Chronic disease pathology analysis method and system based on big data

Abstract

The invention provides a chronic disease pathology analysis method and system based on big data, and relates to the field of medical data analys.The method comprises the steps that physiological index data, medication information and symptom description information are input; processing the symptom description information through a natural language processing model to obtain a symptom description vector; searching a target historical patient according to the symptom description vector and the physiological index data; according to the historical physiological index data and the historical medication information of the target historical patient, training an illness state development prediction model to obtain a trained illness state development prediction model; according to the trained disease development prediction model, the medication information and the physiological index data, determining whether the medication information needs to be adjusted; if the medication information needs to be adjusted, the adjusted reference medication information is determined according to the trained disease development prediction model and the physiological index data. According to the invention, the illness condition of the chronic disease patient can be accurately evaluated, the illness development can be predicted, and accurate medication suggestions can be given.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical data analysis, and in particular to a chronic disease pathological analysis method and system based on big data. BACKGROUND

[0002] In the related art, CN119581029A discloses a chronic disease risk management system based on big data analysis, and specifically relates to the technical field of chronic disease risk management. By obtaining hardware level information, sensor quality information, and user device wearing behavior information of intelligent wearable devices, an error influence evaluation model is constructed according to the hardware level differentiation coefficient, the sensor quality decline index, and the user device wearing behavior coefficient, and an error influence evaluation index is generated. The potential error risk caused by low-cost devices due to low hardware level, sensor quality decline, and nonlinear influence factors of user wearing behavior is intelligently identified. By combining the personalized hardware level differentiation coefficient, the sensor quality decline index, and the user device wearing behavior coefficient, the compensation intensity and method are dynamically adjusted, breaking through the limitations of traditional linear or simple compensation strategies, flexibly optimizing the compensation strategy, and ensuring the data quality of low-cost device users, thereby expanding the applicable population of the chronic disease management system.

[0003] CN118866213A discloses a chronic disease patient management system based on big data analysis, and relates to the technical field of medical management. Specifically, it is a chronic disease patient management system based on big data analysis, which includes a medical data input module, a database, a medical background, and a data analysis module. The medical data input module is interactively connected to the input end of the database, the database is interactively connected to the input end of the data analysis module, and the data analysis module is interactively connected to the input end of the medical background. The medical data input module is used to collect the physiological data and medical information of chronic disease patients. The chronic disease patient management system based on big data analysis can retrieve abnormal data of initial diagnosis patients, compare and analyze the data with historical chronic disease data, facilitate medical personnel to directly determine whether the initial diagnosis patients have chronic diseases, and at the same time, compare and analyze the data of re-visit patients with historical chronic disease data, provide medical advice, and reduce the workload of medical personnel.

[0004] Therefore, the related art can analyze the physiological data of chronic disease patients through the background, but cannot determine the development trend and recovery of chronic diseases, nor can it provide medication suggestions.

[0005] The information disclosed in the background section of the present application is only intended to deepen the understanding of the general background of the present application, and should not be regarded as acknowledging or implying in any form that the information constitutes prior art known to those skilled in the art. SUMMARY

[0006] The application provides a chronic disease pathological analysis method and system based on big data, which can solve the technical problems that the development trend and recovery condition of chronic diseases cannot be determined and drug recommendations cannot be given in the related art.

[0007] According to a first aspect of the application, a chronic disease pathological analysis method based on big data is provided, comprising:

[0008] Enter the physiological index data, drug information and symptom description information of the patient;

[0009] Process the symptom description information through a natural language processing model to obtain a symptom description vector;

[0010] According to the symptom description vector and the physiological index data, find a target historical patient in a historical database;

[0011] According to the historical physiological index data and historical drug information of the target historical patient at multiple time points, train a disease development prediction model to obtain a trained disease development prediction model;

[0012] According to the trained disease development prediction model, the drug information and the physiological index data, determine whether the drug information needs to be adjusted;

[0013] If the drug information needs to be adjusted, according to the trained disease development prediction model and the physiological index data, determine the adjusted reference drug information.

[0014] According to the application, according to the symptom description vector and the physiological index data, find a target historical patient in a historical database, comprising:

[0015] According to the symptom description vector and the historical symptom description vectors of a plurality of historical patients in the historical database, select a candidate historical patient from the plurality of historical patients;

[0016] According to the physiological index data and the historical physiological index data of the candidate historical patient, select a target historical patient from the candidate historical patient.

[0017] According to the application, according to the physiological index data and the historical physiological index data of the candidate historical patient, select a target historical patient from the candidate historical patient, comprising:

[0018] Determine the type, abnormal type and abnormal amplitude of the abnormal physiological index data in the physiological index data;

[0019] Determine the historical type, historical abnormal type and historical abnormal amplitude of the abnormal physiological index data in the historical physiological index data of each candidate historical patient;

[0020] determine an abnormal physiological indicator vector of the patient according to the type of the abnormal physiological indicator data, the abnormal type and the abnormal amplitude;

[0021] determine a historical abnormal physiological indicator vector of each of the candidate historical patients according to the historical type of the abnormal physiological indicator data, the historical abnormal type and the historical abnormal amplitude in the historical physiological indicator data of each of the candidate historical patients;

[0022] select a target historical patient from the candidate historical patients according to the abnormal physiological indicator vector and the historical abnormal physiological indicator vector.

[0023] According to the present application, in the abnormal physiological indicator vector, the component corresponding to the type of the normal physiological indicator data is 0, the component corresponding to the type of the abnormal physiological indicator exceeding the upper limit of the physiological indicator is the abnormal amplitude, and the component corresponding to the type of the abnormal physiological indicator lower than the lower limit of the physiological indicator is the negative abnormal amplitude;

[0024] In the historical abnormal physiological indicator vector, the component corresponding to the type of the normal historical physiological indicator data is 0, the component corresponding to the type of the historical abnormal physiological indicator exceeding the upper limit of the physiological indicator is the historical abnormal amplitude, and the component corresponding to the type of the historical abnormal physiological indicator lower than the lower limit of the physiological indicator is the negative historical abnormal amplitude;

[0025] select a target historical patient from the candidate historical patients according to the abnormal physiological indicator vector and the historical abnormal physiological indicator vector, comprising:

[0026] respectively calculate the cosine similarity of the abnormal physiological indicator vector and the historical abnormal physiological indicator vector of each of the candidate historical patients;

[0027] sort the historical abnormal physiological indicator vectors of each of the candidate historical patients according to the cosine similarity;

[0028] determine the candidate historical patients corresponding to the first preset number of historical abnormal physiological indicator vectors in the sequence obtained by sorting as the target historical patients.

[0029] According to the present application, the disease development prediction model is trained according to the historical physiological indicator data and the historical medication information of the target historical patient at multiple time points, and a trained disease development prediction model is obtained, comprising:

[0030] obtain the historical abnormal physiological indicator vectors of the target historical patient at multiple time points according to the historical physiological indicator data of the target historical patient at multiple time points;

[0031] input the historical abnormal physiological indicator vector and the historical medication information of the target historical patient at the i-th time point into the disease development prediction model to obtain a predicted abnormal physiological indicator vector at the i+1-th time point.

[0032] obtaining a loss function of the disease development prediction model according to the predicted abnormal physiological indicator vector at the i+1th moment and the historical abnormal physiological indicator vector at the i+1th moment;

[0033] training the disease development prediction model according to the loss function of the disease development prediction model, and obtaining a trained disease development prediction model.

[0034] According to the present application, the loss function of the disease development prediction model is obtained according to the predicted abnormal physiological indicator vector at the i+1th moment and the historical abnormal physiological indicator vector at the i+1th moment, comprising:

[0035] According to the formula

[0036]

[0037] obtaining a loss function of the disease development prediction model LOSS, wherein I j,i+1,p is the predicted abnormal physiological indicator vector of the i+1th moment of the jth target historical patient, I j,i+1 is the historical abnormal physiological indicator vector of the i+1th moment of the jth target historical patient, I j,i is the historical abnormal physiological indicator vector of the i+1th moment of the jth target historical patient, W is a preset weight matrix, the preset weight matrix is a diagonal matrix, and the element of the kth row and the kth column in the preset weight matrix is a preset weight of the type of the kth component of the predicted abnormal physiological indicator vector, n j is the number of historical abnormal physiological indicator vectors of the jth target historical patient, N is the number of target historical patients in each training batch, if is a conditional function, i≤n j , j≤N, and i, n j , j and N are all positive integers.

[0038] According to the present application, if it is necessary to adjust the medication information, the adjusted reference medication information is determined according to the trained disease development prediction model and the physiological indicator data, comprising:

[0039] obtaining an abnormal physiological indicator vector according to the physiological indicator data;

[0040] determining a constraint condition of a medication optimization model according to the abnormal physiological indicator vector and the trained disease development prediction model;

[0041] determining an objective function of the medication optimization model according to the trained disease development prediction model;

[0042] solving the medication optimization model according to the constraint condition and the objective function, and obtaining the adjusted reference medication information.

[0043] According to the present application, the constraint condition of the medication optimization model is determined according to the abnormal physiological index vector and the trained disease development prediction model, comprising:

[0044] According to the formula

[0045] f S (I ab ),M p )=I ab,next

[0046]

[0047] U x ≤U x,max

[0048] The constraint condition of the medication optimization model is determined, wherein I ab is an abnormal physiological index vector, M p is a to-be-determined adjusted reference medication information, U1 is the daily dosage of the first drug, U2 is the daily dosage of the second drug, U x is the daily dosage of the xth drug, U n is the daily dosage of the nth drug, U x,max is the upper limit of the daily dosage of the xth drug, I ab,next is the predicted abnormal physiological index vector of the patient at the next time, f S is a mapping function of the trained disease development prediction model, n is the number of drug types, 1≤x≤n, and x and n are positive integers.

[0049] According to the present application, the objective function of the medication optimization model is determined according to the trained disease development prediction model, comprising:

[0050] According to the formula

[0051] minimize|WI ab,next |

[0052]

[0053] The objective function of the medication optimization model is determined, wherein W is a preset weight matrix, the preset weight matrix is a diagonal matrix, the element of the kth row and the kth column in the preset weight matrix is a preset weight of the type of the kth component corresponding to the physiological index data in the abnormal physiological index vector, and minimize is a minimum function.

[0054] According to the second aspect of the present application, a chronic disease pathological analysis system based on big data is provided, comprising:

[0055] An input module is configured to input a plurality of physiological index data, medication information and symptom description information of a patient;

[0056] A symptom description vector module is configured to process the symptom description information by using a natural language processing model to obtain a symptom description vector;

[0057] A searching module is configured to search a target historical patient in a historical database according to the symptom description vector and the physiological index data;

[0058] A training module is configured to train a disease development prediction model according to a plurality of historical physiological index data and historical medication information of the target historical patient to obtain a trained disease development prediction model;

[0059] A judging module is configured to determine whether the medication information needs to be adjusted according to the trained disease development prediction model, the medication information and the physiological index data;

[0060] A reference medication information module is configured to determine adjusted reference medication information according to the trained disease development prediction model and the physiological index data if the medication information needs to be adjusted.

[0061] By using the above technical solutions, the present application can achieve the following technical effects:

[0062] According to the present application, the physiological index data, past medication information and symptom description information of the chronic disease patient can be input, and the historical data of the chronic disease patient similar to the patient is found, so as to train the disease development prediction model, determine the disease development trend of the chronic disease patient based on the trained disease development prediction model, determine whether the medication information needs to be adjusted, and give the adjusted reference medication information for the doctor to refer to, so as to accurately evaluate the disease condition of the chronic disease patient and predict the disease development, and give accurate medication suggestions, and reduce the workload of the doctor. When training the disease development prediction model, the preset weight matrix can be used to weight the prediction abnormal physiological index vector and the historical abnormal physiological index vector, so that the change of the overall health status of the historical patient can be accurately determined, and the loss function value corresponding to the change of the overall health status of the historical patient can be determined by the conditional function, so that the error of the historical abnormal amplitude of the key physiological index is focused on in the case of the overall health status improving, so as to improve the training pertinence and training efficiency, and the error of the historical abnormal amplitude of each physiological index is calculated in the case of the overall health status not improving, so as to improve the overall training and the accuracy of the disease development prediction model. When determining the reference medication information, the medication optimization model can be used to solve the optimal solution of the medication method, and under the constraint of the constraint condition, the medication safety and the prediction accuracy can be met, and the overall medication effect and the medication convenience of the patient can be improved by the two objective functions, so that the accurate and convenient recommended medication method can be obtained, and the workload of the doctor and other professionals can be reduced. BRIEF DESCRIPTION OF DRAWINGS

[0063] Figure 1 An exemplary flowchart of the chronic disease pathological analysis method based on big data according to an embodiment of the present application is shown.

[0064] Figure 2 An exemplary flowchart of finding the target historical patient according to an embodiment of the present application is shown.

[0065] Figure 3 An exemplary flowchart of training the disease development prediction model according to an embodiment of the present application is shown.

[0066] Figure 4 An exemplary flowchart of determining the adjusted reference medication information according to an embodiment of the present application is shown.

[0067] Figure 5 An exemplary block diagram of the chronic disease pathological analysis system based on big data according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0068] The technical solutions of the present application are described in detail below with specific examples. The following specific examples can be combined with each other, and some examples may not be described in detail for the same or similar concepts or processes.

[0069] Figure 1 An exemplary flowchart of a chronic disease pathological analysis method based on big data according to an embodiment of the present application is shown, which comprises:

[0070] Step S1, inputting multiple physiological index data, medication information and symptom description information of a patient;

[0071] Step S2, processing the symptom description information by a natural language processing model to obtain a symptom description vector;

[0072] Step S3, searching for a target historical patient in a historical database according to the symptom description vector and the physiological index data;

[0073] Step S4, training a disease development prediction model according to historical physiological index data and historical medication information of the target historical patient at multiple time points to obtain a trained disease development prediction model;

[0074] Step S5, determining whether the medication information needs to be adjusted according to the trained disease development prediction model, the medication information and the physiological index data;

[0075] Step S6, if the medication information needs to be adjusted, determining the adjusted medication information according to the trained disease development prediction model and the physiological index data.

[0076] The chronic disease pathological analysis method based on big data according to the embodiment of the present application can input multiple physiological index data, past medication information and symptom description information of a chronic disease patient, and search for historical data of a chronic disease patient similar to the patient, thereby training a disease development prediction model to determine the disease development trend of the chronic disease patient based on the trained disease development prediction model, to determine whether the medication information needs to be adjusted, and to give the adjusted reference medication information for the doctor to refer, which can accurately evaluate the disease condition of the chronic disease patient and predict the disease development, and give accurate medication suggestions, thereby reducing the workload of the doctor.

[0077] According to one embodiment of the present application, in step S1, there are many common chronic diseases, such as hypertension, diabetes, coronary heart disease, chronic hepatitis, etc., and when diagnosing and treating chronic diseases, the physiological index data of the patient needs to be checked, such as blood pressure, blood sugar, glutamic-pyruvic transaminase, glutamic-oxalacetic transaminase, uric acid, etc. In addition, the current medication information and symptom description information of the patient can also be obtained. The medication information can include the current medication type and the daily medication amount, and the symptom description information can be in the form of text description information, such as the text record of the doctor's inquiry to the patient and the text record of the patient's answer, etc.

[0078] According to one embodiment of the present application, in step S2, the symptom description information can be processed by a natural language processing model to obtain a symptom description vector. For example, the natural language processing model can filter the key content in the symptom description information, such as filtering the text in the patient's answer text record for describing the pain position and pain frequency, and obtaining the semantic vector of these texts, i.e., the symptom description vector.

[0079] According to one embodiment of the present application, in step S3, based on the symptom description vector and the physiological index data, a target historical patient similar to the above-mentioned patient can be found in the historical database, thereby providing a reference for determining the disease development trend and medication information of the above-mentioned patient.

[0080] Figure 2 An exemplary flowchart of finding a target historical patient according to an embodiment of the present application is shown.

[0081] According to one embodiment of the present application, step S3 includes: step S31, filtering a candidate historical patient from a plurality of historical patients according to the symptom description vector and the historical symptom description vectors of the plurality of historical patients in the historical database; and step S32, filtering a target historical patient from the candidate historical patient according to the physiological index data and the historical physiological index data of the candidate historical patient.

[0082] According to one embodiment of the present application, in step S31, a historical symptom description vector with high similarity to the symptom description vector can be found from the historical symptom description vectors of the plurality of historical patients at a plurality of time points, and the historical patient corresponding to the historical symptom description vector is the candidate historical patient. Each historical patient may have undergone multiple treatments at multiple time points, and the doctor can inquire the symptoms of the historical patient during each treatment, and after processing by the natural language processing model, the historical symptom description vector can be obtained. The cosine similarity of the symptom description vector and the historical symptom description vectors of each historical patient at a plurality of time points can be solved, and the historical patient corresponding to the historical symptom description vector with a cosine similarity higher than a preset similarity threshold (for example, 0.6) is determined as the candidate historical patient.

[0083] According to an embodiment of the present application, in step S32, the target historical patient can be selected from the candidate historical patients based on the physiological index data and the historical physiological index data of the candidate historical patients. For example, the multiple physiological index data can be combined into a vector, and the cosine similarity between the vector and the historical physiological index data of the candidate historical patients is calculated, and the candidate historical patient with higher cosine similarity is selected as the target historical patient.

[0084] According to an embodiment of the present application, step S32 comprises: step S321, determining the type, abnormal type and abnormal amplitude of the abnormal physiological index data in the physiological index data; step S322, determining the historical type, historical abnormal type and historical abnormal amplitude of the abnormal physiological index data in the historical physiological index data of each candidate historical patient; step S323, determining the abnormal physiological index vector of the patient according to the type, abnormal type and abnormal amplitude of the abnormal physiological index data; step S324, determining the historical abnormal physiological index vector of each candidate historical patient according to the historical type, historical abnormal type and historical abnormal amplitude of the abnormal physiological index data in the historical physiological index data of each candidate historical patient; and step S325, selecting the target historical patient from the candidate historical patients according to the abnormal physiological index vector and the historical abnormal physiological index vector.

[0085] According to an embodiment of the present application, in step S321, each physiological index has a certain normal range, if a certain physiological index data is higher than the upper limit of the normal range or lower than the lower limit of the normal range, the certain physiological index data is abnormal physiological index data, the type of the certain abnormal physiological index data can be recorded, and the abnormal type of the certain abnormal physiological index data, in other words, the reason of the certain abnormal physiological index data exceeding the upper limit of the physiological index or being lower than the lower limit of the physiological index, can be recorded. The abnormal amplitude is the amplitude of the abnormal physiological index data exceeding the upper limit of the physiological index or being lower than the lower limit of the physiological index, for example, if the abnormal type is exceeding the upper limit of the physiological index, the abnormal amplitude is calculated as the ratio of the difference between the abnormal physiological index data and the upper limit of the physiological index to the upper limit of the physiological index, and if the abnormal type is lower than the lower limit of the physiological index, the abnormal amplitude is calculated as the ratio of the difference between the lower limit of the physiological index and the abnormal physiological index data to the lower limit of the physiological index.

[0086] According to an embodiment of the present application, in step S322, the historical type of the abnormal physiological index data in the historical physiological index data of the candidate historical patient can be determined by a similar method as determining the type of the abnormal physiological index data, the historical abnormal type can be determined by a similar method as determining the abnormal type, and the historical abnormal amplitude can be determined by a similar method as determining the abnormal amplitude.

[0087] According to an embodiment of the present application, in step S323, in order to highlight the abnormal physiological indicator data, i.e., to focus on the abnormal physiological indicator data in subsequent operations, the abnormal physiological indicator vector of the patient can be determined based on the type of the abnormal physiological indicator data, the abnormal type and the abnormal amplitude, in which the abnormal type and the abnormal amplitude of the abnormal physiological indicator data can be highlighted, and the data of the abnormal physiological indicator vector can be normalized (i.e., by using the abnormal amplitude as a component of the vector) to simplify the operation. In the abnormal physiological indicator vector, the component corresponding to the type of the normal physiological indicator data is 0, the component corresponding to the type of the abnormal physiological indicator whose abnormal type is exceeding the upper limit of the physiological indicator is the abnormal amplitude, and the component corresponding to the type of the abnormal physiological indicator whose abnormal type is lower than the lower limit of the physiological indicator is the negative abnormal amplitude.

[0088] According to an embodiment of the present application, in step S324, similar to the abnormal physiological indicator vector, the historical abnormal physiological indicator vector of the candidate historical patient can be obtained. In the historical abnormal physiological indicator vector, the component corresponding to the type of the normal historical physiological indicator data is 0, the component corresponding to the type of the abnormal physiological indicator whose historical abnormal type is exceeding the upper limit of the physiological indicator is the historical abnormal amplitude, and the component corresponding to the type of the abnormal physiological indicator whose historical abnormal type is lower than the lower limit of the physiological indicator is the negative historical abnormal amplitude.

[0089] According to an embodiment of the present application, in step S325, the target historical patient can be screened based on the abnormal physiological indicator vector and the historical abnormal physiological indicator vector. Thus, the difference of the normal physiological indicator data (e.g., the difference of the normal physiological indicator caused by individual differences such as age, height, weight, etc.) can be ignored, and only the abnormal physiological indicator data is used to screen the target historical patient with similar conditions, which can improve the screening accuracy and reduce the interference of the difference of the normal physiological indicator data.

[0090] According to an embodiment of the present application, step S325 includes: step S3251, calculating the cosine similarity of the abnormal physiological indicator vector and the historical abnormal physiological indicator vector of each candidate historical patient, respectively; step S3252, sorting the historical abnormal physiological indicator vectors of each candidate historical patient according to the cosine similarity; and step S3253, determining the candidate historical patients corresponding to the historical abnormal physiological indicator vectors in the sequence obtained by sorting as the target historical patients in the front pre-set number (e.g., the front 1000).

[0091] According to one embodiment of the present application, cosine similarities of the normal physiological indicator vector and the historical abnormal physiological indicator vectors of the candidate historical patients at different time points can be calculated respectively, and sorted, and the historical abnormal physiological indicator vectors with the highest cosine similarities can be selected from the sorted sequence, and the candidate historical patients corresponding to the historical abnormal physiological indicator vectors are the target historical patients. Since the same candidate historical patient can obtain multiple historical physiological indicator data at different time points, the same candidate historical patient can have multiple historical abnormal physiological indicator vectors, and the target historical patients selected by the above method can be repeated, and thus, the target historical patients can be processed to remove the repetition.

[0092] According to one embodiment of the present application, in step S4, the disease progression prediction model is a BP neural network model, which can predict the health status of the patient at the next time point based on the current physiological indicator data and the medication information. The time interval between adjacent time points can be three months, half a year, one year, etc., and the present application does not limit the time interval between adjacent time points.

[0093] According to one embodiment of the present application, as described above, the abnormal physiological indicator vector can highlight the abnormal type and the abnormal amplitude of the abnormal physiological indicator data, can not only highlight the abnormal physiological indicator data, but also normalize the data of the abnormal physiological indicator vector, can reduce the interference of the normal physiological indicator on the abnormal condition judgment, and can also simplify the operation. Therefore, the information input into the disease progression prediction model is the abnormal physiological indicator vector and the medication information, so as to facilitate the disease progression prediction model to estimate the health status at the next time point. The output of the disease progression prediction model can be the predicted abnormal physiological indicator vector at the next time point, so as to judge whether the type and the number of the abnormal physiological indicator data are reduced and whether the abnormal amplitude is reduced based on the predicted abnormal physiological indicator vector, and thus, the health status of the patient at the next time point can be predicted.

[0094] Figure 3 An exemplary flowchart of training the disease progression prediction model according to an embodiment of the present application is shown.

[0095] According to one embodiment of the present application, step S4 comprises: step S41, obtaining a plurality of time point historical abnormal physiological indicator vectors of the target historical patient according to the historical physiological indicator data of the target historical patient at a plurality of time points; step S42, inputting the i-th time point historical abnormal physiological indicator vector and the historical medication information of the target historical patient into the disease development prediction model to obtain a predicted abnormal physiological indicator vector at an (i+1)-th time point; step S43, obtaining a loss function of the disease development prediction model according to the predicted abnormal physiological indicator vector at the (i+1)-th time point and the historical abnormal physiological indicator vector at the (i+1)-th time point; and step S44, training the disease development prediction model according to the loss function of the disease development prediction model to obtain a trained disease development prediction model.

[0096] According to one embodiment of the present application, in step S41, the historical abnormal physiological indicator vectors of the target historical patient at a plurality of time points can be obtained in a manner similar to the abnormal physiological indicator vector. The specific obtaining manner is not described herein again.

[0097] According to one embodiment of the present application, in step S42, the historical medication information can be in the form of a vector, for example, each component in the vector is the amount of a type of drug used in a day. The i-th time point historical abnormal physiological indicator vector and the historical medication information of the target historical patient are input into the disease development prediction model to obtain a predicted abnormal physiological indicator vector at an (i+1)-th time point, i.e., the health status of the target historical patient at the (i+1)-th time point predicted by the disease development prediction model. The predicted health status can be compared with the actual health status of the target historical patient at the (i+1)-th time point (i.e., the historical abnormal physiological indicator vector at the (i+1)-th time point), so as to determine the error therebetween, and then determine a loss function based on the error, so as to reduce the error in the training process through the loss function and improve the accuracy of the disease development prediction model.

[0098] According to one embodiment of the present application, step S43 comprises: obtaining the loss function LOSS of the disease development prediction model according to formula (1),

[0099]

[0100] wherein I j,i+1,p is the predicted abnormal physiological indicator vector at the (i+1)-th time point of the j-th target historical patient, I j,i+1 is the historical abnormal physiological indicator vector at the (i+1)-th time point of the j-th target historical patient, I j,iis a historical abnormal physiological indicator vector of the jth target historical patient at the i+1th time point, W is a preset weight matrix, the preset weight matrix is a diagonal matrix, and an element in the kth row and the kth column of the preset weight matrix is a preset weight of a type of physiological indicator data corresponding to a kth component in the predicted abnormal physiological indicator vector, n j is a number of historical abnormal physiological indicator vectors of the jth target historical patient, N is a number of target historical patients in each training batch, if is a conditional function, i≤n j , j≤N, and i, n j , j, and N are positive integers.

[0101] According to an embodiment of the present application, the preset weight in the preset weight matrix is a value preset by a professional such as a doctor, which can be used to represent the importance of a piece of physiological indicator data. For example, blood pressure is more important and can be assigned a higher preset weight, and uric acid is only important in a specific population (for example, patients with gout), and its weight can be lower than the preset weight of blood pressure. The preset weight matrix is in the form of a diagonal matrix, and when multiplied by the predicted abnormal physiological indicator and the historical abnormal physiological indicator vector, each component corresponding to a physiological indicator can be weighted, so that the importance of the component corresponding to the key physiological indicator can be highlighted, for example, the historical abnormal amplitude of blood pressure is weighted, so that the influence of abnormal blood pressure is highlighted.

[0102] According to an embodiment of the present application, if (|WI j,i+1 |<|WI j,i |, |WI j,i+1,p -WI j,i+1 |, |I j,i+1,p -I j,i+1 |) is a conditional function, and when |WI j,i+1 |<|WI j,i |, the value of the conditional function is |WI j,i+1,p -WI j,i+1 |, otherwise, the value of the conditional function is |I j,i+1,p -I j,i+1 |. |WI j,i+1 |<|WI j,i| indicates that the magnitude of the weighted historical abnormal physiological indicator vector at time i+1 is less than the magnitude of the weighted historical abnormal physiological indicator vector at time i. In other words, the historical abnormality amplitude of the weighted abnormal physiological indicator is reduced, or even reduced to 0, indicating that the historical medication information is accurate and the drug effect is good. The comparison between the magnitude of the weighted historical abnormal physiological indicator vector at time i+1 and the magnitude of the weighted historical abnormal physiological indicator vector at time i is to highlight the historical abnormality amplitude of key physiological indicator data, thereby reflecting changes in the historical abnormality amplitude of key physiological indicators and more objectively reflecting changes in the overall health level of historical patients. If the medication is accurate, i.e., |WI j,i+1 |<|WI j,i |, then the conditional function value is |WI j,i+1,p -WI j,i+1 | represents the error between the weighted predicted abnormal physiological indicator vector at time i+1 and the weighted historical abnormal physiological indicator vector at time i+1. By using a pre-defined weight matrix, the error in the historical abnormal magnitude of key physiological indicator data can be highlighted, thereby specifically reducing the error in the historical abnormal magnitude of key physiological indicator data during training, improving training efficiency and prediction accuracy. If |WI j,i+1 |≥|WI j,i | indicates that the medication was inaccurate, resulting in a failure to significantly reduce the abnormal amplitude of multiple key physiological indicators, or that while the abnormal amplitude of key physiological indicators decreased, the abnormal amplitude of other physiological indicators increased, causing the overall abnormal amplitude not to decrease. In this case, |I j,i+1,p -I j,i+1 As a conditional function value, if the medication failed to significantly reduce the abnormal amplitude of key physiological indicators or increased the abnormal amplitude of other physiological indicators, then not only may the abnormal amplitude of key physiological indicators change, but the abnormal amplitude of other physiological indicators may also change. Therefore, the focus shifts from the error of the abnormal amplitude of key physiological indicators to the overall error of the abnormal amplitude of all physiological indicators. The goal is to reduce the error of the abnormal amplitude of all physiological indicators during training, thereby improving the accuracy of the disease progression prediction model. On the other hand, if the medication failed to achieve effective treatment, then this historical medication information is not valuable. Therefore, there is no need for targeted training based on the impact of this medication information on key physiological indicators; that is, there is no need to improve the accuracy of the abnormal amplitude of key physiological indicators in this case through weighted methods. In other words, the unweighted |I| can be used directly. j,i+1,p -I j,i+1 | is used as the value of a conditional function.

[0103] According to one embodiment of the present application, the conditional function values corresponding to the plurality of time points of the plurality of target historical patients can be summed to obtain a loss function of the disease development prediction model. The disease development prediction model is trained in step S44 by the loss function to obtain the trained disease development prediction model. For example, the parameters of the disease development prediction model can be adjusted by the loss function through back propagation. The target historical patients can be divided into batches, and the loss function can be solved for each batch of target historical patients, and the disease development prediction model can be trained. After multiple training, the trained disease development prediction model can be obtained.

[0104] In this way, the predicted abnormal physiological indicator vector and the historical abnormal physiological indicator vector can be weighted by the preset weight matrix, so that the overall health status of the historical patient can be accurately determined, and the loss function value corresponding to the change of the different overall health status can be determined by the conditional function, so that the error of the historical abnormal amplitude of the key physiological indicator is focused on in the case of overall health improvement, so as to improve the training pertinence and training efficiency, and the error of the historical abnormal amplitude of each physiological indicator is calculated in the case of overall health improvement, so as to improve the overall training and the accuracy of the disease development prediction model.

[0105] According to one embodiment of the present application, in step S5, the current medication information and physiological indicator data of the patient can be input into the trained disease development prediction model to obtain the predicted abnormal physiological indicator vector of the next time point, and compared with the abnormal physiological indicator vector of the current time point. If the norm of the predicted abnormal physiological indicator vector of the next time point weighted by the preset weight matrix does not decrease compared with the norm of the abnormal physiological indicator vector of the current time point weighted by the preset weight matrix, or the decrease does not reach the expectation, it can be determined that the medication information needs to be adjusted.

[0106] According to one embodiment of the present application, in step S6, the medication information can be adjusted based on the trained disease development prediction model and the physiological indicator data.

[0107] Figure 4 An exemplary flowchart for determining the adjusted reference medication information according to an embodiment of the present application is shown.

[0108] According to an embodiment of the present invention, step S6 includes: step S61, obtaining an abnormal physiological index vector based on physiological index data; step S62, determining the constraints of the medication optimization model based on the abnormal physiological index vector and the trained disease progression prediction model; step S63, determining the objective function of the medication optimization model based on the trained disease progression prediction model; and step S64, solving the medication optimization model based on the constraints and the objective function to obtain adjusted reference medication information.

[0109] According to one embodiment of the present invention, in step S61, the method for determining the abnormal physiological index vector is as described above and will not be repeated here. In step S62, the medication optimization model can be an optimization model such as a nonlinear programming model or a genetic algorithm model. An objective function and constraints can be set for the medication optimization model to find the optimal solution that maximizes the satisfaction of the objective function's described objective under the constraints. Each solution represents a type of medication information, and the optimal solution that maximizes the satisfaction of the objective function's described objective is the reference medication information.

[0110] According to an embodiment of the present invention, step S62 includes: determining the constraints of the medication optimization model based on the abnormal physiological index vector and the trained disease progression prediction model, including: determining the constraints of the medication optimization model according to formulas (2), (3), and (4).

[0111] f S (I ab M p ) = I ab,next (2)

[0112]

[0113] U x ≤U x,max (4)

[0114] Among them, I ab M is a vector of abnormal physiological indicators. p This is the revised reference medication information to be determined. U1 is the daily dosage of the first drug, U2 is the daily dosage of the second drug, and U... x For the daily dosage of drug x, U n U represents the daily dosage of the nth drug. x,max I represents the upper limit of daily dosage for drug x. ab,next f is the vector of predicted abnormal physiological indicators for the patient at the next time step. S Let be the mapping function of the disease progression prediction model after training, n be the number of drug types, 1≤x≤n, and x and n are both positive integers.

[0115] According to one embodiment of the present application, formula (2) represents the operation of the mapping function of the trained disease development prediction model on the abnormal physiological index vector and the to-be-determined adjusted reference medication information, and the predicted abnormal physiological index vector at the next moment can be obtained. The to-be-determined adjusted reference medication information takes different values, and different predicted abnormal physiological index vectors at the next moment can be obtained. Formula (3) is a description of the to-be-determined adjusted reference medication information, that is, a vector composed of the daily dosage of each drug. Formula (4) represents that the daily dosage of each drug cannot exceed the upper limit of the daily dosage.

[0116] According to one embodiment of the present application, step S63 comprises determining the objective function of the medication optimization model according to the trained disease development prediction model, comprising: determining the objective function of the medication optimization model according to formula (5) and formula (6),

[0117] minimize|WI ab,next | (5)

[0118]

[0119] wherein W is a preset weight matrix, the preset weight matrix is a diagonal matrix, and the element of the kth row and the kth column in the preset weight matrix is a preset weight of the type of the kth component of the abnormal physiological index vector. minimize is a minimization function.

[0120] According to one embodiment of the present application, formula (5) is one of the objective functions, and the objective described thereby is to minimize the norm of the weighted predicted abnormal physiological index vector, that is, to reduce the number of types of key abnormal physiological indexes and to reduce the abnormal amplitude of key abnormal physiological index data to the greatest extent. Through the weighting processing, the change of the key physiological index can be focused on, so that the medication suggestion can be given in a targeted manner, and the overall medication effect can be improved.

[0121] According to one embodiment of the present application, formula (6) is also one of the objective functions, and the objective described thereby is to minimize the total daily dosage of the plurality of drugs, to improve the convenience of the patient's medication, and to reduce the probability of the patient forgetting to take the medication.

[0122] According to one embodiment of the present application, in step S64, the medication optimization model can be solved based on the above constraint conditions and the objective function, and the optimal solution obtained is the adjusted reference medication information, that is, the adjusted recommended medication method, which can be provided as a reference for professionals such as doctors.

[0123] In this way, the optimal solution of the medication method can be solved by the medication optimization model, the medication safety and the prediction accuracy can be met under the constraint condition, the overall medication effect and the medication convenience of the patient can be improved through the two objective functions, and thus the accurate and convenient recommended medication method can be obtained, and the workload of the doctors and other professionals can be reduced.

[0124] The chronic disease pathological analysis method based on big data according to the embodiment of the application can input the physiological index data, the past medication information and the symptom description information of the chronic disease patient, find the historical data of the chronic disease patient similar to the patient, train the disease development prediction model, determine the disease development trend of the chronic disease patient based on the trained disease development prediction model, determine whether the medication information needs to be adjusted, and give the adjusted reference medication information for the doctor to refer, so as to accurately evaluate the disease condition of the chronic disease patient, predict the disease development, give accurate medication suggestions, and reduce the workload of the doctors. When training the disease development prediction model, the abnormal physiological index vector and the historical abnormal physiological index vector can be weighted through the preset weight matrix, so that the change of the overall health condition of the historical patient can be accurately determined, and the loss function value corresponding to the change of the overall health condition can be determined through the condition function, so that the error of the historical abnormal amplitude of the key physiological index is focused on in the case of the improvement of the overall health condition, so as to improve the training pertinence and the training efficiency, and the error of the historical abnormal amplitude of each physiological index is calculated in the case of the unimprovement of the overall health condition, so as to improve the overall training and the accuracy of the disease development prediction model. When determining the reference medication information, the optimal solution of the medication method can be solved by the medication optimization model, the medication safety and the prediction accuracy can be met under the constraint condition, the overall medication effect and the medication convenience of the patient can be improved through the two objective functions, and thus the accurate and convenient recommended medication method can be obtained, and the workload of the doctors and other professionals can be reduced.

[0125] Figure 5 An example of a block diagram of a chronic disease pathological analysis system based on big data according to an embodiment of the application is shown, which includes:

[0126] The input module is configured to input the physiological index data, the medication information and the symptom description information of the patient.

[0127] The symptom description vector module is configured to process the symptom description information through a natural language processing model to obtain a symptom description vector.

[0128] The finding module is configured to find a target historical patient in a historical database according to the symptom description vector and the physiological index data.

[0129] The training module is configured to train the disease development prediction model according to historical physiological index data and historical medication information of the target historical patient at multiple time points, and obtain a trained disease development prediction model.

[0130] The determining module is configured to determine whether the medication information needs to be adjusted according to the trained disease development prediction model, the medication information and the physiological index data.

[0131] The reference medication information module is configured to determine adjusted reference medication information according to the trained disease development prediction model and the physiological index data if the medication information needs to be adjusted.

[0132] The present application can be a method, device, system and / or computer program product. The computer program product can include a computer readable storage medium having computer readable program instructions loaded thereon for performing the various aspects of the present application.

[0133] Those skilled in the art should understand that the embodiments of the application shown in the above description and the accompanying drawings are only examples of the application and do not limit the application. The purpose of the application has been fully and effectively achieved. The function and structural principle of the application has been shown and explained in the embodiments, and the embodiments of the application can be modified or changed in any way without departing from the principle.

[0134] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit the present application; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A big data-based chronic disease pathology analysis method, characterized in that, The method comprises the following steps: Enter the physiological index data, medication information and symptom description information of a patient; Process the symptom description information through a natural language processing model to obtain a symptom description vector; Find a target historical patient in a historical database according to the symptom description vector and the physiological index data; Train a disease development prediction model according to the historical physiological index data and historical medication information of the target historical patient at multiple time points to obtain a trained disease development prediction model; Determine whether the medication information needs to be adjusted according to the trained disease development prediction model, the medication information and the physiological index data; If the medication information needs to be adjusted, determine the adjusted reference medication information according to the trained disease development prediction model and the physiological index data.

2. The big data-based chronic disease pathology analysis method of claim 1, wherein, Finding a target historical patient in a historical database according to the symptom description vector and the physiological index data comprises the following steps: Screen a candidate historical patient from a plurality of historical patients according to the symptom description vector and the historical symptom description vectors of the plurality of historical patients in the historical database; Screen a target historical patient from the candidate historical patients according to the physiological index data and the historical physiological index data of the candidate historical patients. 3.The big data-based chronic disease pathology analysis method of claim 2, wherein, Screening a target historical patient from the candidate historical patients according to the physiological index data and the historical physiological index data of the candidate historical patients comprises the following steps: Determine the type, abnormal type and abnormal amplitude of the abnormal physiological index data in the physiological index data; Determine the historical type, historical abnormal type and historical abnormal amplitude of the abnormal physiological index data in the historical physiological index data of each candidate historical patient; Determine an abnormal physiological index vector of the patient according to the type, abnormal type and abnormal amplitude of the abnormal physiological index data; Determine a historical abnormal physiological index vector of each candidate historical patient according to the historical type, historical abnormal type and historical abnormal amplitude of the abnormal physiological index data in the historical physiological index data of each candidate historical patient; Screen a target historical patient from the candidate historical patients according to the abnormal physiological index vector and the historical abnormal physiological index vector. 4.The big data-based chronic disease pathology analysis method of claim 3, wherein, In the abnormal physiological index vector, the component corresponding to the type of normal physiological index data is 0, the component corresponding to the type of abnormal physiological index whose abnormal type is exceeding the upper limit of the physiological index is the abnormal amplitude, and the component corresponding to the type of abnormal physiological index whose abnormal type is lower than the lower limit of the physiological index is the negative abnormal amplitude; In the historical abnormal physiological index vector, the component corresponding to the type of normal historical physiological index data is 0, the component corresponding to the type of abnormal physiological index whose historical abnormal type is exceeding the upper limit of the physiological index is the historical abnormal amplitude, and the component corresponding to the type of abnormal physiological index whose historical abnormal type is lower than the lower limit of the physiological index is the negative historical abnormal amplitude; Screening a target historical patient from the candidate historical patients according to the abnormal physiological index vector and the historical abnormal physiological index vector comprises the following steps: Calculate the cosine similarity of the abnormal physiological index vector and the historical abnormal physiological index vector of each candidate historical patient, respectively; Sort the historical abnormal physiological index vectors of the candidate historical patients according to the cosine similarity; The target historical patient is determined from the candidate historical patients corresponding to the first preset number of historical abnormal physiological indicator vectors in the sorted sequence. 5.The big data-based chronic disease pathology analysis method of claim 4, wherein, The disease development prediction model is trained according to the historical physiological indicator data and the historical medication information of the target historical patient at multiple time points, and a trained disease development prediction model is obtained, including: The historical abnormal physiological indicator vectors of the target historical patient at multiple time points are obtained according to the historical physiological indicator data of the target historical patient at multiple time points. The historical abnormal physiological indicator vector of the target historical patient at the i th time point and the historical medication information are input into the disease development prediction model to obtain a predicted abnormal physiological indicator vector at the i + 1 th time point. The loss function of the disease development prediction model is obtained according to the predicted abnormal physiological indicator vector at the i + 1 th time point and the historical abnormal physiological indicator vector at the i + 1 th time point. The disease development prediction model is trained according to the loss function of the disease development prediction model, and a trained disease development prediction model is obtained. 6.The big data-based chronic disease pathology analysis method of claim 5, wherein, The loss function of the disease development prediction model is obtained according to the predicted abnormal physiological indicator vector at the i + 1 th time point and the historical abnormal physiological indicator vector at the i + 1 th time point, including: According to the formula A loss function LOSS for obtaining a disease progression prediction model, wherein, I j,i+1,p is a predicted abnormal physiological indicator vector of an (i+1)th moment of the jth target historical patient, I j,i+1 is a historical abnormal physiological indicator vector of the (i+1)th moment of the jth target historical patient, I j,i is a historical abnormal physiological indicator vector of the (i+1)th moment of the jth target historical patient, W is a preset weight matrix, the preset weight matrix is a diagonal matrix, and an element of a kth row and a kth column in the preset weight matrix is a preset weight of a type of a kth component in the predicted abnormal physiological indicator vector, n j is a number of the historical abnormal physiological indicator vectors of the jth target historical patient, N is a number of target historical patients in each training batch, if is a conditional function, i≤n j , j≤N, and i, n j , j, and N are positive integers. 7.The big data-based chronic disease pathology analysis method of claim 1, wherein, If the medication information needs to be adjusted, the adjusted reference medication information is determined according to the trained disease development prediction model and the physiological indicator data, including: The abnormal physiological indicator vector is obtained according to the physiological indicator data. The constraint condition of the medication optimization model is determined according to the abnormal physiological indicator vector and the trained disease development prediction model. The objective function of the medication optimization model is determined according to the trained disease development prediction model. The medication optimization model is solved according to the constraint condition and the objective function to obtain the adjusted reference medication information. 8.The big data-based chronic disease pathology analysis method of claim 7, wherein, The constraint condition of the medication optimization model is determined according to the abnormal physiological indicator vector and the trained disease development prediction model, including: According to the formula f S (I ab ,M p )=I ab,next U x ≤u x,max determining the constraint condition of the medication optimization model, wherein, I ab is an abnormal physiological index vector, M p is the adjusted reference medication information to be determined, U1 is the daily dosage of the first drug, U2 is the daily dosage of the second drug, U x is the daily dosage of the xth drug, U n is the daily dosage of the nth drug, U x,max is the upper limit of the daily dosage of the xth drug, I ab,next is the predicted abnormal physiological index vector of the patient at the next time, f S is the mapping function of the trained disease development prediction model, n is the number of drug types, 1≤x≤n, and x and n are positive integers. 9.The big data-based chronic disease pathology analysis method of claim 8, wherein, The objective function of the medication optimization model is determined according to the trained disease development prediction model, including: According to the formula minimize | WI ab,next | The objective function of the medication optimization model is determined, wherein W is a preset weight matrix, the preset weight matrix is a diagonal matrix, and the element of the k th row and the k th column in the preset weight matrix is a preset weight of the type of the k th component in the abnormal physiological indicator vector corresponding to the physiological indicator data, and minimize is a minimization function. 10.A big data-based chronic disease pathology analysis system, characterized in that, It includes: The input module is used for inputting multiple physiological indicator data, medication information and symptom description information of the patient; The symptom description vector module is used for processing the symptom description information through a natural language processing model to obtain a symptom description vector; The search module is used for searching the target historical patient in the historical database according to the symptom description vector and the physiological indicator data; The training module is used for training the disease development prediction model according to the historical physiological indicator data and the historical medication information of the target historical patient at multiple time points, and obtaining a trained disease development prediction model; The judgment module is used for determining whether the medication information needs to be adjusted according to the trained disease development prediction model, the medication information and the physiological indicator data. With reference to the medication information module, if it is necessary to adjust the medication information, the adjusted reference medication information is determined according to the trained disease development prediction model and the physiological index data.

Images