count the mean and standard deviation of each item of pathological data, physical sign data and statistical data of all patients;

[0019] Step S2: Subtract the mean value from each item of the pathological data, physical sign data and statistical data of each patient and divide the result by the standard deviation to obtain standardized pathological data, standardized physical sign data and standardized statistical data, respectively. The specific formulas are as follows:

[0020] ;

[0021] in, express After standardization, Indicates the pathological data, physical sign data and statistical data Item data, represents the mean, Represents standard deviation.

[0022] The data enhancement module further divides the category of death into survival status using KNN classifier, filters the noise by conditional variational autoencoder combined with improved Gaussian kernel function, and then constructs new integrated data. The specific steps are as follows:

[0023] Step D1: define the majority class and the minority class: the integrated data of patients whose survival status is alive is taken as the majority class, and the integrated data of patients whose survival status is dead is taken as the minority class;

[0024] Step D2: Use KNN classifier to divide the minority class into difficult class and easy class: Take survival status as the target object of KNN classifier classification task, use integrated data of all patients to train KNN classifier, after training, use all minority class samples as test data set, use KNN classifier to classify the test data set, and use all samples classified as survival as difficult class, and all samples classified as death as easy class;

[0025] Step D3: Construct and train a conditional variational autoencoder: The conditional variational autoencoder specifically includes an encoder, a latent space, a decoder, and a classifier. An objective function is constructed, and the conditional variational autoencoder is trained to maximize the objective function. The formula of the objective function is as follows:

[0026] ;

[0027] in, represents the objective function, represents the latent variable distribution of the encoder output, express The log-likelihood of represents the reconstructed distribution of the decoder output, represents the prior distribution of the latent variable, represents the KL divergence term of the variational upper bound, The distribution of variables representing the output of the classifier;

[0028] Step D4: Use the improved Gaussian kernel function to filter the noise of the difficult and easy classes. The specific steps are as follows:

[0029] Step D41: Calculate the density of each sample of the difficult class and the easy class in the latent space by improving the Gaussian kernel function. The specific formula is as follows:

[0030] ;

[0031] in, and Respectively and The corresponding value of samples in the latent space, express The density of Indicates The total number of samples corresponding to the category, represents the width of the improved Gaussian kernel function, Represents improved Gaussian kernel function;

[0032] The specific formula of the improved Gaussian kernel function is as follows:

[0033] ;

[0034] in, Indicates that the independent variable is The improved Gaussian kernel function, is the independent variable, represents weight;

[0035] Step D42: define a first threshold and a second threshold, remove samples with a density less than the first threshold in the difficult class, remove samples with a density less than the second threshold in the easy class, and use all remaining samples of the difficult class and the easy class as the screened integrated data;

[0036] Step D5: Generate new integrated data based on the screened integrated data. The specific steps are as follows:

[0037] Step D51: traverse the filtered integrated data and use the traversed integrated data as the current sample;

[0038] Step D52: Calculate the Euclidean distance between the current sample and other integrated data of the same category, and select P samples with the shortest Euclidean distance as neighboring samples;

[0039] Step D53: randomly select a sample from the neighboring samples, and use the sample and the current sample as the benchmark to generate new integrated data. The basic information of the new integrated data is consistent with the current sample. The specific formula is as follows:

[0040] ;

[0041] in, Indicates the first part of the new integrated data except basic information The value of the item data, and Represents the current sample and randomly selected neighboring samples in addition to the basic information The value of the item data, Represents a random number between 0 and 1;

[0042] Step D54: Define the maximum number of samples, and repeat steps D51-D53 until the total number of new integrated data generated reaches the maximum number of samples.

[0043] The risk prediction module uses a random forest model to predict the patient's risk of death from cardiovascular surgery. The specific steps are as follows:

[0044] Step A1: Constructing a data set: The data set specifically includes standardized pathological data, standardized physical sign data, standardized statistical data and basic information of all patients;

[0045] Step A2: Supplementing the data set: Obtaining statistical data of each new integrated data through the statistical data calculation module, and then obtaining standardized statistical data through the data preprocessing module, adding the standardized statistical data to the new integrated data, and adding the new integrated data to the data set;

[0046] Step A3: Training the random forest model: The random forest model is trained using the data set. The random forest model classifies and predicts the survival status of each data item in the data set. The random forest model is evaluated using the cross-validation method.

[0047] Step A4: Use the random forest model to predict the surgical mortality risk of new patients: Use the data acquisition module, statistical data calculation module, and data preprocessing module to obtain the standardized pathological data, standardized vital sign data, standardized statistical data, and basic information of the new patients, and input them into the random forest model to obtain the predicted results of survival status.

[0048] The beneficial results achieved by the present invention using the above scheme are as follows:

[0049] (1) In order to solve the problem that in the data enhancement process of cardiovascular surgery risk prediction, samples are easily disturbed by noise, thus generating unreliable data and reducing the accuracy of prediction, the present invention proposes a data enhancement module, and uses the KNN classifier in the data enhancement module to further divide the minority class according to the prediction results to better distinguish the noise data, and then uses the improved Gaussian kernel function in the latent space of the conditional variational autoencoder to calculate the density of each minority class data, and filters the noise data with lower density, thereby ensuring the authenticity and reliability of the generated data and improving the accuracy of cardiovascular surgery risk prediction.

[0050] (2) To address the problem that the differences between patient samples in the Department of Cardiology are small and noise data is difficult to filter, first, the present invention trains a KNN classifier to predict the survival status of patients after surgery, and further divides minority class samples into easy and difficult classes according to the difficulty of prediction, thereby increasing the degree of feature differentiation of minority class patient samples. Secondly, the present invention proposes an improved Gaussian kernel function, and calculates the density of patient data in the latent space for each survival category of death through the improved Gaussian kernel function. The improved Gaussian kernel function introduces an adaptive strategy based on the original function, which dynamically adjusts the behavior of the kernel function according to the gap between patient data. When the gap is large, the Gaussian kernel function is used for calculation, and when the gap is small, a linearization strategy is used for weighted calculation, thereby improving the sensitivity of the kernel function to similar patient data, making the noise data in the patient sample data set easier to be screened and filtered.

[0051] (3) In order to address the problem of high dimensionality of patient samples in the Department of Cardiology and low computational efficiency when using kernel functions for density estimation, the present invention uses a conditional variational autoencoder to learn the distribution of patient data and filter out the noise in the minority class in the latent space, thereby reducing the patient's high-dimensional data to a low-dimensional space, thereby retaining the original characteristics of the patient data while reducing the computational complexity. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a module diagram of a cardiovascular medicine preoperative physical sign assessment system proposed by the present invention;

[0053] Figure 2 Flowchart for generating new samples for the data augmentation module.

[0054] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0055] 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 only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0056] Example 1, see Figure 1 , Figure 1 This is a module diagram of a cardiovascular medicine preoperative physical sign assessment system proposed by the present invention. In this embodiment, an overall architecture of a cardiovascular medicine preoperative physical sign assessment system is proposed, and the specific contents are as follows:

[0057] The present invention provides a cardiovascular medicine preoperative physical sign assessment system, which specifically includes: a data acquisition module, a statistical data calculation module, a data preprocessing module, a data integration module, a data enhancement module and a risk prediction module, wherein the output end of the data acquisition module is respectively connected to the input ends of the statistical data calculation module, the data preprocessing module and the data integration module, the output end of the statistical data calculation module is connected to the input end of the data preprocessing module, the output end of the data preprocessing module is respectively connected to the input ends of the data integration module and the risk prediction module, the output end of the data integration module is connected to the input end of the data enhancement module, and the output end of the data enhancement module is connected to the input end of the risk prediction module;

[0058] The data acquisition module acquires the patient's pathological data, vital sign data and basic information, and specifically includes a pathological data acquisition unit, a vital sign data acquisition unit and a basic information acquisition unit;

[0059] The pathological data acquisition unit acquires pathological data of the patient;

[0060] The vital sign data collection unit collects the vital sign data of the patient;

[0061] The basic information acquisition unit acquires the basic information and survival status of the patient, where the survival status refers to whether the patient survives the cardiovascular surgery, and the survival status includes survival and death;

[0062] The statistical data calculation module calculates statistical data based on the patient's pathological data and physical sign data;

[0063] The data preprocessing module performs standardization processing on the patient's pathological data, physical sign data and statistical data to obtain standardized pathological data, standardized physical sign data and standardized statistical data;

[0064] The data integration module integrates the standardized pathological data, standardized vital sign data, basic information and survival status into integrated data;

[0065] The data enhancement module further divides the category of death into survival status using KNN classifier, filters the noise by conditional variational autoencoder combined with improved Gaussian kernel function, and then constructs new integrated data;

[0066] The risk prediction module uses a random forest model to predict the patient's risk of death from cardiovascular surgery.

[0067] In this embodiment, the pathological data specifically include: blood urea nitrogen, hematocrit, white blood cell count, platelet count, lactate concentration, glucose concentration, sodium ion concentration, potassium ion concentration, magnesium ion concentration, arterial oxygen saturation, pH, arterial oxygen partial pressure, arterial carbon dioxide partial pressure, inspired oxygen volume fraction, alkaline phosphatase, alanine aminotransferase, aspartate aminotransferase, and total cholesterol.

[0068] The vital signs data specifically include: heart rate, systolic blood pressure, non-invasive systolic blood pressure, diastolic blood pressure, non-invasive diastolic blood pressure, mean arterial pressure, non-invasive mean arterial pressure, respiratory rate, and body temperature.

[0069] The statistical data specifically include: the maximum value, minimum value and average value of each pathological data and each physical sign data.

[0070] Basic information includes: gender and age.

[0071] The KNN classifier, conditional variational autoencoder and random forest model in this embodiment are written in Python language, the development platform is PyCharm, and the scikit-learn machine learning development library is used.

[0072] Embodiment 2: Based on the above embodiment, this embodiment implements a data preprocessing module, and the specific contents are as follows:

[0073] The data preprocessing module performs standardization processing on the patient's pathological data, physical sign data and statistical data to obtain standardized pathological data, standardized physical sign data and standardized statistical data. The specific steps of the standardization processing are as follows:

[0074] Step S1: Count the mean and standard deviation of each item of pathological data, physical sign data and statistical data of all patients;

[0075] Step S2: Subtract the mean value from each item of the pathological data, physical sign data and statistical data of each patient and divide the result by the standard deviation to obtain standardized pathological data, standardized physical sign data and standardized statistical data, respectively. The specific formulas are as follows:

[0076] ;

[0077] in, express After standardization, Indicates the pathological data, physical sign data and statistical data Item data, represents the mean, Represents standard deviation.

[0078] Example 3, see Figure 2 , Figure 2 The flowchart of generating new samples for the data enhancement module. This embodiment implements the data enhancement module based on the above embodiment. The specific contents are as follows:

[0079] The data enhancement module further divides the category of death into survival status using KNN classifier, filters the noise by conditional variational autoencoder combined with improved Gaussian kernel function, and then constructs new integrated data. The specific steps are as follows:

[0080] Step D1: define the majority class and the minority class: the integrated data of patients whose survival status is alive is taken as the majority class, and the integrated data of patients whose survival status is dead is taken as the minority class;

[0081] Step D2: Use KNN classifier to divide the minority class into difficult class and easy class: Take survival status as the target object of KNN classifier classification task, use integrated data of all patients to train KNN classifier, after training, use all minority class samples as test data set, use KNN classifier to classify the test data set, and use all samples classified as survival as difficult class, and all samples classified as death as easy class;

[0082] Step D3: Construct and train a conditional variational autoencoder: The conditional variational autoencoder specifically includes an encoder, a latent space, a decoder, and a classifier. The encoder maps the input data to the latent space, and the encoder reconstructs the data in the latent space into the original data. The classifier predicts the category of the sample based on the input data, constructs an objective function, and trains the conditional variational autoencoder with the goal of maximizing the objective function. The formula of the objective function is as follows:

[0083] ;

[0084] in, represents the objective function, represents the latent variable distribution of the encoder output, express The log-likelihood of represents the reconstructed distribution of the decoder output, represents the prior distribution of the latent variable, represents the KL divergence term of the variational upper bound, The distribution of variables representing the output of the classifier;

[0085] Step D4: Use the improved Gaussian kernel function to filter the noise of the difficult and easy classes. The specific steps are as follows:

[0086] Step D41: Calculate the density of each sample of the difficult class and the easy class in the latent space by improving the Gaussian kernel function. The specific formula is as follows:

[0087] ;

[0088] in, and Respectively and The corresponding value of samples in the latent space, express The density of Indicates The total number of samples corresponding to the category, represents the width of the improved Gaussian kernel function, Represents improved Gaussian kernel function;

[0089] The specific formula of the improved Gaussian kernel function is as follows:

[0090] ;

[0091] in, Indicates that the independent variable is The improved Gaussian kernel function, is the independent variable, represents the weight, and in this embodiment, the weight is set to 0.8;

[0092] Step D42: define a first threshold and a second threshold, remove samples with a density less than the first threshold in the difficult class, remove samples with a density less than the second threshold in the easy class, and use all remaining samples of the difficult class and the easy class as the screened integrated data;

[0093] Step D5: Generate new integrated data based on the screened integrated data. The specific steps are as follows:

[0094] Step D51: traverse the filtered integrated data and use the traversed integrated data as the current sample;

[0095] Step D52: Calculate the Euclidean distance between the current sample and other integrated data of the same category, and select the five samples with the shortest Euclidean distance as neighboring samples. The calculation formula of the Euclidean distance is as follows:

[0096] ;

[0097] in, express and The Euclidean distance between two integrated data, represents the total number of data items of the integrated data, and Respectively represent The value of the item data;

[0098] Step D53: randomly select a sample from the neighboring samples, and use the sample and the current sample as the benchmark to generate new integrated data. The basic information of the new integrated data is consistent with the current sample. The specific formula is as follows:

[0099] ;

[0100] in, Indicates the first part of the new integrated data except basic information The value of the item data, and Represents the current sample and randomly selected neighboring samples in addition to the basic information The value of the item data, Represents a random number between 0 and 1;

[0101] Step D54: define the maximum number of samples, and repeat steps D51-D53 until the total number of newly generated integrated data reaches the maximum number of samples. In this embodiment, the maximum number of samples is equal to the difference between the majority class and the minority class.

[0102] Embodiment 4: Based on the above embodiment, this embodiment implements a risk prediction module, and the specific contents are as follows:

[0103] The risk prediction module uses a random forest model to predict the patient's risk of death from cardiovascular surgery. The specific steps are as follows:

[0104] Step A1: Constructing a data set: The data set specifically includes standardized pathological data, standardized physical sign data, standardized statistical data and basic information of all patients;

[0105] Step A2: Supplementing the data set: Obtaining statistical data of each new integrated data through the statistical data calculation module, and then obtaining standardized statistical data through the data preprocessing module, adding the standardized statistical data to the new integrated data, and adding the new integrated data to the data set;

[0106] Step A3: Training the random forest model: The random forest model is trained using the data set. The random forest model classifies and predicts the survival status of each data item in the data set. The random forest model is evaluated using the cross-validation method.

[0107] Step A4: Use the random forest model to predict the surgical mortality risk of new patients: Use the data acquisition module, statistical data calculation module, and data preprocessing module to obtain the standardized pathological data, standardized vital sign data, standardized statistical data, and basic information of the new patients, and input them into the random forest model to obtain the predicted results of survival status.

Claims

1. A cardiovascular medicine preoperative physical sign assessment system, characterized by: Specifically include: Data collection module, statistical data calculation module, data preprocessing module, data integration module, data enhancement module and risk prediction module; The data acquisition module acquires the patient's pathological data, vital sign data and basic information, and specifically includes a pathological data acquisition unit, a vital sign data acquisition unit and a basic information acquisition unit; The basic information acquisition unit acquires the basic information and survival status of the patient, where the survival status refers to whether the patient survives the cardiovascular surgery, and the survival status includes survival and death; The statistical data calculation module calculates statistical data based on the patient's pathological data and physical sign data; The data preprocessing module performs standardization processing on the patient's pathological data, physical sign data and statistical data to obtain standardized pathological data, standardized physical sign data and standardized statistical data; The data integration module integrates the standardized pathological data, standardized vital sign data, basic information and survival status into integrated data; The data enhancement module uses a KNN classifier to reclassify the category of death as the survival status, filters the noise through a conditional variational autoencoder combined with an improved Gaussian kernel function, and then constructs new integrated data; The risk prediction module uses a random forest model to predict the patient's risk of death from cardiovascular surgery.

2. A cardiovascular medicine preoperative physical sign assessment system according to claim 1, characterized in that: The data enhancement module uses the KNN classifier to reclassify the category of death as the survival status, and filters the noise through the conditional variational autoencoder combined with the improved Gaussian kernel function, and then constructs the new integrated data in the following specific steps: Step D1: define the majority class and the minority class: the integrated data of patients whose survival status is alive is taken as the majority class, and the integrated data of patients whose survival status is dead is taken as the minority class; Step D2: Use KNN classifier to divide the minority class into difficult class and easy class; Step D3: Construct and train a conditional variational autoencoder: The conditional variational autoencoder specifically includes an encoder, a latent space, a decoder, and a classifier. An objective function is constructed, and the conditional variational autoencoder is trained to maximize the objective function; Step D4: Use the improved Gaussian kernel function to filter the noise of the difficult and easy classes to obtain the filtered integrated data; Step D5: Generate new integrated data based on the screened integrated data.

3. A cardiovascular medicine preoperative physical sign assessment system according to claim 2, characterized in that: The specific content of step D2 using the KNN classifier to divide the minority class into a difficult class and an easy class is: taking the survival status as the target object of the KNN classifier classification task, using the integrated data of all patients to train the KNN classifier, after the training is completed, using all minority class samples as the test data set, using the KNN classifier to classify the test data set, and all minority class samples with the classification result of survival are regarded as the difficult class, and all minority class samples with the classification result of death are regarded as the easy class.

4. A cardiovascular medicine preoperative physical sign assessment system according to claim 2, characterized in that: The specific steps of step D4 using the improved Gaussian kernel function to filter the noise of the difficult class and the easy class are as follows: Step D41: Calculate the density of each sample of the difficult class and the easy class in the latent space by improving the Gaussian kernel function. The specific formula is as follows: ； in, and Respectively and The corresponding value of samples in the latent space, express The density of Indicates The total number of samples corresponding to the category, represents the width of the improved Gaussian kernel function, Represents improved Gaussian kernel function; Step D42: define a first threshold and a second threshold, remove samples in the difficult class whose density is less than the first threshold, remove samples in the easy class whose density is less than the second threshold, and use all remaining samples of the difficult class and the easy class as the screened integrated data.

5. A cardiovascular preoperative physical sign assessment system according to claim 4, characterized in that: The specific formula of the improved Gaussian kernel function is as follows: ； in, Indicates that the independent variable is The improved Gaussian kernel function, is the independent variable, Represents weight.

6. A cardiovascular medicine preoperative physical sign assessment system according to claim 2, characterized in that: The specific steps of step D5 for generating new integrated data based on the screened integrated data are as follows: Step D51: traverse the filtered integrated data and use the traversed integrated data as the current sample; Step D52: Calculate the Euclidean distance between the current sample and other integrated data of the same category, and select P samples with the shortest Euclidean distance as neighboring samples; Step D53: randomly select a sample from the neighboring samples, and generate new integrated data based on the sample and the current sample; Step D54: Define the maximum number of samples, and repeat steps D51-D53 until the total number of new integrated data generated reaches the maximum number of samples.

7. A cardiovascular medicine preoperative physical sign assessment system according to claim 6, characterized in that: The basic information of the new integrated data is consistent with the current sample.

Images