A method and system for intelligently generating doctor portraits for full course of disease management

Abstract

The present invention discloses a method and system for intelligently generating doctor portraits for whole course management, which relates to the technical field of medical management. The method obtains clinical data, academic achievement data and patient treatment records corresponding to a number of doctors and encapsulates and generates corresponding data sets respectively. The data sets corresponding to the several doctors are cleaned and integrated to generate corresponding normalized data sets. Feature analysis and extraction are performed on the normalized data sets to obtain personalized diagnosis and treatment feature information corresponding to each doctor. A corresponding doctor portrait model is constructed according to the personalized diagnosis and treatment feature information corresponding to the several doctors, and personalized course description information corresponding to the doctor portrait model is generated. The treatment course of the current patient is divided into a number of sub-process courses, and the patient treatment-related data of each sub-process course and the several personalized course description information are matched in sequence, so as to match the corresponding doctor for each sub-process course.

Description

Technical Field

[0001] The present invention relates to the field of medical management technology, and in particular to a method and system for intelligently generating doctor portraits for full course of disease management. Background Art

[0002] Whole-course management refers to the coordinated medical services and support provided by medical institutions, medical personnel and related service providers to patients throughout the entire process from initial disease discovery to rehabilitation or terminal care. Its purpose is to improve medical quality, reduce medical costs, enhance patient satisfaction, and promote the improvement of patients' health and quality of life by integrating medical services and resources at all stages. Whole-course management emphasizes cross-disciplinary, cross-institutional and cross-stage collaboration and coordination, involving services in diagnosis, treatment, rehabilitation, follow-up, health education, psychological support, etc., aiming to achieve all-round care and support for patients.

[0003] In the traditional process of diagnosing a disease, the entire course of treatment is often handled by a single doctor. Different doctors are only good at handling a certain stage of the treatment. How to reasonably divide the patient's treatment course and match the most suitable doctor to each stage is a difficult problem that needs to be solved in the current traditional disease diagnosis process. Summary of the invention

[0004] In order to solve the above problems, the purpose of the present invention is to provide a method and system for intelligently generating doctor portraits for full-course disease management.

[0005] The purpose of the present invention can be achieved by the following technical solution: A method for intelligently generating a doctor's portrait for full course of disease management, comprising the following steps:

[0006] Step S1: Acquire clinical data, academic achievement data, and patient treatment records corresponding to several doctors, and encapsulate and generate data sets corresponding to several doctors;

[0007] Step S2: performing data cleaning and data integration on the data sets corresponding to the several doctors, and generating corresponding several normalized data sets, performing feature analysis and extraction on the several normalized data sets, and then obtaining the personalized diagnosis and treatment feature information corresponding to each doctor;

[0008] Step S3: construct a corresponding doctor portrait model according to the personalized diagnosis and treatment feature information corresponding to a number of doctors, and generate corresponding personalized course of disease description information according to the doctor portrait model of each doctor;

[0009] Step S4: Divide the current patient's treatment course into several sub-process courses, match them in sequence according to the patient treatment-related data of each sub-process course and the personalized course description information corresponding to several doctor portrait models, and then match the corresponding doctor for each sub-process course.

[0010] Furthermore, the process of obtaining clinical data, academic achievement data, and patient treatment records corresponding to several doctors and encapsulating and generating data sets corresponding to several doctors includes:

[0011] Several doctors are numbered and recorded as i, i = 1, 2, 3, ..., n, where n is a natural number greater than 0, and the clinical data, academic achievement data and patient treatment records corresponding to the several doctors are obtained, and the clinical data, academic achievement data and patient treatment records corresponding to the doctor numbered i are recorded as D1[i], D2[i] and D3[i] respectively. Starting from i = 1, the D1[i], D2[i] and D3[i] corresponding to the doctor numbered i are encapsulated in turn until the encapsulation of D1[i], D2[i] and D3[i] corresponding to the doctor i = n is completed, and then the corresponding data sets of several doctors are encapsulated and generated;

[0012] The data set corresponding to the doctor numbered i is denoted as data[i], and data[i] = {D1[i], D2[i], D3[i]}, where the clinical data D1[i] includes all clinical operation records corresponding to the doctor numbered i, the academic achievement data D2[i] includes all academic journals, papers and conference reports published by the doctor numbered i, and the patient diagnosis and treatment records D3[i] include the diagnosis and treatment records of all patients consulted by the doctor numbered i.

[0013] Furthermore, the process of cleaning and integrating the data sets corresponding to several doctors and generating several corresponding normalized data sets includes:

[0014] Setting a first operation node and a second operation node, wherein the first operation node is used to clean data sets corresponding to a plurality of doctors, and the second operation node is used to integrate data for a plurality of cleaned data sets;

[0015] The first operation node is used to obtain the node processing values ​​of D1[i] and D3[i] in each doctor's corresponding data set, and they are recorded as p and q respectively. The cleaning thresholds corresponding to the node processing values ​​p and q are set, and they are recorded as μ and τ respectively. All clinical data with p≤μ and all patient diagnosis and treatment records with q≤τ are marked as useless data, and all useless data are eliminated. All clinical data with p>μ are marked as reference data 1, and all patient diagnosis and treatment records with q>τ are marked as reference data 2. The reference data 1 and reference data 2 generated after data cleaning of each data set are summarized as the cleaning data set corresponding to each doctor.

[0016] The second operation node obtains the cleaned data set corresponding to each doctor, and obtains the academic author signature and document code in the academic achievement data corresponding to each doctor, thereby forming a collective authentication identifier for each doctor, converting the data formats of the reference data one and reference data two corresponding to each cleaned data set into a preset standard format, and assigning different reference value weights to the reference data one and reference data two, and recording the corresponding reference value weights as λ1 and λ2 respectively, thereby integrating the academic achievement data, reference data one and reference data two corresponding to each doctor to generate a normalized data set corresponding to each doctor.

[0017] Furthermore, the process of performing feature analysis and extraction on several normalized data sets to obtain the personalized diagnosis and treatment feature information corresponding to each doctor includes:

[0018] Obtain the reference value coefficient of each doctor's corresponding normalized data set and record it as γ, where γ = p*λ1 + q*λ2. Use machine learning technology to build a feature analysis program, and let the programmer write the information value interval and feature extraction field corresponding to the feature analysis program. The information value interval includes a high reference value interval, a general reference value interval, and a low reference value interval, which are marked as Ω1, Ω2, and Ω3 respectively;

[0019] Determine the subordinate relationship between γ and Ω1, Ω2 and Ω3 respectively, and then associate different diagnosis and treatment reference levels with the normalized data set corresponding to γ. The diagnosis and treatment reference levels include level 1, level 2 and level 3, and level 1>level 2>level 3;

[0020] When γ∈Ω1, the diagnosis and treatment reference level associated with the corresponding normalized data set is level one;

[0021] When γ∈Ω2, the diagnosis and treatment reference level associated with the corresponding normalized data set is level 2;

[0022] When γ∈Ω3, the diagnosis and treatment reference level associated with the corresponding normalized data set is level three;

[0023] When the normalized data sets corresponding to different diagnosis and treatment reference levels are input into the feature analysis program, the feature analysis program processes each normalized data set into several data fields through semantic segmentation in the order of diagnosis and treatment reference levels from high to low, and matches the several data fields corresponding to each normalized data set with the feature extraction fields, marks the data fields that match the feature extraction fields as valid fields, and marks the data fields that do not match the feature extraction fields as invalid fields, and then summarizes all the valid fields corresponding to each normalized data set as the personalized diagnosis and treatment feature information corresponding to each associated doctor.

[0024] Furthermore, the process of constructing a corresponding doctor portrait model according to the personalized diagnosis and treatment characteristic information corresponding to a number of doctors includes:

[0025] The personalized diagnosis and treatment characteristic information corresponding to each doctor includes the disease type, disease diagnosis accuracy, disease treatment effect and patient satisfaction. The disease type corresponding to each doctor is used as the distinguishing mark of their respective portraits;

[0026] Constructing the initial model architecture of the doctor portrait model corresponding to each doctor through the convolutional neural network technology, and taking the disease diagnosis accuracy, disease treatment effect and patient satisfaction as the first modeling feature item, the second modeling feature item and the third modeling feature item of the initial model architecture corresponding to each doctor, and constructing the corresponding first activation function, second activation function and third activation function for the first modeling feature item, the second modeling feature item and the third modeling feature item respectively;

[0027] Set the vector activation thresholds corresponding to the first activation function, the second activation function and the third activation function, and record them as JH1, JH2 and JH3 respectively, obtain the effect coefficient of the disease treatment effect, and use the numerical value of the disease diagnosis accuracy, the effect coefficient and the numerical value of the patient's satisfaction with the treatment as the judgment values ​​corresponding to the first modeling feature item, the second modeling feature item and the third modeling feature item respectively, and record their respective judgment values ​​as PD1, PD2 and PD3 respectively;

[0028] When PD1≥JH1, a first modeling vector corresponding to the first modeling feature item is generated by a first activation function, and a first image layer is constructed according to the first modeling vector. When PD1<JH1, the first activation function is not operated.

[0029] When PD2<JH2, a second modeling vector corresponding to the second modeling feature item is generated by a second activation function, and a second image layer is constructed according to the second modeling vector. When PD2≥JH2, the second activation function is not operated.

[0030] When PD3≥JH3, a third modeling vector corresponding to the third modeling feature item is generated by a third activation function, and a third image layer is constructed according to the third modeling vector. When PD3<JH3, the third activation function is not operated.

[0031] When the first portrait layer, the second portrait layer and the third portrait layer all exist, the first portrait layer, the second portrait layer and the third portrait layer are spliced ​​and fused from top to bottom in sequence to generate a doctor portrait model corresponding to each doctor. When any of the first portrait layer, the second portrait layer and the third portrait layer does not exist, the doctor portrait model of the corresponding doctor is not constructed.

[0032] Furthermore, the process of generating corresponding personalized disease course description information according to the doctor portrait model of each doctor includes:

[0033] Obtain the doctor portrait model corresponding to each doctor, and obtain the model measurement parameters corresponding to the doctor portrait model. The model measurement parameters include model generalization rate, model prediction rate and model fit, which are marked as η1, η2 and η3 respectively. Set the expected target values ​​corresponding to the model generalization rate, model prediction rate and model fit respectively, and record them as ρ1, ρ2 and ρ3 respectively.

[0034] When η1≥ρ1, η2≥ρ2 and η3≥ρ3, the personalized disease course description information is directly generated through the doctor portrait model corresponding to each doctor. When any one of the conditions η1≥ρ1, η2≥ρ2 or η3≥ρ3 is not met, the doctor portrait model is trained until η1≥ρ1, η2≥ρ2 and η3≥ρ3 are met at the same time, and then the personalized disease course description information is generated through the doctor portrait model corresponding to each doctor.

[0035] Furthermore, the process of dividing the current patient's treatment course into several sub-processes includes:

[0036] Obtain the treatment course corresponding to the current patient, and divide the treatment course into several different types of sub-process courses, the types of sub-process courses include diagnosis and evaluation sub-process courses, treatment plan formulation sub-process courses, treatment plan implementation sub-process courses, follow-up management sub-process courses and rehabilitation intervention sub-process courses;

[0037] Each patient's different sub-processes correspond to corresponding patient treatment-related data, and the patient treatment-related data includes diagnostic assessment data, treatment plan, implementation operation data after the treatment plan is implemented, follow-up data, and rehabilitation guidance data;

[0038] Among them, the diagnosis and evaluation sub-process course corresponds to the diagnosis and evaluation data, the treatment plan formulation sub-process course corresponds to the treatment plan, the treatment plan implementation sub-process course corresponds to the implementation operation data after the implementation of the treatment plan, the follow-up management sub-process course corresponds to the follow-up data, and the rehabilitation intervention sub-process course corresponds to the rehabilitation guidance data.

[0039] Furthermore, the process of matching the corresponding doctor for each sub-process course of disease according to the patient treatment-related data of each sub-process course of disease and the personalized course of disease description information corresponding to several doctor portrait models is performed in sequence, and then the corresponding doctor is matched for each sub-process course of disease, including:

[0040] A portrait matching period is set. During the portrait matching period, the patient treatment-related data corresponding to the sub-process disease course of each patient are matched with the personalized disease course description information corresponding to several doctor portrait models in sequence, and then the several sub-process disease courses corresponding to each patient are matched to the corresponding doctors. The doctors corresponding to all the sub-process disease courses matched to each patient are merged, and then a personalized treatment doctor team corresponding to each patient is established. The personalized treatment doctor team corresponding to each patient will hand over the operations of the corresponding sub-process disease course to the corresponding doctors of each sub-process disease course in the execution order of the sub-process disease course.

[0041] Furthermore, a doctor portrait intelligent generation system for full course of disease management includes:

[0042] The data collection module is used to obtain the clinical data, academic achievement data and patient treatment records corresponding to several doctors, and encapsulate and generate data sets corresponding to several doctors;

[0043] The data processing module is used to clean and integrate the data sets corresponding to several doctors, generate several corresponding normalized data sets, perform feature analysis and extraction on the several normalized data sets, and then obtain the personalized diagnosis and treatment feature information corresponding to each doctor;

[0044] A portrait construction module is used to construct a corresponding doctor portrait model based on the personalized diagnosis and treatment feature information corresponding to a number of doctors, and generate corresponding personalized course of disease description information based on the doctor portrait model of each doctor;

[0045] The disease course matching module is used to divide the current patient's treatment course into several sub-process courses, and match them in sequence according to the patient treatment-related data of each sub-process course and the personalized disease course description information corresponding to several doctor portrait models, and then match the corresponding doctor for each sub-process course.

[0046] Compared with the prior art, the beneficial effects of the present invention are as follows: by obtaining clinical data, academic achievement data and patient treatment records corresponding to several doctors, and encapsulating them into data sets corresponding to each doctor, performing data cleaning and data integration on the data sets corresponding to each doctor to generate a corresponding normalized data set, and then performing feature analysis and extraction on the data sets to obtain personalized diagnosis and treatment feature information corresponding to each doctor, building a corresponding doctor portrait model according to the personalized diagnosis and treatment feature information, and generating corresponding personalized course of disease description information according to the doctor portrait model of each doctor, dividing the treatment course of the current patient into several sub-process courses, matching the patient treatment-related data of each sub-process course and the personalized course of disease description information corresponding to several doctor portrait models in sequence, and then matching the corresponding doctor for each sub-process course. On the one hand, encapsulating, cleaning and integrating clinical data, academic achievement data and patient treatment records provides more dimensional data for building the corresponding doctor portrait model. On the other hand, dividing the entire treatment course of the current patient into several sub-process courses, matching each sub-process course with a corresponding appropriate doctor, to a certain extent makes each treatment stage of the current patient more reasonable, which is helpful for the diagnosis and treatment of the patient's disease and post-illness rehabilitation. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0048] like Figure 1 As shown, a method for intelligently generating a doctor portrait for full course of disease management includes the following steps:

[0049] Step S1: Acquire clinical data, academic achievement data, and patient treatment records corresponding to several doctors, and encapsulate and generate data sets corresponding to several doctors;

[0050] Step S2: performing data cleaning and data integration on the data sets corresponding to the several doctors, and generating corresponding several normalized data sets, performing feature analysis and extraction on the several normalized data sets, and then obtaining the personalized diagnosis and treatment feature information corresponding to each doctor;

[0051] Step S3: construct a corresponding doctor portrait model according to the personalized diagnosis and treatment feature information corresponding to a number of doctors, and generate corresponding personalized course of disease description information according to the doctor portrait model of each doctor;

[0052] Step S4: Divide the current patient's treatment course into several sub-process courses, match them in sequence according to the patient treatment-related data of each sub-process course and the personalized course description information corresponding to several doctor portrait models, and then match the corresponding doctor for each sub-process course.

[0053] It should be further explained that the process of obtaining the clinical data, academic achievement data and patient treatment records corresponding to several doctors and encapsulating and generating the data sets corresponding to several doctors includes:

[0054] Number several doctors and record the number as i, then i = 1, 2, 3, ..., n, where n is a natural number greater than 0, set the data collection period, record the data collection period as T 采集 , then T 采集 =[t1, t2], where t1 is the start time of the data collection period and t2 is the end time of the data collection period;

[0055] At the start time corresponding to t1, obtain the clinical data, academic achievement data and patient diagnosis and treatment records corresponding to several doctors, and record the clinical data, academic achievement data and patient diagnosis and treatment records corresponding to the doctor numbered i as D1[i], D2[i] and D3[i] respectively;

[0056] At the end time corresponding to t2, starting from i=1, the clinical data D1[i], academic achievement data D2[i] and patient treatment records D3[i] corresponding to the doctor numbered i are encapsulated in turn until the encapsulation of D1[i], D2[i] and D3[i] corresponding to the doctor i=n is completed, and then the data sets corresponding to several doctors are encapsulated and generated;

[0057] The data set corresponding to the doctor numbered i is recorded as data[i], then data[i] = {D1[i], D2[i], D3[i]}, where the clinical data D1[i] includes all clinical operation records corresponding to the doctor numbered i, the academic achievement data D2[i] includes all academic journals, papers and conference reports published by the doctor numbered i, and the patient treatment records D3[i] include the treatment records corresponding to all patients consulted by the doctor numbered i, and the treatment records include medical history information, drug prescription information, surgery information and medical advice information.

[0058] It should be further explained that the process of cleaning and integrating the data sets corresponding to several doctors and generating several corresponding normalized data sets includes:

[0059] A first operation node and a second operation node are set, wherein the first operation node is used to perform data cleaning on data sets corresponding to a plurality of doctors, thereby generating a cleaned data set corresponding to each doctor, and the second operation node is used to perform data integration on the plurality of cleaned data sets, thereby generating a normalized data set corresponding to each doctor;

[0060] The node processing values ​​of D1[i] and D3[i] in each doctor's corresponding data set are obtained by the first operation node processing, and are recorded as p and q respectively, where the node processing value p represents the operation success rate of all clinical operation records in the clinical data corresponding to D1[i], and the node processing value q represents the cure rate of several diagnosis and treatment records in the patient treatment records corresponding to D3[i] that successfully cure the corresponding patients;

[0061] Where p = X / X 总 , q=Y / Y 总 , X 总 represents the number of all clinical operations, X represents the number of clinical operations successfully performed, and Y 总 represents the number of all medical records, and Y represents the number of medical records that successfully cured the corresponding patients;

[0062] Set the cleaning thresholds corresponding to the node processing values ​​p and q, and record them as μ and τ respectively. Mark all clinical data with p≤μ and all patient diagnosis and treatment records with q≤τ as useless data, and remove all useless data. Mark all clinical data with p>μ as reference data 1, and all patient diagnosis and treatment records with q>τ as reference data 2. Summarize the reference data 1 and reference data 2 generated after data cleaning for each data set, and then use them as the cleaning data set corresponding to each doctor.

[0063] The second operation node obtains the cleaned data set corresponding to each doctor, and obtains the academic author signature and document code in the academic achievement data corresponding to each doctor, thereby forming a unique set authentication identifier corresponding to each doctor;

[0064] The data formats of reference data one and reference data two corresponding to each cleaned data set are converted into the preset standard format, and different reference value weights are assigned to reference data one and reference data two. The reference value weight of reference data one is denoted as λ1, and the reference value weight of reference data two is denoted as λ2. Then, the academic achievement data, reference data one and reference data two corresponding to each doctor are integrated to generate the corresponding normalized data set for each doctor.

[0065] It should be further explained that the process of performing feature analysis and extraction on several normalized data sets to obtain the personalized diagnosis and treatment feature information corresponding to each doctor includes:

[0066] Obtain the reference value coefficient of each doctor's corresponding normalized data set, and record the reference value coefficient as γ, then γ=p*λ1+q*λ2, construct a feature analysis program through machine learning technology, and use the feature analysis program to perform feature analysis and extraction on several normalized data sets;

[0067] The programmer writes the information value interval and feature extraction field corresponding to the feature analysis program, wherein the information value interval includes a high reference value interval, a general reference value interval, and a low reference value interval, which are marked as Ω1, Ω2, and Ω3, respectively. The feature extraction field is used to extract personalized feature information from each normalized data set;

[0068] Among them, Ω1>Ω2>Ω3, that is, the interval minimum value of Ω1 is equal to the interval maximum value of Ω2, and the interval minimum value of Ω2 is equal to the interval maximum value of Ω3. For example: Ω1=[b, a], Ω2=[c, b), Ω3=[0, c), 0<c<b<a;

[0069] Determine the subordinate relationship between γ and Ω1, Ω2 and Ω3 respectively, and then associate different diagnosis and treatment reference levels with the normalized data set corresponding to γ, wherein the diagnosis and treatment reference levels include level one, level two and level three, and level one>level two>level three;

[0070] When γ∈Ω1, the diagnosis and treatment reference level associated with the corresponding normalized data set is level one;

[0071] When γ∈Ω2, the diagnosis and treatment reference level associated with the corresponding normalized data set is level 2;

[0072] When γ∈Ω3, the diagnosis and treatment reference level associated with the corresponding normalized data set is level three;

[0073] When the normalized data sets corresponding to different diagnosis and treatment reference levels are input into the feature analysis program, the feature analysis program processes each normalized data set into several data fields through semantic segmentation in the order of diagnosis and treatment reference levels from high to low, and matches several data fields corresponding to each normalized data set with the feature extraction fields, and marks the data fields that match the feature extraction fields as valid fields, and marks the data fields that do not match the feature extraction fields as invalid fields;

[0074] Then, all valid fields corresponding to each normalized data set are summarized as the personalized diagnosis and treatment feature information corresponding to each associated doctor, all invalid fields corresponding to each normalized data set are encapsulated into a preset recovery file, and the set authentication identifier corresponding to each doctor is marked in the corresponding recovery file.

[0075] It should be noted that the feature extraction field is a pre-defined personalized diagnosis and treatment data for the diagnosis of the disease, which is used to measure the treatment style and treatment characteristics of different doctors when treating the disease. Different groups of doctors can then be preliminarily matched through the feature extraction field. When the data field in the normalized data set corresponding to the doctor is converted to characters and any feature extraction field is converted to characters, if the character position and number of characters are all the same, then the data field is consistent with the feature extraction field; otherwise, the data field is inconsistent with the feature extraction field.

[0076] It should be further explained that the process of constructing a corresponding doctor portrait model based on the personalized diagnosis and treatment feature information corresponding to a number of doctors includes:

[0077] The personalized diagnosis and treatment characteristic information corresponding to each doctor includes the disease type, disease diagnosis accuracy, disease treatment effect and patient satisfaction. The disease type corresponding to each doctor is used as the distinguishing mark of their respective portraits;

[0078] Constructing the initial model architecture of the doctor portrait model corresponding to each doctor through the convolutional neural network technology, and taking the disease diagnosis accuracy, disease treatment effect and patient satisfaction as the first modeling feature item, the second modeling feature item and the third modeling feature item of the initial model architecture corresponding to each doctor, constructing a first activation function for the first modeling feature item, constructing a second activation function for the second modeling feature item, and constructing a third activation function for the third modeling feature item;

[0079] The vector activation thresholds corresponding to the first activation function, the second activation function and the third activation function are set, and are respectively recorded as JH1, JH2 and JH3. The numerical value of the disease diagnosis accuracy is used as the judgment value of the first modeling feature item, and is recorded as PD1. The effect coefficient corresponding to the disease treatment effect is used as the judgment value of the second modeling feature item, and is recorded as PD2. The numerical value of the patient's satisfaction with the treatment is used as the judgment value of the third modeling feature item, and is recorded as PD3.

[0080] The effect coefficient corresponding to the disease treatment effect is generated as follows: PD2 = (α*β) / (α+β), where α represents the recovery period of the corresponding disease of the patient diagnosed and treated by the doctor, and β represents the number of postoperative recurrences of the disease. It should be noted that the smaller the value of the effect coefficient PD2, the better the treatment effect on the corresponding disease;

[0081] When PD1≥JH1, it means that the current doctor's disease diagnosis accuracy for the type of disease he is good at is qualified, and the first modeling vector corresponding to the first modeling feature item is generated through the first activation function, and the first portrait layer is constructed according to the first modeling vector. When PD1<JH1, it means that the current doctor's disease diagnosis accuracy for the type of disease he is good at is unqualified, and the first activation function is not operated;

[0082] When PD2<JH2, it means that the current doctor's treatment effect on the disease type that he is good at is qualified, and the second modeling vector corresponding to the second modeling feature item is generated through the second activation function, and the second portrait layer is constructed according to the second modeling vector. When PD2≥JH2, it means that the current doctor's treatment effect on the disease type that he is good at is unqualified, and the second activation function is not operated;

[0083] When PD3≥JH3, it means that the current doctor's satisfaction with the patients of the disease type he / she is good at is qualified, and the third modeling vector corresponding to the third modeling feature item is generated through the third activation function, and the third portrait layer is constructed according to the third modeling vector. When PD3<JH3, it means that the current doctor's satisfaction with the patients of the disease type he / she is good at is unqualified, and the third activation function is not operated;

[0084] When the first portrait layer, the second portrait layer and the third portrait layer all exist, the first portrait layer, the second portrait layer and the third portrait layer are spliced ​​and fused in sequence from top to bottom, thereby generating a doctor portrait model corresponding to each doctor;

[0085] When any of the first portrait layer, the second portrait layer and the third portrait layer does not exist, the doctor portrait model of the corresponding doctor will not be constructed.

[0086] It should be noted that JH1 is the minimum value that satisfies the qualified accuracy of disease diagnosis, JH2 is the limited value that achieves the expected disease treatment effect, that is, when the effect coefficient is less than the limited value JH2, it is judged to have achieved the ideal treatment effect for the current disease, and JH3 is the minimum value that satisfies the qualified patient satisfaction.

[0087] It should be further explained that the process of generating corresponding personalized disease description information based on the doctor portrait model of each doctor includes:

[0088] Obtain a doctor portrait model corresponding to each doctor, and obtain model measurement parameters corresponding to the doctor portrait model, wherein the model measurement parameters include model generalization rate, model prediction rate, and model fit, and are marked as η1, η2, and η3 respectively;

[0089] Set the expected target values ​​corresponding to the model generalization rate, model prediction rate and model fit, and record them as ρ1, ρ2 and ρ3 respectively. When η1≥ρ1, η2≥ρ2 and η3≥ρ3, the corresponding personalized course of disease description information is directly generated through the doctor portrait model corresponding to each doctor. When any one of the conditions η1≥ρ1, η2≥ρ2 or η3≥ρ3 is not met, the doctor portrait model is trained until η1≥ρ1, η2≥ρ2 and η3≥ρ3 are met at the same time, and then the corresponding personalized course of disease description information is generated through the doctor portrait model corresponding to each doctor.

[0090] The personalized disease course description information is used to characterize the disease course stage in which each doctor specializes in the patient's corresponding treatment course, that is, the doctor corresponding to each doctor portrait model has the best treatment effect on a certain disease course stage.

[0091] It should be further explained that the process of dividing the current patient's treatment course into several sub-process courses includes:

[0092] Obtain the treatment course corresponding to the current patient, and divide the treatment course into several different types of sub-process courses, the types of sub-process courses include diagnosis and evaluation sub-process courses, treatment plan formulation sub-process courses, treatment plan implementation sub-process courses, follow-up management sub-process courses and rehabilitation intervention sub-process courses;

[0093] Each patient's different sub-processes correspond to corresponding patient treatment-related data, and the patient treatment-related data includes diagnostic assessment data, treatment plan, implementation operation data after the treatment plan is implemented, follow-up data, and rehabilitation guidance data;

[0094] Among them, the diagnosis and evaluation sub-process course corresponds to the diagnosis and evaluation data, the treatment plan formulation sub-process course corresponds to the treatment plan, the treatment plan implementation sub-process course corresponds to the implementation operation data after the implementation of the treatment plan, the follow-up management sub-process course corresponds to the follow-up data, and the rehabilitation intervention sub-process course corresponds to the rehabilitation guidance data.

[0095] It should be noted that the diagnostic assessment data corresponding to the course of the diagnosis and assessment sub-process include the patient's medical history information, examination reports and clinical diagnosis information. The examination reports include blood test reports, urine test reports and imaging reports, etc. The clinical diagnosis information is used to determine the type and severity of the disease; the treatment plans corresponding to the treatment plan formulation sub-process course of the disease include surgical treatment and drug treatment, as well as the risk factors corresponding to the treatment plans; the implementation operation data after the implementation of the treatment plan include drug use records, surgical records and patient observation records; follow-up data include the follow-up time, frequency and content of the patient, so as to regularly monitor the progression of the patient's disease and the treatment effect. The follow-up content is used to determine whether the patient is likely to have complications; rehabilitation guidance data is used to prevent and intervene in possible complications after treatment, minimize the impact of complications, and formulate corresponding rehabilitation guidance plans for patients to help them recover their health faster.

[0096] It should be further explained that the process of matching the corresponding doctor for each sub-process disease course is carried out in sequence according to the patient treatment-related data of each sub-process disease course and the personalized disease course description information corresponding to several doctor portrait models, and then matching the corresponding doctor for each sub-process disease course includes:

[0097] Set a portrait matching period, and within the portrait matching period, sequentially match the patient treatment-related data corresponding to the sub-process disease course of each patient with the personalized disease course description information corresponding to several doctor portrait models;

[0098] The treatment-related data of each patient's different sub-processes of the disease course are sequentially input into the pre-set bag-of-words model, and then disassembled into a number of corresponding matching word segments 1; the personalized disease course description information corresponding to each doctor's portrait model is sequentially input into the bag-of-words model, and then disassembled into a number of matching word segments 2 corresponding to the personalized disease course description information of each doctor;

[0099] Get the overlap rate between matching segment 1 and matching segment 2, and record the overlap rate as R, then R = Num1 / Num2, where Num1 is the number of successful matches between matching segment 1 and matching segment 2, and Num2 is the number of matching segment 2. Set the matching threshold and record it as R 阈 , when R ≥ R 阈 When R < R 阈 , no matching is performed;

[0100] Repeat the above operation to match the corresponding doctors for several sub-processes corresponding to each patient, merge the doctors corresponding to all sub-processes matched to each patient, and then establish a personalized treatment doctor team for each patient. The personalized treatment doctor team corresponding to each patient will hand over the operations of the corresponding sub-process to the doctors corresponding to each sub-process in the order of execution of the sub-process, and finally complete the full course management for each patient.

[0101] It should be noted that the successful matching of the matching word segment one and the matching word segment two is explained as follows: the matching word segment one and the matching word segment two are converted into strings. If the number and position of each character included in the corresponding strings of the matching word segment one and the matching word segment two are equal, the matching is successful, otherwise, the matching fails.

[0102] The present invention also provides a doctor portrait intelligent generation system for full course of disease management, the system comprising:

[0103] The data collection module is used to obtain the clinical data, academic achievement data and patient treatment records corresponding to several doctors, and encapsulate and generate data sets corresponding to several doctors;

[0104] The data processing module is used to clean and integrate the data sets corresponding to several doctors, generate several corresponding normalized data sets, perform feature analysis and extraction on the several normalized data sets, and then obtain the personalized diagnosis and treatment feature information corresponding to each doctor;

[0105] A portrait construction module is used to construct a corresponding doctor portrait model based on the personalized diagnosis and treatment feature information corresponding to a number of doctors, and generate corresponding personalized course of disease description information based on the doctor portrait model of each doctor;

[0106] The disease course matching module is used to divide the current patient's treatment course into several sub-process courses, and match them in sequence according to the patient treatment-related data of each sub-process course and the personalized disease course description information corresponding to several doctor portrait models, and then match the corresponding doctor for each sub-process course.

[0107] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A method for intelligently generating doctor portraits for full course of disease management, characterized in that: The following steps are involved: Step S1: Acquire clinical data, academic achievement data, and patient treatment records corresponding to several doctors, and encapsulate and generate data sets corresponding to several doctors; Step S2: performing data cleaning and data integration on the data sets corresponding to the several doctors, and generating corresponding several normalized data sets, performing feature analysis and extraction on the several normalized data sets, and then obtaining the personalized diagnosis and treatment feature information corresponding to each doctor; Step S3: construct a corresponding doctor portrait model according to the personalized diagnosis and treatment feature information corresponding to a number of doctors, and generate corresponding personalized course of disease description information according to the doctor portrait model of each doctor; Step S4: Divide the current patient's treatment course into several sub-process courses, match them in sequence according to the patient treatment-related data of each sub-process course and the personalized course description information corresponding to several doctor portrait models, and then match the corresponding doctor for each sub-process course; The process of building a corresponding doctor portrait model based on the personalized diagnosis and treatment feature information corresponding to several doctors includes: The personalized diagnosis and treatment characteristic information corresponding to each doctor includes the disease type, disease diagnosis accuracy, disease treatment effect and patient satisfaction. The disease type corresponding to each doctor is used as the distinguishing mark of their respective portraits; Constructing the initial model architecture of the doctor portrait model corresponding to each doctor through the convolutional neural network technology, and taking the disease diagnosis accuracy, disease treatment effect and patient satisfaction as the first modeling feature item, the second modeling feature item and the third modeling feature item of the initial model architecture corresponding to each doctor, and constructing the corresponding first activation function, second activation function and third activation function for the first modeling feature item, the second modeling feature item and the third modeling feature item respectively; Set the vector activation thresholds corresponding to the first activation function, the second activation function and the third activation function, and record them as JH1, JH2 and JH3 respectively, obtain the effect coefficient of the disease treatment effect, and use the numerical value of the disease diagnosis accuracy, the effect coefficient and the numerical value of the patient's satisfaction with the treatment as the judgment values ​​corresponding to the first modeling feature item, the second modeling feature item and the third modeling feature item respectively, and record their respective judgment values ​​as PD1, PD2 and PD3 respectively; When PD1≥JH1, a first modeling vector corresponding to the first modeling feature item is generated by a first activation function, and a first image layer is constructed according to the first modeling vector. When PD1<JH1, the first activation function is not operated. When PD2<JH2, a second modeling vector corresponding to the second modeling feature item is generated by a second activation function, and a second image layer is constructed according to the second modeling vector. When PD2≥JH2, the second activation function is not operated. When PD3≥JH3, a third modeling vector corresponding to the third modeling feature item is generated by a third activation function, and a third image layer is constructed according to the third modeling vector. When PD3<JH3, the third activation function is not operated. When the first portrait layer, the second portrait layer and the third portrait layer all exist, the first portrait layer, the second portrait layer and the third portrait layer are spliced ​​and fused in sequence from top to bottom to generate a doctor portrait model corresponding to each doctor; when any of the first portrait layer, the second portrait layer and the third portrait layer does not exist, the doctor portrait model of the corresponding doctor is not constructed; The process of generating corresponding personalized disease course description information based on the doctor portrait model of each doctor includes: Obtain the doctor portrait model corresponding to each doctor, and obtain the model measurement parameters corresponding to the doctor portrait model. The model measurement parameters include model generalization rate, model prediction rate and model fit, which are marked as η1, η2 and η3 respectively. Set the expected target values ​​corresponding to the model generalization rate, model prediction rate and model fit respectively, and record them as ρ1, ρ2 and ρ3 respectively. When η1≥ρ1, η2≥ρ2 and η3≥ρ3, the corresponding personalized course of disease description information is directly generated through the doctor portrait model corresponding to each doctor. When any one of the conditions η1≥ρ1, η2≥ρ2 or η3≥ρ3 is not met, the doctor portrait model is trained until η1≥ρ1, η2≥ρ2 and η3≥ρ3 are met at the same time, and then the corresponding personalized course of disease description information is generated through the doctor portrait model corresponding to each doctor; The process of matching the patient treatment-related data of each sub-process disease course with the personalized disease course description information corresponding to several doctor portrait models in sequence and then matching the corresponding doctor for each sub-process disease course includes: A portrait matching period is set. During the portrait matching period, the patient treatment-related data corresponding to the sub-process disease course of each patient are matched with the personalized disease course description information corresponding to several doctor portrait models in sequence, and then the several sub-process disease courses corresponding to each patient are matched to the corresponding doctors. The doctors corresponding to all the sub-process disease courses matched to each patient are merged, and then a personalized treatment doctor team is established for each patient. The personalized treatment doctor team corresponding to each patient will hand over the operations of the corresponding sub-process disease course to the doctors corresponding to each sub-process disease course in the execution order of the sub-process disease course.

2. According to claim 1, a method for intelligently generating doctor portraits for full course of disease management is characterized in that: The process of obtaining clinical data, academic achievement data, and patient treatment records corresponding to several doctors and encapsulating and generating data sets corresponding to several doctors includes: Several doctors are numbered and recorded as i, i = 1, 2, 3, ..., n, where n is a natural number greater than 0, and the clinical data, academic achievement data and patient treatment records corresponding to the several doctors are obtained, and the clinical data, academic achievement data and patient treatment records corresponding to the doctor numbered i are recorded as D1[i], D2[i] and D3[i] respectively. Starting from i = 1, the D1[i], D2[i] and D3[i] corresponding to the doctor numbered i are encapsulated in turn until the encapsulation of D1[i], D2[i] and D3[i] corresponding to the doctor i = n is completed, and then the corresponding data sets of several doctors are encapsulated and generated; The data set corresponding to the doctor numbered i is denoted as data[i], and data[i] = {D1[i], D2[i], D3[i]}, where the clinical data D1[i] includes all clinical operation records corresponding to the doctor numbered i, the academic achievement data D2[i] includes all academic journals, papers and conference reports published by the doctor numbered i, and the patient diagnosis and treatment records D3[i] include the diagnosis and treatment records of all patients consulted by the doctor numbered i.

3. The method for intelligently generating doctor portraits for full course of disease management according to claim 2, characterized in that: The process of cleaning and integrating the data sets corresponding to several doctors and generating several corresponding normalized data sets includes: Setting a first operation node and a second operation node, wherein the first operation node is used to clean data sets corresponding to a plurality of doctors, and the second operation node is used to integrate data for a plurality of cleaned data sets; The first operation node is used to obtain the node processing values ​​of D1[i] and D3[i] in each doctor's corresponding data set, and they are recorded as p and q respectively. The cleaning thresholds corresponding to the node processing values ​​p and q are set, and they are recorded as μ and τ respectively. All clinical data with p≤μ and all patient diagnosis and treatment records with q≤τ are marked as useless data, and all useless data are eliminated. All clinical data with p>μ are marked as reference data 1, and all patient diagnosis and treatment records with q>τ are marked as reference data 2. The reference data 1 and reference data 2 generated after data cleaning of each data set are summarized as the cleaning data set corresponding to each doctor. The second operation node obtains the cleaned data set corresponding to each doctor, and obtains the academic author signature and document code in the academic achievement data corresponding to each doctor, thereby forming a collective authentication identifier for each doctor, converting the data formats of the reference data one and reference data two corresponding to each cleaned data set into a preset standard format, and assigning different reference value weights to the reference data one and reference data two, and recording the corresponding reference value weights as λ1 and λ2 respectively, thereby integrating the academic achievement data, reference data one and reference data two corresponding to each doctor to generate a normalized data set corresponding to each doctor.

4. The method for intelligently generating doctor portraits for full course of disease management according to claim 3, characterized in that: The process of performing feature analysis and extraction on several normalized data sets to obtain the personalized diagnosis and treatment feature information corresponding to each doctor includes: Obtain the reference value coefficient of each doctor's corresponding normalized data set and record it as γ, where γ = p*λ1 + q*λ2. Use machine learning technology to build a feature analysis program, and let the programmer write the information value interval and feature extraction field corresponding to the feature analysis program. The information value interval includes a high reference value interval, a general reference value interval, and a low reference value interval, which are marked as Ω1, Ω2, and Ω3 respectively; Determine the subordinate relationship between γ and Ω1, Ω2 and Ω3 respectively, and then associate different diagnosis and treatment reference levels with the normalized data set corresponding to γ. The diagnosis and treatment reference levels include level 1, level 2 and level 3, and level 1>level 2>level 3; When γ∈Ω1, the diagnosis and treatment reference level associated with the corresponding normalized data set is level one; When γ∈Ω2, the diagnosis and treatment reference level associated with the corresponding normalized data set is level 2; When γ∈Ω3, the diagnosis and treatment reference level associated with the corresponding normalized data set is level three; When the normalized data sets corresponding to different diagnosis and treatment reference levels are input into the feature analysis program, the feature analysis program processes each normalized data set into several data fields through semantic segmentation in the order of diagnosis and treatment reference levels from high to low, and matches the several data fields corresponding to each normalized data set with the feature extraction fields, marks the data fields that match the feature extraction fields as valid fields, and marks the data fields that do not match the feature extraction fields as invalid fields, and then summarizes all the valid fields corresponding to each normalized data set as the personalized diagnosis and treatment feature information corresponding to each associated doctor.

5. The method for intelligently generating doctor portraits for full course of disease management according to claim 4, characterized in that: The process of dividing the current patient's treatment process into several sub-process processes includes: Obtain the treatment course corresponding to the current patient, and divide the treatment course into several different types of sub-process courses, the types of sub-process courses include diagnosis and evaluation sub-process courses, treatment plan formulation sub-process courses, treatment plan implementation sub-process courses, follow-up management sub-process courses and rehabilitation intervention sub-process courses; Each patient's different sub-processes correspond to corresponding patient treatment-related data, and the patient treatment-related data includes diagnostic assessment data, treatment plan, implementation operation data after the treatment plan is implemented, follow-up data, and rehabilitation guidance data; Among them, the diagnosis and evaluation sub-process course of disease corresponds to the diagnosis and evaluation data, the treatment plan formulation sub-process course of disease corresponds to the treatment plan, the treatment plan implementation sub-process course of disease corresponds to the implementation operation data after the implementation of the treatment plan, the follow-up management sub-process course of disease corresponds to the follow-up data, and the rehabilitation intervention sub-process course of disease corresponds to the rehabilitation guidance data.

6. A doctor portrait intelligent generation system for full course of disease management, used to implement the doctor portrait intelligent generation method according to any one of claims 1 to 5, characterized in that: The system includes: The data collection module is used to obtain the clinical data, academic achievement data and patient treatment records corresponding to several doctors, and encapsulate and generate data sets corresponding to several doctors; The data processing module is used to clean and integrate the data sets corresponding to several doctors, generate several corresponding normalized data sets, perform feature analysis and extraction on the several normalized data sets, and then obtain the personalized diagnosis and treatment feature information corresponding to each doctor; A portrait construction module is used to construct a corresponding doctor portrait model based on the personalized diagnosis and treatment feature information corresponding to a number of doctors, and generate corresponding personalized course of disease description information based on the doctor portrait model of each doctor; The disease course matching module is used to divide the current patient's treatment course into several sub-process courses, and match them in sequence according to the patient treatment-related data of each sub-process course and the personalized disease course description information corresponding to several doctor portrait models, and then match the corresponding doctor for each sub-process course.

Images