A dynamic disease model library analysis method and system
Through the analysis method of the dynamic disease model library, the problems of high misdiagnosis rate and insufficient data quality in clinical diagnosis are solved, and the rapid and accurate push of diagnostic results is achieved. It is suitable for all types of medical institutions and improves diagnostic efficiency and accuracy.
Patent Information
- Application Number
- CN201711424873.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2017-12-25
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2037-12-25
AI Technical Summary
The prior art has problems such as high misdiagnosis rate, limited experience of doctors and insufficient data quality in clinical diagnosis, resulting in low diagnostic accuracy and efficiency, especially in small and medium-sized medical institutions.
Using the analysis method of the dynamic disease model library, a disease model is established through an expert group to distinguish between strong, medium and weak relationship symptoms, form a disease derivation formula, update the model and push possible diagnostic results, and analyze it in combination with medical record data to improve diagnostic accuracy.
Reduce the misdiagnosis rate, improve diagnosis speed and accuracy, and is suitable for all types of medical institutions. Regardless of the amount of data, it can dynamically adjust the disease model and improve data utilization efficiency.
Smart Images

Figure CN108091399B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of medical data analysis, and in particular to an analysis method and system for a dynamic disease model library. Background Art
[0002] According to literature research, the three main causes of misdiagnosis in clinical diagnosis and treatment are insufficient doctor experience, failure to conduct the most powerful examination items, and lack of consensus in questioning and examination. There are several manifestations of insufficient doctor experience, such as doctors with less years of experience, or strong professional fields but unfamiliar with other fields; failure to conduct the most powerful examination is when the attending physician has a preliminary impression or preliminary diagnosis before diagnosis, and in order to further confirm the results, additional examination items will be added to verify the initial speculation. At this time, due to the complexity of the disease, it is very easy for the examination application issued to be not the most urgent, resulting in economic waste and time delay; failure to reach a consensus is that the relationship between the examination results and the disease is difficult to determine when the condition is complicated, and the opinions of various experts are not consistent, and in the end, it is often necessary to consult multiple departments to determine the diagnosis.
[0003] To address the above issues, currently, intelligent triage or clinical decision support software is mostly used in clinical practice. These two types of systems are mainly based on two types of information processing methods: one is based on static clinical knowledge processing methods, that is, establishing a risk factor table according to the manifestations of the disease, and making initial diagnosis opinions by comparing them one by one. This method is similar to the empirical judgment method of clinical experts; the other is to use the multi-factor regression analysis method of big data to process historical medical records, dynamically analyze all disease-related symptoms, obtain the corresponding prediction formula, and then input the patient's condition data for predictive analysis.
[0004] The above two methods, static clinical knowledge base method and big data method, have their own advantages in specific situations, but they also have certain limitations as follows:
[0005] Although the static knowledge base method is more effective for some common diseases with easily identifiable symptoms, there are still many diseases whose symptom combination patterns are not clear. At the same time, as human society's understanding of diseases and detection methods continue to develop, knowledge is also constantly enriched. Static knowledge bases are often summarized clinically and then applied and promoted after academic discussions, and their timeliness is low.
[0006] The multi-factor regression method relies on the analysis of big data, and naturally has certain requirements for data quality and quantity. Due to the rapid development of domestic medical information and the complexity and diversity of information systems, the current historical data quality is not high and the information is not comprehensive. Therefore, it is extremely difficult for most small and medium-sized medical institutions to adopt this type of method to carry out implementation work. Therefore, from an objective point of view, not all hospitals have this condition. At the clinical application level, the current popularity and usability are greatly reduced. Summary of the invention
[0007] The technical problem to be solved by the present invention is to provide an analysis method and system for a dynamic disease model library.
[0008] The present invention adopts the following technical solution:
[0009] The present invention provides an analysis method for a dynamic disease model library, and the specific steps are as follows:
[0010] S1. The expert group established a disease model based on the disease type and corresponding symptoms;
[0011] Establish a disease list, and establish a symptom list under the disease list, wherein the symptom list is divided into three categories: strong relationship symptom S, medium relationship symptom M, and weak relationship symptom W;
[0012] According to the law of disease diagnosis, the disease derivation formula of disease and symptoms is established:
[0013] O(x)=f(m1,m2,…,m n ,w1,w2,…,w i )=b1X1+b2X2+…+b n X n +p1W1+p2w2…+p i w i;
[0014] m n It means that there are n symptoms m in the middle relational symptom sequence. The middle relational symptom sequence is represented as an array (m1, m2…m n ), represented by M, w i The weak relationship symptom sequence has j symptoms w. The weak symptom sequence is represented as an array (w1, w2…w n ), represented by W; b1…b n It represents the weight value of the disease symptoms in the formula when diagnosing the disease, and the value is between 0 and 1, p1, p2…p i Refers to the coefficient of the weak relationship symptom referenced by the formula, and the coefficient value is between 0 and 1;
[0015] S2, selecting all appropriate medical records to form a valid medical record set, and inputting the valid medical record set into the disease model;
[0016] S3. When the number of valid medical records reaches a certain number, the disease model is updated and a new disease model is finally obtained;
[0017] S4. Enter the patient's relevant symptom information into the model. The model will judge and analyze the symptom information and the disease list, and use the disease derivation formula to calculate the possible diseases and probabilities, and push the various diseases and corresponding probabilities to the user end of the clinician.
[0018] Furthermore, the specific steps of selecting a valid medical record set include:
[0019] 1) Extract all the symptoms appearing in the disease model from the medical record content to form a list Q of the extraction results for each medical record;
[0020] 2) Obtain a list of disease symptoms from the disease model;
[0021] 3) Then compare the list Q from 1) with the symptom list obtained from 2). Only when all the strongly related symptoms in the symptom list appear in the list Q, the medical record can enter the subsequent processing; otherwise, it is an invalid medical record;
[0022] 4) Then substitute the nodes in list Q into the disease derivation formula to calculate the probability of illness. When the probability is greater than the set minimum threshold, this medical record can be included in the valid medical record set of the disease.
[0023] Furthermore, the range of the minimum threshold is 0-1.
[0024] Furthermore, the step S3 of updating the disease model is as follows:
[0025] 1) Collect the corresponding symptom result list Q of each medical record in the valid medical records together as a parameter for analysis and processing;
[0026] 2) After the above analysis and processing, the correlation between symptoms and diseases is obtained, and they are sorted from high to low according to the correlation, and a symptom sequence Y with weighted values is obtained;
[0027] The symptom sequence Y is expressed as (y1: r1, y2: r1, y3: r3...y k :r k ), where y is the name of the symptom, r is the relevant weight value, and the relevance of the installed diseases is arranged from high to low, where k represents the total number of all symptoms in the medical record model;
[0028] 3) Obtain the symptom list of the corresponding disease from the disease model, and compare each of the middle-relation symptoms in the symptom list with the symptoms in the symptom sequence Y one by one. First, exclude the strong-relation symptoms of the disease. Then, if it is found that there is a middle-relation symptom in the symptom sequence Y that does not appear in the disease model and is ranked before any existing symptom in the middle-relation symptom in the disease model, add it to the original middle-relation symptom to form a new middle-relation symptom M. new, and remove the corresponding symptoms from the list of weakly related symptoms of the disease in the disease model;
[0029] 4) Use the M obtained in 3) new and W new Compare the nodes of the symptom list Q of all valid medical records of the corresponding disease, and compare each symptom in Q with M new Symptoms and W new The intersection nodes of the top N symptoms are extracted to form a new list L, where N can be configured in the system to form a list L set;
[0030] 5) Substitute the new list L set into the disease derivation formula to obtain a new derivation formula. The new derivation formula and the new M new The list of diseases constitutes the new disease model.
[0031] Furthermore, the step S3 of updating the disease model is as follows: the step S3 of updating the disease model when the number of valid medical record sets reaches a certain number, and the number range is more than 1000.
[0032] A dynamic clinical disease model system is also disclosed, including a dynamic inference engine, which is respectively connected to the information input terminal and the user terminal for communication, and is used to receive medical record information sent by the information input terminal, wherein the medical record information includes clinical data of patient information and symptom information; for undiagnosed medical record information, possible diseases and probability information are deduced according to the disease model in the dynamic inference engine, and then the disease and probability information are sent to the corresponding user terminal as a reference for the diagnosis of the corresponding doctor; for confirmed medical record information, the dynamic inference engine determines the validity of the medical record according to the disease model, and transmits the medical record determined to be valid to the medical record storage;
[0033] The dynamic clinical disease model system also includes a medical record repository, which is used to receive all medical record information collected by the information input terminal that has been diagnosed and determined to be valid medical records by the dynamic reasoning engine;
[0034] The dynamic reasoning engine extracts the medical record set of the corresponding disease from the medical record repository, and extracts the corresponding symptoms in each medical record. Whenever the medical record set reaches a specified level, the dynamic reasoning engine starts analyzing the correlation between the corresponding disease and the symptoms. The dynamic reasoning engine sorts the new symptom list according to the analyzed symptom correlation, and adjusts the disease model according to the new symptom list.
[0035] Furthermore, the dynamic reasoning engine includes a dynamic reasoning module, a disease model storage module, a symptom extraction module and a case analysis module, and the dynamic reasoning module is bidirectionally connected to the disease model storage module, the symptom extraction module and the case analysis module respectively;
[0036] A dynamic reasoning module is used to substitute the disease symptom list obtained by the disease symptom extraction module into the disease model storage module to obtain the disease model and calculate the probability of related diseases;
[0037] A disease model storage module, used to store a disease model, wherein the disease model includes a disease list, a symptom list and a disease derivation formula model;
[0038] A symptom extraction module, used to extract symptom information related to the disease from the medical record information of the patient or the confirmed medical record in the medical record storage, and process the symptom information into a symptom list Q for analysis by the case analysis module;
[0039] The case analysis module is used to compare the symptom list Q with the symptom list of the disease model of the corresponding disease. When the strongly related symptoms in the symptom list of the disease model all appear in the symptom list Q, the probability of the disease is calculated; and it is used to perform symptom correlation analysis on each disease to obtain the symptom sequence Y, and pass the symptom sequence Y to the weight regulator.
[0040] Furthermore, the dynamic clinical disease model system also includes a weight regulator for receiving the symptom sequence Y of the case analysis module, correcting the weight value in the disease model, and comparing the symptom list with the symptoms in the symptom sequence Y one by one to form a new disease model.
[0041] Furthermore, the symptom list Q is represented in an array as follows:
[0042] Q(disease name, medical record number) = {(symptom 1, value 1), (symptom 2, value 2) ..., (symptom n, value n)};
[0043] Each node in the symptom list consists of two parts: name and value. The value form is divided into three categories: (1) yes / no, (2) numerical value, and (3) text sequence.
[0044] Furthermore, the number of nodes in the list Q is equal to the number of symptoms in the symptom library of the disease model.
[0045] Furthermore, the strong relationship symptom S, the medium relationship symptom M and the weak relationship symptom W are reflected in the disease model in the form of a symptom list, as follows:
[0046] The symptom list is represented in array form as {(symptom 1, value 1), (symptom 2, value 2)…(symptom i, value i)…(symptom n, value n)};
[0047] Each node in the symptom list consists of two parts: name and value. The value form is divided into three categories: (1) yes / no, (2) numerical value, and (3) text sequence.
[0048] Compared with the prior art, the present invention has the following beneficial effects:
[0049] The present invention takes reducing the probability of misdiagnosis as its starting point, and provides clinical assistance to clinicians from the perspective of breaking the experience limitations of individual doctors. It is an information processing technology method and equipment that can improve the accuracy and speed of diagnosis and treatment while increasing patient satisfaction and reducing the cost of seeing a doctor.
[0050] The present invention uses a universal dynamic disease model library to effectively integrate the static clinical knowledge base with the big data method to form a dynamic disease model, so that this method can be used to work regardless of the size of the hospital and the amount of historical data. At the same time, existing and new medical data can be converted into factors that promote disease model optimization at any time, and the disease model library can be dynamically adjusted.
[0051] The present invention can break through the limitation of doctors' experience, improve the speed and accuracy of diagnosis, and at the same time help improve the quality of new clinical data, convert clinical data into experience in real time, and make the hospital's data truly usable wealth. BRIEF DESCRIPTION OF THE DRAWINGS
[0052] Figure 1 It is a schematic diagram of the process of dynamic disease model analysis method;
[0053] Figure 2 Schematic diagram of the dynamic clinical disease model system modules. DETAILED DESCRIPTION
[0054] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and do not limit the present invention.
[0055] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.
[0056] Reference Figure 1 As shown, the above method is described in detail below:
[0057] Step 1: The expert group establishes a disease model based on the disease type and corresponding symptoms;
[0058] A disease list is established, and a symptom list is established under the disease list. The symptom list is divided into three categories: strong relationship symptoms, medium relationship symptoms, and weak relationship symptoms.
[0059] In this embodiment:
[0060] The disease list includes heart failure, ischemic cardiomyopathy, adenomyosis and prostate, etc., forming a complete disease list. At the same time, a symptom list is established under the corresponding specific disease. The symptom list here refers to the characteristics related to the definition and diagnosis of the relevant disease, including detailed information such as medical history, surgical history, allergy history, symptoms, signs, examinations, tests, living habits, gender, social living environment, etc. Take heart failure as an example:
[0061] The list of symptoms of heart failure includes: age, age group, history of diabetes, hyperlipidemia, ischemic cerebrovascular disease, chronic obstructive pulmonary disease, diastolic blood pressure, serum protein, serum sodium, Ln blood cholesterol, fever, dystonia and chills, etc.
[0062] Strong relationship symptoms: Indicated by S, it refers to the facts that must exist when diagnosing a disease, such as the symptoms and signs that must exist, and the patient's physical reactions. It may also be a natural person's attribute, such as gender, and so on. For example, starting from the definition of the disease, such as male disease, the patient must be male, not female patients;
[0063] Medium-term symptoms: Indicated by M, these are the symptoms that clinical experts have reached a consensus on and must be confirmed when making a diagnosis. However, for each specific patient, only some of the symptoms may appear, but the possibility of all symptoms appearing cannot be ruled out.
[0064] Weak relationship symptoms: Indicated by W, it means that this symptom may or may not exist. It can be used as a reference when diagnosing a disease, but it is not a mandatory indicator.
[0065] At the same time, it should be pointed out that as humans gain a deeper understanding of diseases, the strength of the relationship between symptoms and diseases can be transformed, especially in different populations, different regions, different periods or different living habits. Although strong or medium relationship symptoms are not included in the definition of the disease, they may be transformed into a higher relationship. Therefore, a dynamic disease model is needed that can be transformed in real time according to the transformation of relationship symptoms to help clinicians quickly diagnose related diseases and increase accuracy.
[0066] For example, the distinction between strong, medium and weak relationship symptoms of heart failure:
[0067] Strong relationship symptoms: coronary heart disease, hypertension
[0068] Symptoms: Orthopnea, hepatojugular reflux, tachycardia, jugular venous distension, pulmonary rales, cardiomegaly, pulmonary edema, bilateral lower limb edema
[0069] Weak relationship symptoms: shortness of breath after activity, pleural effusion, enlarged lung hilum, gallop rhythm of third heart sound, hepatomegaly
[0070] This embodiment uses heart failure as an example to explain in detail as follows:
[0071] The S list, M list, and W list all represent symptom lists, which are represented in array form as {(symptom 1, value 1), (symptom 2, value 2)…(symptom i, value i)…(symptom n, value n)}. Each node in the list consists of two parts: name (symptom n) and value (value n). The value form is divided into three categories: (1) yes / no, (2) numerical value, and (3) text sequence (dictionary value). For example, the symptom list of strong relationship symptoms of heart failure, S (heart failure) = {(history of coronary heart disease, yes), (history of hypertension, yes)}
[0072] Symptom list of related symptoms, M (heart failure) = {(orthopnea, yes), (hepatojugular reflux sign, yes), (tachycardia, yes), (jugular venous distention, yes), (pulmonary rales, yes), (cardiac hypertrophy, yes), (cardiac hypertrophy, yes), (pulmonary edema, yes), (edema of both lower limbs, yes)}
[0073] Symptom list of weak relationship symptoms, W (heart failure) = {(shortness of breath after activity, yes), (pleural effusion, yes), (enlarged hilum of lung, yes), (third heart sound gallop, yes), (enlarged liver, yes)}.
[0074] According to the law of disease diagnosis, the disease derivation formula of disease and symptoms is established:
[0075] O(x)=f(m1,m2,…,m n; w1,w2,…,w i )=b1X1+b2X2+…+b n X n +p1W1+p2w2…+p i w i;
[0076] m n It means that there are n symptoms m in the middle relational symptom sequence. The middle relational symptom sequence is represented as an array (m1, m2…m n ), w i The weak relationship symptom sequence has j symptoms w, and the medium relationship symptom sequence is represented as an array (w1, w2, …, w i )。 b1…b n It represents the weight value of the disease symptoms in the formula when diagnosing the disease, and the value is between 0 and 1, p1, p2…p i Refers to the coefficient of the weak relationship symptom referenced by the formula, and the coefficient value is between 0 and 1;
[0077] For example, in one of the heart failure medical records, the medium-relation symptom sequence and the weak-relation symptom sequence are substituted into the disease derivation formula:
[0078] O (suffering from heart failure) = 0.50 * hepatojugular reflux + 0.25 * orthopnea + 0.15 * pleural effusion + 0.1 * bilateral lower limb edema = 0.50 * 1 + 0.25 * 1 + 0.15 * 1 + 0.1 * 1 = 1
[0079] S2, select all appropriate medical records to form a valid medical record set, and input the valid medical record set into the disease model;
[0080] The specific steps of selecting a valid medical record set include:
[0081] 1) Extract the descriptive information and values of all symptoms appearing in the disease model from the medical record content to form a list Q of the extraction results for each medical record;
[0082] The detailed list Q here is represented as an array as Q(disease name, medical record number)={(symptom 1, value 1), (symptom 2, value 2)..., (symptom n, value n)}; the number of nodes in the list Q is equal to the number of symptoms in the symptom library of the disease model. There are three types of symptom values: (1) numerical value (2) present / absent / not present (3) text sequence. The numerical value refers to the value expressed in digital form. For example, if the weight is 57kg, the weight is the symptom and 57 is the value. Before the array is established, each result will be a numerical value. Symptoms specify the unit of annotation, and the data is converted according to the unit when taking values; yes / no / no expression is whether the symptom is found. Taking fever as an example, in different medical records, the possible situations are: yes: corresponds to the patient having this symptom, no: indicates that the symptom is clearly denied, and no: indicates that there is no description in the medical record; text sequence means that the result value is in text form, similar to the expression of a dictionary. For example, in some medical records, when the patient's gender is a symptom, male is one of its values, and in other medical records, the patient's gender may be female. Taking heart failure as an example, we can get a vector table for each medical record.
[0083] For example, the 100th medical record of heart failure is expressed as, Q(heart failure, 100) = {(orthopnea, yes), (hepatojugular reflux sign, no), (tachycardia, yes), (jugular venous distension, yes), (fever, no), (age group, elderly), …}.
[0084] 2) Obtain a list of disease symptoms from the disease model;
[0085] That is, S list, M list and W list.
[0086] 3) Then compare the list Q from 1) with the symptom list obtained from 2). Only when all the strongly related symptoms in the symptom list appear in the list Q, the medical record can enter the subsequent processing; otherwise, it is an invalid medical record;
[0087] What is pointed out here is the comparison between list Q and list S. When the symptoms in list S of the disease do not appear in list Q, the medical record is not a valid medical record of the disease and the next step of processing will be abandoned; otherwise, when all the symptoms in list S of the disease appear in list Q, the medical record is a valid medical record of the disease and the next step of processing will be carried out.
[0088] Compare Q(heart failure, 100) = {(orthopnea, yes), (hepatojugular reflux sign, no), (tachycardia, yes), (jugular venous distension, yes), (fever, no), (age group, elderly), …} with S(heart failure) = {(history of coronary heart disease, yes), (history of hypertension, yes)} in detail. The 100th medical record contains a history of coronary heart disease and hypertension, so step 4) can be carried out.
[0089] 4) Then substitute the nodes in the list Q into the disease derivation formula to calculate the probability of illness. When the probability is greater than the set minimum threshold, the range of the minimum threshold is 0-1, and this medical record can be included in the valid medical record set of the disease.
[0090] The lowest threshold value set in this embodiment is 0.6. At this time, the estimated disease probability of the list Q (heart failure, 100) in this embodiment is greater than 0.6, so the medical record can be classified into the valid medical record set of heart failure.
[0091] S3. When the number of valid medical records reaches a certain number, the disease model is updated and a new disease model is finally obtained;
[0092] In this embodiment, updating the disease model when the valid medical record set reaches a certain number means that when the number of valid medical records associated with heart failure reaches a certain number, the update processing flow described below can be started, and the number is set to two types: initial level and incremental level, both of which can be adjusted. For example, in this embodiment, the initial level is set to 10,000 and the incremental level is set to 1,000. When the number of suitable medical records reaches 10,000, the update flow of the derivation formula of the disease is started for the first time. When it reaches 11,000, 12,000, 13,000, and so on, the update flow is continuously started.
[0093] In this embodiment, the step S3 of updating the disease model is specifically as follows:
[0094] 1) Collect the corresponding symptom results (heart failure) list Q of each medical record in the valid medical records together as parameters for analysis and processing;
[0095] The analysis and processing here mainly uses software with single factor analysis calculation algorithms or processing equipment with interface services, such as servers with SPSS software or other data analysis software.
[0096] 2) After the above analysis and processing, the correlation between symptoms and diseases is obtained, and they are sorted from high to low according to the correlation, and a symptom sequence Y with weighted values is obtained;
[0097] The symptom sequence Y is expressed as (y1: r1, y2: r1, y3: r3...y k :r k ), where y is the name of the symptom, r is the relevant weight value, and they are arranged from high to low according to the relevance of the disease, and k represents the total number of all symptoms in the medical record model;
[0098] In this embodiment, according to the different analysis software used, k is the total number of all symptoms in the medical record model, and r represents the relevant weight value, which ranges from 0 to 1. For example, for heart failure, the symptom sequence of heart failure is exemplified as Y (heart failure) = (age: 0.8, diastolic blood pressure: 0.7, pulmonary edema 0.45, diabetes: 0.5, jugular vein distension: 0.2, fever: 0.01...).
[0099] 3) Obtain the symptom list of the corresponding disease from the disease model, and compare each of the middle-relation symptoms in the symptom list with the symptoms in the symptom sequence Y one by one. First, exclude the strong-relation symptoms of the disease. Then, if it is found that there is a middle-relation symptom in the symptom sequence Y that does not appear in the disease model and is ranked before any existing symptom in the middle-relation symptom in the disease model, add it to the original middle-relation symptom to form a new middle-relation symptom M. new , and remove the corresponding symptoms from the list of weakly related symptoms of the disease in the disease model;
[0100] For example, in this embodiment, heart failure is taken as an example. In the disease model, the original middle relation feature of heart failure is M (heart failure) = {(orthopnea, yes), (hepatojugular reflux sign, yes), (tachycardia, yes), (jugular venous distention, yes), (pulmonary rales, yes), (cardiac hypertrophy, yes), (cardiac hypertrophy, yes), (pulmonary edema, yes), (edema of both lower limbs, yes)}. It is now found that the age group is before the jugular venous distention, so it is added to the middle relation feature and updated to M. new (Heart failure) = {(orthopnea, yes), (age group, elderly), (hepatojugular reflux sign, yes), (tachycardia, yes), (jugular venous distention, yes), (pulmonary rales, yes), (cardiac hypertrophy, yes), (cardiac hypertrophy, yes), (pulmonary edema, yes), (edema of both lower limbs, yes)}; At the same time, it was found that the fever symptom was not higher than any of the middle-related symptoms in M, so fever was classified as a weak-related symptom W of heart failure. newmiddle.
[0101] 4) Use the M obtained in 3) new and W new Compare the nodes of the symptom list Q of all valid medical records of the corresponding disease, and compare each symptom in Q with M new Symptoms and W new The intersection nodes of the top N symptoms are extracted to form a new list L, where N can be configured in the system to form a list L set;
[0102] As in step 3) above, M new (heart failure) and the node comparison of Q(heart failure, 100) of the 100th valid medical record of heart failure. new The symptoms in (heart failure) that appear in Q(heart failure, 100) are selected, and the specified number of weakly related symptoms are added to form a new table L. The new table L is as follows:
[0103] L(heart failure, 100)={(orthopnea, yes),(hepatojugular reflux sign, no),(tachycardia, yes),(jugular venous distension, yes),(age group, elderly), …}.
[0104] The new list Q set is analyzed and processed using software with a single factor analysis algorithm or an interface service processing device, such as a server with SPSS software or other data analysis software. Finally, the symptoms are re-analyzed and arranged, and the middle-relationship symptoms and weak-relationship symptoms are specifically re-differentiated. Finally, a new disease derivation formula is obtained:
[0105] O(x)=f(m1,m2,…,m n; w1,w2,…,w i )=b1X1+b2X2+…+b n X n +p1W1+p2w2…+p i w i;
[0106] Here, the transformation of the relationship symptoms is real-time, so the disease model updates the disease derivation formula in real time according to the transformation situation to effectively ensure the accuracy. For example, the symptoms of the above age group are transformed from weak relationship symptoms to medium relationship symptoms, which has a great impact on the results of the derivation formula.
[0107] It should be noted here that the strength of the relationship between symptoms and diseases can be transformed, especially in different populations, different regions, different periods or different living habits. Although strong or medium relationship symptoms are not included in the definition of the disease, it is possible to transform to a higher relationship. Therefore, a dynamic disease model is needed that can be transformed in real time according to the transformation of relationship symptoms to help clinicians quickly diagnose related diseases and increase accuracy.
[0108] 5) Substitute the new list Q set into the disease derivation formula to obtain a new derivation formula. The new derivation formula and the new M new The list of diseases constitutes the new disease model.
[0109] S4. Enter the patient's relevant symptom information into the model. The model will judge and analyze the symptom information and the disease list, and use the disease derivation formula to calculate the possible diseases and probabilities, and push the various diseases and corresponding probabilities to the user end of the clinician.
[0110] After the clinician sees the patient, he or she first collects the patient's relevant symptom information, including the patient's past medical history and current medical history. After summarizing all the patient's relevant symptoms through questioning, examination and testing, the clinician enters the relevant symptom information into the disease model. The disease model will calculate the possible diseases and the probabilities of the diseases based on the input symptom information. The possible diseases mentioned here refer to more than one disease, and the diseases and the probabilities of the diseases are provided for the reference of the clinician.
[0111] Reference Figure 2 As shown, a dynamic clinical disease model system is also disclosed, including a dynamic inference engine, which is respectively connected to the information input terminal and the user terminal for communication,
[0112] Receive medical record information sent by the information input terminal, the medical record information includes clinical data of patient information and symptom information; for undiagnosed medical record information, deduce possible diseases and probability information based on the disease model in the dynamic reasoning engine, and then send the disease and probability information to the corresponding user terminal as a reference for the diagnosis of the corresponding doctor; for confirmed medical record information, the dynamic reasoning engine determines the validity of the medical record based on the disease model, and transfers the medical record determined to be valid to the medical record storage;
[0113] The dynamic clinical disease model system also includes a medical record repository, which is used to receive all medical record information collected by the information input terminal that has been diagnosed and determined to be valid medical records by the dynamic reasoning engine;
[0114] The dynamic reasoning engine extracts the medical record set of the corresponding disease from the medical record repository, and extracts the corresponding symptoms in each medical record. Whenever the medical record set reaches a specified level, the dynamic reasoning engine starts the correlation analysis of the corresponding disease and symptoms. The dynamic reasoning engine sorts the new symptom list according to the analyzed symptom correlation, and adjusts the disease model according to the new symptom list.
[0115] Here, the information input terminal generally refers to the computer or smart handheld device in the hospital that can communicate with this system, such as the computer of the doctor's workstation. The user terminal communication refers to the PC or smart handheld device used by the clinician, which is mainly used to receive the results obtained by the dynamic reasoning machine. The dynamic reasoning machine can also be considered as the background system server.
[0116] It should be pointed out that the communication connection mentioned here includes wired communication connection and invalid communication connection. In this embodiment, the dynamic reasoning engine adopts a communication connection with terminal equipment related to the hospital. The terminal equipment includes the hospital's display screen and PC and other equipment.
[0117] Furthermore, the dynamic reasoning engine includes a dynamic reasoning module, a disease model storage module, a symptom extraction module and a case analysis module, and the dynamic reasoning module is bidirectionally connected to the disease model storage module, the symptom extraction module and the case analysis module respectively;
[0118] A dynamic reasoning module is used to substitute the disease symptom list obtained by the disease symptom extraction module into the disease model storage module to obtain the disease model and calculate the probability of related diseases;
[0119] A disease model storage module, used to store a disease model, wherein the disease model includes a disease list, a symptom list and a disease derivation formula model;
[0120] A symptom extraction module, used to extract symptom information related to the disease from the medical record information of the patient or the confirmed medical record in the medical record storage, and process the symptom information into a symptom list Q for analysis by the case analysis module;
[0121] The case analysis module is used to compare the symptom list Q with the symptom list of the disease model of the corresponding disease. When the strongly related symptoms in the symptom list of the disease model all appear in the symptom list Q, the probability of the disease is calculated; and it is used to perform symptom correlation analysis on each disease to obtain the symptom sequence Y, and pass the symptom sequence Y to the weight regulator.
[0122] Furthermore, the dynamic clinical disease model system also includes a weight regulator for receiving the symptom sequence Y of the case analysis module, correcting the weight value in the disease model, and comparing the symptom list with the symptoms in the symptom sequence Y one by one to form a new disease model.
[0123] Furthermore, the symptom extraction module, case analysis module, weight regulator and dynamic inference engine should all run on at least one CPU, but it does not exclude the application in distributed computing, such as SAAS model deployment.
[0124] Furthermore, the disease model storage module and the medical record repository may be stored in an independent DB manner or in a distributed storage device.
[0125] The medical assistance system constructed by the present invention is not restricted by the scale of the medical institution, the time it has been in business, and the previous level of informatization. It can be applied to clinical practice based on the initial state of the disease model, and can use the historical data and new data of the medical institution to continuously revise the disease model during the subsequent opening of the medical institution. The revised model can be applied to clinical practice again, thus forming a closed-loop data chain, which makes the production and utilization of data harmonious and unified, and available immediately. At the same time, this data chain is also an open data chain that can incorporate data from external hospitals.
[0126] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for analyzing a dynamic disease model library, characterized in that: The specific steps are as follows: S1. The expert group established a disease model based on disease types and corresponding symptoms; Establish a disease list, and establish a symptom list under the disease list, wherein the symptom list is divided into three categories: a strong relationship symptom sequence S, a medium relationship symptom sequence M, and a weak relationship symptom sequence W; According to the law of disease diagnosis, the disease derivation formula of disease and symptoms is established: O(x)=f(m1,m2,…,m n ,w1,w2,…,w i )=b1X1+b2X2+…+b n X n +p1W1+p2w2…+p i w i; m n It means that there are n symptoms m in the relational symptom sequence M. The relational symptom sequence M is represented as an array (m1, m2…m n ), w i It means that the weak relationship symptom sequence W has i symptoms w. The weak relationship symptom sequence W is represented as an array (w1, w2…w i ); b1…b n It represents the weight value of the disease symptoms in the formula when diagnosing the disease, and the value is between 0 and 1, p1, p2…p i Refers to the coefficient of the weak relationship symptom referenced by the formula, and the coefficient value is between 0 and 1; S2, selecting all appropriate medical records to form a valid medical record set, and inputting the valid medical record set into the disease model; The specific steps of selecting a valid medical record set include: 1) Extract all the symptoms appearing in the disease model from the medical record content to form a symptom list Q of each medical record extraction result; 2) Obtain the list of symptoms of the confirmed diseases and the corresponding derivation formulas of the corresponding medical records from the disease model; 3) Then compare the list Q in 1) with the symptom list obtained in 2). Only when all the strong relationship symptoms in the model symptom list appear in the list Q, the medical record can enter the subsequent processing; otherwise, it is an invalid medical record; 4) Then substitute the symptom nodes in the list Q into the disease derivation formula to calculate the probability of illness. When the probability is greater than the set minimum threshold, this medical record can be included in the valid medical record set of the disease; S3. When the number of valid medical records reaches a certain number, the disease model of the corresponding disease is updated, and finally a new disease model is obtained; the specific steps of updating the disease model are as follows: 1) Collect all the symptom results Q lists in the valid medical records of the same disease as parameters for analysis and processing; 2) After the above analysis and processing, the correlation between symptoms and diseases is obtained, and they are sorted from high to low according to the correlation, and a symptom sequence Y with weighted values is obtained; the symptom sequence Y is expressed as (y1: r1, y2: r1, y3: r3...y k :r k ), where y is the symptom, r is the relevant weight value, and they are arranged from high to low according to their relevance to the disease, and k represents the total number of all symptoms in the medical record model; 3) Obtain a list of symptoms of the corresponding disease from the disease model, and compare the middle-relation symptoms in the symptom list with the symptoms in the symptom sequence Y one by one. First, exclude the strong-relation symptoms of the disease. Then, if it is found that there are middle-relation symptoms in the symptom sequence Y that do not appear in the disease model and are ranked before any existing symptoms in the middle-relation symptoms in the disease model, add them to the original middle-relation symptoms to form a new middle-relation symptom sequence M. new , and remove the corresponding symptoms from the weak relationship symptom sequence W of the disease model to form a new weak relationship symptom sequence W new ; 4) Use the M obtained in 3) new and W new Compare each Q with the nodes of the symptom list Q of all valid medical records of the corresponding disease one by one, and compare each Q with M new The intersection of the symptoms in the symptom node and the intersection of Q and W new The intersection symptom nodes of the top N symptoms are extracted to form a new list L, where N can be configured in the system to form a list L set; 5) Substitute the new list L set into the disease derivation formula to obtain a new derivation formula. The new derivation formula and the new M new and W new The list of symptoms constitutes a new disease model; S4. Enter the patient's relevant symptom information into the model. The model first selects a list of diseases that meet the strong relationship symptoms in the disease model based on the symptom information, and then uses the disease derivation formula of the corresponding disease list to calculate the various possible diseases and probabilities, and pushes the various diseases and corresponding probabilities to the user end of the clinician.
2. The method for analyzing a dynamic disease model library according to claim 1, characterized in that: The range value of the minimum threshold is 0-1.
3. The method for analyzing a dynamic disease model library according to claim 1, characterized in that: The step S3 of updating the disease model is as follows: when the number of valid medical record sets reaches a certain number, the disease model is updated in the step S3, and the number range is more than 1000.
4. A dynamic clinical disease model system, characterized in that: The system uses the analysis method of the dynamic disease model library according to any one of claims 1 to 3, and the system comprises: A dynamic inference engine, the dynamic inference engine is connected to the information input terminal and the user terminal for communication, Receive medical record information sent by the information input terminal, the medical record information includes clinical data of patient information and symptom information; for undiagnosed medical record information, deduce possible diseases and probability information based on the disease model in the dynamic reasoning engine, and then send the disease and probability information to the corresponding user terminal as a reference for the diagnosis of the corresponding doctor; for confirmed medical record information, the dynamic reasoning engine determines the validity of the medical record based on the disease model, and transfers the medical record determined to be valid to the medical record storage; The dynamic clinical disease model system also includes a medical record repository, which is used to receive all medical record information collected by the information input terminal that has been diagnosed and determined to be valid medical records by the dynamic reasoning engine; The dynamic reasoning engine extracts the medical record set of the corresponding disease from the medical record repository, and extracts the corresponding symptoms in each medical record. Whenever the medical record set reaches a specified level, the dynamic reasoning engine starts analyzing the correlation between the corresponding disease and the symptoms. The dynamic reasoning engine sorts the new symptom list according to the analyzed symptom correlation, and adjusts the disease model according to the new symptom list.
5. The dynamic clinical disease model system according to claim 4, characterized in that: The dynamic reasoning engine includes a dynamic reasoning module, a disease model storage module, a symptom extraction module and a case analysis module, and the dynamic reasoning module is bidirectionally connected with the disease model storage module, the symptom extraction module and the case analysis module respectively; A dynamic reasoning module is used to substitute the disease symptom list obtained by the disease symptom extraction module into the disease model storage module to obtain the disease model and calculate the probability of related diseases; A disease model storage module, used to store a disease model, wherein the disease model includes a disease list, a symptom list and a disease derivation formula model; A symptom extraction module, used to extract symptom information related to the disease from the medical record information of the patient or the confirmed medical record in the medical record storage, and process the symptom information into a symptom list Q for analysis by the case analysis module; The case analysis module is used to compare the symptom list Q with the symptom list of the disease model of the corresponding disease. When the strongly related symptoms in the symptom list of the disease model all appear in the symptom list Q, the probability of the disease is calculated; and it is used to perform symptom correlation analysis on each disease to obtain the symptom sequence Y, and pass the symptom sequence Y to the weight regulator.
6. The dynamic clinical disease model system according to claim 4, characterized in that: The dynamic clinical disease model system also includes a weight regulator for receiving the symptom sequence Y of the case analysis module, correcting the weight value in the disease model, and comparing the symptom list with the symptoms in the symptom sequence Y one by one to form a new disease model.
7. The dynamic clinical disease model system according to claim 5, characterized in that: The symptom list Q is represented in array form as follows: Q(disease name, medical record number) = {(symptom 1, value 1), (symptom 2, value 2) ..., (symptom n, value n)}; Each node in the symptom list consists of two parts: name and value. The value can be divided into three categories: (1) yes / no, (2) numerical value, and (3) text sequence.
8. The dynamic clinical disease model system according to claim 4, characterized in that: The strong relationship symptom sequence S, the medium relationship symptom sequence M and the weak relationship symptom sequence W are embodied in the disease model in the form of a symptom list, as follows: The symptom list is represented in array form as {(symptom 1, value 1), (symptom 2, value 2)...(symptom i, value i)...(symptom n, value n)}; Each node in the symptom list consists of two parts: name and value. The value can be divided into three categories: (1) yes / no, (2) numerical value, and (3) text sequence.
Citation Information
Cited By
Multi-modal tumor data intelligent fusion and dynamic typing system
CN121148646A
