Disclosure of Invention
In view of this, the present application provides a method, an apparatus, a device, and a storage medium for optimizing a medical record set, so as to correct a small number of medical records in the medical record set that have unreasonable diagnoses, thereby improving the quality and reliability of the medical record set, and the technical scheme is as follows:
a medical record set optimization method comprises the following steps:
correcting medical records in which unreasonable diagnosis exists in a target medical record set respectively by adopting a plurality of different medical record correction modes to obtain a plurality of corrected medical record sets;
acquiring a disease knowledge graph corresponding to the target medical record set and disease knowledge graphs corresponding to the corrected medical record sets respectively, wherein the disease knowledge graph comprises a plurality of nodes and edges among the nodes, each node represents one disease in the diagnosis result of each medical record in the corresponding medical record set, and the edges between the two nodes represent the relationship between the two corresponding diseases;
and determining an optimal medical record set from the plurality of corrected medical record sets according to the disease knowledge graph corresponding to the target medical record set and the disease knowledge graphs corresponding to the plurality of corrected medical record sets respectively, wherein the optimal medical record set is used as the optimized medical record set corresponding to the target medical record set.
Optionally, obtaining a disease knowledge graph corresponding to a medical record set includes:
extracting disease names from the diagnosis results of the medical records in the medical record set to obtain a disease set consisting of the extracted disease names;
and acquiring a disease knowledge graph corresponding to the medical record set from a pre-constructed disease knowledge graph according to the disease set, wherein the pre-constructed disease knowledge graph comprises a plurality of nodes respectively representing various diseases and edges among the nodes.
Optionally, the determining an optimal medical record set from the plurality of corrected medical record sets according to the disease knowledge graph corresponding to the target medical record set and the disease knowledge graphs corresponding to the plurality of corrected medical record sets respectively includes:
for each corrected medical record set, determining a correction effect characteristic value corresponding to the corrected medical record set according to the symptom words in the medical record contained in the target medical record set, the symptom words in the medical record contained in the corrected medical record set and the disease knowledge maps respectively corresponding to the target medical record and the corrected medical record set so as to obtain correction effect characteristic values respectively corresponding to the plurality of corrected medical record sets;
and determining the corrected medical record set corresponding to the largest correction effect characteristic value in the correction effect characteristic values respectively corresponding to the plurality of corrected medical record sets as the optimal medical record set.
Optionally, the determining, according to the symptom words in the medical records included in the target medical record set, the symptom words in the medical records included in the corrected medical record set, and the disease knowledge graphs respectively corresponding to the target medical record set and the corrected medical record set, the correction effect characterization values corresponding to the corrected medical record set includes:
determining the quality score of the disease knowledge graph corresponding to the target medical record set according to the symptom words in the medical records contained in the target medical record set and according to whether the diseases are matched with the corresponding symptoms of the diseases, wherein the quality score is used as a quality characterization value of the target medical record set;
determining the quality score of the disease knowledge graph corresponding to the corrected medical record set according to the symptom words in the medical record contained in the corrected medical record set and taking the quality score as the quality characteristic value of the corrected medical record set according to whether the disease is matched with the corresponding symptom;
and determining a correction effect characteristic value corresponding to the corrected medical record set according to the quality characteristic value of the target medical record set, the quality characteristic value of the corrected medical record set and the number of corrected medical records in the corrected medical record set.
Optionally, determining the quality score of the disease knowledge graph corresponding to a medical record set according to the symptom words in the medical record included in the medical record set based on whether the disease is matched with the corresponding symptom, including:
extracting symptom words from each medical record in the medical record set, and forming a symptom word total set by the extracted symptom words;
determining a symptom distribution vector of each disease in a disease knowledge graph corresponding to the medical record set according to the symptom word total set, wherein the symptom distribution vector of a disease is composed of the disease and the co-occurrence condition characteristic values of the symptom words in the symptom word total set in the medical record set;
and determining the quality score of the disease knowledge graph corresponding to the medical record set according to the symptom distribution vector of each disease in the disease knowledge graph corresponding to the medical record set based on whether the diseases are matched with the corresponding symptoms.
Optionally, the relationship between the two diseases is one of an upper-lower relationship, an evolution relationship and an identification relationship;
the determining the quality score of the disease knowledge graph corresponding to the medical record set according to the symptom distribution vector of each disease in the disease knowledge graph corresponding to the medical record set comprises:
determining the quality score of the disease knowledge graph corresponding to the medical record set on the upper-lower relation according to the edges representing the upper-lower relation in the disease knowledge graph corresponding to the medical record set and the symptom distribution vector of the disease connected by the edges representing the upper-lower relation;
determining the quality score of the disease knowledge graph corresponding to the medical record set on the evolution relation according to the edge representing the evolution relation in the disease knowledge graph corresponding to the medical record set and the symptom distribution vector of the disease connected with the edge representing the evolution relation;
determining the quality score of the disease knowledge graph corresponding to the medical record set on the identification relationship according to the side representing the identification relationship in the disease knowledge graph corresponding to the medical record set and the symptom distribution vector of the disease connected with the side representing the identification relationship;
determining the quality score of the disease knowledge graph corresponding to the medical record set on the disease symptom number according to the disease symptom distribution vector in the disease knowledge graph corresponding to the medical record set;
and determining the quality score of the disease knowledge graph corresponding to the medical record set according to the quality scores of the disease knowledge graph corresponding to the medical record set on the upper and lower relationships, the evolution relationship, the identification relationship and the disease symptom number respectively.
Optionally, the determining, according to the side representing the upper-lower relationship in the disease knowledge graph corresponding to the medical record set and the symptom distribution vector of the disease connected to the side representing the upper-lower relationship, the quality score of the disease knowledge graph corresponding to the medical record set in the upper-lower relationship includes:
taking the side representing the upper-lower relation in the disease knowledge graph corresponding to the medical record set as a first side:
for each disease connected with each first edge, determining a symptom word set corresponding to the disease according to the state distribution vector of the disease and the symptom word set;
and determining the quality score of the disease knowledge graph corresponding to the medical record set on the upper and lower relation according to whether the symptom word set corresponding to the lower disease is a subset of the symptom word set corresponding to the upper disease in the two diseases connected by each first edge.
Optionally, the determining, according to the edge representing the evolutionary relationship in the disease knowledge graph corresponding to the medical record set and the symptom distribution vector of the disease connected to the edge representing the evolutionary relationship, the quality score of the disease knowledge graph corresponding to the medical record set on the evolutionary relationship includes:
taking the edge representing the evolutionary relationship in the disease knowledge graph corresponding to the medical record set as a second edge:
for each disease connected by each second edge, determining a symptom word set corresponding to the disease according to the state distribution vector of the disease and the symptom word set;
acquiring the intersection of the symptom word sets respectively corresponding to the two diseases connected by each second edge to obtain the common symptom word set of the two diseases connected by each second edge;
determining a symptom severity change consistency characteristic value of the two diseases connected by each second edge on the common symptom of the two diseases connected by each second edge according to the common symptom word set of the two diseases connected by each second edge;
and determining the quality score of the disease knowledge graph corresponding to the medical record set on the evolutionary relationship according to the symptom severity change consistency characteristic value of the two diseases connected by each second edge on the common symptoms of the two diseases.
Optionally, the determining, according to the side representing the identification relationship and the symptom distribution vector of the disease connected to the side representing the identification relationship in the disease knowledge graph corresponding to the medical record set, a quality characterization value of the medical record set on the identification relationship includes:
and taking the side representing the identification relationship in the disease knowledge graph corresponding to the medical record set as a third side:
determining a symptom distribution difference representation value of the two diseases connected by each third edge according to the symptom distribution vectors of the two diseases connected by each third edge;
and determining the quality score of the disease knowledge graph corresponding to the medical record set on the identification relationship according to the symptom distribution difference characterization values of the two diseases connected by each third edge.
Optionally, the determining, according to the symptom distribution vector of each disease in the disease knowledge graph corresponding to the medical record set, a quality score of the disease knowledge graph corresponding to the medical record set on the number of symptoms of the disease includes:
respectively determining the number of the symptom words in the symptom word total set, which do not belong to each disease in the disease knowledge graph corresponding to the medical record set, according to the symptom distribution vector of each disease in the disease knowledge graph corresponding to the medical record set:
and determining the quality score of the disease knowledge graph corresponding to the medical record set on the disease symptom number according to the number of the symptom words in the disease knowledge graph not belonging to each disease in the medical record set corresponding to the symptom word total set.
Optionally, the medical record set optimizing method further includes:
acquiring a disease course set from the optimal medical record set, wherein one disease course in the disease course set consists of all medical records of one patient, and all medical records in one disease course are sequenced according to the treatment time;
and determining unreasonable medical records from the courses in the course set according to whether the medical records are reasonable in the course, and correcting the determined unreasonable medical records.
Optionally, the determining unreasonable medical records from the medical processes in the medical process set based on whether the medical records are reasonable in the medical processes where the medical records are located, and correcting the determined unreasonable medical records includes:
for each course in the course set, if the diagnosis result of the medical record in the course includes a primary disease and at least one secondary disease, and the primary disease and the secondary disease satisfy three conditions, determining an unreasonable medical record for the medical record of the secondary disease according to the diagnosis result in the course, and correcting the diagnosis result of the unreasonable medical record into the primary disease:
wherein the three conditions include: the diagnosis result is that the number ratio of the medical records of the main diseases is larger than a preset main diagnosis ratio threshold; the secondary disease and the primary disease have one of a relationship among a superior-inferior relationship, an evolutionary relationship and an identification relationship; the diagnosis results are continuous medical records of the same minor disease.
Optionally, the determining unreasonable medical records from the medical processes in the medical process set based on whether the medical records are reasonable in the medical processes where the medical records are located, and correcting the determined unreasonable medical records includes:
selecting a target disease course from the disease course set, and forming the target disease course set by the selected target disease course, wherein the diagnosis result of the former part of medical records in the target disease course is a first disease, the diagnosis result of the latter part of medical records is a second disease, and the second disease is evolved from the first disease;
if the number of the disease courses in the target disease course set is larger than a preset number threshold, determining whether an unreasonable medical record exists in each disease course in the target disease course set according to the time span and the time span ratio of the first disease of each disease course in the target disease course set, and correcting the unreasonable medical record when the unreasonable medical record exists in the disease course, wherein the time span ratio of the first disease of one disease course is the ratio of the time span of the first disease of the disease course to the time span of the second disease of the disease course.
Optionally, the determining, according to the time span and the time span ratio of the first disease of each disease course in the target disease course set, whether an unreasonable medical record exists in each disease course in the target disease course set, and correcting the unreasonable medical record when an unreasonable medical record exists in a disease course, includes:
calculating a mean value of the time span of the first disease of each disease course in the target disease course set, calculating a standard deviation of the time span proportion of the first disease of each disease course in the target disease course set, and determining an acceptable time span and an acceptable time span proportion of the first disease according to the calculated mean value and standard deviation;
for each course in the set of target courses:
determining whether an unreasonable medical record exists in the course of the disease according to the time span and the time span ratio of the first disease of the course of the disease and the acceptable time span ratio of the first disease;
if the unreasonable medical record exists in the course of the disease, the unreasonable medical record in the course of the disease is determined according to the acceptable time span and the acceptable time span ratio of the first disease and the time span of the course of the disease, and the diagnosis result of the unreasonable medical record in the course of the disease is corrected into a second disease of the course of the disease.
A medical record set optimizing apparatus comprising: the system comprises a first medical record correction module, a disease knowledge graph acquisition module and an optimal medical record set determination module;
the first medical record correction module is used for correcting medical records which are not reasonably diagnosed in the target medical record set respectively by adopting a plurality of different medical record correction modes to obtain a plurality of corrected medical record sets;
the disease knowledge graph acquisition module is used for acquiring a disease knowledge graph corresponding to the target medical record set and disease knowledge graphs corresponding to the corrected medical record sets respectively, wherein the disease knowledge graph comprises a plurality of nodes and edges among the nodes, each node represents one disease in the diagnosis result of each medical record in the corresponding medical record set, and the edges between the two nodes represent the relationship between the two corresponding diseases;
and the optimal medical record set determining module is used for determining an optimal medical record set from the plurality of corrected medical record sets according to the disease knowledge graph corresponding to the target medical record set and the disease knowledge graphs corresponding to the plurality of corrected medical record sets respectively, and using the optimal medical record set as the optimized medical record set corresponding to the target medical record set.
Optionally, the medical record set optimizing apparatus further includes: the medical course set acquisition module and the second medical record correction module;
the disease course set acquisition module is used for acquiring a disease course set from the optimal medical record set, wherein one disease course in the disease course set consists of all medical records of one patient with one attack, and all medical records in one disease course are sequenced according to the treatment time;
and the second medical record correction module is used for determining unreasonable medical records from the medical processes in the medical process set according to whether the medical records are reasonable in the medical processes in which the medical records are positioned, and correcting the determined unreasonable medical records.
A medical record set optimizing device comprising: a memory and a processor;
the memory is used for storing programs;
the processor is configured to execute the program to implement each step of the medical record set optimizing method described in any one of the above.
A readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the medical record set optimizing method according to any one of the above.
According to the above scheme, the medical record set optimization method provided by the application comprises the steps of firstly adopting a plurality of different medical record correction modes to respectively correct medical records which are unreasonably diagnosed in a target medical record set to obtain a plurality of corrected medical record sets, then obtaining a disease knowledge graph corresponding to the target medical record set and disease knowledge graphs corresponding to the corrected medical record sets, and finally determining an optimal medical record set from the corrected medical record sets according to the disease knowledge graph corresponding to the target medical record set and the disease knowledge graphs corresponding to the corrected medical record sets respectively to be used as the optimized medical record set corresponding to the target medical record set, so that the medical record set optimization method provided by the application can realize the optimization of the medical record sets, and the optimized medical record set corresponding to the target medical record set is the optimal medical record set in the corrected medical record sets obtained by adopting the plurality of correction modes to correct the target medical record set, therefore, the optimized medical record set corresponding to the target medical record set has good quality and high reliability, and subsequent analysis and mining are performed based on the medical record set, so that good analysis and mining effects can be obtained.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
At present, some schemes for finding suspected misdiagnosed medical records from a medical record set exist, and some schemes for automatically giving diagnosis and correction opinions also exist, wherein the schemes for finding the suspected misdiagnosed medical records are mostly outlier mining schemes based on statistics, the general idea of the schemes is to vectorize texts of the medical records in the medical record set, then use a statistical method to find outlier vectors from numerous vectors, the medical records corresponding to the outlier vectors are the suspected misdiagnosed medical records, and the schemes for automatically giving diagnosis and correction suggestions mainly include that a plurality of similar medical records are calculated and most of the medical records in diagnosis are used as correction suggestions.
The inventor of the present invention finds out through research on the two methods: the scheme for finding out the suspected misdiagnosed medical record does not fully excavate the medical information behind the medical record, but only uses a general statistical method, so that the accuracy is not high, and the interpretability is relatively poor. The proposal for automatically giving the diagnosis and correction suggestions has low accuracy, for example, some diagnoses are very similar and need to be distinguished by different diagnosis names (namely disease names), the different diagnosis names needing to be distinguished often cause confusion in clinic, even partial misdiagnosis medical records can be caused by the confusion, so that the accuracy cannot be effectively guaranteed by adopting a method of similar medical records for correcting the misdiagnosis medical records.
Based on the findings, the inventor of the present invention has conducted an in-depth study, and in the study process, medical information behind medical records is fully mined, and at the same time, it is found that, in addition to the fact that medical records may be unreasonable, when the medical records are reasonable, the situation that medical records are unreasonable in the whole course of a patient may also exist.
The medical record set optimizing method provided by the application can be applied to a terminal with data processing capacity (such as a PC, a notebook computer, a smart phone, a PAD and the like) and can also be applied to a server (which can be a server, a plurality of servers or a server cluster), and the terminal or the server with data processing capacity can acquire and optimize the medical record set. The medical record set optimizing method provided by the present application is described below by the following embodiments.
First embodiment
Referring to fig. 1, a schematic flow chart of a medical record optimizing method provided in an embodiment of the present application is shown, where the method may include:
step S101: and correcting the medical records which are not reasonably diagnosed in the target medical record set respectively by adopting a plurality of different medical record correction modes to obtain a plurality of corrected medical record sets.
The target medical record set is a medical record set to be optimized, most medical records in the target medical record set are medical records with reasonable diagnosis results, and the minority medical records are medical records with unreasonable diagnosis results. It should be noted that an unreasonable diagnosis result of a medical record means that the diagnosis result of the medical record does not match the symptom described in the medical record.
The medical record correction method in the present application may be a medical record correction method based on a diagnosis prediction algorithm, that is, a medical record prediction algorithm is used to predict the diagnosis result of each medical record in a medical record set, the predicted diagnosis result is compared with the diagnosis result recorded in the corresponding medical record, and if the two diagnosis results are not the same, the diagnosis result recorded in the medical record is corrected to the predicted diagnosis result.
It should be noted that the various medical record correction methods mentioned in this embodiment exist in two cases: the first situation is that the diagnosis prediction algorithms adopted by the medical record correction modes are different, for example, the target medical record set is corrected by adopting three medical record correction modes, namely a medical record correction mode based on a diagnosis prediction algorithm 1, a medical record correction mode based on a diagnosis prediction algorithm 2 and a medical record correction mode based on a prediction algorithm 3; in the second case, the same diagnosis prediction algorithm is used for each medical record correction mode, but the parameters of the diagnosis prediction algorithm are different in each medical record correction mode, for example, the diagnosis prediction algorithm is an algorithm based on a neural network model, and the parameters of the neural network model are different in each medical record correction mode, so that it can be understood that the diagnosis prediction effect is different due to the different parameters of the diagnosis prediction algorithm, and further the correction effect is different.
Step S102: and acquiring a disease knowledge graph corresponding to the target medical record set and disease knowledge graphs corresponding to the plurality of corrected medical record sets respectively.
The disease knowledge graph comprises a plurality of nodes and edges between the nodes, one node represents one disease in the diagnosis result of each medical record in the corresponding medical record set, and the edge between two nodes represents the relationship between two corresponding diseases. Illustratively, if the medical records in a medical record set have a total of 50 diseases in the diagnosis result, the disease knowledge graph corresponding to the medical record set includes 50 nodes, and one node represents a disease.
Step S103: and determining an optimal medical record set from the plurality of corrected medical record sets as the optimized medical record set corresponding to the target medical record set according to the disease knowledge graph corresponding to the target medical record set and the disease knowledge graphs corresponding to the plurality of corrected medical record sets respectively.
Specifically, the quality score of the disease knowledge graph corresponding to the target medical record set can be determined according to the target medical record set, the quality score of the disease knowledge graph corresponding to each corrected medical record set can be determined according to each corrected medical record set, so that the score of the disease knowledge graph corresponding to the target medical record set and the scores of the disease knowledge graphs corresponding to the corrected medical record sets are obtained, and then the optimal medical record set can be determined from the corrected medical record sets according to the score of the disease knowledge graph corresponding to the target medical record set and the scores of the disease knowledge graphs corresponding to the corrected medical record sets. It should be noted that the higher the score of the disease knowledge graph is, the better the disease knowledge graph is, and the better the disease knowledge graph is, the better the quality of the corresponding medical record set is.
The medical record set optimizing method provided by the embodiment of the application comprises the steps of firstly adopting a plurality of different medical record correction modes to respectively correct medical records which are not reasonably diagnosed in a target medical record set to obtain a plurality of corrected medical record sets, then obtaining a disease knowledge graph corresponding to the target medical record set and disease knowledge graphs corresponding to the plurality of corrected medical record sets respectively, and finally determining an optimal medical record set from the plurality of corrected medical record sets according to the disease knowledge graph corresponding to the target medical record set and the disease knowledge graphs corresponding to the plurality of corrected medical record sets respectively to be used as the optimized medical record set corresponding to the target medical record set, so that the medical record set optimizing method provided by the embodiment of the application can realize the optimization of the medical record sets, and the optimized medical record set corresponding to the target medical record set is the optimal medical record set in the plurality of corrected medical record sets obtained by adopting the plurality of medical record correction modes to correct the target medical record set, therefore, the optimized medical record set corresponding to the target medical record set has good quality and high reliability, and subsequent analysis and mining are performed based on the medical record set, so that good analysis and mining effects can be obtained.
Second embodiment
This embodiment is similar to the "step S102: and acquiring a disease knowledge graph corresponding to the target medical record set and a disease knowledge graph corresponding to each of the plurality of corrected medical record sets for introduction.
Since the process of acquiring the disease knowledge graph corresponding to each medical record set is similar, this embodiment takes one medical record set as an example, and introduces the process of acquiring the disease knowledge graph corresponding to the medical record set.
Referring to fig. 2, a flow chart of acquiring a disease knowledge graph corresponding to a medical record set is shown, which may include:
step S201: the disease names are extracted from the diagnosis results of the medical records in the medical record set to obtain a disease set consisting of the extracted disease names.
Since there are usually many medical records in a medical record set, there is a high possibility that the disease names in the diagnosis results of two or more medical records are the same disease name, and therefore, after the disease names are extracted from the diagnosis results of the medical records, deduplication processing can be performed, and the disease names after deduplication processing are combined into a disease set.
Step S202: and acquiring a disease knowledge graph corresponding to the medical record set from a pre-constructed disease knowledge graph according to the disease set.
The embodiment pre-constructs a disease knowledge graph covering various diseases as much as possible, and the disease knowledge graph comprises a plurality of nodes respectively representing various diseases and edges between the nodes. In the above embodiments, it is mentioned that the edge between two nodes represents the relationship between two corresponding diseases, and it should be noted that the relationship between two diseases may be one of a top-bottom relationship, an evolution relationship and an identification relationship.
In the above-lower relationship, if the medical meaning of disease a completely covers the medical meaning of disease B, disease a and disease B are said to have the upper-lower relationship, and disease a is the upper-level disease of disease B and disease B is the lower-level disease of disease a, for example, "pneumonia" is the upper-level disease of "chronic pneumonia".
Regarding the evolutionary relationship, if the disease a is likely to develop into the disease B and the disease B is unlikely to develop into the disease a within the same disease period of the same patient, the disease a and the disease B have an evolutionary relationship and the disease a can evolve into the disease B or the disease B can evolve from a.
For identification, if the clinical manifestations of disease a and disease B are highly similar but different in one or more clinical manifestations, and usually they need to be distinguished according to the one or more clinical manifestations, a and B are said to have an identification relationship, e.g., "myocardial infarction" and "aortic dissection", which usually need to be carefully distinguished.
Referring to fig. 3, a schematic diagram of an example of a disease knowledge graph is shown, where nodes in the graph represent diseases, a connection line between two nodes is an edge representing a relationship between two diseases, an undirected edge represents an identification relationship, a dashed directed edge represents an evolutionary relationship, an arrow of a dashed directed edge points to a later disease, for example, if a disease B evolves from a disease a, an arrow of a directed edge between the two points to the disease B, a solid directed edge represents an up-down relationship, and an arrow of a directed edge points to an upper disease, for example, if the disease a is an upper disease of the disease B, an arrow of a directed edge between the two points to the disease a. It should be noted that the disease knowledge map shown in fig. 3 is merely an example, and is not intended to represent the disease knowledge map.
After the disease set is obtained in step S201, the disease knowledge graph corresponding to the medical record set is obtained from the pre-constructed disease knowledge graph according to the disease set, specifically, first, the nodes representing the diseases in the disease set in the pre-constructed disease knowledge graph are lit, then the edges connected with two lit nodes are lit, and finally, the unlit nodes and the unlit edges are deleted, and the remaining part is the disease knowledge graph corresponding to the medical record set.
And respectively processing the target medical record set and the plurality of corrected medical record sets according to the mode to obtain the disease knowledge graph corresponding to the target medical record set and the disease knowledge graph corresponding to the plurality of corrected medical record sets.
Third embodiment
This embodiment is similar to the "step S103: and determining an optimal medical record set from the plurality of corrected medical record sets according to the disease knowledge graph corresponding to the target medical record set and the disease knowledge graphs corresponding to the plurality of corrected medical record sets respectively, and introducing the optimal medical record set as the optimized medical record set corresponding to the target medical record set.
Referring to fig. 4, a flowchart illustrating a specific implementation process of step S103 may include:
step S401, for each corrected medical record set, according to the symptom words in the medical records contained in the target medical record set, the symptom words in the medical records contained in the corrected medical record set, and the disease knowledge maps respectively corresponding to the target medical record and the corrected medical record set, determining the correction effect characteristic values corresponding to the corrected medical record set to obtain the correction effect characteristic values respectively corresponding to the plurality of corrected medical record sets.
The process of determining the characteristic value of the correction effect corresponding to the corrected medical record set according to the symptom word in the medical record contained in the target medical record set, the symptom word in the medical record contained in the corrected medical record set, and the disease knowledge graph corresponding to the target medical record set and the corrected medical record set respectively may include:
step S4011a, based on whether the disease and the corresponding symptom are matched, determining a quality score of the disease knowledge graph corresponding to the target medical record set according to the symptom word in the medical record included in the target medical record set, as a quality characterization value of the target medical record set.
Step S4011b, based on whether the disease and the corresponding symptom are matched, determining a quality score of the disease knowledge graph corresponding to the corrected medical record set according to the symptom word in the medical record included in the corrected medical record set, and using the quality score as a quality characterization value of the corrected medical record set.
The method and the device for evaluating the medical record set based on the symptom words in the medical record set start from the angle that whether the diseases are matched with the corresponding symptoms or not, and evaluate the disease knowledge graph corresponding to the medical record set according to the symptom words in the medical record set.
It should be noted that the quality score of the disease knowledge graph corresponding to a medical record set can reflect the quality of the disease knowledge graph corresponding to the medical record set, the higher the quality score of the disease knowledge graph corresponding to the medical record set is, the better the disease knowledge graph corresponding to the medical record set is, and the better the disease knowledge graph corresponding to the medical record set is, the better the quality of the medical record set is, the higher the reliability is, for this reason, the quality score of the disease knowledge graph corresponding to the medical record set is taken as the quality characterization value of the medical record set.
And S4012, determining a correction effect characteristic value corresponding to the corrected medical record set according to the quality characteristic value of the target medical record set, the quality characteristic value of the corrected medical record set and the number of corrected medical records in the corrected medical record set.
Assuming that the target medical record set is B, the corrected medical record set is Bi' (i is 1,2, …, M is the total number of corrected medical records), the quality Score of the disease knowledge graph corresponding to the target medical record set (i.e. the quality characterization value of the target medical record set) is Score, and the corrected medical record set B isi' quality scores of the corresponding disease profiles are Scorei' (i.e., corrected medical record set B)i' quality characterization value of), the corrected medical record set Bi' corresponding correction effect characterization values are:
Scorei′-Score-×numi correction of medical record(1)
Wherein, numCorrection of medical recordIs BiIn the point of' correction of the number of medical records, since the vast majority of B are medical records with reasonable diagnosis results, the corrected medical records should not be too many, from which point of view, in ScoreiThe number of corrected medical records is subtracted on the basis of' -Score as a penalty for the number of corrected medical records.
And S402, determining the corrected medical record set corresponding to the largest correction effect characteristic value in the correction effect characteristic values respectively corresponding to the plurality of corrected medical record sets as the optimal medical record set.
Will max (Score)1′-Score-×num1 Correction of medical record,Score2′-Score-×num2 Correction of medical record,…,ScoreM′-Score-×numM Correction of medical record) And determining the corresponding corrected medical record set as the optimal medical record set.
Fourth embodiment
As can be seen from the third embodiment, in the process of determining the optimal medical record set, the score of the disease knowledge graph corresponding to the target medical record set and the score of the disease knowledge graph corresponding to each corrected medical record set need to be determined.
Referring to fig. 5, a schematic flow chart illustrating a process of determining a quality score of a disease knowledge graph corresponding to a medical record set according to symptom words in medical records included in the medical record set is shown, and the process may include:
step S501: and (4) extracting symptom words from each medical record in the medical record set, and forming a symptom word total set by the extracted symptom words.
The symptom words extracted from each medical record in the medical record set may be repeated, therefore, after all symptom words are extracted, the duplication removing processing can be carried out, and the symptom words after the duplication removing processing form a symptom word total set { symp }1symp2,…,sympNAnd N is the total number of symptom words contained in the symptom word total set. Preferably, the symptom word set { symp }1symp2,…,sympNThe symptom words in the Chinese characters can be ranked in a certain order, for example, in a Pinyin order.
Step S502: and determining the symptom distribution vector of each disease in the disease knowledge graph corresponding to the medical record set according to the symptom word total set.
The disease symptom distribution vector is composed of the disease and the co-occurrence condition characteristic values of the symptom words in the symptom word total set in the medical record included in the medical record set.
Specifically, for each disease d in the disease knowledge graph corresponding to the medical record set: firstly, respectively determining the co-occurrence frequency of each symptom word in the disease d and symptom word general set in the medical record set, namely obtaining the cococurd,1、cooccurd,2、…、cooccurd,N(ii) a Then, by coccurd,1~cooccurd,NForming a vector as a co-occurrence vector corresponding to the disease d; then, the co-occurrence vector corresponding to disease d is normalized according to the following formula:
normalizing the co-occurrence vector corresponding to the disease d to obtain the element of ratiod,1、ratiod,2、…、ratiod,NThe vector of (4) is used as a symptom distribution vector of the disease d.
Step S503: and determining the quality score of the disease knowledge graph corresponding to the medical record set according to the symptom distribution vector of each disease in the disease knowledge graph corresponding to the medical record set.
Specifically, the process of determining the quality score of the disease knowledge graph corresponding to the medical record set according to the symptom distribution vector of each disease in the disease knowledge graph corresponding to the medical record set may include:
step S5031 a: and determining the quality score of the disease knowledge graph corresponding to the medical record set on the upper-lower relation according to the edges representing the upper-lower relation in the disease knowledge graph corresponding to the medical record set and the symptom distribution vector of the disease connected by the edges representing the upper-lower relation.
Step S5031 b: and determining the quality score of the disease knowledge graph corresponding to the medical record set on the evolution relation according to the edge representing the evolution relation in the disease knowledge graph corresponding to the medical record set and the symptom distribution vector of the disease connected with the edge representing the evolution relation.
Step S5031 c: and determining the quality score of the disease knowledge graph corresponding to the medical record set on the identification relationship according to the side representing the identification relationship in the disease knowledge graph corresponding to the medical record set and the symptom distribution vector of the disease connected with the side representing the identification relationship.
Step S5031 d: and determining the quality score of the disease knowledge graph corresponding to the medical record set on the disease symptom number according to the symptom distribution vector of each disease in the disease knowledge graph corresponding to the medical record set.
It should be noted that, in this embodiment, the execution order of S5031a to S5031d is not limited, and S5031a to S5031d may be executed in any sequential order or in parallel.
Step S5032: and determining the quality score of the disease knowledge graph corresponding to the medical record set according to the quality scores of the disease knowledge graph corresponding to the medical record set on the upper and lower relationships, the evolution relationship, the identification relationship and the disease symptom number respectively.
In a possible implementation manner, the quality scores of the disease knowledge graph corresponding to the medical record set on the upper-lower relationship, the evolution relationship, the identification relationship and the disease symptom number can be directly summed, and the summed score is used as the quality score of the disease knowledge graph corresponding to the medical record set.
In another possible implementation manner, weights may be preset respectively for the upper-lower relationship, the evolutionary relationship, the identification relationship, and the number of symptoms of a disease, then quality scores of the disease knowledge graph corresponding to the medical record set on the upper-lower relationship, the evolutionary relationship, the identification relationship, and the number of symptoms of the disease are weighted and summed according to the preset weights, and the weighted and summed Score is used as the quality Score of the disease knowledge graph corresponding to the medical record setKnowledge map of diseaseNamely:
Scoreknowledge map of disease=α×ScoreUpper and lower parts+β×ScoreEvolution of+γ×ScoreAuthentication+×ScoreNumber of symptoms(3)
α, β and gamma are weights set for upper and lower relationships, evolutionary relationships, discriminatory relationships and the number of symptoms of diseases in turn, α, β and gamma are set for practical application, and Score is setUpper and lower parts、ScoreEvolution of、ScoreAuthentication、ScoreNumber of symptomsThe quality scores of the disease knowledge graph corresponding to the medical record set in the upper-lower relation, the quality score in the evolution relation, the quality score in the identification relation and the quality score in the disease symptom number are sequentially obtained.
Fifth embodiment
The present embodiment describes a specific implementation process of steps S5031 a-S5031 d in the above embodiment.
First, a "step S5031 a: determining the quality score of the disease knowledge graph corresponding to the medical record set in the upper-lower relationship according to the edges representing the upper-lower relationship in the disease knowledge graph corresponding to the medical record set and the symptom distribution vector of the disease connected by the edges representing the upper-lower relationship, which may include:
step a1, regarding the side representing the upper and lower relation in the disease knowledge graph corresponding to the medical record set as the first side, and for each disease connected by each first side, determining the symptom word set corresponding to the disease according to the state distribution vector and the symptom word total set of the disease, so as to obtain the symptom word set corresponding to each disease connected by each first side.
For each disease connected by each first edge, acquiring symptom words corresponding to elements which are not 0 in the symptom distribution vector of the disease from the symptom word total set, and forming a symptom word set corresponding to the disease to obtain a symptom word set corresponding to each disease connected by each first edge.
Step a2, determining the quality score of the disease knowledge graph corresponding to the medical record set on the upper and lower relationship according to whether the symptom word set corresponding to the lower disease is the subset of the symptom word set corresponding to the upper disease in the two diseases connected by each first edge.
It should be noted that the symptoms of the lower disease should be a subset of the upper disease, but medical records with unreasonable diagnostic results may disrupt this relationship. Under the condition of a large number of medical records, symptoms corresponding to each disease can be fully exposed, and the upper-level diseases completely cover the medical meanings of the lower-level diseases, so that the symptom set of the medical records is necessarily a superset of the symptom word set of the lower-level diseases, while a few medical records with unreasonable diagnosis results possibly introduce completely irrelevant symptoms for the lower-level diseases, and the completely irrelevant symptoms can not be contained in the upper-level diseases, so that the lower-level disease symptom set is not a subset of the upper-level disease symptom set any more.
In view of the above, the present application scores the quality of the disease knowledge graph corresponding to the medical record set in the upper-lower relationship based on whether the symptom word set corresponding to the lower disease is a subset of the symptom word set corresponding to the upper disease in the two diseases connected by the first edge (edge representing the upper-lower relationship).
Specifically, for two diseases connected at each first edge, if the symptom word set corresponding to the lower disease is a subset of the symptom word set corresponding to the upper disease, determining that the quality characteristic value of the disease knowledge graph corresponding to the medical record set on the first edge is 1, otherwise, determining that the quality characteristic value of the disease knowledge graph corresponding to the medical record set on each first edge is 0, so as to obtain the quality characteristic value of the disease knowledge graph corresponding to the medical record set on each first edge; then summing the quality characterization values of the disease knowledge graph corresponding to the medical record set on each first edge, and determining the ratio of the summed value to the total number of the first edges as the Score of the disease knowledge graph corresponding to the medical record set in the upper-lower relationUpper and lower parts。
Next, "step S5031 b: determining the implementation process of the quality score of the disease knowledge graph corresponding to the medical record set on the evolutionary relationship according to the edge representing the evolutionary relationship in the disease knowledge graph corresponding to the medical record set and the symptom distribution vector of the disease connected with the edge representing the evolutionary relationship, which may include:
and b1, taking the edge representing the evolutionary relationship in the disease knowledge graph corresponding to the medical record set as a second edge, and determining a symptom word set corresponding to the disease according to the state distribution vector and the symptom word total set of the disease aiming at each disease connected by each second edge.
For each disease connected by each second edge, the symptom words corresponding to the elements not being 0 in the symptom distribution vector of the disease can be obtained from the symptom word total set to form a symptom word set corresponding to the disease, so as to obtain a symptom word set corresponding to each disease connected by each second edge.
And b2, acquiring the intersection of the symptom word sets respectively corresponding to the two diseases connected by each second edge to obtain the common symptom word set of the two diseases connected by each second edge.
The common symptom words of the two diseases connected by the second edge can be obtained by solving the intersection of the symptom word sets respectively corresponding to the two diseases connected by the second edge.
And b3, determining the symptom severity change consistency characteristic value of the two diseases connected by each second edge on the common symptom of the two diseases connected by each second edge according to the common symptom word set of the two diseases connected by each second edge.
It should be noted that if disease a can evolve into disease B, the severity of each symptom shared by disease a and disease B should change relatively uniformly as disease a evolves into disease B. This is because, the disease evolution is a medical process driven by a certain medical mechanism, rather than a random process, and clinical manifestations such as symptoms are usually external reflections of the medical mechanism, so that the change trend of the severity of a single symptom should be relatively consistent when the disease evolves. For example, the "cold" may evolve into "pneumonia" if not treated in time, which is usually accompanied by an exacerbation of symptoms such as "cough", but not manifested as an exacerbation of "cough" in some patients and a reduction of "cough" in others, essentially because "cough" is an external reflection of a dysfunction of the human body, particularly a dysfunction of the respiratory system, rather than a random phenomenon. It is emphasized that, in relation to the disease progression, the trends in severity between symptoms are not necessarily synchronized. For example, "cold" evolves into "pneumonia," which is usually aggravated by "cough", "weakness", "chest distress", etc., while "runny nose" is alleviated. This is because each disease has its own pathogenesis and characteristics, and even if disease a evolves into disease B, which is an exacerbation, this does not represent an exacerbation of all common symptoms.
However, medical records with unreasonable diagnostic results may impair the consistency of the change in symptom severity over the common symptoms of both diseases that have an evolutionary relationship. For example, most of the medical records in the medical records with reasonable diagnosis results reflect that the "cold" is generally aggravated by "cough" when evolving into "pneumonia", and if a patient evolves from the "cold" into "pneumonia" within a certain course of disease but "cough" is relieved, the medical records reflecting the evolution from the "cold" into "pneumonia" weaken the overall consistency level.
Based on this, under the same other conditions, for a disease knowledge graph, the more consistent the change trend of the severity of common symptoms corresponding to the evolution among diseases is, the better the disease knowledge graph is.
Assuming that two diseases connected by a second side i are respectively disease a and disease B, and disease B can evolve from disease a, the consistency characteristic value of the change of symptom severity of disease a and disease B on the common symptom words of both diseases can be determined according to the consistency characteristic value of the change of symptom severity of disease a and disease B on each symptom word in the common symptom word set of both diseases, wherein the consistency characteristic value fit (S, a, B) of the change of symptom severity of disease a and disease B on one symptom word S in the common symptom word set S of both diseases can be determined by the following formula:
wherein, numThe same course evolves from A to B and s aggravatesThe number of medical record pairs which satisfy the following conditions in the medical record set is as follows: the two medical records are medical records of the same patient in one course of disease; the diagnosis result of the medical record with early treatment time in the two medical records is disease A, and the diagnosis result of the medical record with late treatment time is disease B; both cases contain the symptomatic word s, and s becomes worse, num when disease A evolves into disease BSame course evolves from A to B and s alleviatesThe meaning of (A) is similar, and the description is omitted here.
It should be noted that, a disease course mentioned in the present application is composed of all medical records of a patient with one disease occurrence, all medical records in a disease course are ordered according to the treatment time, and the medical records in a disease course should satisfy the following two conditions: condition 1, two diseases arbitrarily adjacent to each otherThe calendar date interval is less than a preset date interval threshold, i.e. datadiff (med)i,medi+1)<thresholdDate intervalWherein datadiff (med)i,medi+1) Is the time interval between the ith medical record and the (i + 1) th medical record, thresholdDate intervalIs a date interval threshold; in condition 2, the similarity between any two adjacent medical records is greater than a preset similarity threshold, i.e., similarity (med)i,medi+1)>thresholdDegree of similarityWherein, similarity (med)i,medi+1) Is the similarity of the ith medical record and the (i + 1) th medical record, thresholdDegree of similarityIs a similarity threshold.
After obtaining the fit (s, a, B), it may be determined whether the fit (s, a, B) is greater than a preset symptom severity change threshold, and if so, it is determined that the contribution value of the symptom word s on the consistency of the symptom severity change is equal to consistency of the symptom severity change, that is, the contribution value is equal to 1, otherwise, the contribution value is equal to 0.
The contribution value of each symptom word in the common symptom word set S of the disease A and the disease B on the consistency of the change of the symptom severity can be obtained through the process, then the contribution values of all the symptom words in the common symptom words S of the disease A and the disease B on the consistency of the change of the symptom severity are summed, and the ratio of the summed value and the total number of the symptom words contained in the common symptom word set S is used as the representation value of the change of the symptom severity of the disease A and the disease B on the common symptoms.
Step b4, determining the quality score of the disease knowledge graph corresponding to the medical record set on the evolutionary relationship according to the symptom severity change consistency characteristic value of the two diseases connected by each second edge on the common symptoms of the two diseases.
Specifically, the symptom severity change consistency characteristic values of the two diseases connected by each second edge on the common symptoms of the two diseases can be summed, the ratio of the summed value and the total number of the second edges is determined as Score of the disease knowledge graph corresponding to the medical record set on the evolution relationEvolution of。
Next, "step S5031 c: determining the quality score of the disease knowledge graph corresponding to the medical record set in the identification relationship according to the side representing the identification relationship in the disease knowledge graph corresponding to the medical record set and the symptom distribution vector of the disease connected with the side representing the identification relationship, which may include:
and c1, taking the side representing the identification relationship in the disease knowledge graph corresponding to the medical record set as a third side, and determining the symptom distribution difference representation value of the two diseases connected by the third side according to the symptom distribution vector of the two diseases connected by each third label.
It should be noted that, if the disease a and the disease B have a differential relationship, the symptom distributions of the two diseases are highly similar but still appear differently because the disease a and the disease B are not the same disease after all, and the differential relationship exists, which means that the two diseases can be distinguished by one or more symptoms, therefore, the symptom distributions of the two diseases are not completely the same and the two diseases can be actually distinguished by the difference, but the difference can be weakened by a case history with an unreasonable diagnosis result.
In view of this, the present application assesses the quality score of a disease knowledge graph on discriminatory relationships by the differences in symptom distribution among diseases with discriminatory relationships. For a disease knowledge graph, the greater the difference in symptom distribution among diseases having an identification relationship, the better the disease knowledge graph, and correspondingly, the higher the quality score of the disease knowledge graph on the identification relationship.
The symptom distribution of a disease can be characterized by the symptom distribution vector of the disease, based on which, the symptom distribution difference characteristic value of two diseases connected by each third edge can be determined according to the symptom distribution vector of two diseases connected by each third edge, specifically, the standard deviation of the symptom distribution vector of two diseases connected by each third edge can be calculated as the symptom distribution difference characteristic value, so that the symptom distribution difference characteristic value of two diseases connected by each third edge can be obtained.
And c2, determining the quality score of the disease knowledge graph corresponding to the medical record set on the identification relationship according to the symptom distribution difference characterization values of the two diseases connected by each third edge.
Specifically, can be prepared bySumming the symptom distribution difference characterization values of the two diseases connected by the third edges, determining the ratio of the sum obtained value to the total number of the third edges as a quality Score of the disease knowledge graph corresponding to the medical record set in the discrimination relationshipAuthentication。
Finally, a "step S5031 d: and determining the specific implementation process of the quality score of the disease knowledge graph corresponding to the medical record set on the disease symptom number according to the disease symptom distribution vector in the disease knowledge graph corresponding to the medical record set.
It should be noted that the number of different symptom words of a disease is not too large, however, a medical record with an unreasonable diagnosis may result in a disease with more different symptom words.
The number of different symptoms of a disease is not too large because although there may be a plurality of causes to induce the same disease, and thus a disease may theoretically correspond to several or even dozens of different symptoms, each disease has a limited range after all, and the corresponding symptoms are not too numerous, for example, the symptoms of "pneumonia" are limited to the respiratory tract system, and at most extend to the symptoms of "weakness" and the like without specific organ direction, and the symptoms of "abdominal pain", "numbness of lower limbs" and the like are never generated.
However, a few medical records with an unreasonable diagnosis may introduce completely unrelated symptoms for the current disease, so that the disease is associated with an upsurge in the number of different symptoms, for example, if one medical record has symptoms "cough", "strength", "chest distress" but is misdiagnosed as diabetes, then these unrelated symptoms are directly introduced for diabetes. It should be noted that, under the assumption of "massive medical records", it is reasonable to assume that the corresponding symptoms of each disease are fully exposed, and the case of "not appearing due to small data volume" is excluded.
Based on the above analysis, it is known that, under the same other conditions, the less the number of different symptoms of a disease is, the better the disease profile is.
In view of this, the present application proposes that the quality score of the disease knowledge graph corresponding to the medical record set on the number of symptoms of the disease can be determined according to the symptom distribution vector of each disease in the disease knowledge graph corresponding to the medical record set as follows:
and d1, respectively determining the number of the symptom words in the symptom word total set, which do not belong to each disease in the disease knowledge graph corresponding to the medical record set, according to the symptom distribution vector of each disease in the disease knowledge graph corresponding to the medical record set.
Wherein, the symptom word total set is a set composed of symptom words extracted from the medical records contained in the medical record set.
Specifically, for each disease in the disease knowledge graph corresponding to the medical record set, the number of elements which are 0 in the symptom distribution vector of the disease is counted, and the number is used as the number of symptom words which do not belong to the disease in the symptom word total set.
And d2, determining the quality score of the disease knowledge graph corresponding to the medical record set on the disease symptom number according to the number of the symptom words in the symptom word total set, which do not belong to each disease in the disease knowledge graph corresponding to the medical record set.
Specifically, for each disease in the disease knowledge graph corresponding to the medical record set, calculating the ratio of the number of symptom words not belonging to the disease in the symptom word total set to the total number of symptom words contained in the symptom word total set, and taking the ratio as the number of non-symptom words corresponding to the disease; the number ratios of the non-symptom words corresponding to each disease are summed up, and the sum value is used as the quality Score of the disease knowledge graph corresponding to the medical record set on the symptom number of the diseaseNumber of symptoms。
Sixth embodiment
When the embodiment corrects the medical records in the target medical record set, starting from a single medical record, and considering whether the single medical record is reasonable or not, an unreasonable medical record is corrected, however, there is a special clinical scene:
some medical records are reasonable by themselves when viewed alone, but are unreasonable when they are put into a certain course of a certain patient, that is, there may be some unreasonable medical records in the "optimal medical record set" in step S103 of the above embodiment, and in order to further improve the quality of the medical record set, the following scheme is proposed in the present application:
and step S103, acquiring a course set from the optimal medical record set obtained in the step S103, determining unreasonable medical records from the courses in the course set according to whether the medical records are reasonable in the course, and correcting the unreasonable medical records.
One common clinical situation is: for the tentative treatment of complex conditions.
The tentative treatment of a complex condition means that most of medical records of a patient are medical records with a diagnosis result of disease a in a longer course of disease, a small number of continuous medical records with a diagnosis result of disease b are interspersed, and disease b can be tentatively treated in the treatment process on the premise that the exclusion is different from the synonym of disease name.
Clinically, the reason for this is often that the effect of the treatment of disease a is not as expected, and to exclude the possibility of "suffering from other diseases", disease b is diagnosed briefly and treated as disease b, and then, due to the worse effect or other medical indications, it is confirmed that disease b is not, and then disease a is returned. Therefore, the medical record with the diagnosis result of non-disease a is unreasonable medical record, and the diagnosis result thereof is disease a.
It should be noted that the disease b is usually a disease that cannot be eliminated by a routine examination such as a blood drawing test, and in addition, in a real case, it may be necessary to eliminate not only one possible disease b but also the diseases c and d, that is, a small and continuous medical record with the diagnosis result of the disease c and a small and continuous medical record with the diagnosis result of the disease d may be interspersed in the course of the disease.
It is emphasized that, in the above-described case of excluding other diseases by heuristic treatment, the medical records of the disease to be excluded should be grouped together during the course of the disease, and since the medical records are ordered in time, the form of the course should be { …, m }b,mb,mb,mb,…}(mbMedical record for disease b as a diagnostic result) The case history that the diagnosis result is disease a or the case history that the diagnosis result is other diseases is inserted in the case history that the diagnosis result is disease b does not appear, and the disease a and the disease b are directly related medically and are not unrelated diseases.
In view of the above situation, the implementation process of determining unreasonable medical records from the medical processes in the medical process set and correcting the determined unreasonable medical records based on whether the medical records are reasonable in the medical processes in which the medical records are located can include:
for each course in the course set, if the diagnosis result of the medical record in the course includes a primary disease and at least one secondary disease, and the primary disease and the secondary disease satisfy three conditions, determining the medical record with the diagnosis result of the medical record in the course as the secondary disease as the unreasonable medical record, and correcting the diagnosis result of the unreasonable medical record as the primary disease.
The three conditions described above include: the diagnosis result is that the number ratio of the medical records of the main diseases is larger than a preset main diagnosis ratio threshold; the secondary disease and the primary disease have one of the relationship of up-down relationship, evolution relationship and identification relationship; the diagnosis results are continuous medical records of the same minor disease. The first condition is that most of the medical records in the course of disease should be the medical records with the diagnosis result of the main disease, the second condition is that the disease to be eliminated should have medical relation with the main disease during the tentative treatment, and the third condition is that the medical records with the diagnosis result of the same secondary disease should be close together, and the diagnosis result of other diseases should not be interspersed in the middle.
Another common clinical situation is: delayed findings with respect to disease progression status.
The delayed discovery of the disease evolution condition means that a patient gradually evolves from a disease a to another disease b in a certain course of a disease due to some reason (for example, the condition is aggravated due to poor treatment, the curative effect is good, but the disease is not completely cured directly, and the like), and the patient usually knows the afterward after the disease evolves clinically, and usually realizes the evolution after the poor treatment effect is discovered or a new symptom is discovered and derived.
It can be understood that the patient gradually evolves from disease a to disease b in a certain course, and the concept of time is inevitably involved, and the disease knowledge map does not consider the time dimension, where the time dimension is an extrinsic reflection and measure of the change of physical function and health condition, although for a specific patient, the time point of the disease course is objectively unknown in which patient evolves from disease a to disease b, but from the big data statistics, the duration of the disease evolution has an approximate range, if the disease course does not significantly meet the range, it can be reasonably speculated that the disease course delays to find the evolution of the disease, that is, in the case of the diagnosis result being a, part of the medical history later is unreasonable, because during this period, the patient has turned to disease b, based on this, the diagnosis result of such unreasonable medical records can be corrected to disease b.
In view of this, the implementation process of determining the unreasonable medical records from the medical processes in the medical process set and correcting the determined unreasonable medical records based on whether the medical records are reasonable in the medical processes may include:
and e1, selecting a target course from the course set, and forming the target course set by the selected target course.
Wherein the diagnosis result of the former part of the medical record in the target course is a first disease, the diagnosis result of the latter part of the medical record is a second disease, and the second disease evolves from the first disease.
Assuming that the first disease is disease a and the second disease is disease b, the target course of disease is in the form of { m }a,ma,ma,…,ma,mb,mb,mb,…,mb}(maFor the case history of disease a as a diagnostic result, mbA medical history of disease b as a diagnostic result), and disease b has evolved from disease a.
Step e2, if the number of disease courses in the target disease course set is larger than the preset number threshold, determining whether an unreasonable medical record exists in each disease course in the target disease course set according to the time span and the time span ratio of the first disease of each disease course in the target disease course set, and correcting the unreasonable medical record when the unreasonable medical record exists in the disease course.
Assume a disease course in the target set of disease courses is { m }a,ma,ma,…,ma,mb,mb,mb,…,mbAnd (c) the first disease of the disease course is a, the second disease is b, each medical record in the disease course has a date, and the date interval from the medical record with the first diagnosis result of a to the previous day of the medical record with the first diagnosis result of b can be used as the time span of the disease a, daygap (a), and the date interval from the medical record with the first diagnosis result of b to the medical record with the last diagnosis result of b can be used as the time span of the disease b, daygap (b).
The ratio of the time span of the first disease of a disease course to the time span of the second disease is the ratio of the time span of the first disease of the disease course to the time span of the second disease, and assuming that the first disease is disease a and the second disease is disease b, the time span of the first disease is daygap (a), the time span of the second disease is daygap (b), and the ratio of the time span of the first disease to daygap _ ratio (a/b) is daygap (a)/daygap (b).
Specifically, according to the time span and the time span ratio of the first disease of each course in the target course set, it is determined whether an unreasonable medical record exists in each course in the target course set, and the unreasonable medical record exists in the course, and the process of correcting the unreasonable medical record may include:
step e21, averaging the time span of the first disease of each course in the target course set, and calculating the standard deviation of the ratio of the time span of the first disease of each course in the target course set.
Assuming that the target course set includes P courses, the time spans of the first disease of the P medical records are averaged, and the standard deviation of the ratio of the time spans of the first disease of the P medical records is calculated.
And e22, determining the acceptable time span and the acceptable time span ratio of the first disease according to the obtained mean value and standard deviation.
Specifically, the acceptable time span and acceptable time span ratio for the first disease can be calculated according to the following formula:
acceptable_daygap(a)=avg_daygap(a)+N*std_daygap(a) (5)
acceptable_daygap_ratio(a)=avg_daygap_ratio(a)+N*std_daygap_ratio(a/b) (6)
wherein, acceptable _ daygap (a) is acceptable time span of the first disease, acceptable _ daygap _ ratio (a) is acceptable time span ratio of the first disease, avg _ daygap (a) is mean value obtained by step e21, std _ daygap (a) is standard deviation obtained by step e21, and N can be determined according to business experience and data distribution.
Step e23, for each course M in the target course set, performing:
and e231, determining whether the unreasonable medical record exists in the disease process according to the time span and the time span ratio of the first disease of the disease process M and the acceptable time span ratio of the first disease.
Specifically, it can be determined whether the time span daygap (a) of the first disease a of the disease course is greater than or equal to the acceptable time span acceptable _ daygap (a) of the first disease a, or whether the time span proportion daygap _ ratio (a) of the first disease a is greater than or equal to the acceptable time span proportion acgpap _ ratio (a) of the first disease a, and if so, it is determined that the disease course lags to find the evolution status, i.e. there is an unreasonable medical record in the disease course.
And e232, if the unreasonable medical record exists in the course of the disease, determining the unreasonable medical record in the course of the disease according to the acceptable time span and the acceptable time span ratio of the first disease and the time span of the course of the disease, and correcting the diagnosis result of the unreasonable medical record in the course of the disease into the second disease of the course of the disease.
Specifically, the date of the first medical record in the course of the disease can be obtained0Then acceptable _ daygap (a) and acceptable time span according to the acceptable time span of the first diseaseDegree-occupying ratio acceptable _ daygap _ ratio (a) and time span daygap (M) (date interval between the first medical record and the last medical record in the disease course M) of the disease course, and determining time increment delta; finally date0And determining the medical record with the diagnosis result of a in the medical records after + delta as an unreasonable medical record, and correcting the determined unreasonable medical record diagnosis result as the disease b. Wherein the time increment Delta can be determined according to the following formula:
Delta=max(acceptable_daygap(a),acceptable_daygap_ratio(a)*daygap(M)) (7)
the medical record set optimizing method provided by the embodiment can not only start from a single medical record per se and correct unreasonable medical records, but also start from a course of a patient and correct unreasonable medical records in the course of the disease, namely the corrected medical record set has better quality and higher reliability.
Seventh embodiment
In this embodiment, a medical record optimizing apparatus corresponding to the medical record optimizing method provided in the foregoing embodiment is provided, please refer to fig. 6, which shows a schematic structural diagram of the medical record optimizing apparatus, and the method may include: a medical record correction module 601, a disease knowledge graph acquisition module 602 and an optimal medical record set determination module 603.
The medical record correction module 601 is configured to respectively correct medical records in the target medical record set, which have an unreasonable diagnosis, by using a plurality of different medical record correction modes, so as to obtain a plurality of corrected medical record sets.
A disease knowledge graph obtaining module 602, configured to obtain a disease knowledge graph corresponding to the target medical record set and disease knowledge graphs corresponding to the plurality of corrected medical record sets, respectively.
The disease knowledge graph comprises a plurality of nodes and edges between the nodes, each node represents one disease in the diagnosis result of each medical record in the corresponding medical record set, and the edges between the two nodes represent the relationship between the two corresponding diseases.
An optimal medical record set determining module 603, configured to determine an optimal medical record set from the plurality of corrected medical record sets according to the disease knowledge graph corresponding to the target medical record set and the disease knowledge graphs corresponding to the plurality of corrected medical record sets, where the optimal medical record set is used as an optimized medical record set corresponding to the target medical record set.
Preferably, the disease knowledge graph acquiring module 602 is specifically configured to extract disease names from the diagnosis results of the medical records in a medical record set when acquiring a disease knowledge graph corresponding to the medical record set, so as to obtain a disease set composed of the extracted disease names; and acquiring a disease knowledge graph corresponding to the medical record set from a pre-constructed disease knowledge graph according to the disease set. Wherein the pre-constructed disease knowledge graph comprises a plurality of nodes representing various diseases respectively and edges between the nodes.
Preferably, the optimal medical record set determining module 603 may include: a modification effect characteristic value determining submodule and an optimal medical record set determining submodule.
And the correction effect characteristic value determining module is used for determining the correction effect characteristic values corresponding to the corrected medical record sets according to the symptom words in the medical records contained in the target medical record set, the symptom words in the medical records contained in the corrected medical record set and the disease knowledge maps respectively corresponding to the target medical record and the corrected medical record sets so as to obtain the correction effect characteristic values respectively corresponding to the corrected medical record sets.
And the medical record set determining module is used for determining the corrected medical record set corresponding to the maximum correction effect characteristic value in the correction effect characteristic values respectively corresponding to the plurality of corrected medical record sets as the optimal medical record set.
Optionally, the modifying effect characteristic value determining module includes: a quality evaluation submodule and a correction effect characteristic value determination submodule.
The quality evaluation submodule is used for determining the quality score of the disease knowledge graph corresponding to the target medical record set according to the symptom words in the medical records contained in the target medical record set and taking the quality score as the quality characteristic value of the target medical record set according to whether the diseases are matched with the corresponding symptoms; and determining the quality score of the disease knowledge graph corresponding to the corrected medical record set according to the symptom words in the medical record contained in the corrected medical record set and according to whether the disease is matched with the corresponding symptom, and taking the quality score as the quality characterization value of the corrected medical record set.
And the correction effect characteristic value determining submodule is used for determining a correction effect characteristic value corresponding to the corrected medical record set according to the quality characteristic value of the target medical record set, the quality characteristic value of the corrected medical record set and the number of the corrected medical records in the corrected medical record set.
Optionally, the quality evaluation sub-module includes: a symptom word total set obtaining sub-module, a symptom distribution vector determining sub-module and a quality score determining sub-module.
And the symptom word total set acquisition submodule is used for extracting symptom words from each medical record in the medical record set, and the extracted symptom words form a symptom word total set.
And the symptom distribution vector determining submodule is used for determining the symptom distribution vector of each disease in the disease knowledge graph corresponding to the medical record set according to the symptom word total set. The disease symptom distribution vector is composed of the disease and the co-occurrence condition characteristic values of the symptom words in the symptom word total set in the medical record included in the medical record set.
And the quality score determining submodule is used for determining the quality score of the disease knowledge graph corresponding to the medical record set according to the symptom distribution vector of each disease in the disease knowledge graph corresponding to the medical record set based on whether the disease is matched with the corresponding symptom.
Optionally, the relationship between the two diseases is one of an upper-lower relationship, an evolution relationship and an identification relationship;
the quality score determination submodule includes: an upper and lower relation score determining submodule, an evolution relation score determining submodule, a discrimination relation score determining submodule, a symptom number score determining submodule and a total score determining submodule.
And the upper and lower relation score determining submodule is used for determining the quality score of the disease knowledge graph corresponding to the medical record set on the upper and lower relation according to the side representing the upper and lower relation in the disease knowledge graph corresponding to the medical record set and the symptom distribution vector of the disease connected with the side representing the upper and lower relation.
And the evolutionary relationship score determining submodule is used for determining the quality score of the disease knowledge graph corresponding to the medical record set on the evolutionary relationship according to the edge representing the evolutionary relationship in the disease knowledge graph corresponding to the medical record set and the symptom distribution vector of the disease connected with the edge representing the evolutionary relationship.
And the identification relation score determining submodule is used for determining the quality score of the disease knowledge graph corresponding to the medical record set on the identification relation according to the side representing the identification relation in the disease knowledge graph corresponding to the medical record set and the symptom distribution vector of the disease connected with the side representing the identification relation.
And the symptom number score determining submodule is used for determining the quality score of the disease knowledge graph corresponding to the medical record set on the symptom number of the disease according to the symptom distribution vector of each disease in the disease knowledge graph corresponding to the medical record set.
And the total score determining submodule is used for determining the quality score of the disease knowledge graph corresponding to the medical record set according to the quality scores of the disease knowledge graph corresponding to the medical record set on the upper-lower relationship, the evolution relationship, the identification relationship and the disease symptom number respectively.
Optionally, the superior-inferior relation score determining submodule is specifically configured to take an edge representing the superior-inferior relation in the disease knowledge graph corresponding to the medical record set as the first edge: for each disease connected with each first edge, determining a symptom word set corresponding to the disease according to the state distribution vector of the disease and the symptom word set; and determining the quality score of the disease knowledge graph corresponding to the medical record set on the upper and lower relation according to whether the symptom word set corresponding to the lower disease is a subset of the symptom word set corresponding to the upper disease in the two diseases connected by each first edge.
Optionally, the evolutionary relationship score determining submodule is specifically configured to use an edge representing an evolutionary relationship in the disease knowledge graph corresponding to the medical record set as a second edge: for each disease connected by each second edge, determining a symptom word set corresponding to the disease according to the state distribution vector of the disease and the symptom word set; acquiring the intersection of the symptom word sets respectively corresponding to the two diseases connected by each second edge to obtain the common symptom word set of the two diseases connected by each second edge; determining a symptom severity change consistency characteristic value of the two diseases connected by each second edge on the common symptom of the two diseases connected by each second edge according to the common symptom word set of the two diseases connected by each second edge; and determining the quality score of the disease knowledge graph corresponding to the medical record set on the evolutionary relationship according to the symptom severity change consistency characteristic value of the two diseases connected by each second edge on the common symptoms of the two diseases.
Optionally, the identification relationship score determining submodule is specifically configured to use, as the third side, a side representing the identification relationship in the disease knowledge graph corresponding to the medical record set: determining a symptom distribution difference representation value of the two diseases connected by each third edge according to the symptom distribution vectors of the two diseases connected by each third edge; and determining the quality score of the disease knowledge graph corresponding to the medical record set on the identification relationship according to the symptom distribution difference characterization values of the two diseases connected by each third edge.
Optionally, the symptom number score determining submodule is specifically configured to determine, according to the symptom distribution vector of each disease in the disease knowledge graph corresponding to the medical record set, the number of the symptom words in the symptom word total set that do not belong to each disease in the disease knowledge graph corresponding to the medical record set: and determining the quality score of the disease knowledge graph corresponding to the medical record set on the disease symptom number according to the number of the symptom words in the disease knowledge graph not belonging to each disease in the medical record set corresponding to the symptom word total set.
Optionally, the medical record set optimizing apparatus provided in the embodiment of the present application may further include: a medical course set acquisition module and a second medical record correction module.
And the disease course set acquisition module is used for acquiring a disease course set from the optimal medical record set, wherein one disease course in the disease course set consists of all medical records of one patient with one attack, and all medical records in one disease course are sequenced according to the treatment time.
And the second medical record correction module is used for determining unreasonable medical records from the medical processes in the medical process set according to whether the medical records are reasonable in the medical processes in which the medical records are positioned, and correcting the determined unreasonable medical records.
Optionally, the second medical record correcting module is specifically configured to, for each course in the course set, determine, if the diagnosis result of the medical record in the course includes a primary disease and at least one secondary disease, and the primary disease and the secondary disease satisfy three conditions, an unreasonable medical record for the medical record of the secondary disease according to the diagnosis result in the course, and correct the diagnosis result of the unreasonable medical record into the primary disease:
wherein the three conditions include: the diagnosis result is that the number ratio of the medical records of the main diseases is larger than a preset main diagnosis ratio threshold; the secondary disease and the primary disease have one of a relationship among a superior-inferior relationship, an evolutionary relationship and an identification relationship; the diagnosis results are continuous medical records of the same minor disease.
Optionally, the second medical record modification module is specifically configured to select a target course from the course set, and form the target course set by the selected target course, where a diagnosis result of a previous part of medical records in the target course is a first disease, a diagnosis result of a later part of medical records in the target course is a second disease, and the second disease evolves from the first disease; if the number of the disease courses in the target disease course set is larger than a preset number threshold, determining whether an unreasonable medical record exists in each disease course in the target disease course set according to the time span and the time span ratio of the first disease of each disease course in the target disease course set, and correcting the unreasonable medical record when the unreasonable medical record exists in the disease course, wherein the time span ratio of the first disease of one disease course is the ratio of the time span of the first disease of the disease course to the time span of the second disease of the disease course.
Optionally, the second medical record correcting module determines whether an unreasonable medical record exists in each disease course in the target disease course set according to the time span and the time span ratio of the first disease of each disease course in the target disease course set, and when an unreasonable medical record exists in a disease course, corrects the unreasonable medical record, and is specifically configured to calculate a mean value of the time span of the first disease of each disease course in the target disease course set, calculate a standard deviation of the time span ratio of the first disease of each disease course in the target disease course set, and determine an acceptable time span and an acceptable time span ratio of the first disease according to the calculated mean value and standard deviation; for each course in the set of target courses: determining whether an unreasonable medical record exists in the course of the disease according to the time span and the time span ratio of the first disease of the course of the disease and the acceptable time span ratio of the first disease; if the unreasonable medical record exists in the course of the disease, the unreasonable medical record in the course of the disease is determined according to the acceptable time span and the acceptable time span ratio of the first disease and the time span of the course of the disease, and the diagnosis result of the unreasonable medical record in the course of the disease is corrected into a second disease of the course of the disease.
The medical record set optimizing device provided by the embodiment of the application can not only start from a single medical record to correct unreasonable medical records, but also start from a course of a patient to correct unreasonable medical records in the course of a disease.
Eighth embodiment
The embodiment also provides a medical record optimizing device, please refer to fig. 7, which shows a schematic structural diagram of the medical record optimizing device, and the medical record optimizing device may include: at least one processor 701, at least one communication interface 702, at least one memory 703 and at least one communication bus 704;
in the embodiment of the present application, the number of the processor 701, the communication interface 702, the memory 703 and the communication bus 704 is at least one, and the processor 701, the communication interface 702 and the memory 703 complete mutual communication through the communication bus 704;
the processor 701 may be a central processing unit CPU, or an application specific Integrated circuit asic, or one or more Integrated circuits configured to implement embodiments of the present invention, or the like;
the memory 703 may include a high-speed RAM memory, and may further include a non-volatile memory (non-volatile memory) or the like, such as at least one disk memory;
wherein the memory stores a program and the processor can call the program stored in the memory, the program for:
correcting medical records in which unreasonable diagnosis exists in a target medical record set respectively by adopting a plurality of different medical record correction modes to obtain a plurality of corrected medical record sets;
acquiring a disease knowledge graph corresponding to the target medical record set and disease knowledge graphs corresponding to the corrected medical record sets respectively, wherein the disease knowledge graph comprises a plurality of nodes and edges among the nodes, each node represents one disease in the diagnosis result of each medical record in the corresponding medical record set, and the edges between the two nodes represent the relationship between the two corresponding diseases;
and determining an optimal medical record set from the plurality of corrected medical record sets according to the disease knowledge graph corresponding to the target medical record set and the disease knowledge graphs corresponding to the plurality of corrected medical record sets respectively, wherein the optimal medical record set is used as the optimized medical record set corresponding to the target medical record set.
Alternatively, the detailed function and the extended function of the program may be as described above.
Ninth embodiment
The present embodiment provides a readable storage medium storing a program adapted to be executed by a processor, the program being configured to:
correcting medical records in which unreasonable diagnosis exists in a target medical record set respectively by adopting a plurality of different medical record correction modes to obtain a plurality of corrected medical record sets;
acquiring a disease knowledge graph corresponding to the target medical record set and disease knowledge graphs corresponding to the corrected medical record sets respectively, wherein the disease knowledge graph comprises a plurality of nodes and edges among the nodes, each node represents one disease in the diagnosis result of each medical record in the corresponding medical record set, and the edges between the two nodes represent the relationship between the two corresponding diseases;
and determining an optimal medical record set from the plurality of corrected medical record sets according to the disease knowledge graph corresponding to the target medical record set and the disease knowledge graphs corresponding to the plurality of corrected medical record sets respectively, wherein the optimal medical record set is used as the optimized medical record set corresponding to the target medical record set.
Finally, it should also be noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.