Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
In the medical field, due to the huge data volume of medical record data and the lack of strict data monitoring in the early stage, the information in the medical record is very redundant and disordered. Taking the "cold" as an example, the diagnosis names in practice may be "cold", "acute upper respiratory infection", "upper sense", "acute upper sense", etc., and the different diagnosis names may represent the same disease in practice. In order to facilitate the query, management and use of the later medical record data, standardization is required for the unnormal diagnosis names in the medical records.
Currently, the normalization of diagnostic names is mainly based on manual labeling, or by comparing the similarity of two diagnostic names with a preset similarity threshold value, whether the two diagnostic names represent the same disease is determined. The method has the advantages of high labor consumption, low efficiency, high subjectivity and low accuracy of the labeling result; the latter relies on the judgment of the similarity threshold completely, but the current technology cannot guarantee the accuracy of the setting of the similarity threshold, and the relationship between diagnostic names is determined only by the similarity threshold is not scientific, so that the accuracy and the reliability of standardization cannot be guaranteed. In this regard, the embodiment of the invention provides a diagnostic name standardization method to improve the accuracy and reliability of diagnostic name standardization.
Fig. 1 is a flow chart of a diagnostic name normalization method according to an embodiment of the present invention, as shown in fig. 1, where the method includes:
at step 110, a plurality of diagnostic names are determined.
The diagnosis name is a diagnosis name which needs to be standardized, and the diagnosis name can be extracted from medical record data.
And 120, adjusting the candidate synonym relation between every two diagnostic names in the plurality of diagnostic names based on the similarity and the medical relation between every two diagnostic names in the plurality of diagnostic names to obtain the final synonym relation between every two diagnostic names.
Specifically, for any two diagnostic names, the synonym relationship between the two may be "yes" or "no", where "yes" indicates that the two diagnostic names are synonyms, and "no" indicates that the two diagnostic names are non-synonyms. The candidate synonym relationship represents the synonym relationship before and during the adjustment of the synonym relationship between each two diagnostic names based on the similarity and medical relationship between each two diagnostic names, and the final synonym relationship represents the synonym relationship after the adjustment is completed. The candidate synonym relationships between each two diagnostic names may be randomly generated before the candidate synonym relationships between each two diagnostic names are adjusted based on the similarity and medical relationships between each two diagnostic names.
The similarity between any two diagnosis names is used to represent the similarity degree of the information corresponding to the two diagnosis names, where the information may be medical record data corresponding to the two diagnosis names, or may be specific attributes of the medical record data corresponding to the two diagnosis names, such as a main complaint, an actual medical history, a past medical history, an allergy history, a physical examination, an auxiliary examination, a medical department, etc., which is not limited in particular in the embodiment of the present invention.
The medical relation between any two diagnosis names is used for representing the relation of the two diagnosis names in the medical field, and the medical relation can be the upper and lower relation of diseases corresponding to the two diagnosis names in a disease system in the medical field, the sequence of the diseases corresponding to the two diagnosis names on a patient, or the difference of the time of common onset of the diseases corresponding to the two diagnosis names, and the like.
The similarity and medical relationship between any two diagnostic names are associated with the synonym relationship between the two diagnostic names. For example, the higher the similarity, the higher the probability that the synonym relationship between the two diagnostic names is "yes", and if the diseases corresponding to the two diagnostic names in the medical relationship are reflected in the patient in the sequence or concurrence, or if the diseases corresponding to the two diagnostic names are in the upper-lower relationship, the probability that the synonym relationship between the two diagnostic names is "yes" is reduced.
Here, the candidate synonym relationship is adjusted by the mutual restriction of the similarity between every two diagnosis names and the medical relationship, so that the defect caused by judging the synonym relationship only by the similarity is avoided.
For example, for any two diagnostic names "acute upper respiratory infection" and "viral pharyngitis", the candidate synonym relationship between the two is "yes", the similarity between the two is 70%, and in the medical relationship between the two, "acute upper respiratory infection" is the upper concept of "viral pharyngitis", so that although the similarity between the two is higher, the candidate synonym relationship between the two is still regulated due to the limitation of the upper and lower relationships, and the final synonym relationship is "no".
Step 130, determining a standardized diagnosis name corresponding to each diagnosis name based on the final synonym relationship between every two diagnosis names.
Specifically, after determining the final synonym relationship between every two diagnostic names, several groups of diagnostic names which are synonyms, such as "spinal cord lesion", "spinal cord dysfunction", "spinal cord disease" and "neuromyelopathy", can be obtained as a group of synonyms, one diagnostic name can be selected from the group of synonyms as a standardized diagnostic name of the group of synonyms, and the standardized diagnostic name corresponding to each diagnostic name can be obtained based on the same method, thereby realizing the standardization of diagnostic names.
According to the method provided by the embodiment of the invention, the synonym relation between the diagnosis names is adjusted based on the similarity and the medical relation between every two diagnosis names, so that the diagnosis name standardization is realized, the medical knowledge is fused into the determining process of the synonym relation, and the similarity is mutually restricted, so that the accuracy and the reliability of the diagnosis name standardization are improved.
Based on the above embodiment, the medical relationship between any two diagnostic names includes at least one of a time series relationship, an upper-lower relationship, and a time distribution difference of the onset between the two diagnostic names.
Specifically, for any two diagnosis names, the time sequence relationship between the two diagnosis names, that is, the sequence of the diseases corresponding to the two diagnosis names on the same patient, is shown. The timing relationship can be extracted from a large amount of medical record data, where two diagnostic names are most likely not synonyms for the medical record of the same patient if another diagnostic name must appear some time before any diagnostic name appears. Here, the length of time for judging whether or not there is a timing relationship between the two diagnostic names may be determined based on business experience.
The upper and lower relationship of any two diagnosis names, namely the upper and lower relationship of the diseases corresponding to the two diagnosis names under the disease system in the medical field, if any diagnosis name corresponding symptom word set contains another diagnosis name corresponding symptom word set, the two diagnosis names may have the upper and lower relationship, and are not likely to be synonyms.
The difference in the time distribution of any two diagnosis names, that is, the difference in the time distribution of the diseases corresponding to the two diagnosis names, may be divided in months, quarterly, or other time units. If there is a significant difference in the distribution of the preset time units between the diseases corresponding to the two diagnostic names, the two diagnostic names are most likely not synonyms.
Based on any of the above embodiments, fig. 2 is a flowchart of a final synonym relationship determination method according to the embodiments of the present disclosure, as shown in fig. 2, and step 120 specifically includes:
and step 121, adjusting the candidate synonym relation between every two diagnostic names in the plurality of diagnostic names until the global function value obtained correspondingly reaches the maximum value.
And step 122, using the candidate synonym relation between every two diagnosis names corresponding to the maximum value as the final synonym relation between every two diagnosis names.
Wherein the global function value is determined based on the benchmark function value and the medical penalty function value for each two diagnostic names; the benchmark function value of any two diagnostic names is determined based on the candidate synonym relationship and the similarity between the two diagnostic names, and the medical penalty function value of any two diagnostic names is determined based on the candidate synonym relationship and the medical relationship between the two diagnostic names.
Specifically, the purpose of adjusting the candidate synonym relationship is to maximize the global function value, so that the final synonym relationship between every two diagnostic names is the candidate synonym relationship between every two diagnostic names corresponding to the maximum global function value.
For any two diagnostic names, the benchmark function values of the two can be obtained based on the candidate synonym relationship and the similarity between the two. Since the higher the similarity is, the higher the probability that the candidate synonym relationship between the two diagnostic names is "yes", the lower the similarity is, the higher the probability that the candidate synonym relationship between the two diagnostic names is "no", the reference function value determined based on the above rule can be regarded as the score of the rule between the candidate synonym relationship and the similarity, and when the candidate synonym relationship is "yes", the higher the similarity is, the higher the reference function value is.
For any two diagnostic names, the medical penalty function values for the two can be derived based on the candidate synonym relationship and the medical relationship between the two. Here, the medical penalty function value may be regarded as a penalty score based on a limitation of the medical relationship to the candidate synonym relationship. When there is a medical relationship between the two that limits the candidate synonym relationship of the two to "no", the medical penalty function value may be correspondingly increased. For example, if the diseases corresponding to the two diagnosis names in the medical relationship are reflected in the existence sequence or concurrence of the diseases on the patient, the probability of the candidate synonym relationship between the two diagnosis names being "yes" is reduced, and if the candidate synonym relationship at the moment is "yes", the medical penalty function value is increased.
Based on the baseline function value and the medical penalty function value for each two diagnostic names, a global function value may be determined. Wherein, the higher the reference function value of every two diagnosis names is, the higher the global function value is; the higher the medical penalty function value per two diagnostic names, the lower the global function value. Under the condition that the basic function value and the medical penalty value are mutually restricted, the global function value is maximized by adjusting the candidate synonym relation between every two diagnosis names, and the candidate synonym relation between every two diagnosis names corresponding to the maximum global function value is used as the final synonym relation between every two diagnosis names.
According to the method provided by the embodiment of the invention, the medical penalty function value is determined based on the candidate synonym relation and the medical relation between the diagnosis names, so that the medical penalty function value is applied to the determination of the global function value, the value of the candidate synonym relation is restricted by the medical relation, and the accurate determination of the synonym relation between the diagnosis names is realized.
Based on any of the above embodiments, the medical penalty function value includes at least one of a disease timing penalty function value, a context penalty function value, and a time distribution difference penalty function value; wherein the disease time sequence penalty function value of any two diagnosis names is determined based on the time sequence relation and the candidate synonym relation between the two diagnosis names; the upper and lower relation penalty function values of any two diagnosis names are determined based on the upper and lower relation between the two diagnosis names and the candidate synonym relation; the time distribution difference penalty function value for any two diagnostic names is determined based on the time distribution difference of onset between the two diagnostic names and the candidate synonym relationship.
Specifically, if the time sequence relationship between any two diagnosis names is "present" and the candidate synonym relationship is "yes", it is obvious that the time sequence relationship and the candidate synonym relationship are contrary, and the disease time sequence penalty function value is correspondingly set to a preset value, so that the disease time sequence penalty is effective, and the global function value is constrained; in other cases, for example, when the timing relationship is "present" and the candidate synonym relationship is "no", or when the timing relationship is "absent", the disease timing penalty function value is set to zero, and the disease timing penalty is not validated.
If the upper and lower relationship between any two diagnosis names is 'existing', and the candidate synonym relationship is 'yes', obviously the upper and lower relationship is opposite to the candidate synonym relationship, correspondingly setting the upper and lower relationship punishment function value as a preset value, so that the upper and lower relationship punishment is effective, and the global function value is restrained; in other cases, for example, when the context is "present" and the candidate synonym relationship is "no", or when the context is "absent", the context penalty function value is set to zero, and the context penalty is not effective.
If the attack time distribution difference between any two diagnosis names is 'existing', and the candidate synonym relation is 'yes', obviously the attack time distribution difference and the candidate synonym relation are opposite, correspondingly setting a time distribution difference punishment function value to a preset value, so that the time distribution difference punishment is effective, and the global function value is restrained; in the other cases, for example, when the attack time distribution difference is "present" and the candidate synonym relationship is "no", or when the attack time distribution difference is "absent", the time distribution difference penalty function value is set to zero, and the time distribution difference penalty is not effective.
Based on any of the above embodiments, the candidate synonym relationship between any two diagnostic names p and q, p and q is expressed as Exist (p, q), exist (p, q) =1 corresponding to the candidate synonym relationship being "yes" and Exist (p, q) =0 corresponding to the candidate synonym relationship being "no". The similarity between p and q can be expressed as Prob (p, q), and the resulting p and q reference function values S can be expressed as:
S＝Prob(p，q)*Exist(p，q)
based on any of the above embodiments, the timing relationship between p and q is denoted as T (p, q):
thus, the disease timing penalty function values S1 for p and q are expressed as:
S1＝T(p，q)*(T(p，q)∧Exist(p，q))
wherein T (p, q) ∈exist (p, q) represents the intersection of the timing relationship and the candidate synonym relationship, s1=1 if and only if both T (p, q) and Exist (p, q) are 1, the disease timing penalty is effective, otherwise s1=0, the disease timing penalty is ineffective.
Based on any of the above embodiments, the upper and lower relationships between p and q are expressed as
Where set (P) and set (Q) are symptomatic word sets of P and Q, respectively.
Thus, the upper and lower relation penalty function values S2 of p and q are expressed as:
in the method, in the process of the invention,representing the intersection of the context relationship and the candidate synonym relationship if and only if +.>And when Exist (p, q) is 1, the upper and lower relation penalty is effective, if not, the upper and lower relation penalty is ineffective, and if not, the upper and lower relation penalty is ineffective.
Based on any of the above examples, the difference in the time distribution of onset between p and q is expressed as R (p, q):
thus, the time distribution difference penalty function value S3 of p and q is expressed as:
S3＝R(p，q)*(R(p，q)∧Exist(p，q))
wherein R (p, q) ζexist (p, q) represents an intersection of the attack time distribution difference and the candidate synonym relationship, s3=1, the time distribution difference penalty takes effect if and only if R (p, q) and Exist (p, q) are both 1, otherwise s3=0, the time distribution difference penalty takes effect.
Based on any of the above embodiments, the value of R (p, q) may be used to compare the standard deviation diff (p, q) of the difference between the time distributions of the two diagnostic names with a preset standard deviation threshold, which is specifically expressed as:
wherein diff (p, q) =std (montarate (p) -montarate (q)), montarate (p) and montarate (q) respectively represent the incidence of diagnostic names p and q in each natural month with the year as a period, and montarate (p) and montarate (q) are 12-dimensional vectors.
The standard deviation threshold can be obtained by the following formula:
wherein D is a set containing each diagnosis name, and N is the number of diagnosis names contained in D.
For example, diagnosis name p is acute upper respiratory tract infection, diagnosis name q is hypertension:
MonthRate(p)
＝c(0.08，0.09，0.28，0.21，0.04，0.04，0.02，0.01，0.04，0.04，0.07，0.08)
MonthRate(q)
＝c(0.09，0.08，0.06，0.07，0.09，0.09，0.08，0.09，0.10，0.08，0.08，0.09)
from this, diff (p, q) =std (montarate (p) -montarate (p))=0.09.
Based on any of the above embodiments, the global function value is specifically determined based on the benchmark function value and the medical penalty function value for each two diagnostic names, and at least one of the synonym transfer penalty function value, the similarity penalty function value, and the similarity threshold penalty function value for each two diagnostic names.
The synonym transfer penalty function value for any two of the diagnostic names is determined based on the synonym relationship between the any two diagnostic names and the remaining diagnostic names, respectively; the similarity penalty function value of any two diagnostic names is determined based on the synonym relationship and the similarity between the any two diagnostic names; the similarity threshold penalty function value for any two diagnostic names is determined based on the synonym relationship and similarity between any two diagnostic names and a preset similarity threshold.
In particular, synonyms are transitive in nature, and B and C are most likely synonyms given that a and B are synonyms for each other and that a and C are synonyms for each other. Based on the rule, if the candidate synonym relation between the two diagnostic names is 'no', the candidate synonym relation between the two diagnostic names and the rest of diagnostic names is obtained, so that whether the diagnostic names with the candidate synonym relation of the two diagnostic names being 'yes' exist or not is judged, if the candidate synonym relation of the two diagnostic names is obviously contrary to the transitivity of the synonym, a synonym transfer penalty function value is determined, and the synonym transfer penalty is enabled to be effective, so that the global function value is restrained; if not, the synonym transfer penalty function value is set to zero, and the synonym transfer penalty is not effective.
The corresponding relation exists between the similarity and the candidate synonym relation, the higher the similarity is, the higher the probability that the candidate synonym relation between two diagnosis names is 'yes', the lower the similarity is, the higher the probability that the candidate synonym relation between two diagnosis names is 'no', and based on the rule, the global function value is constrained by the similarity penalty for any two diagnosis names. Here, when the candidate synonym relationship is yes, the higher the similarity is, the smaller the similarity penalty function value is, and the lower the similarity is, the larger the similarity penalty function value is; when the candidate synonym relation is 'no', the higher the similarity is, the larger the similarity penalty function value is, and the lower the similarity is, the smaller the similarity penalty function value is.
The corresponding candidate synonym relationship may be determined, typically by comparing the magnitudes of the similarity and a preset similarity threshold. For any two diagnosis names, when the similarity between the two diagnosis names is larger than a preset similarity threshold, the probability of the candidate synonym relationship being 'yes' is larger, and the probability of the candidate synonym relationship being 'no' is smaller; when the similarity is smaller than a preset similarity threshold, the probability of the candidate synonym relationship being 'yes' is smaller, and the probability of the candidate synonym relationship being 'no' is larger. Based on the rule, if the similarity is greater than a preset similarity threshold, if the candidate synonym relation is no, or if the similarity is less than the preset similarity threshold, if the candidate synonym relation is yes, obviously the candidate synonym is contrary to a comparison rule based on the preset similarity threshold, and a similarity threshold punishment function value is correspondingly set, so that the similarity threshold punishment is effective, and the global function value is restrained; if not, the similarity threshold penalty function value is set to zero, and the similarity threshold penalty is not effective.
The method provided by the embodiment of the invention not only can be used for determining the global function value, but also can be used for determining the global function value by at least one of the synonym transfer penalty function value, the similarity penalty function value and the similarity threshold penalty function value, so that the value of the candidate synonym relationship is restricted by the medical relationship, and meanwhile, the value of the candidate synonym relationship is restricted by the similarity judgment and the rule of the synonym relationship, thereby further improving the accuracy of determining the final synonym relationship.
Based on any embodiment, when the candidate synonym relation of any two diagnostic names is no and there are a plurality of transfer diagnostic names whose candidate synonym relation with the two diagnostic names is yes, the synonym transfer penalty function value of the two diagnostic names is the maximum value of the penalty corresponding to each transfer diagnostic name; the penalty score corresponding to any one of the transfer diagnostic names is determined based on the similarity between the transfer diagnostic name and the two diagnostic names, respectively.
Specifically, for any two diagnostic names p and q, either one of the transmitted diagnostic names ri The ith transfer diagnosis name of the p and q transfer diagnosis names, in particular the diagnosis name which has a 'yes' relation with the candidate synonyms of the two diagnosis names, namely Exist (p, r)i )＝1，Exist(q，ri )＝1。
If the candidate synonym relationship exists (p, q) =0 between p and q, for any transfer diagnosis name ri Candidate diagnosis name ri The corresponding penalty is based on p, q, p, ri Between and q, ri Similarity between, wherein p, ri Similarity between Prob (p, ri ) Q, ri Similarity between Prob (q, ri ) The higher ri The higher the corresponding penalty score.
Based on any of the above embodiments, for ri The corresponding penalty score is expressed as:
(Prob(p，ri )+Prob(ri ，q)-Prob(p，q))*(1-Exist(p，q))
wherein, if the candidate synonym relation between p and q is "yes", the penalty point corresponds to 0, and if the candidate synonym relation between p and q is "no", the penalty point corresponds to Prob (p, ri )Prob(ri Q) -Prob (p, q). And taking the maximum value of the penalty for all the transfer diagnosis names, and obtaining the synonym transfer penalty function value S4 as follows:
S4＝|max((Prob(p，ri )+Prob(ri ，q)-Prob(p，q))*(1-Exist(p，q)))|
based on any of the above embodiments, the similarity penalty function value S5 of p and q may be expressed as an absolute difference between the candidate synonym relationship Exist (p, q) of p and q and the similarity Prob (p, q) of p and q, specifically:
S5＝|Exist(p，q)-Prob(p，q)|
based on any of the above embodiments, it is assumed that the preset similarity threshold is THGlobal Setting W (p, q) for representing the similarity Prob (p, q) of p and q and a preset similarity threshold value as THGlobal Specifically, the size of (3) is:
the similarity threshold penalty function value S6 for p and q is thus obtained as:
in the method, in the process of the invention,for the nand symbol, when W (p, q) =1 and Exist (p, q) =0, or W (p, q) =0 and Exist (p, q) =1, that is, when the similarity is greater than the preset similarity threshold and the candidate synonym relationship is "no", or when the similarity is less than the preset similarity threshold and the candidate synonym relationship is "yes", s6=1, the similarity threshold penalty is effective.
Based on any of the above embodiments, in the method, the global function value may be obtained based on the benchmark function value S of each two diagnosis names, and the disease timing penalty function value S1, the upper-lower relationship penalty function value S2, the time distribution difference penalty function value S3, the synonym transfer penalty function value S4, the similarity penalty function value S5, and the similarity threshold penalty function value S6, and the objective function for achieving the maximization of the global function value may be specifically expressed as the following formula:
maxmize(S-α*S1-β*S2-γ*S3-δ*S4-ε*S5-θ*S6)
where α, β, γ, δ, ε, and θ are weights set in advance to correspond to S1, S2, S3, S4, S5, and S6, respectively.
Based on any of the above embodiments, fig. 3 is a flowchart of a standardized diagnosis name determining method according to an embodiment of the present invention, as shown in fig. 3, step 130 specifically includes:
Step 131, determining a plurality of synonym diagnostic name sets based on the final synonym relationship between each two diagnostic names.
Specifically, after obtaining the final synonym relationship between every two diagnosis names, the synonym of each diagnosis name can be determined, so as to obtain a plurality of synonym diagnosis name sets. Here, any synonym diagnosis name set includes a plurality of diagnosis names, and any diagnosis name in the set is at least synonymous with one diagnosis name in the set.
Step 132, if a plurality of bridge diagnosis names exist in the Ren Yitong sense word diagnosis name set, the bridge diagnosis name with the highest similarity score is used as the standardized diagnosis name of the synonym diagnosis name set; otherwise, the diagnosis name with the highest similarity score in the synonym diagnosis name set is used as the standardized diagnosis name of the synonym diagnosis name set; wherein the similarity score for any one of the diagnostic names is determined based on the similarity between that diagnostic name and the diagnostic name for which each final synonym relationship is.
Specifically, in any synonym diagnosis name set, if deleting one diagnosis name results in the diagnosis name which is not synonymous with each of the rest diagnosis names in the set, the deleted diagnosis name is taken as the bridge diagnosis name. The bridge diagnosis names are in the synonym diagnosis name set and play a role of communicating all diagnosis names in the set.
In general, in the set of synonym diagnostic names, there are two cases, one is that there are a plurality of bridge diagnostic names in the set of synonym diagnostic names, and the other is that there are no bridge diagnostic names in the set of synonym diagnostic names.
Fig. 4 is a schematic structural diagram of a set of synonym diagnostic names with bridge diagnostic names provided by an embodiment of the present invention, fig. 5 is a schematic structural diagram of a set of synonym diagnostic names without bridge diagnostic names provided by an embodiment of the present invention, in fig. 4 and 5, each node corresponds to one diagnostic name in the set of synonym diagnostic names, a connection line between two nodes indicates that a final synonym relationship between two diagnostic names is yes, that is, two nodes are synonyms, and a value on a connection line between nodes, that is, a similarity between two diagnostic names. In fig. 4, A, B is the bridge diagnostic name, and in fig. 5, the bridge diagnostic name is not present.
When the bridge diagnosis names exist, calculating the similarity score of each bridge diagnosis name, and taking the bridge diagnosis name with the highest similarity score as a standardized diagnosis name; when the bridge diagnosis names do not exist, the similarity score of each diagnosis name is calculated, and the diagnosis name with the highest similarity score is used as the standardized diagnosis name. The similarity score is determined based on the similarity between the diagnosis name and the diagnosis name with the final synonym relationship being "yes", that is, the similarity score may be determined based on the value on the node connection corresponding to the diagnosis name.
Further, the similarity score may be a combination of the similarity between the diagnosis name and each diagnosis name with the final synonym relationship of "yes", for example, in fig. 4, the similarity score of node a is 7.3, and the similarity score of node B is 6.67, so the diagnosis name corresponding to node a is regarded as the standardized diagnosis name. In fig. 5, the node with the highest similarity score is a, and the diagnosis name corresponding to the node a is taken as the standardized diagnosis name.
Based on any of the above embodiments, the similarity between any two diagnostic names is determined based on the diagnostic attributes corresponding to any two diagnostic names, and the diagnostic attribute corresponding to any one diagnostic name is extracted from the medical record data corresponding to the diagnostic name.
Specifically, a large amount of medical record data corresponding to any diagnosis name can be collected in advance, and a plurality of diagnosis attributes corresponding to the diagnosis name can be extracted therefrom. Here, the diagnostic attribute may include at least one of a main complaint, an existing medical history, an allergy history, a past medical history, an auxiliary examination, and a department of medical care.
The similarity Prob (p, q) between the diagnosis names p and q can be obtained by weighting the similarity of the respective diagnosis attributes corresponding to the diagnosis names p and q, for example, the related information corresponding to the diagnosis names p and q includes Main complaint Main, current medical history, allergy history, previous medical history, auxiliary examination Auxiliary, department of diagnosis Dep, and the similarity prob_main (p, q) of the Main complaint Main is calculated as follows, taking the Main complaint Main as an example:
Wherein, p_main and q_main respectively represent p and q complaints, p-Main n q_main and p_main q_main are respectively intersection and union of the two, and the cart represents the number of elements in the collection.
Based on similar formulas, the similarity prob_current (p, q) of the Current medical history, the similarity prob_allergy (p, q) of the Allergy history, the similarity prob_previous (p, q) of the Previous medical history, the similarity prob_auxliary (p, q) of the Auxiliary examination, and the similarity prob_dep (p, q) of the medical department can be obtained respectively.
Then, the similarity Prob (p, q) between p and q can be obtained by weighting based on the similarity of each related information.
Based on any of the above embodiments, a diagnostic name normalization method includes the steps of:
firstly, a large amount of medical record data is acquired, and a plurality of diagnosis names which need to be standardized and diagnosis attributes corresponding to each diagnosis name are extracted from the large amount of medical record data. The similarity between each two diagnostic names is calculated based on the diagnostic attributes of each two diagnostic names. In addition, in combination with medical knowledge, the time series relationship, the upper and lower relationship and the difference in the distribution of the time of onset between every two diagnosis names are determined.
And secondly, calculating a global function value based on the similarity, the time sequence relation, the upper and lower relation and the attack time distribution difference between every two diagnosis names, and adjusting the candidate synonym relation between every two diagnosis names with the maximum global function value as a target until the maximum value of the global function value is obtained.
And taking the candidate synonym relation between every two diagnosis names corresponding to the maximum value as the final synonym relation between every two diagnosis names.
After determining the final synonym relationship between each two diagnostic names, a number of sets of synonym diagnostic names may be determined based on the final synonym relationship between each two diagnostic names. Aiming at Ren Yitong sense word diagnosis names, if bridge diagnosis names exist, calculating similarity scores of each bridge diagnosis name, and taking the bridge diagnosis name with the highest similarity score as a standardized diagnosis name; otherwise, the similarity score of each diagnosis name is calculated, and the diagnosis name with the highest similarity score is used as the standardized diagnosis name.
Based on any one of the above embodiments, fig. 6 is a schematic structural diagram of a diagnostic name normalization device according to an embodiment of the present invention, and as shown in fig. 6, the diagnostic name normalization device includes a diagnostic name determining unit 610, a synonym relationship determining unit 620, and a normalization unit 630;
wherein the diagnosis name determining unit 610 is configured to determine a plurality of diagnosis names;
the synonym relationship determination unit 620 is configured to adjust candidate synonym relationships between every two diagnostic names based on the similarity and the medical relationship between every two diagnostic names in the plurality of diagnostic names, so as to obtain a final synonym relationship between every two diagnostic names;
The normalization unit 630 is configured to determine a normalized diagnosis name corresponding to each diagnosis name based on the final synonym relationship between each two diagnosis names.
According to the device provided by the embodiment of the invention, the synonym relation between the diagnosis names is adjusted based on the similarity and the medical relation between every two diagnosis names, so that the diagnosis name standardization is realized, the medical knowledge is fused into the determining process of the synonym relation, and the similarity is mutually restricted, so that the accuracy and the reliability of the diagnosis name standardization are improved.
Based on any of the above embodiments, the medical relationship between any two diagnostic names includes at least one of a time series relationship, an up-down relationship, and a difference in distribution of time of onset between the any two diagnostic names.
Based on any of the above embodiments, the synonym relationship determination unit 620 is specifically configured to:
the candidate synonym relation between every two diagnostic names in the plurality of diagnostic names is adjusted until the global function value obtained correspondingly reaches the maximum value;
the candidate synonym relation between every two diagnosis names corresponding to the maximum value is used as the final synonym relation between every two diagnosis names;
wherein the global function value is determined based on the benchmark function value and the medical penalty function value for each two diagnostic names; the benchmark function value of any two diagnostic names is determined based on the candidate synonym relationship and the similarity between the any two diagnostic names, and the medical penalty function value of any two diagnostic names is determined based on the candidate synonym relationship and the medical relationship between the any two diagnostic names.
Based on any one of the above embodiments, the medical penalty function value includes at least one of a disease timing penalty function value, an upper-lower relationship penalty function value, and a time distribution difference penalty function value;
wherein the disease timing penalty function value for any two diagnostic names is determined based on the timing relationship between the any two diagnostic names and the candidate synonym relationship;
the upper and lower relation penalty function values of any two diagnosis names are determined based on the upper and lower relation and the candidate synonym relation between the any two diagnosis names;
the time distribution difference penalty function value for any two diagnostic names is determined based on the time distribution difference of the onset between the any two diagnostic names and the candidate synonym relationship.
Based on any of the above embodiments, the global function value is specifically determined based on a benchmark function value and a medical penalty function value for each two diagnostic names, and at least one of a synonym transfer penalty function value, a similarity penalty function value, and a similarity threshold penalty function value for each two diagnostic names;
the synonym transfer penalty function value for any two diagnostic names is determined based on the candidate synonym relationship between the any two diagnostic names and the remaining diagnostic names, respectively;
The similarity penalty function value of any two diagnostic names is determined based on the candidate synonym relationship and the similarity between the any two diagnostic names;
the similarity threshold penalty function value for any two diagnostic names is determined based on the candidate synonym relationship and similarity between the any two diagnostic names, and a preset similarity threshold.
Based on any one of the above embodiments, when the candidate synonym relationship of any two diagnostic names is no, and there are a plurality of transfer diagnostic names whose candidate synonym relationship with the two diagnostic names is yes, the synonym transfer penalty function value of the any two diagnostic names is the maximum value of the penalty score corresponding to each transfer diagnostic name;
the penalty score corresponding to any one of the transmitted diagnostic names is determined based on the similarity between the any one of the transmitted diagnostic names and the any two of the diagnostic names, respectively.
Based on any of the above embodiments, the normalization unit 630 is specifically configured to:
determining a plurality of synonym diagnosis name sets based on the final synonym relationship between every two diagnosis names;
if a plurality of bridge diagnosis names exist in the Ren Yitong sense word diagnosis name set, the bridge diagnosis name with the highest similarity score is used as the standardized diagnosis name of any synonym diagnosis name set;
Otherwise, taking the diagnosis name with the highest similarity score in any synonym diagnosis name set as the standardized diagnosis name of any synonym diagnosis name set;
wherein the similarity score for any one of the diagnostic names is determined based on the similarity between the any one of the diagnostic names and each of the final synonym relationships being a positive diagnostic name.
Based on any of the above embodiments, the similarity between any two diagnostic names is determined based on the diagnostic attributes corresponding to the any two diagnostic names, and the diagnostic attribute corresponding to any one diagnostic name is extracted from the medical record data corresponding to the any one diagnostic name.
Fig. 7 is a schematic structural diagram of an electronic device according to an embodiment of the present invention, as shown in fig. 7, the electronic device may include: processor 710, communication interface (Communications Interface) 720, memory 730, and communication bus 740, wherein processor 710, communication interface 720, memory 730 communicate with each other via communication bus 740. Processor 710 may invoke logic commands in memory 730 to perform the following method: determining a plurality of diagnostic names; based on the similarity and the medical relation between every two diagnostic names in the plurality of diagnostic names, the candidate synonym relation between every two diagnostic names is adjusted, and the final synonym relation between every two diagnostic names is obtained; based on the final synonym relationship between each two diagnostic names, a normalized diagnostic name corresponding to each diagnostic name is determined.
In addition, the logic commands in the memory 730 described above may be implemented in the form of software functional units and may be stored in a computer readable storage medium when sold or used as a stand alone product. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in the form of a software product stored in a storage medium, comprising several commands for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
Embodiments of the present invention also provide a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, is implemented to perform the methods provided by the above embodiments, for example, comprising: determining a plurality of diagnostic names; based on the similarity and the medical relation between every two diagnostic names in the plurality of diagnostic names, the candidate synonym relation between every two diagnostic names is adjusted, and the final synonym relation between every two diagnostic names is obtained; based on the final synonym relationship between each two diagnostic names, a normalized diagnostic name corresponding to each diagnostic name is determined.
The apparatus embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
From the above description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus necessary general hardware platforms, or of course may be implemented by means of hardware. Based on this understanding, the foregoing technical solution may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a computer readable storage medium, such as ROM/RAM, a magnetic disk, an optical disk, etc., including several commands for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in the respective embodiments or some parts of the embodiments.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.