Monitoring method of chronic obstructive pulmonary disease based on multidimensional data analysis

Abstract

The present invention relates to the field of medical data mining technology, and specifically to a method for monitoring chronic obstructive pulmonary disease based on multidimensional data analysis. The present invention obtains the reference value of each characteristic parameter of each symptom characteristic for classification and assessment of the disease by analyzing, adjusts the information gain of each symptom characteristic for the classification result, and obtains the characteristic importance of each symptom characteristic; further combines the mutual influence between different symptom characteristics to obtain the contribution weight of each decision tree in the random forest; and then inputs the case sample to be estimated into the random forest to obtain the disease condition estimation result of the monitored patient. The present invention accurately obtains the characteristic importance of each symptom characteristic by analyzing the influence of individual differences of patients, and then combines the relevant influence between the symptom characteristics to evaluate the construction and classification effect of each decision tree, thereby accurately obtaining the classification result of each case sample to be estimated, and improving the disease condition monitoring effect of the random forest algorithm on patients with chronic obstructive pulmonary disease.

Description

Technical Field

[0001] The present invention relates to the technical field of medical data mining, and in particular to a method for monitoring chronic obstructive pulmonary disease based on multidimensional data analysis. Background Art

[0002] COPD is a chronic lung disease characterized by airflow obstruction. If not effectively managed, it may lead to other complications and an increased risk of death, and will also increase the medical burden. Therefore, the monitoring of the condition of COPD patients cannot be ignored. At present, by analyzing the multidimensional disease indicators in a large number of historical cases of COPD patients, the random forest algorithm can be used to classify and estimate whether the patient's condition will worsen. When the disease is estimated to worsen, the relevant medical staff will be warned to intervene in the diagnosis and treatment, thereby effectively managing the changes in the condition of COPD patients.

[0003] However, in the prior art, in the process of monitoring and predicting the worsening of the patient's condition, the random forest algorithm averages the classification results of each decision tree to obtain the final classification result; but in the process of building each decision tree, the selection of samples and features is random. The random selection of features may lead to different prediction accuracy of each decision tree, that is, each decision tree has a different impact on the final classification result. Directly using the average result of the classification results of all decision trees may lead to low accuracy of the final classification result, thereby affecting the accuracy of the classification prediction of the worsening of the condition of patients with chronic obstructive pulmonary disease, resulting in poor disease monitoring effect on patients with chronic obstructive pulmonary disease. Summary of the invention

[0004] In order to solve the technical problem that the existing monitoring effect of chronic obstructive pulmonary disease patients is poor, the purpose of the present invention is to provide a method for monitoring chronic obstructive pulmonary disease based on multidimensional data analysis. The technical solution adopted is as follows:

[0005] Obtain all case samples of the disease to be monitored and the disease condition label of each case sample, wherein the case sample includes characteristic parameters under each disease characteristic; construct a random forest based on all the case samples and the corresponding disease condition labels;

[0006] Under each characteristic parameter of each disease characteristic, according to the consistency of the disease label corresponding to the same case sample, obtain the reference value of each characteristic parameter of each disease characteristic for the classification and assessment of the disease; according to the reference value of each characteristic parameter of each disease characteristic, adjust the information gain of each disease characteristic for the random forest classification result, and obtain the characteristic importance of each disease characteristic; according to the mutual influence between different disease characteristics, combined with the characteristic importance of each disease characteristic, obtain the contribution weight of each decision tree in the random forest to the random forest classification result;

[0007] Obtain a sample of cases to be estimated from the monitored patient; input the sample of cases to be estimated into a random forest, and obtain an estimated result of the condition of the monitored patient by combining the contribution weight of each decision tree in the random forest.

[0008] Furthermore, the method for obtaining the reference value includes:

[0009] Under each characteristic parameter of each disease characteristic, the same case samples are divided into a subset, and the consistency of the case samples in each subset corresponding to the disease label is comprehensively considered to obtain the disease discrete parameter of each characteristic parameter; the disease discrete parameters are negatively correlated and mapped to obtain the reference value of each characteristic parameter of each disease characteristic for the classification and assessment of the disease.

[0010] Furthermore, the method for obtaining the discrete parameters of the disease condition includes:

[0011] In each subset, the total number of the same disease condition labels is counted respectively, and the discrete parameters of each subset are obtained according to the discrete situation of the total number corresponding to each disease condition label; the discrete parameters of all subsets are averaged to obtain the disease condition discrete parameters of each characteristic parameter.

[0012] Furthermore, the method for obtaining the feature importance includes:

[0013] When calculating the information gain of each disease feature for the random forest classification result, the conditional probability of each subset in the conditional entropy is weighted using the reference value to obtain the weighted conditional entropy, thereby adjusting the information gain and obtaining the feature importance of each disease feature.

[0014] Furthermore, the method for obtaining the contribution weight includes:

[0015] Taking any disease feature as a target feature and any non-target disease feature as a reference feature, based on the information gain calculation method, according to the mutual influence between the target feature and the reference feature, the target influence parameter of the target feature on the reference feature and the reference influence parameter of the reference feature on the target feature are obtained; according to the target influence parameter and the reference influence parameter, the mutual influence parameters between the corresponding different disease features are obtained;

[0016] Obtain all combinations of different disease features in each decision tree of the random forest; in each combination, obtain the combined feature importance of the disease features corresponding to each combination according to the feature importance of each disease feature and the corresponding mutual influence parameter; obtain the contribution weight of the corresponding decision tree by combining the combined feature importance of all combinations in the comprehensive decision tree.

[0017] Furthermore, the method for obtaining the mutual influence parameters includes:

[0018] Calculating the information entropy of the target feature and the reference feature in all the case samples respectively; calculating the weighted conditional entropy of the target feature and the reference feature respectively;

[0019] The difference between the information entropy of the reference feature and the weighted conditional entropy of the target feature is used as the target influencing parameter; the difference between the information entropy of the target feature and the weighted conditional entropy of the reference feature is used as the reference influencing parameter;

[0020] The average of the target influencing parameter and the reference influencing parameter is used as the mutual influencing parameter.

[0021] Furthermore, the method for obtaining the importance of the combined features includes:

[0022] In each combination, the sum of the feature importances of all disease features is subtracted from the mutual influence parameters between the corresponding disease features to obtain the combined feature importance.

[0023] Furthermore, the method for obtaining the condition prediction result includes:

[0024] In the random forest, the decision trees with the same disease classification results are divided into one category; the contribution weights of all decision trees in each disease classification result are accumulated to obtain the reference coefficient of each disease classification result; and the disease estimation result is obtained by combining all the reference coefficients.

[0025] Furthermore, the specific method of obtaining the condition prediction result by integrating all the reference coefficients includes:

[0026] The disease classification result with the largest reference coefficient is used as the disease estimation result.

[0027] Furthermore, the condition label includes aggravated and not aggravated.

[0028] The present invention has the following beneficial effects:

[0029] The present invention obtains all case samples and the disease label of each case sample, and the case sample includes the characteristic parameters under each disease feature; a random forest is constructed based on all case samples and the corresponding disease label, so as to input the cases to be estimated under the multidimensional disease features of the monitored patients into the constructed random forest for classification, so as to evaluate whether their disease conditions are aggravated; under each characteristic parameter of each disease feature, according to the consistency of the disease label corresponding to the same case sample, the reference value of each characteristic parameter of each disease feature for the classification and evaluation of the disease condition is obtained, and the reference value reflects the influence of individual differences of patients under each characteristic parameter of each disease feature, so as to facilitate the subsequent adjustment of information entropy to accurately evaluate each characteristic Importance of features; According to the reference value of each characteristic parameter of each symptom feature, adjust the information gain of each symptom feature to the random forest classification result, and obtain the feature importance of each symptom feature; According to the mutual influence between different symptom features, combined with the feature importance of each symptom feature, obtain the contribution weight of each decision tree in the random forest to the random forest classification result, and the contribution weight is combined with the mutual influence between the symptom features to evaluate the construction and classification effect of the decision tree, so as to accurately determine its contribution weight; obtain the case sample to be estimated of the monitored patient; input the case sample to be estimated into the random forest, and obtain the condition estimation result of the monitored patient in combination with the contribution weight of each decision tree in the random forest. The present invention accurately obtains the feature importance of each symptom feature by analyzing the influence of individual differences of patients, and then evaluates the construction and classification effect of each decision tree according to the relevant influence between the symptom features, and obtains its contribution weight in the random forest, so as to accurately obtain the classification result of each case sample to be estimated, and improve the condition monitoring effect of the random forest algorithm on the patient. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0031] Figure 1 A flowchart of a method for monitoring chronic obstructive pulmonary disease based on multidimensional data analysis provided by one embodiment of the present invention;

[0032] Figure 2 A flow chart of a method for obtaining contribution weights provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0033] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation, structure, features and effects of a chronic obstructive pulmonary disease monitoring method based on multidimensional data analysis proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0034] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0035] A specific scheme of a method for monitoring chronic obstructive pulmonary disease based on multidimensional data analysis provided by the present invention is described in detail below with reference to the accompanying drawings.

[0036] See also Figure 1 , which shows a flow chart of a method for monitoring chronic obstructive pulmonary disease based on multidimensional data analysis provided by an embodiment of the present invention, specifically comprising:

[0037] Step S1, obtaining all case samples of the disease to be monitored and the disease label of each case sample, the case sample includes characteristic parameters under each disease characteristic; constructing a random forest based on all case samples and corresponding disease labels.

[0038] In one embodiment of the present invention, chronic obstructive pulmonary disease is taken as an example of a disease to be monitored; in order to monitor the condition of patients with chronic obstructive pulmonary disease, the cases to be estimated under the multidimensional disease characteristics of the monitored patients are input into the constructed random forest for classification to evaluate whether their condition has worsened; therefore, in one embodiment of the present invention, a large number of case samples of chronic obstructive pulmonary disease are first obtained, and each case sample is marked and assigned a disease label based on the historical diagnosis and treatment results of the medical record samples by disease experts, wherein the disease labels include worsening and non-aggravated, and the number of case samples is 10,000; then a random forest is constructed based on all case samples and corresponding disease labels.

[0039] It should be noted that the construction of random forest is an existing technology well known to those skilled in the art and will not be described in detail here.

[0040] Among them, in each case sample, the disease characteristics of each patient collected include at least age, smoking history, respiratory rate, heart rate, oxygen saturation, history of exacerbations, number of coughs, and sputum color; since the smoking history, history of exacerbations and sputum color are usually in text format, they are digitized to facilitate subsequent analysis.

[0041] As an example, age, respiratory rate, heart rate, oxygen saturation and number of coughs are the actual collected values ​​during each patient monitoring process. They are dimensionless and directly used as characteristic parameters under the corresponding symptom characteristics. Smoking history and aggravation history are usually yes or no. "History of smoking" and "history of aggravation" are represented by 1, and "no history of smoking" and "no history of aggravation" are represented by 0 to obtain the corresponding characteristic parameters. For sputum color, the color characteristics are quantified through color plates or color contrast paper to obtain characteristic parameters.

[0042] It should be noted that in other embodiments of the present invention, the implementer may also set the number of case samples and the types of symptom characteristics by himself, and may further subdivide the disease label corresponding to each case sample according to the degree of disease aggravation to obtain multiple labels; the implementer may also estimate the aggravation of other non-chronic obstructive pulmonary diseases based on the same inventive concept.

[0043] Step S2, under each characteristic parameter of each disease feature, according to the consistency of the disease label corresponding to the same case sample, obtain the reference value of each characteristic parameter of each disease feature for the classification and assessment of the disease; according to the reference value of each characteristic parameter of each disease feature, adjust the information gain of each disease feature for the random forest classification result, and obtain the characteristic importance of each disease feature; according to the mutual influence between different disease features and the characteristic importance of each disease feature, obtain the contribution weight of each decision tree in the random forest to the random forest classification result.

[0044] Taking into account the multidimensional disease characteristics, the greater the possibility that the symptoms and disease development conditions of the patients corresponding to the same case samples are similar; however, there are individual differences among patients, and it cannot be guaranteed that the aggravation of the condition of the patients corresponding to the same case samples is consistent; and due to the complexity and diversity of the patients' conditions, each disease characteristic covers a variety of characteristic parameters. Therefore, the embodiment of the present invention first obtains the reference value of each characteristic parameter of each disease characteristic for the classification and assessment of the disease according to the consistency of the disease label corresponding to the same case sample under each characteristic parameter of each disease characteristic.

[0045] The reference value reflects the influence of individual differences of patients under each characteristic parameter of each disease feature. The lower the consistency of the disease labels of the same case samples corresponding to a certain characteristic parameter, the greater the degree to which the characteristic parameter is affected by individual differences of patients, and the lower its reference value. In the subsequent analysis of the information gain of each disease feature on the random forest classification results, the accuracy of the information gain obtained is lower.

[0046] Preferably, in one embodiment of the present invention, the method for obtaining the reference value includes:

[0047] Under each characteristic parameter of each disease characteristic, the same case samples are divided into a subset, and the consistency of the disease labels corresponding to the case samples in each subset is combined to obtain the disease discrete parameters of each characteristic parameter; the disease discrete parameters are negatively correlated and mapped to obtain the reference value of each characteristic parameter of each disease characteristic for the classification and assessment of the disease.

[0048] The method for obtaining discrete parameters of the disease condition includes:

[0049] In each subset, the total number of the same disease labels is counted respectively, and the discrete parameters of each subset are obtained according to the discrete situation of the total number corresponding to each disease label; the discrete parameters of all subsets are averaged to obtain the disease discrete parameters of each characteristic parameter.

[0050] As an example, the case sample is represented in the form of a vector, each vector element is a characteristic parameter under each disease feature, and the vector is constructed in the order of age, smoking history, respiratory rate, heart rate, oxygen saturation, aggravation history, number of coughs, and sputum color. For example, the corresponding vector of a case sample is (15, 1, 12, 70, 95, 0, 2, 3), and the characteristic parameter of the fixed age disease feature is 15. All the same case samples with the age characteristic parameter of 15 are divided into a subset, and multiple subsets can be obtained, wherein the characteristic parameters of the disease features of different case samples in the same subset are consistent, while the characteristic parameters of the disease features of case samples in different subsets are not consistent; for example, the case samples in the first subset are all (15, 1, 12, 70, 95, 0, 2, 3), and the case samples in the second subset are all (15, 2, 22, 70, 95, 0, 2, 3), and so on, all subsets corresponding to each characteristic parameter of each disease feature can be obtained;

[0051] The total number of each disease condition label in each subset is counted separately. For example, in a subset corresponding to the age characteristic parameter of 15, the total number of aggravated disease condition labels is 100, and the total number of non-aggravated disease condition labels is 800. The difference between the total number of these two disease condition labels is large, that is, the total number of labels is more concentrated in the non-aggravated disease condition label, so the discrete parameter of this subset is small; specifically, the absolute value of the difference between the total number of these two disease condition labels is used as the discrete parameter of the corresponding subset, and then the discrete parameters of all subsets are averaged to obtain the disease condition discrete parameter of each characteristic parameter; then the disease condition discrete parameter is used as In Perform negative correlation mapping to adjust the logic, so as to obtain the reference value of each characteristic parameter of each disease feature for the classification and assessment of the disease condition. The natural constant is an exponential function with a base; when the label distribution is more discrete and inconsistent, the discrete parameter of the disease is larger, the influence of individual differences is greater, and the reference value of the characteristic parameter is lower; other negative correlation mapping methods can also be used in other embodiments of the present invention, which will not be repeated here.

[0052] In other embodiments of the present invention, considering that the disease condition labels may include not only two but multiple types, the total number of each disease condition label in each subset of each characteristic parameter is counted, and the variance of the total number is calculated to obtain the discrete parameter of the subset; the larger the variance, the more discrete the distribution of different labels, the greater the impact of individual differences of patients in the subset on the type of disease condition label, and the larger the corresponding discrete parameter; further, the discrete parameters of all subsets are averaged to obtain the disease condition discrete parameter of each characteristic parameter.

[0053] Considering that when evaluating the impact of each symptom feature on the classification and evaluation of the disease, that is, determining the disease label, information gain is usually used to measure the impact or contribution of each symptom feature on the disease classification result; the larger the information gain, the greater the impact or contribution of the symptom feature on the disease classification; and considering that the essence of information gain is to calculate the information entropy by measuring the degree of confusion of the data set, and then divide the data set into multiple subsets based on the characteristic parameters of the fixed symptom feature, evaluate the information entropy of the corresponding subsets, that is, the conditional entropy, and finally evaluate the impact or contribution of each symptom feature on the disease classification based on the difference between the information entropy and the corresponding conditional entropy of all subsets; information gain reflects the degree of reduction in the degree of confusion of the data set. The greater the reduction, the greater the information gain, and the greater the accuracy of each symptom feature in classifying the disease.

[0054] In the process of evaluating conditional entropy, due to the influence of individual differences among patients, the reference value of each characteristic parameter of each disease feature is different. Some disease features have high reference values ​​under corresponding characteristic parameters, which are less affected by individual differences and have low impact on the evaluation of conditional entropy. Conversely, characteristic parameters of disease features with low reference values ​​may introduce large errors in the evaluation of conditional entropy, thereby affecting the accuracy of the evaluation of the information gain, i.e., the importance of the features, of each disease feature. Therefore, after obtaining the reference value of each characteristic parameter of each disease feature for the classification and evaluation of the disease, the embodiment of the present invention can use the reference value to adjust the information gain, thereby obtaining accurate feature importance.

[0055] Preferably, in one embodiment of the present invention, the method for obtaining feature importance includes:

[0056] When calculating the information gain of each disease feature for the random forest classification result, the conditional probability of each subset in the conditional entropy is weighted using the reference value to obtain the weighted conditional entropy, thereby adjusting the information gain and obtaining the feature importance of each disease feature.

[0057] It should be noted that the calculation process of information gain is an existing technology well known to those skilled in the art. Here, only the adjustment calculation formula is briefly described to obtain the feature importance; the calculation formula is:

[0058] ; In the formula, For the The characteristic importance of the first disease feature, that is, The information gain of the disease features on the random forest classification results; The information entropy of all case samples corresponding to the disease labels; For the The serial number of the characteristic parameter of the disease characteristic; For the The total number of characteristic parameter types of disease characteristics; For the Symptoms of the first The reference value of the characteristic parameters; For the Symptoms of the first The total number of case samples in the subset corresponding to each characteristic parameter; For the The inclusion of symptoms The serial number of the characteristic parameters included; For the Symptoms of the first The reference value of the characteristic parameters; For the Symptoms of the first The total number of case samples in the subset corresponding to each characteristic parameter; For the Symptoms of the first The weighted conditional probability of the subset corresponding to the feature parameters, where For the Symptoms of the first The conditional probability of the subset corresponding to the characteristic parameters; For the Symptoms of the first The information entropy of the subset corresponding to the feature parameter.

[0059] In the above formula, Weighted conditional entropy of disease features Relative to the conditional entropy in the prior art Improvements are made to integrate the reference value of each characteristic parameter of each disease feature to adjust the conditional entropy, reduce the impact of individual differences among patients, thereby obtaining the adjusted information gain and accurately obtaining the characteristic importance of each disease feature.

[0060] Considering that the greater the feature importance of the disease feature in the decision tree, the greater its contribution to the classification is considered, and this type of disease feature will be given priority in the construction of the decision tree; however, if the decision tree contains many highly correlated disease features, these related features will provide similar information to the decision tree, which may lead to overfitting, that is, the decision tree over-relies on the related features and ignores other equally important but relatively independent features, thereby affecting the construction and classification effect of the decision tree; therefore, the embodiment of the present invention obtains the contribution weight of each decision tree in the random forest to the random forest classification result based on the mutual influence between different disease features and the feature importance of each disease feature.

[0061] Preferably, in one embodiment of the present invention, the method for obtaining the contribution weight includes:

[0062] See also Figure 2 , which shows a flow chart of a method for obtaining contribution weights provided by an embodiment of the present invention, including:

[0063] Step S201, taking any disease feature as the target feature and any non-target disease feature as the reference feature, based on the information gain calculation method, according to the mutual influence between the target feature and the reference feature, obtain the target influence parameter of the target feature on the reference feature, and the reference influence parameter of the reference feature on the target feature; according to the target influence parameter and the reference influence parameter, obtain the mutual influence parameters between the corresponding different disease features.

[0064] In an embodiment of the present invention, the importance of each symptom feature to the classification result is evaluated by taking into account the degree of reduction in the degree of confusion of the classification result before and after the division of the corresponding subsets of each symptom feature. It also indirectly reflects the change in the degree of confusion of the classification result as the corresponding subsets of each feature parameter are divided. Based on this idea, the information gain can be used to indirectly evaluate the mutual influence relationship between different symptom features. The greater the information gain, the greater the mutual influence relationship between the symptom features.

[0065] Furthermore, since information gain is asymmetric and has measurement directionality, that is, the information gain of the target feature to the reference feature may be different from the information gain of the reference feature to the target feature, the embodiment of the present invention will separately obtain the target influence parameter and the reference influence parameter, and then comprehensively obtain the mutual influence parameter to evaluate the mutual influence relationship between the target feature and the reference feature.

[0066] In a preferred embodiment of the present invention, the calculation method of the mutual influence parameter is:

[0067] Calculate the information entropy of the target feature and the reference feature in all case samples respectively; Calculate the weighted conditional entropy of the target feature and the reference feature respectively;

[0068] The difference between the information entropy of the reference feature and the weighted conditional entropy of the target feature is used as the target influence parameter; the difference between the information entropy of the target feature and the weighted conditional entropy of the reference feature is used as the reference influence parameter;

[0069] The mean of the target influence parameter and the reference influence parameter is taken as the mutual influence parameter.

[0070] The calculation formula of the target impact parameter is: ;in, is the sequence number of the target feature, is the serial number of the reference feature; is the information gain of the target feature relative to the reference feature, i.e., the target influence parameter; is the information entropy of the reference feature in all case samples; For the The sequence number of the characteristic parameter of the target feature; For the The total number of characteristic parameter types of target features; For the The target feature The reference value of the characteristic parameters; For the The target feature The total number of case samples in the subset corresponding to each characteristic parameter; For the The inclusion of target features The serial number of the characteristic parameters included; For the The target feature The reference value of the characteristic parameters; For the The target feature The total number of case samples in the subset corresponding to each characteristic parameter; For the The target feature The information entropy of the subset corresponding to the characteristic parameters; For the The weighted conditional entropy of the target features.

[0071] Similarly, the reference influence parameter can be obtained. Its calculation formula is similar to that of the target influence parameter, which will not be described here.

[0072] The larger the target influence parameter, the greater the information gain of the target feature for the reference feature, which indirectly indicates that the change of the target feature has a greater impact on the change of the reference feature; similarly, the larger the reference influence parameter, the greater the corresponding information gain, which indirectly indicates that the change of the reference feature has a greater impact on the change of the target feature.

[0073] Then, considering the asymmetry and measurement directionality of information gain, the target influence parameter and the reference influence parameter are averaged to obtain the mutual influence parameter between the target feature and the reference feature.

[0074] In other embodiments of the present invention, the implementer may also directly construct a target feature parameter sequence corresponding to the target feature, and a reference feature parameter sequence corresponding to the reference feature, and use relevant measurement methods to evaluate the relevant influencing parameters between the target feature and the reference feature. Specifically, the correlation coefficient, Pearson correlation coefficient, etc. may be used, but since its value ranges from -1 to 1, it is necessary to perform absolute value processing to obtain the relevant influencing parameters; considering that there are many case samples, the implementer may also divide it into multiple feature parameter sequence sub-segments, evaluate the absolute value of the correlation coefficient between the corresponding feature parameter sequence sub-segments, and then calculate the average or take the mode to obtain the relevant influencing parameters.

[0075] In other embodiments of the present invention, since relative entropy is used to measure the degree of difference between probability distributions, the smaller the difference in feature distribution, the smaller the relative entropy, which indirectly reflects that the corresponding distribution changes of the two features are more similar; therefore, the implementer can also introduce the idea of ​​relative entropy to evaluate the mutual influence relationship between the target feature and the reference feature, and based on the same idea of ​​weighting the conditional entropy, introduce the reference feature to adjust the probability distribution, and then use the relative entropy calculation formula to evaluate the target influence parameter of the target feature on the reference feature, and the reference influence parameter of the reference feature on the target feature. Since the smaller the relative entropy, the greater the possibility of mutual change between the features, it is also necessary to perform negative correlation mapping on the target influence parameter and the reference influence parameter respectively, and average the negative correlation mapping results to obtain the relevant influence parameters, which will not be repeated here.

[0076] At this point, the relevant influence parameters between all different disease features can be obtained. The relevant influence parameters reflect the overlap and correlation between different disease features. If there are disease features with strong relevant influence in the decision tree, it may lead to poor classification effect of the decision tree. Therefore, in the following step S202, the importance of the combined features of different disease features will be further analyzed to evaluate the construction and classification effects of the corresponding decision tree. The worse the construction and classification effects are, the lower the contribution weight of the decision tree should be when all decision trees are comprehensively evaluated to obtain an accurate final estimation result.

[0077] Step S202, obtaining all combinations of different disease features in each decision tree of the random forest; in each combination, obtaining the combined feature importance of the disease features corresponding to each combination according to the feature importance of each disease feature and the corresponding mutual influence parameter; and obtaining the contribution weight of the corresponding decision tree by comprehensively considering the combined feature importance of all combinations in the decision tree.

[0078] As an example, firstly, any combination of two different disease features in each decision tree is obtained, the combined feature importance of each combination is obtained, and then the combined feature importances of all combinations are accumulated to obtain the contribution weight of the corresponding decision tree.

[0079] Among them, in a preferred embodiment of the present invention, the method for obtaining the importance of the combined features includes:

[0080] In each combination, the sum of the feature importance of each disease feature is subtracted from the interaction parameter between the corresponding disease features to obtain the combined feature importance.

[0081] Step S3, obtaining a sample of cases to be estimated for the monitored patient; inputting the sample of cases to be estimated into the random forest, combining the contribution weight of each decision tree in the random forest, and obtaining the estimated result of the condition of the monitored patient.

[0082] In one embodiment of the present invention, the acquired samples of cases to be estimated of the monitored patients are input into the constructed random forest, and the classification results of each decision tree for the samples of cases to be estimated can be obtained; but since the construction and classification effects of each decision tree are different, it is also necessary to evaluate the estimated results of the monitored patients' condition based on the contribution weight of each decision tree.

[0083] It should be noted that the method and content of obtaining the case samples to be estimated are consistent with the case samples involved in constructing the random forest, and the acquisition process will not be repeated here.

[0084] Preferably, in one embodiment of the present invention, the method for obtaining the condition prediction result includes:

[0085] In the random forest, decision trees with the same disease classification results are divided into one category; the contribution weights of all decision trees in each disease classification result are accumulated to obtain the reference coefficient of each disease classification result; all reference coefficients are combined to obtain the disease prediction result; among them, the disease classification result with the largest reference coefficient is used as the disease prediction result.

[0086] As an example, a total of 100 decision trees are constructed in the random forest, of which 80 decision trees classify the estimated case samples as aggravated, and 20 decision trees classify the estimated case samples as not aggravated. The contribution weights of the 80 decision trees corresponding to the aggravated results and the contribution weights of the 20 decision trees corresponding to the not aggravated results are added together to obtain the corresponding reference coefficients. Assuming that the reference coefficient of the aggravated result is greater than the reference coefficient of the not aggravated result, the final condition estimation result is aggravated.

[0087] At this point, relevant medical staff can regularly monitor and obtain samples of cases to be estimated for each patient, and use the constructed random forest to quickly estimate the condition of each patient, thereby alerting relevant medical staff to intervene in diagnosis and treatment, effectively managing changes in the condition of patients with chronic obstructive pulmonary disease and improving monitoring results.

[0088] In summary, the embodiment of the present invention constructs a random forest based on all case samples and corresponding disease labels; further analyzes and obtains the reference value of each characteristic parameter of each disease feature for the classification and evaluation of the disease, thereby adjusting the information gain of each disease feature to the random forest classification result, and obtaining the characteristic importance of each disease feature; further combining the mutual influence between different disease features, obtaining the contribution weight of each decision tree in the random forest to the random forest classification result; inputting the case sample to be estimated into the random forest, combining the contribution weight of each decision tree in the random forest, and obtaining the disease prediction result of the monitored patient. The present invention accurately obtains the characteristic importance of each disease feature by analyzing the influence of individual differences among patients, and then evaluates the construction and classification effect of each decision tree according to the related influence between the disease features, and obtains its contribution weight in the random forest, thereby accurately obtaining the classification result of each case sample to be estimated, and improving the effect of the random forest algorithm on the patient's disease monitoring.

[0089] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0090] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A method for monitoring chronic obstructive pulmonary disease based on multidimensional data analysis, characterized in that: The method comprises: Obtain all case samples of the disease to be monitored and the disease condition label of each case sample, wherein the case sample includes characteristic parameters under each disease characteristic; construct a random forest based on all the case samples and the corresponding disease condition labels; Under each characteristic parameter of each disease characteristic, according to the consistency of the disease label corresponding to the same case sample, obtain the reference value of each characteristic parameter of each disease characteristic for the classification and assessment of the disease; according to the reference value of each characteristic parameter of each disease characteristic, adjust the information gain of each disease characteristic for the random forest classification result, and obtain the characteristic importance of each disease characteristic; according to the mutual influence between different disease characteristics, combined with the characteristic importance of each disease characteristic, obtain the contribution weight of each decision tree in the random forest to the random forest classification result; Obtain a sample of cases to be estimated for the monitored patient; input the sample of cases to be estimated into a random forest, and obtain an estimated result of the condition of the monitored patient by combining the contribution weight of each decision tree in the random forest; The method for obtaining the reference value includes: Under each characteristic parameter of each disease characteristic, the same case samples are divided into a subset, and the consistency of the case samples in each subset corresponding to the disease label is comprehensively considered to obtain the disease discrete parameter of each characteristic parameter; the disease discrete parameters are negatively correlated and mapped to obtain the reference value of each characteristic parameter of each disease characteristic for the classification and assessment of the disease; The method for obtaining the feature importance includes: When calculating the information gain of each disease feature to the random forest classification result, the conditional probability of each subset in the conditional entropy is weighted by using the reference value to obtain the weighted conditional entropy, thereby adjusting the information gain and obtaining the feature importance of each disease feature; The method for obtaining the contribution weight includes: Taking any disease feature as a target feature and any non-target disease feature as a reference feature, based on the information gain calculation method, according to the mutual influence between the target feature and the reference feature, the target influence parameter of the target feature on the reference feature and the reference influence parameter of the reference feature on the target feature are obtained; according to the target influence parameter and the reference influence parameter, the mutual influence parameters between the corresponding different disease features are obtained; Obtain all combinations of different disease features in each decision tree of the random forest; in each combination, according to the feature importance of each disease feature and the corresponding mutual influence parameter, obtain the combined feature importance of the disease features corresponding to each combination; comprehensively consider the combined feature importance of all combinations in the decision tree to obtain the contribution weight of the corresponding decision tree; In the random forest, the decision trees with the same disease classification results are divided into one category; the contribution weights of all decision trees in each disease classification result are accumulated to obtain the reference coefficient of each disease classification result; and the disease estimation result is obtained by combining all the reference coefficients.

2. A method for monitoring chronic obstructive pulmonary disease based on multidimensional data analysis according to claim 1, characterized in that: The method for obtaining the discrete parameters of the disease condition includes: In each subset, the total number of the same disease condition labels is counted respectively, and the discrete parameters of each subset are obtained according to the discrete situation of the total number corresponding to each disease condition label; the discrete parameters of all subsets are averaged to obtain the disease condition discrete parameters of each characteristic parameter.

3. The method for monitoring chronic obstructive pulmonary disease based on multidimensional data analysis according to claim 1, characterized in that: The method for obtaining the mutual influence parameters includes: Calculating the information entropy of the target feature and the reference feature in all the case samples respectively; calculating the weighted conditional entropy of the target feature and the reference feature respectively; The difference between the information entropy of the reference feature and the weighted conditional entropy of the target feature is used as the target influencing parameter; the difference between the information entropy of the target feature and the weighted conditional entropy of the reference feature is used as the reference influencing parameter; The average of the target influencing parameter and the reference influencing parameter is used as the mutual influencing parameter.

4. The method for monitoring chronic obstructive pulmonary disease based on multidimensional data analysis according to claim 1, characterized in that: The method for obtaining the importance of the combined features includes: In each combination, the sum of the feature importances of all disease features is subtracted from the mutual influence parameters between the corresponding disease features to obtain the combined feature importance.

5. The method for monitoring chronic obstructive pulmonary disease based on multidimensional data analysis according to claim 1, characterized in that: The specific method of obtaining the condition prediction result by integrating all the reference coefficients includes: The disease classification result with the largest reference coefficient is used as the disease estimation result.

6. The method for monitoring chronic obstructive pulmonary disease based on multidimensional data analysis according to claim 1, characterized in that: The condition labels include exacerbated and non-exacerbated.

Images