Prediction method and device of disease probability and electronic equipment

Abstract

The invention provides a disease probability prediction method and apparatus, and an electronic device. The method comprises the steps of obtaining at least one target keyword associated with a target disease and disease description information of a target patient; for each target keyword, marking the target keyword according to whether the target keyword exists in the disease description information or not to obtain a first marking result; and inputting the disease description information and the first marking result corresponding to each target keyword into a pre-trained BRF model to output the probability that the target patient suffers from the target disease, in this way, at least one target keyword associated with the target disease is pre-determined, and the probability that the target patient suffers from the target disease is determined on the basis of the disease description information of the target patient. After each target keyword is marked, the first marking result corresponding to each target keyword and the disease description information are jointly used as the input of the BRF model, so that the comprehensiveness and accuracy of evaluating whether the target patient suffers from the target disease can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of disease prediction, and in particular to a method, device and electronic equipment for predicting disease probability. Background Art

[0002] Systemic lupus erythematosus is a complex autoimmune disease. SLE patients are often at risk of multiple serious complications, including kidney disease, cardiovascular disease, and nervous system damage. These complications have a significant impact on patients' survival rate and quality of life. Predicting the risk of systemic lupus erythematosus helps to adopt individualized treatment plans and intervention measures early. Traditional SLE risk assessment methods mainly rely on the doctor's experience and judgment, or are based on simple statistical models. This method is difficult to comprehensively and accurately assess a patient's risk of developing systemic lupus erythematosus. Summary of the invention

[0003] The purpose of the present invention is to provide a method, device and electronic equipment for predicting the probability of disease, so as to comprehensively and accurately assess the risk of a patient suffering from a disease.

[0004] The present invention provides a method for predicting the probability of disease, which includes: obtaining at least one target keyword associated with a target disease and disease description information of a target patient; for each target keyword, marking the target keyword according to whether the target keyword exists in the disease description information, and obtaining a first marking result corresponding to the target keyword; inputting the disease description information and the first marking result corresponding to each target keyword into a pre-trained BRF model, so as to output the probability of the target patient suffering from the target disease through the BRF model.

[0005] Furthermore, the target keywords are determined in the following manner: obtaining an initial medical data set; wherein the initial medical data set includes first diagnosis and treatment data of multiple patients; according to preset fields, filtering out second diagnosis and treatment data corresponding to preset fields from the first diagnosis and treatment data corresponding to each patient; wherein the preset fields are fields associated with the target disease; inputting the second diagnosis and treatment data of each patient into a preset word segmentation model respectively to perform word segmentation processing on each second diagnosis and treatment data to obtain word segmentation results corresponding to each second diagnosis and treatment data; screening and processing the word segmentation results corresponding to all the second diagnosis and treatment data to obtain the target keywords.

[0006] Furthermore, the step of screening the word segmentation results corresponding to all the second medical data to obtain the target keyword includes: screening out the first word segmentation result associated with the target disease from the word segmentation results corresponding to all the second medical data; judging whether there is at least one word segmentation combination in the first word segmentation result; wherein the multiple word segmentations in each word segmentation combination have the same meaning; if there is at least one word segmentation combination, for each word segmentation combination, replacing the multiple word segmentations in the word segmentation combination with a designated word segmentation; wherein the designated word segmentation has the same meaning as each word segmentation in the word segmentation combination; and obtaining the target keyword based on the first word segmentation result and each designated word segmentation.

[0007] Furthermore, the BRF model is trained in the following manner: for each second medical data, if the word segmentation result corresponding to the second medical data contains at least one target keyword, the second medical data is determined as sample medical data; for each sample medical data, each target keyword is marked according to whether each target keyword exists in the sample medical data, and the second marking result corresponding to each target keyword is obtained; multiple second marking results corresponding to all sample medical data are used as the first data set; data balancing is performed on the first data set to obtain a second data set; wherein the second data set includes multiple sample data; each sample data is marked with first marking information; the first marking information is used to indicate whether the patient corresponding to the sample data suffers from the target disease; based on the second data set, the current sample data is determined, and the current sample data is input into the initial model to output the prediction result corresponding to the current sample data through the initial model; based on the first marking information and the prediction result corresponding to the current sample data, the loss value is calculated, the initial model is updated based on the loss value, the next set of sample data is used as the new current sample data, and the step of inputting the current sample data into the initial model is repeated until the loss value converges to obtain a trained BRF model.

[0008] Furthermore, the step of performing data equalization processing on the first data set to obtain the second data set includes: processing the first data set through a preset generative adversarial network to obtain a third data set; resampling the third data set to obtain a second data set; wherein, in the second data set, the number of sample data of patients suffering from the target disease is the same as the number of sample data of patients not suffering from the target disease.

[0009] Furthermore, the method also includes: performing data mining processing on the third data set according to the FP-Growth algorithm, and outputting the correlation results between each target keyword and other target keywords.

[0010] Furthermore, the preset fields include: diagnosis of disease, description of condition, and disease code.

[0011] The present invention provides a device for predicting the probability of disease, which includes: an acquisition module, used to acquire at least one target keyword associated with a target disease, and disease description information of a target patient; a marking module, used to mark each target keyword according to whether the target keyword exists in the disease description information, and obtain a first marking result corresponding to the target keyword; and an output module, used to input the disease description information and the first marking result corresponding to each target keyword into a pre-trained BRF model, so as to output the probability of the target patient suffering from the target disease through the BRF model.

[0012] An electronic device provided by the present invention includes a processor and a memory, wherein the memory stores machine executable instructions that can be executed by the processor, and the processor executes the machine executable instructions to implement any of the above-mentioned methods for predicting the probability of disease.

[0013] The present invention provides a machine-readable storage medium, which stores machine-executable instructions. When the machine-executable instructions are called and executed by a processor, the machine-executable instructions prompt the processor to implement any of the above-mentioned methods for predicting the probability of disease.

[0014] The method, device and electronic device for predicting the probability of disease provided by the present invention obtain at least one target keyword associated with the target disease and the disease description information of the target patient; for each target keyword, the target keyword is marked according to whether the target keyword exists in the disease description information, and a first marking result corresponding to the target keyword is obtained; the disease description information and the first marking result corresponding to each target keyword are input into a pre-trained BRF model to output the probability that the target patient suffers from the target disease through the BRF model. This method predetermines at least one target keyword associated with the target disease. After marking each target keyword based on the disease description information of the target patient, the first marking result corresponding to each target keyword and the disease description information are used together as the input of the BRF model, which can improve the comprehensiveness and accuracy of evaluating whether the target patient suffers from the target disease. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 A flowchart of a method for predicting the probability of illness provided by an embodiment of the present invention; Figure 2 A flowchart of another method for predicting the probability of illness provided by an embodiment of the present invention; Figure 3 A schematic diagram of the structure of a device for predicting the probability of illness provided by an embodiment of the present invention; Figure 4 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] Systemic lupus erythematosus (SLE) is a complex autoimmune disease with unknown etiology and diverse and unpredictable symptoms, which poses many challenges to clinical diagnosis and treatment. SLE patients are often at risk of multiple serious complications, including kidney disease, cardiovascular disease, and nervous system damage, which have a significant impact on patients' survival rate and quality of life.

[0019] The symptoms of SLE are complex, highly variable, and significantly different among different patients, which adds challenges to the accurate prediction of its complications. Current prediction models are mostly based on single or limited clinical variables, and fail to fully utilize the complex disease-related characteristics, making it difficult to comprehensively and accurately assess the risk of complications in patients. In addition, data imbalance is common: the amount of data for a few serious complications is small, which increases the difficulty of model prediction, especially the lack of a systematic solution for generating a few target samples, resulting in poor model prediction accuracy and stability. Therefore, the performance of traditional models is still insufficient in accurately identifying a few high-risk cases.

[0020] In summary, the prior art has the following objective disadvantages: Traditional SLE risk assessment methods mainly rely on doctors' experience or simple statistical models, which makes it difficult to comprehensively and accurately assess patients' risk of developing systemic lupus erythematosus and to efficiently handle the complex correlations of SLE complications. Existing prediction technologies lack an effective data supplementation mechanism for a small number of complication samples, resulting in poor prediction results for a small number of complications. In terms of data processing and model application, most existing technologies do not provide a simple, user-friendly input and prediction interface, making it difficult to achieve rapid deployment and clinical application. In terms of complication association analysis, existing technologies mainly rely on traditional data mining methods, but lack fine-grained mining of complex complication associations, and fail to fully utilize the advantages of deep learning and generative models to supplement data.

[0021] Based on this, the embodiments of the present invention provide a method, device and electronic device for predicting the probability of disease. The technology can be applied to applications that require predicting the risk of a patient suffering from a target disease.

[0022] To facilitate understanding of this embodiment, a method for predicting the probability of illness disclosed in an embodiment of the present invention is first introduced. Figure 1 As shown, the method comprises the following steps: Step S102, obtaining at least one target keyword associated with the target disease and disease description information of the target patient.

[0023] The above-mentioned target disease can be any disease, for example, it can be systemic lupus erythematosus, hypertension, etc.; the above-mentioned target keywords usually include symptoms, complications, diagnosis and treatment prescriptions, etc. corresponding to the target disease, and the number of target keywords corresponding to different target diseases is usually different; the above-mentioned disease description information can be a diagnosis in natural language form, for example, it can include case history, medication history, etc., and the disease description information can be compiled by the doctor based on the target patient's self-description of the condition; in actual implementation, when it is necessary to evaluate the probability that the target patient suffers from the target disease, it is usually necessary to first obtain at least one target keyword corresponding to the target disease, as well as the disease description information of the target patient. Step S104: for each target keyword, the target keyword is marked according to whether the target keyword exists in the disease description information, and a first marking result corresponding to the target keyword is obtained.

[0024] Each target keyword can be extracted from the above-mentioned disease description information to determine whether each target keyword exists in the disease description information. According to the extraction situation of each target keyword, each target keyword is marked respectively to obtain a first marking result corresponding to the target keyword. The first marking result is used to indicate whether the target keyword exists in the disease description information.

[0025] Step S106, inputting the disease description information and the first labeling result corresponding to each target keyword into a pre-trained BRF model, so as to output the probability that the target patient suffers from the target disease through the BRF model.

[0026] The above-mentioned BRF (Balanced Random Forest) model is an integrated learning method, which combines the prediction results of multiple decision trees to improve the accuracy and stability of the model. In actual implementation, the above-mentioned disease description information and the first labeling result corresponding to each target keyword can be input into a pre-trained BRF model to predict the probability that the target patient suffers from the target disease, for example, the possibility of suffering from the target disease is 52%.

[0027] The above-mentioned method for predicting the probability of disease obtains at least one target keyword associated with the target disease and the disease description information of the target patient; for each target keyword, the target keyword is marked according to whether the target keyword exists in the disease description information, and a first marking result corresponding to the target keyword is obtained; the disease description information and the first marking result corresponding to each target keyword are input into a pre-trained BRF model to output the probability that the target patient suffers from the target disease through the BRF model. This method predetermines at least one target keyword associated with the target disease. After marking each target keyword based on the disease description information of the target patient, the first marking result corresponding to each target keyword and the disease description information are used together as input to the BRF model, which can improve the comprehensiveness and accuracy of evaluating whether the target patient suffers from the target disease.

[0028] The embodiment of the present invention further provides another method for predicting the probability of illness, which is implemented on the basis of the method in the above embodiment, and comprises the following steps: Step 1: obtaining at least one target keyword associated with the target disease and disease description information of the target patient; The target keywords are determined by following steps 10 to 13: Step 10, obtaining an initial medical data set; wherein the initial medical data set includes first diagnosis and treatment data of multiple patients; The first diagnosis and treatment data of the above-mentioned multiple patients may be medical-related data of multiple patients corresponding to a specified time period in a specified area, for example, medical-related data in a medical insurance data set of a certain city in the past five years. In actual implementation, in order to comprehensively and accurately determine the target keywords associated with the target disease, the first diagnosis and treatment data of the initial medical multiple patients may be obtained on demand.

[0029] Step 11, according to the preset fields, filter out the second diagnosis and treatment data corresponding to the preset fields from the first diagnosis and treatment data corresponding to each patient; wherein the preset fields are fields associated with the target disease.

[0030] The above-mentioned preset fields are usually fields associated with the target disease, for example, they may include: diagnosis of the disease, description of the condition, disease code, etc.; wherein, the diagnosis of the disease can be understood as the patient's final disease diagnosis result, which may be the target disease or other diseases; the above-mentioned condition description may be a diagnosis in natural language obtained by the doctor based on the target patient's self-description of the condition; the above-mentioned disease code may refer to the code in the International Classification of Diseases (ICD) system, which is a standard method for classifying and coding diseases, injuries and poisonings; in actual implementation, the first diagnosis and treatment data of the multiple patients obtained above usually include data corresponding to multiple different fields, some fields are associated with disease assessment, such as description of the condition, etc., and some are not related to disease assessment, such as the cost spent by the patient, etc.; in order to reduce the amount of data calculation, the preset fields can be pre-set, and the first diagnosis and treatment data of each patient can be screened according to the preset fields to screen out the second diagnosis and treatment data corresponding to the preset fields.

[0031] Step 12: input the second medical data of each patient into a preset word segmentation model to perform word segmentation processing on each piece of the second medical data to obtain a word segmentation result corresponding to each piece of the second medical data.

[0032] The above-mentioned preset word segmentation model can be an LTP (Language Technology Platform) model. The LTP model can provide rich, efficient and accurate natural language processing including Chinese word segmentation, part-of-speech tagging, named entity recognition, dependency syntax analysis, semantic role labeling, etc.; in actual implementation, the second medical data of each patient can be input into the preset word segmentation model respectively, and each second medical data can be segmented separately through the preset word segmentation model to obtain the word segmentation results corresponding to each second medical data.

[0033] Step 13, screening the word segmentation results corresponding to all the second medical data to obtain target keywords.

[0034] The step 13 can be implemented by the following steps 130 to 133: Step 130, screening out the first segmentation result associated with the target disease from the segmentation results corresponding to all the second diagnosis and treatment data; Step 131, determining whether there is at least one word segmentation combination in the first word segmentation result; wherein multiple words in each word segmentation combination have the same meaning; Step 132: if there is at least one participle combination, for each participle combination, multiple participles in the participle combination are replaced with a designated participle; wherein the designated participle has the same meaning as each participle in the participle combination; Step 133: Obtain a target keyword based on the first word segmentation result and each designated word segmentation.

[0035] In actual implementation, for each second medical data, the word segmentation result corresponding to the second medical data usually contains multiple word segmentations, each word segmentation may be related to the target disease or may not be related to the target disease. In order to reduce the amount of data calculation and improve the prediction accuracy, the first word segmentation result associated with the target disease can be screened out from the word segmentation results corresponding to all the second medical data. Considering that the same meaning may have different descriptions in actual applications, the word segmentations with the same meaning can be combined into a word segmentation combination. Therefore, after obtaining the first word segmentation result, it can be determined whether there is at least one word segmentation combination in the first word segmentation result; for example, the first word segmentation result contains headache and headache. Although headache and headache are two different descriptions, they express the same meaning. Therefore, headache and headache can be used as a word segmentation combination. If there are one or more participle combinations, for each participle combination, multiple participles in the participle combination can be replaced with a specified participle. The specified participle can be any participle in the participle combination. Of course, other participles can also be selected as replacements, as long as the specified participle has the same meaning as the multiple participles in the participle combination. After replacing each participle combination in the first participle result with a corresponding specified participle, the target keyword can be obtained; for example, a total of 200 participles are collected in the first participle result. After the above replacement process, 33 target keywords are finally obtained. For example, the target keywords corresponding to systemic lupus erythematosus include: rash, edema, ambrisentan, combined white blood cells, tadalafil, urination, methotrexate, etc.; in this way, the uniqueness of each participle in the target keyword can be guaranteed.

[0036] Step 2: For each target keyword, mark the target keyword according to whether the target keyword exists in the disease description information to obtain a first marking result corresponding to the target keyword.

[0037] Step three, input the disease description information and the first labeling result corresponding to each target keyword into a pre-trained BRF model to output the probability that the target patient suffers from the target disease through the BRF model.

[0038] For example, the natural language description of the target patient is obtained as follows: the patient recently suffered from respiratory tract infection, chest tightness and shortness of breath, hypothyroidism, proteinuria, weakened lower limb venous function, severe osteoporosis, and took Ursofibrate, amlodipine besylate and tadalafil, etc., with thrombocytopenia and leukocytopenia; the above 33 target keywords are extracted and marked whether they appear. The pre-trained BRF model is called for prediction, and the prediction result is {False: 51.50, True: 48.50}, that is, the possibility of suffering from SLE is about 48.50%, which should be taken seriously in clinical or other application scenarios.

[0039] The above BRF model can be trained by following the steps 30 and 35: Step 30: for each piece of second medical data, if the word segmentation result corresponding to the second medical data contains at least one target keyword, the second medical data is determined as sample medical data. Step 31 : for each piece of sample diagnosis and treatment data, each target keyword is marked respectively according to whether each target keyword exists in the sample diagnosis and treatment data, so as to obtain a second marking result corresponding to each target keyword.

[0040] Step 32: taking the plurality of second labeling results corresponding to all sample diagnosis and treatment data as the first data set.

[0041] According to each target keyword obtained as described above, and the word segmentation results corresponding to each second medical data, the second medical data containing one or more target keywords are retrieved, and the multiple second medical data retrieved can be used as multiple sample medical data. In each sample medical data, it is traversed to see whether each target keyword exists, and the corresponding second marking result is marked. For example, if a target keyword exists, it can be indicated by "yes", if it does not exist, it can be indicated by "no", etc.; the multiple second marking results corresponding to all the sample medical data can be used as a first data set, and the form of the first data set is a table characterized by the target keywords. For example, if the number of target keywords is 33, then the first data set is a table characterized by 33 target keywords.

[0042] Step 33, performing data equalization processing on the first data set to obtain a second data set; wherein the second data set includes multiple sample data; each sample data is annotated with first annotation information; the first annotation information is used to indicate whether the patient corresponding to the sample data suffers from the target disease.

[0043] The step 33 can be specifically implemented through the following steps 330 to 332 .

[0044] Step 330: Process the first data set through a preset generative adversarial network to obtain a third data set.

[0045] Step 331, resample the third data set to obtain a second data set; wherein, in the second data set, the number of sample data of patients suffering from the target disease is the same as the number of sample data of patients not suffering from the target disease.

[0046] Step 332: Perform data mining on the third data set according to the FP-Growth algorithm, and output the correlation results between each target keyword and other target keywords.

[0047] The above-mentioned Generative Adversarial Network (GAN) is a deep learning model, which is mainly used to generate new data samples, which are very similar to the training data. The above-mentioned FP-Growth (Frequent Pattern Growth) is an algorithm for frequent pattern mining, which is particularly suitable for processing large data sets. It is mainly used to mine frequent item sets and association rules, and is widely used in market basket analysis, recommendation systems and other data mining tasks. In actual implementation, among all the sample diagnosis and treatment data corresponding to the above-mentioned first data set, only a very small number of patients may be diagnosed with the target disease, and other patients may be diagnosed with other diseases. The proportion of patients with the target disease is quite different. Therefore, it is necessary to generate more data through the preset generative adversarial network to obtain the third data set; for example, the first data set corresponds to 140,000 data, and only 171 data corresponding to patients are diagnosed with SLE, and the proportion of patients with the disease is very different. After the generative adversarial network is used, more data can be generated, for example, 5,000 data diagnosed with SLE can be added. In the final third data set, there are 5,171 data diagnosed with SLE and 137,157 data diagnosed with not SLE.

[0048] In actual implementation, SMOTE (Synthetic Minority Over-sampling Technique) can be used for resampling. SMOTE is a resampling method used to solve the problem of unbalanced data sets, especially in classification tasks. The core idea of ​​SMOTE is to enhance minority samples by generating synthetic samples, thereby improving the performance of the classifier. Through SMOTE resampling, the characteristics of minority samples can be highlighted to achieve data balancing. For example, continuing with the above example, the sample distribution in the second data set obtained after resampling is Counter({False: 137157, True: 137157}). After the above steps, the data with SLE and the data without SLE in the second data set have achieved a balanced effect. Specifically, the data volume ratio can be 1:1.

[0049] During the training process of the BRF model, after obtaining the above-mentioned third data set, the correlation between each target keyword and other target keywords can be verified according to the FP-Growth algorithm. For example, if the target keywords include symptoms, complications, etc., the correlation between each symptom and complication can be verified. The correlation result is mainly used as an auxiliary conclusion.

[0050] Step 34, determining the current sample data based on the second data set, and inputting the current sample data into the initial model, so as to output the prediction result corresponding to the current sample data through the initial model.

[0051] Step 35, based on the first annotation information and prediction results corresponding to the current sample data, calculate the loss value, update the initial model based on the loss value, use the next set of sample data as the new current sample data, and repeat the step of inputting the current sample data into the initial model until the loss value converges to obtain a trained BRF model.

[0052] The above BRF is an improved random forest algorithm designed to deal with unbalanced classification problems. It combines the advantages of random forests and sample balancing technology to improve the classification performance of minority class samples. It is also widely used in the medical field, such as disease prediction and diagnosis, and identification of rare disease cases. In actual implementation, the current sample data can be determined based on the second data set. For example, taking the number of target keywords as 33, the second data set includes multiple sample data, each of which includes the labeling results corresponding to the 33 target keywords. The first sample data can be used as the current sample data and input into the initial model to update the weight parameters in the initial model. The second sample data continues to be used as the current sample data, and the above process is repeated until the loss value converges, and a trained BRF model can be obtained.

[0053] For easier understanding, see Figure 2 A flowchart of another method for predicting the probability of illness is shown, which extracts diagnostic information from a city's medical insurance data set (corresponding to the above-mentioned preset fields, the second diagnostic data corresponding to the preset fields are screened out from the first diagnostic data corresponding to each patient), performs word segmentation processing on the diagnostic information through the LTP model, and obtains target keywords through screening. For each second diagnostic data, if the word segmentation result corresponding to the second diagnostic data contains at least one target keyword, the second diagnostic data is determined as sample diagnostic data. For each sample diagnostic data, each target keyword is marked separately according to whether it exists in the sample diagnostic data, so as to obtain each target keyword for each The corresponding second labeling results are used, and the multiple second labeling results corresponding to all the sample diagnosis and treatment data are used as the first data set. If the first data set is extremely unbalanced, the GANs adversarial network is used to make up for the problem of fewer diseased data samples to obtain a third data set. The third data set can be processed by data mining according to the FP-Growth algorithm to output the correlation results between each target keyword and other target keywords as auxiliary conclusions; the third data set is resampled by SMOTE to balance the data to obtain the second data set; the BRF model is trained based on the second data set, and the patient's diagnosis and treatment information is input into the BRF model to predict the risk probability of the patient suffering from the target disease.

[0054] The following is an actual verification result. In the test set, there are 34,333 data, of which 34,287 are undiagnosed SLE patients and 46 are diagnosed SLE patients. In the model prediction results, for 34,287 patients without SLE, only 1 was misdiagnosed with SLE (or said to have a higher risk of SLE); for 46 patients with SLE, 37 were predicted to have SLE and 9 were not diagnosed with SLE (or said to have a lower risk of SLE). In general, for SLE, a rare disease in medicine, the prediction accuracy reached 80.4%, which can be considered a good model training effect.

[0055] On the user side, the demand side can directly input disease description information (including case history and medication history) in natural language form through the user and pass it to the BRF model to return the prediction results, which is relatively simple and flexible.

[0056] The above-mentioned disease probability prediction method automatically analyzes the patient's disease description information by combining the LTP model (Lexicalized Tree Projection) in natural language processing, and integrates target keywords containing complications and symptom characteristics of the target disease to further enhance the contextual accuracy of the data. It can provide a more efficient and rich input information source for training the BRF model.

[0057] The innovative introduction of the FP-Growth (Frequent Itemset Growth) algorithm can efficiently discover the frequent association rules between complications of the target disease, capture the association patterns between complications, and provide multi-dimensional correlation features for the model, thereby improving the accuracy of BRF model prediction.

[0058] Generative adversarial networks (GANs) are used to generate realistic synthetic data for minority class samples. Through equalization processing, the prediction bias problem caused by insufficient minority samples is innovatively solved, effectively enhancing the generalization ability of the model. This method introduces SMOTE (synthetic minority oversampling technology) to balance the data set before model training to avoid overfitting to majority class samples, so that model training can fully capture the feature distribution of different categories and achieve prediction stability and accuracy. This method innovatively uses the BRF model to enhance the recognition ability of minority classes while achieving classification tasks, ensuring high prediction performance in the case of unbalanced samples.

[0059] Taking SLE as an example, the target disease is SLE. SLE complications often develop rapidly, and once they occur, they are difficult to treat clinically and have a poor prognosis. For example, complications such as lupus nephritis and lupus encephalopathy can cause irreversible organ damage, seriously affecting the patient's prognosis. This method helps to adopt individualized treatment plans and intervention measures early by predicting the risk of systemic lupus erythematosus, thereby delaying or avoiding disease deterioration and gaining the best treatment window for patients. This method can also significantly reduce the consumption of medical resources by early identification of high-risk patients, effectively improve treatment outcomes, and reduce hospitalization rates and long-term complications caused by complications.

[0060] The embodiment of the present invention provides a device for predicting the probability of disease. Figure 3 As shown, the device includes: an acquisition module 30, used to acquire at least one target keyword associated with the target disease and the disease description information of the target patient; a marking module 31, used to mark each target keyword according to whether the target keyword exists in the disease description information, and obtain a first marking result corresponding to the target keyword; an output module 32, used to input the disease description information and the first marking result corresponding to each target keyword into a pre-trained BRF model, so as to output the probability of the target patient suffering from the target disease through the BRF model.

[0061] The above-mentioned device for predicting the probability of illness predetermines at least one target keyword associated with the target disease. After marking each target keyword based on the disease description information of the target patient, the first marking result and the disease description information corresponding to each target keyword are used together as the input of the BRF model, which can improve the comprehensiveness and accuracy of evaluating whether the target patient suffers from the target disease.

[0062] Furthermore, the device also includes a target keyword determination module, which is used to: obtain an initial medical data set; wherein the initial medical data set includes first diagnosis and treatment data of multiple patients; according to a preset field, filter out second diagnosis and treatment data corresponding to a preset field from the first diagnosis and treatment data corresponding to each patient; wherein the preset field is a field associated with a target disease; input the second diagnosis and treatment data of each patient into a preset word segmentation model respectively to perform word segmentation processing on each second diagnosis and treatment data to obtain a word segmentation result corresponding to each second diagnosis and treatment data; filter the word segmentation results corresponding to all the second diagnosis and treatment data to obtain the target keyword.

[0063] Furthermore, the target keyword determination module is also used to: screen out the first segmentation result associated with the target disease from the segmentation results corresponding to all the second diagnosis and treatment data; determine whether there is at least one segmentation combination in the first segmentation result; wherein the multiple segmentations in each segmentation combination have the same meaning; if there is at least one segmentation combination, for each segmentation combination, replace the multiple segmentations in the segmentation combination with a designated segmentation; wherein the designated segmentation has the same meaning as each segmentation in the segmentation combination; and obtain the target keyword based on the first segmentation result and each designated segmentation.

[0064] Furthermore, the device also includes a BRF model training module, which is used to: for each second medical data, if the word segmentation result corresponding to the second medical data contains at least one target keyword, determine the second medical data as sample medical data; for each sample medical data, mark each target keyword separately according to whether each target keyword exists in the sample medical data, and obtain a second marking result corresponding to each target keyword; use multiple second marking results corresponding to all sample medical data as a first data set; perform data balancing on the first data set to obtain a second data set; wherein, the first The two data sets include multiple sample data; each sample data is annotated with first annotation information; the first annotation information is used to indicate whether the patient corresponding to the sample data suffers from the target disease; the current sample data is determined based on the second data set, and the current sample data is input into the initial model to output the prediction result corresponding to the current sample data through the initial model; based on the first annotation information and the prediction result corresponding to the current sample data, the loss value is calculated, the initial model is updated based on the loss value, the next set of sample data is used as the new current sample data, and the step of inputting the current sample data into the initial model is repeated until the loss value converges to obtain a trained BRF model.

[0065] Furthermore, the BRF model training module is used to: process the first data set through a preset generative adversarial network to obtain a third data set; resample the third data set to obtain a second data set; wherein, in the second data set, the number of sample data of patients suffering from the target disease is the same as the number of sample data of patients not suffering from the target disease.

[0066] Furthermore, the BRF model training module is used to: perform data mining processing on the third data set according to the FP-Growth algorithm, and output the correlation results between each target keyword and other target keywords.

[0067] Furthermore, the preset fields include: diagnosis of disease, description of condition, and disease code.

[0068] The device for predicting the probability of illness provided in the embodiment of the present invention has the same implementation principle and technical effects as those in the aforementioned embodiment of the method for predicting the probability of illness. For the sake of brief description, for matters not mentioned in the embodiment of the device for predicting the probability of illness, reference may be made to the corresponding contents in the aforementioned embodiment of the method for predicting the probability of illness.

[0069] The embodiment of the present invention further provides an electronic device, see Figure 4 As shown, the electronic device includes a processor 130 and a memory 131 , wherein the memory 131 stores machine executable instructions that can be executed by the processor 130 , and the processor 130 executes the machine executable instructions to implement the above-mentioned method for predicting the probability of illness.

[0070] Further, Figure 4 The electronic device shown further includes a bus 132 and a communication interface 133 , and the processor 130 , the communication interface 133 and the memory 131 are connected via the bus 132 .

[0071] The memory 131 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk storage. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 133 (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. may be used. The bus 132 may be an ISA bus, a PCI bus, or an EISA bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0072] The processor 130 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 130. The above processor 130 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in conjunction with the embodiments of the present invention can be directly embodied as a hardware decoding processor for execution, or a combination of hardware and software modules in the decoding processor for execution. The software module may be located in a storage medium mature in the art, such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 131, and the processor 130 reads the information in the memory 131 and completes the steps of the method of the above embodiment in combination with its hardware.

[0073] An embodiment of the present invention also provides a machine-readable storage medium, which stores machine-executable instructions. When the machine-executable instructions are called and executed by a processor, the machine-executable instructions prompt the processor to implement the above-mentioned disease probability prediction method. The specific implementation can be found in the method embodiment, which will not be repeated here.

[0074] The computer program product of the disease probability prediction method, device and electronic device provided in the embodiments of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the method described in the previous method embodiments. The specific implementation can be found in the method embodiments, which will not be repeated here.

[0075] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the probability of illness, characterized in that: The method comprises: Acquire at least one target keyword associated with a target disease and disease description information of a target patient; For each of the target keywords, marking the target keyword according to whether the target keyword exists in the disease description information, and obtaining a first marking result corresponding to the target keyword; The disease description information and the first labeling result corresponding to each target keyword are input into a pre-trained BRF model, so as to output the probability that the target patient suffers from the target disease through the BRF model.

2. The method according to claim 1, characterized in that The target keywords are determined in the following way: Acquire an initial medical data set; wherein the initial medical data set includes first diagnosis and treatment data of multiple patients; According to a preset field, second diagnosis and treatment data corresponding to the preset field is screened out from the first diagnosis and treatment data corresponding to each patient; wherein the preset field is a field associated with the target disease; Inputting the second diagnosis and treatment data of each of the patients into a preset word segmentation model respectively, so as to perform word segmentation processing on each piece of the second diagnosis and treatment data, and obtain a word segmentation result corresponding to each piece of the second diagnosis and treatment data; The word segmentation results corresponding to all the second medical data are screened to obtain the target keywords.

3. The method according to claim 2, characterized in that The step of screening the word segmentation results corresponding to all the second medical data to obtain the target keyword includes: Filtering out the first segmentation result associated with the target disease from the segmentation results corresponding to all the second diagnosis and treatment data; Determine whether there is at least one word segmentation combination in the first word segmentation result; wherein the multiple words in each word segmentation combination have the same meaning; If there is at least one participle combination, for each participle combination, multiple participles in the participle combination are replaced with a designated participle; wherein the designated participle has the same meaning as each participle in the participle combination; A target keyword is obtained based on the first word segmentation result and each of the designated words.

4. The method according to claim 2, characterized in that: The BRF model is trained in the following way: For each piece of the second medical data, if the word segmentation result corresponding to the second medical data contains at least one of the target keywords, the second medical data is determined as sample medical data; For each piece of the sample diagnosis and treatment data, marking each of the target keywords respectively according to whether each of the target keywords exists in the sample diagnosis and treatment data, and obtaining a second marking result respectively corresponding to each of the target keywords; Taking the plurality of second labeling results respectively corresponding to all the sample diagnosis and treatment data as the first data set; Performing data balancing processing on the first data set to obtain a second data set; wherein the second data set includes a plurality of sample data; each of the sample data is annotated with first annotation information; the first annotation information is used to indicate whether the patient corresponding to the sample data suffers from the target disease; Determine current sample data based on the second data set, and input the current sample data into an initial model, so as to output a prediction result corresponding to the current sample data through the initial model; Based on the first annotation information corresponding to the current sample data and the prediction result, the loss value is calculated, the initial model is updated based on the loss value, the next set of sample data is used as the new current sample data, and the step of inputting the current sample data into the initial model is repeated until the loss value converges to obtain the trained BRF model.

5. The method according to claim 4, characterized in that The step of performing data equalization processing on the first data set to obtain a second data set includes: Processing the first data set by a preset generative adversarial network to obtain a third data set; The third data set is resampled to obtain a second data set; wherein, in the second data set, the number of sample data of patients suffering from the target disease is the same as the number of sample data of patients not suffering from the target disease.

6. The method according to claim 5, characterized in that The method further comprises: According to the FP-Growth algorithm, data mining processing is performed on the third data set to output the correlation results between each of the target keywords and other target keywords.

7. The method according to claim 2, characterized in that The preset fields include: diagnosis of disease, description of condition, and disease code.

8. A device for predicting the probability of disease, characterized in that: The device comprises: An acquisition module, used to acquire at least one target keyword associated with a target disease and disease description information of a target patient; A marking module, configured to mark each target keyword according to whether the target keyword exists in the disease description information, and obtain a first marking result corresponding to the target keyword; The output module is used to input the disease description information and the first labeling result corresponding to each target keyword into a pre-trained BRF model, so as to output the probability that the target patient suffers from the target disease through the BRF model.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores machine executable instructions that can be executed by the processor, and the processor executes the machine executable instructions to implement the method for predicting the probability of illness according to any one of claims 1 to 7.

10. A machine-readable storage medium, characterized in that: The machine-readable storage medium stores machine-executable instructions. When the machine-executable instructions are called and executed by a processor, the machine-executable instructions prompt the processor to implement the method for predicting the probability of illness described in any one of claims 1-7.

Images