CN109509517A - A kind of medical test Index for examination modified method automatically - Google Patents
A kind of medical test Index for examination modified method automatically Download PDFInfo
- Publication number
- CN109509517A CN109509517A CN201811204706.4A CN201811204706A CN109509517A CN 109509517 A CN109509517 A CN 109509517A CN 201811204706 A CN201811204706 A CN 201811204706A CN 109509517 A CN109509517 A CN 109509517A
- Authority
- CN
- China
- Prior art keywords
- index
- cluster
- standard
- synonymous
- similarity
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
- G16H10/60—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for patient-specific data, e.g. for electronic patient records
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H15/00—ICT specially adapted for medical reports, e.g. generation or transmission thereof
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H70/00—ICT specially adapted for the handling or processing of medical references
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Public Health (AREA)
- Primary Health Care (AREA)
- Physics & Mathematics (AREA)
- Medical Informatics (AREA)
- General Health & Medical Sciences (AREA)
- Epidemiology (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention belongs to computer application field, a kind of medical test Index for examination modified method automatically, this method specifically: data prediction is carried out to the index set of input, obtains the index set that capital and small letter is unified, unit is unified are disclosed;According to the literal feature of index, index gathering is obtained by density-based algorithms;The synonymous index of cluster internal standard index name is found out using two sorting algorithms, it synonymous index and standard index title is carried out index maps to obtain index to be aligned result, it is iteratively repeated the step for being left index non-synonymous, until being only to remain 1 index in synonymous index or cluster in all clusters;Artificial correction and mapping processing, obtain by standardized index name.Experiment shows that this method F1-score can achieve 85.26%.
Description
Technical field
The invention belongs to medical information process fields, more specifically more particularly to a kind of medical test Index for examination from
Move modified method.
Background technique
Due to historical reasons, each hospital is not quite similar about the same appellation for examining Index for examination.Only with " serum sodium "
For, just there are more than 10 kinds of " Na ion concentration ", " NA+ ", " arterial blood sodium ", " blood sodium (Na) " etc. different sayings.Due at present simultaneously
Without complete available index thesaurus to carry out index mapping, this problem has seriously affected interregional medical information
Interconnectivity and sharing.Index for examination is examined to do standardization in area medical health platform as a result, by the same of each hospital
The different appellations of index are mapped to unified title, just seem most important.However, due to examining Index for examination to be related to
A large amount of medical knowledge, the index system of each hospital is numerous and complicated numerous and jumbled in addition, carries out labor standard to it by medical professional
Change, needs to take a substantial amount of time and energy.Therefore, the standardized algorithm for how designing an inspection Index for examination, becomes
Key point.
The standardization issue for examining Index for examination, can be regarded as an entity alignment problem, i.e., by medical treatment & health platform
In actual measurement index be mapped on standard index.It is aligned about entity, method that there are two main classes at present, is different knowledge bases respectively
Entity link in example match and text between middle entity between entity and knowledge base entity.The former often utilizes knowledge base
Middle entity attributes information carries out example match, and the latter often utilizes in text entity in the contextual information of entity and knowledge base
Attribute information carries out entity link.However, Index for examination is examined to be present among electronic health record, only corresponding value and value
Range may be not present attribute information, while also not possess contextual information;Importantly, current China and none standard
Knowledge base provides the title of all indexs.In short, the prior art is all difficult to solve that the standardization of Index for examination is examined to ask
Topic.
The invention proposes a kind of medical test Index for examination modified methods automatically, the experimental results showed that, in Shanghai City 8
On the experimental data set of family's tertiary hospitals, the F1-score of final mapping result can achieve 85.26%.
Summary of the invention
In view of this, the invention discloses a kind of medical test Index for examination modified methods automatically.Its concrete scheme is such as
Under:
Achievement data pretreatment: pre-processing achievement data, realizes that capital and small letter is unified, unit is unified and index reference
Value is extracted;
Index cluster: using the literal feature of index, by density-based algorithms, it is one that different indexs, which is gathered,
Each and every one index cluster, to reduce the alignment range of index;
Two classification in cluster: a title is determined for each cluster, and finds out cluster internal standard name using two sorting algorithms
The synonymous index claimed carries out index mapping, for being left index non-synonymous, is screened out from it a new title, continues
The lookup of synonymous index is carried out using two sorting algorithms, such iteration carries out, until being in synonymous index or cluster in all clusters
Only until surplus 1 index;
Artificial correction and mapping: processing is modified to index alignment result by medical professional again and mapping is handled.
Achievement data in case history is excluded choosing and fills out item, mainly include index in required item by achievement data pretreatment stage
The fields such as title, abbreviation, reference value, unit, affiliated check item, Index for examination result, abnormal index prompt.Wherein, affiliated inspection
Item is looked into because each Hospital Standard is different, Index for examination result is because it varies with each patient, abnormal index prompt is not because having finger for its value
It marks discrimination and loses the meaning as criterion feature.Therefore, available field is only limitted to index name, contracting substantially
It writes, reference value and unit this 4.Data prediction, mainly unified metric capital and small letter, unified metric unit are carried out to index,
And extract index reference value.
Index clustering phase is gathered different indexs in index cluster one by one using density-based algorithms.Base
Carry out cluster dividing according to the tightness degree of sample distribution in the clustering algorithm of density, it mainly investigates the connectivity of sample, and
It can connect and obtain final index alignment result on the basis of sample by continuous extended clustering cluster.
The present invention is based on DBSCAN algorithm, service index title and its abbreviation carry out index cluster.Specifically, it gives and refers to
Mark set D={ x1,x2,...,xn},Wherein,Indicate the finger of i-th of index
Entitling claims,Indicate the title abbreviation of i-th of index,Indicate the index unit of i-th of index,Indicate i-th of index
Index reference value, define ε-neighborhood and kernel object are as follows:
1 (ε-neighborhood) is defined for xi∈ D, its ε-neighborhood be in data set D with xiDistance be not more than ε all samples
This, i.e. N ε (xi)={ xi∈D|dist(xi,xj)≤ε}。
2 (kernel objects) are defined if xiε-neighborhood in include at least minPts sample, i.e., | N ε (xi)|≥
MinPts, then xiIt is a kernel object.
Particularly, it when determining ε-neighborhood, provides and combines distance distjoint(xi,xj): by achievement data xi、xjIt is divided into two
Part calculates, and calculates the index name of multi-hot form (different dimensions indicates different Chinese characters in 0-1 vector) firstCOS distance:
Then parameter is abridgedEditing distance:
WhereinIt is index abbreviationString length,Indicate byIt is inserted into, replaced
It changes, delete operation changes intoRequired minimal action number.Finally, using comprehensive two distances of harmonic average obtain combining away from
From:
Clustering algorithm is constantly extended to the outside from kernel object, and then generates clustering cluster, collects C={ C1,C2,...,
Cm, wherein index cluster CiIn include that index name and index are abridged.
Since cluster is a unsupervised learning process, there may be two problems for it: 1) gathering the index reality for cluster
Medical implication is different on border, but because title is close or abbreviation is similar and is classified as cluster;2) some outliers are neither core
Object, and cannot be accessed by kernel object, because without being clustered.It is therefore possible to use any in the following two kinds method
Two methods of a kind of or simultaneous selection post-process cluster result.
Method one, unit verifying.Assuming that the unit of synonymous index is identical, then can carry out to every cluster index single
The index of not commensurate, is separated into different clusters by position verifying.
Method two, outlier are recommended.For the outlier not being clustered, it is contemplated that outlier and other clusters all distances compared with
Far, it is likely that outlier is established an individual cluster by an inherently completely new index respectively.
Two sorting phase in cluster determines a title for each cluster, and utilizes two sorting algorithms by index in cluster
It is divided into two class of synonymous index and index non-synonymous of title, synonymous index and cluster internal standard index name are subjected to index
Mapping obtains index alignment result;For index cluster CiIn remaining index non-synonymous, filter out one from remaining index list
A new standard index title continues with lookup and index mapping process that two sorting algorithms carry out synonymous index, iteration weight
This multiple process, until being only to remain 1 index in synonymous index or cluster in all clusters.
Particularly, post-processing amendment is carried out to index alignment result for convenience of medical professional, it is contemplated that standard index
It should be most common index, the present invention is using the most index of frequency of occurrence in cluster as standard index.
Two classification in cluster method particularly includes:
Firstly, determining a standard index title for each index cluster Ci in index gathering C;Then, according to knowledge base
Data enhancing is carried out to standard index title, obtains the synonymous index of standard index title;Then, to the finger in an index cluster
Entitling, which claims to abridge with index, utilizes longest common subsequence similarity, Jaccard similarity, cosine similarity, Editing similarity
Method obtains similarity score, obtains piecemeal using piecemeal marking characterization method to the index reference value in an index cluster and obtains
Point, using similarity score and piecemeal score as two sorting algorithm of characteristic use judge whether be standard index title synonymous finger
Mark;Finally, synonymous index is mapped as standard index title, index alignment result is obtained.
The present invention devises two classification of 2 category features for index, is similarity feature and piecemeal marking feature respectively:
The first is characterized in that similarity feature, similarity feature include: longest common subsequence similarity, Jaccard phase
Like degree, cosine similarity, Editing similarity.This category feature mainly consider in cluster each actual measurement index and standard index and its
The title similarity and abbreviation similarity of all synonyms.For the convenience of description, (the abbreviation similarity by taking title similarity as an example
It is similarly), we provide that it is x that index name is surveyed in clusterna, standard index name set isWherein subscript n is the total number of standard index and its synonymous index.
Longest common subsequence similarityWherein | xna| it is
The string length of index name is surveyed,Indicate the maximum common subsequence of two indices title.This
Similarity can be determined that the index of similar hyponymy, such as " blood glucose " and " blood glucose (emergency treatment) " in longest common subsequence phase
Like in degree be 1.
Jaccard similarityThis similarity can be determined that title
Sequentially different indexs, for example the Jaccard similarity of " B-typeNatriuretic Peptide " and " natriuretic peptide Type B " is 1.
Cosine similarityWherein xnaWithIt is
Multi-hot form (different dimensions indicates different Chinese characters in 0-1 vector).This measuring similarity is two multi-
The cosine angle of the index name of hot form, it is influenced smaller by format issues such as similar intermediate insertion "-".
Editing similarityWherein | xna| refer to entitling
Claim xnaString length,It indicates by xnaIt is changed into through insertion, replacement, delete operationIt is required most
Lack number of operations, wherein xnaWithIt is multi-hot form (different dimensions indicates different Chinese characters in 0-1 vector).
This measuring similarity be two multi-hot forms index name editing distance.
Second of piecemeal marking feature.Since Different hospital is to the same index, have in the bound setting of reference value
When can be slightly different, therefore the phenomenon that corresponding to multiple reference values there is an index name in practicing.To cope with this problem,
The present invention is aligned block algorithm using knowledge base entity, proposes the index piecemeal marking algorithm based on reference value.Piecemeal was given a mark
Journey consists of two parts: firstly, each reference value for standard index finds a most like actual measurement index reference therewith
Value;Then, the matching piecemeal between these most like reference values, building actual measurement index and standard index.
Specifically, a certain actual measurement index x in cluster is given, the reference value set corresponding to it isWhereinThe i-th kind of reference range and standard for indicating actual measurement index x refer to
Mark (and its synonymous index) refer to value setWhereinIndicate standard index s
I-th kind of reference range.The present invention gives two indices reference value xrefAnd sref, reference value similarity formula are as follows:
For each reference value of standard indexFound out from cluster one withMost like actual measurement index
Reference valueSo thatAnd the two indexs are formed
Reference value pairAccording to reference value pairIndex set can be constructed to pi=(Xi, Si), wherein Xi
It is for all reference valuesActual measurement index set, SiIt is for all reference valuesStandard index and its synonymous finger
Target set.Give two reference values pairWithReference value is defined to similarity:
Wherein simp_cos(X1, X2) indicate index set X1、X2It is (different in 0-1 vector to be expressed as one-hot form
Dimension indicates different indexs) both rear cosine similarity.
According to below with reference to the accompanying drawings becoming to detailed description of illustrative embodiments, other feature of the invention and aspect
It is clear.
Detailed description of the invention
Reader is after having read a specific embodiment of the invention referring to attached drawing, it will more clearly understands of the invention
Various aspects.Wherein,
Fig. 1 shows an embodiment according to the present invention, test rating title modified flow diagram automatically;
Fig. 2 shows an embodiment according to the present invention, the flow diagrams of index cluster;
Fig. 3 shows an embodiment according to the present invention, the flow diagram of two classification in cluster;
Fig. 4 is the schematic diagram using piecemeal disclosed by the invention marking characterization method to reference value to similarity calculation.
Specific embodiment
In order to keep techniques disclosed in this application content more detailed with it is complete, can refer to attached drawing and of the invention following
Various specific embodiments, identical label represents the same or similar component in attached drawing.However, those skilled in the art
It should be appreciated that embodiment provided hereinafter is not intended to limit the invention covered range.In addition, attached drawing is used only for
It is schematically illustrated, and is drawn not according to its full size.The embodiment of the present invention one discloses a kind of medical test inspection
Index automatic correcting method is looked into, it is shown in Figure 1, this method comprises:
Step S1 carries out data prediction to the index set of input, obtains capital and small letter unification, the index abbreviation of index name
Capital and small letter is unified, index set that index unit is unified
Wherein,Indicate the index name of i-th of index,Indicate the title abbreviation of i-th of index,Indicate i-th of index
Index unit,Indicate the index reference value of i-th of index;
Step S2, according to the literal feature that index name in the index set D and index are abridged, by based on density
Different index names and index abbreviation are clustered obtain index gathering C={ C respectively by clustering algorithm1,C2,...,Cm, wherein
Index cluster CiIn include that index name and index are abridged;
Step S3 is each index cluster C in the index gathering CiIt determines a standard index title, and utilizes two points
Class algorithm finds out the synonymous index of cluster internal standard index name, by the synonymous index and the cluster internal standard index name into
Row index maps to obtain index alignment as a result, for index cluster CiIn remaining index non-synonymous, sieved from remaining index list
A new standard index title is selected, lookup and index mapping process that two sorting algorithms carry out synonymous index are continued with,
It is iteratively repeated step S3, until being only to remain 1 index in synonymous index or cluster in all clusters;
Step S4 carries out artificial correction to index alignment result and mapping is handled, obtains by standardized index
Title.
Particularly, post-processing amendment is carried out to index alignment result for convenience of medical professional, it is contemplated that standard index
It should be most common index, the most index name of frequency of occurrence in entitled each the index cluster Ci of standard index.
Two sorting algorithms can rise decision tree, logistic regression, naive Bayesian, support vector machines or random for gradient
Any one in forest algorithm.
Inventive embodiments two disclose a kind of specific medical test Index for examination automatic correcting method, compared to upper one
Embodiment, the present embodiment technical method have done further instruction and optimization for step S2, as shown in Figure 2.Specifically:
Step S21 gives field parameter ε and minPts and index set D;
Step S22 calculates xiε-field N ε (xi), wherein N ε (xi) expression formula are as follows:
Nε(xi)={ xi∈D|dist(xi,xj)≤ε }, wherein dist (xi,xj) it is index xiAnd xjJoint distance, table
Up to formula are as follows:
WhereinRefer to entitling
ClaimWithCOS distance, expression formula are as follows:
It is index abbreviationWithEditor
Distance, expression formula are as follows:
WhereinRefer to the string length of index abbreviation,Indicate byThrough insertion, replaces, deletes
Except operation changes intoRequired minimal action number.
Step S23, will | N ε (xi) | all x of >=minPtsiIt is added in kernel object collection T;
Step S24 randomly selects a kernel object o from the kernel object collection T, will | N ε (o) | >=minPts's
All indexs are added to index cluster CiIn, then calculate the index cluster CiIn each index ε-field N ε (xj), when | N ε
(xj) | by N ε (x when >=minPtsj) all indexs be added to the index cluster CiIn, until being newly added to the index cluster Ci
In each index ε-field | N ε (xj) | < minPts, finally by the index cluster CiAll kernel objects of middle appearance exist
It is deleted in the kernel object collection T;
Step S25 repeats step S24 and obtains the index gathering C={ C1,C2,...,Cm}。
It should be strongly noted that step S26 can be increased after the present embodiment step S25, to C={ C1,C2,...,
CmIn every cluster index carry out index unit verifying, by the index decomposition of different index units be different clusters.
Step S27 can also be increased after the step S25 of embodiment two, to each not in the finger that peels off of any cluster
Mark, establishes an individual cluster respectively.
Wherein, performance is more preferable when the value that parameter ε in field is 0.35, minPts is 3.As shown in table 1, of the invention based on close
It is poly- that the F1-score of the clustering algorithm of degree is apparently higher than k mean cluster, mean shift algorithm, gauss hybrid models and cohesion level
Class, increase rate is 10% or more.
The performance comparison of the different clustering algorithms of table 1
Inventive embodiments three disclose a kind of specific medical test Index for examination automatic correcting method, compared to upper one
Embodiment, the present embodiment technical method have done further instruction and optimization for step S3, as shown in Figure 3.Specifically:
It step 31, is each index cluster C in the index gathering CiDetermine a standard index title;
Step 32, data enhancing is carried out to standard index title according to existing knowledge base, obtains the same of standard index title
Adopted index;
Step 33, in one index cluster index name and index abbreviation utilize longest common subsequence similarity
Method obtains similarity score simlcs(xna,Sna), similarity score sim is obtained using Jaccard similarity based methodjac(xna,
Sna), similarity score sim is obtained using cosine similarity methodcos(xna,Sna), it is obtained using Editing similarity method similar
Spend score simmed(xna,Sna), the index reference value in one index cluster is divided using piecemeal marking characterization method
Block score, by simlcs(xna,Sna)、simjac(xna,Sna)、simcos(xna,Sna)、simmed(xna,Sna) and the piecemeal score
Judge whether as two sorting algorithm of characteristic use be standard index title synonymous index, wherein
Wherein,Subscript n is the total number of standard index and its same index, | xna| it is real
The string length of index name is surveyed,Indicate the maximum common subsequence of two indices title,Indicate byIt is changed into through insertion, replacement, delete operationRequired minimal action number;
Step 34, synonymous index is mapped as standard index title, obtains index alignment result.
Since medical professional is difficult to enumerate all synonymous indexs without foundation, in addition some indexs might have and name
Claim unrelated synonym (such as " b-type natriuretic peptide " and " brain natriuretic peptide "), therefore in terms of data set generation, except by medical speciality
Personnel mark the synonymous index in part for utilizing SNOMED CT knowledge base, LOINC knowledge base, hundred except classifier training manually
The approach such as degree encyclopaedia 3 carry out the synonym of draw standard index for training.Wherein, SNOMED CT knowledge base is library all over Britain, mesh
Before have no Chinese version, it is therefore desirable to by Baidu translation, Tencent translation, Iciba translation etc. translation tools English index is turned over
It is translated into Chinese index.Wherein, even if to the same index, translation tool is also possible to that different translation results can be obtained, therefore
Translation itself is also one of the approach for obtaining synonym.Table 2 gives " b-type natriuretic peptide " and shows through the enhanced synonymous index of data
Example.
The synonymous index example of table 2
Further, piecemeal score is obtained using piecemeal marking characterization method to the index reference value in one index cluster
Process, comprising:
For each reference value of standard index sFound from cluster one withThe ginseng of most like actual measurement index
Examine valueSo thatAnd the two indexs are formed into reference value pairWherein, the formula of reference value similarity is calculated are as follows:
According to reference value pairIndex set is constructed to pi=(Xi,Si), wherein XiIt is for all reference valuesActual measurement
The set of index, SiIt is for all reference valuesStandard index and its synonymous index set;
The reference value of two reference values pair is calculated to similarity, formula are as follows:
Wherein, simp_cos(X1,X2) indicate to incite somebody to action
Index set X1And X2The cosine similarity both being expressed as after 0-1 vector;
When two reference values pair similarity be greater than threshold values θ, i.e.,When, index set will be surveyed
Close X1、X2With standard index set S1、S2It is included in the same piecemeal BiIn;
To any one block BiMarking, obtains piecemeal score scorei, formula are as follows:
WhereinFor specific gravity shared by block Plays index, S '
For the set of all standard indexs, α is weight parameter, block BiIn all indexs share the same scoreiScore;
When an index appears in M piecemeal simultaneously, then according to the weight beta of different massesiThe weighting for calculating index is flat
Equal score
When an index is not in any one block, then piecemeal score scoreiIt is 0.
Shown in Fig. 4, standard reference valueFor section [0,100], most like actual measurement reference valueFor section [0,
100], therefore it corresponds to index set to for p1=(X1, S1)=({ A, B }, { a, b }).Similarly, standard reference value
Corresponding index set is to p2=(X2, S2)=({ A, B, C }, { a, b }).
As a result,
Threshold values θ is 0.7, α 0.6, βiIt is as shown in table 3 for the experimental result of 1/M, wherein the title of feature field, abbreviation
Title similarity feature, abbreviation similarity feature and reference value piecemeal marking feature are respectively indicated with reference value.
The performance comparison of 3 different classifications algorithm of table
From table 3 it is observed that special when using title similarity feature, abbreviation similarity feature and reference value piecemeal to give a mark
Sign, when being aided with GBDT classifier, classifying quality is best, and F1 value is up to 85.26%.No matter which kind of classifier, be substantially with
Increasing for number of features, classifying quality is become better and better, and when using whole three classes characteristic of division, classifying quality reaches best.
Above, a specific embodiment of the invention is described with reference to the accompanying drawings.But those skilled in the art
It is understood that without departing from the spirit and scope of the present invention, can also make to a specific embodiment of the invention each
Kind change and replacement.These changes and replacement are all fallen within the scope of the invention as defined in the claims.
Claims (10)
1. a kind of medical test Index for examination modified method automatically, which is characterized in that the described method comprises the following steps:
Step S1 carries out data prediction to the index set of input, and the capital and small letter for obtaining index name is unified, index abbreviation
Index set D={ the x that capital and small letter is unified, index unit is unified1,x2,...,xn},
Wherein,Indicate the index name of i-th of index,Indicate the title abbreviation of i-th of index,Indicate i-th of index
Index unit,Indicate the index reference value of i-th of index;
Step S2 passes through density clustering according to the literal feature that index name in the index set D and index are abridged
Different index names and index abbreviation are clustered obtain index gathering C={ C respectively by algorithm1,C2,...,Cm, wherein index
Cluster CiIn include that index name and index are abridged;
Step S3 is each index cluster C in the index gathering CiIt determines a standard index title, and is calculated using two classification
Method finds out the synonymous index of cluster internal standard index name, and the synonymous index and the cluster internal standard index name are referred to
Mark mapping obtains index alignment as a result, for index cluster CiIn remaining index non-synonymous, filtered out from remaining index list
One new standard index title continues with lookup and index mapping process that two sorting algorithms carry out synonymous index, iteration
Step S3 is repeated, until being only to remain 1 index in synonymous index or cluster in all clusters;
Step S4 carries out artificial correction to index alignment result and mapping is handled, obtains by standardized index name.
2. the method as described in claim 1, according to the literal feature that index name in the index set D and index are abridged,
By density-based algorithms, different index names and index abbreviation are clustered obtain index gathering C={ C respectively1,
C2,...,CmThe step of, specifically:
Step S21 gives field parameter ε and minPts and index set D;
Step S22 calculates xiε-field N ε (xi), wherein N ε (xi) expression formula are as follows:
Nε(xi)={ xi∈D|dist(xi,xj)≤ε }, wherein dist (xi,xj) it is index xiAnd xjJoint distance, expression formula
Are as follows:
Wherein
It is index nameWithCOS distance, expression formula are as follows:
It is index abbreviationWithEditing distance, expression formula are as follows:
WhereinRefer to the string length of index abbreviation,Indicate byThrough insertion, replacement, delete operation
It changes intoRequired minimal action number;
Step S23, will | N ε (xi) | all x of >=minPtsiIt is added in kernel object collection T;
Step S24 randomly selects a kernel object o from the kernel object collection T, will | N ε (o) | >=minPts's is all
Index is added to index cluster CiIn, then calculate the index cluster CiIn each index ε-field N ε (xj), when | N ε (xj)|≥
By N ε (x when minPtsj) all indexs be added to the index cluster CiIn, until being newly added to the index cluster CiIn it is each
ε-field of index | N ε (xj) | < minPts, finally by the index cluster CiAll kernel objects of middle appearance are in the core
It is deleted in heart object set T;
Step S25 repeats step S24 and obtains the index gathering C={ C1,C2,...,Cm}。
3. method according to claim 2, which is characterized in that increase following steps after step S25:
Step S26, to C={ C1,C2,...,CmIn every cluster index carry out index unit verifying, by different index units
Index decomposition is different cluster.
4. method according to claim 2, which is characterized in that increase following steps after step S25:
Step S27 establishes an individual cluster to each not in the index that peels off of any cluster respectively.
5. method according to claim 2, which is characterized in that the field parameter ε is 0.35, and the value of the minPts is 3.
6. the method as described in claim 1, which is characterized in that the process of step S3, comprising:
It step 31, is each index cluster C in the index gathering CiDetermine a standard index title;
Step 32, data enhancing is carried out to standard index title according to existing knowledge base, obtains the synonymous finger of standard index title
Mark;
Step 33, in one index cluster index name and index abbreviation utilize longest common subsequence similarity based method
Obtain similarity score simlcs(xna,Sna), similarity score sim is obtained using Jaccard similarity based methodjac(xna,Sna),
Similarity score sim is obtained using cosine similarity methodcos(xna,Sna), similarity, which is obtained, using Editing similarity method obtains
Divide simmed(xna,Sna), piecemeal is obtained using piecemeal marking characterization method to the index reference value in one index cluster and is obtained
Divide score ', by simlcs(xna,Sna)、simjac(xna,Sna)、simcos(xna,Sna)、simmed(xna,Sna) and the piecemeal obtain
Be allocated as being characterized using two sorting algorithms judge whether be standard index title synonymous index, wherein
Wherein,Subscript n is the total number of standard index and its same index, | xna| refer to for actual measurement
The string length that entitling claims,Indicate the maximum common subsequence of two indices title,Table
Show byIt is changed into through insertion, replacement, delete operationRequired minimal action number;
Step 34, synonymous index is mapped as standard index title, obtains index alignment result.
7. method as claimed in claim 6, which is characterized in that obtain the mistake of piecemeal score using piecemeal marking characterization method
Journey, comprising:
For each reference value of standard index sFound from cluster one withThe reference value of most like actual measurement indexSo thatAnd the two indexs are formed into reference value pairWherein, the formula of reference value similarity is calculated are as follows:
According to reference value pairIndex set is constructed to pi=(Xi,Si), wherein XiIt is for all reference valuesActual measurement index
Set, SiIt is for all reference valuesStandard index and its synonymous index set;
The reference value of two reference values pair is calculated to similarity, formula are as follows:
Wherein, simp_cos(X1,X2) indicate index
Set X1And X2The cosine similarity both being expressed as after 0-1 vector;
When two reference values pair similarity be greater than threshold values θ, i.e.,When, index set X will be surveyed1、
X2With standard index set S1、S2It is included in the same piecemeal BiIn;
To any one block BiMarking, obtains piecemeal score scorei, formula are as follows:
WhereinFor specific gravity shared by block Plays index, S ' is institute
There is the set of standard index, α is weight parameter, block BiIn all indexs share the same scoreiScore;
When an index appears in M piecemeal simultaneously, then according to the weight beta of different massesiCalculate the weighted average score of index
When an index is not in any one block, then piecemeal score score ' is 0.
8. the method for claim 7, which is characterized in that the threshold values θ is 0.7, and the α is 0.6, the βiFor 1/M.
9. the method as described in claim 1, which is characterized in that entitled each the index cluster C of the standard indexiInterior appearance
The most index name of the frequency.
10. the method as described in claim 1, which is characterized in that two sorting algorithm is that gradient rises decision tree, logic is returned
Return, any one in naive Bayesian, support vector machines or random forests algorithm.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811204706.4A CN109509517A (en) | 2018-10-16 | 2018-10-16 | A kind of medical test Index for examination modified method automatically |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811204706.4A CN109509517A (en) | 2018-10-16 | 2018-10-16 | A kind of medical test Index for examination modified method automatically |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109509517A true CN109509517A (en) | 2019-03-22 |
Family
ID=65746696
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811204706.4A Withdrawn CN109509517A (en) | 2018-10-16 | 2018-10-16 | A kind of medical test Index for examination modified method automatically |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109509517A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111104400A (en) * | 2019-12-24 | 2020-05-05 | 天津新开心生活科技有限公司 | Data normalization method and device, electronic equipment and storage medium |
CN111860359A (en) * | 2020-07-23 | 2020-10-30 | 江苏食品药品职业技术学院 | Point cloud classification method based on improved random forest algorithm |
CN112768058A (en) * | 2021-01-22 | 2021-05-07 | 武汉大学 | Method and device for processing medical data of metering information type |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106933806A (en) * | 2017-03-15 | 2017-07-07 | 北京大数医达科技有限公司 | The determination method and apparatus of medical synonym |
CN107093005A (en) * | 2017-03-24 | 2017-08-25 | 北明软件有限公司 | The method that tax handling service hall's automatic classification is realized based on big data mining algorithm |
CN107818124A (en) * | 2017-03-03 | 2018-03-20 | 平安医疗健康管理股份有限公司 | Data matching method and device |
US20180089300A1 (en) * | 2016-09-23 | 2018-03-29 | International Business Machines Corporation | Merging synonymous entities from multiple structured sources into a dataset |
CN108491406A (en) * | 2018-01-23 | 2018-09-04 | 深圳市阿西莫夫科技有限公司 | Information classification approach, device, computer equipment and storage medium |
-
2018
- 2018-10-16 CN CN201811204706.4A patent/CN109509517A/en not_active Withdrawn
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180089300A1 (en) * | 2016-09-23 | 2018-03-29 | International Business Machines Corporation | Merging synonymous entities from multiple structured sources into a dataset |
CN107818124A (en) * | 2017-03-03 | 2018-03-20 | 平安医疗健康管理股份有限公司 | Data matching method and device |
CN106933806A (en) * | 2017-03-15 | 2017-07-07 | 北京大数医达科技有限公司 | The determination method and apparatus of medical synonym |
CN107093005A (en) * | 2017-03-24 | 2017-08-25 | 北明软件有限公司 | The method that tax handling service hall's automatic classification is realized based on big data mining algorithm |
CN108491406A (en) * | 2018-01-23 | 2018-09-04 | 深圳市阿西莫夫科技有限公司 | Information classification approach, device, computer equipment and storage medium |
Non-Patent Citations (4)
Title |
---|
YAN ZHUANG ET AL.: ""Hike:A hybrid human-machine method for entity Alignment in Large-Scale Knowledge Bases"", 《PROCEEDINGS OF THE 2017 ACM ON CONF ON INFORMATION AND KNOWLEDGE MANAGEMENT》 * |
周保兴著: "《三维激光扫描技术及其在变形监测中的应用》", 31 January 2018 * |
朱灿等: "实体解析技术综述与展望", 《计算机科学》 * |
栗伟等: "一种面向医学短文本的自适应聚类方法", 《东北大学学报(自然科学版)》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111104400A (en) * | 2019-12-24 | 2020-05-05 | 天津新开心生活科技有限公司 | Data normalization method and device, electronic equipment and storage medium |
CN111860359A (en) * | 2020-07-23 | 2020-10-30 | 江苏食品药品职业技术学院 | Point cloud classification method based on improved random forest algorithm |
CN112768058A (en) * | 2021-01-22 | 2021-05-07 | 武汉大学 | Method and device for processing medical data of metering information type |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Christe et al. | Computer-aided diagnosis of pulmonary fibrosis using deep learning and CT images | |
Akter et al. | Prediction of cervical cancer from behavior risk using machine learning techniques | |
CN110059697B (en) | Automatic lung nodule segmentation method based on deep learning | |
Sharma et al. | Determining similarity in histological images using graph-theoretic description and matching methods for content-based image retrieval in medical diagnostics | |
CN112633601B (en) | Method, device, equipment and computer medium for predicting disease event occurrence probability | |
JP2014029644A (en) | Similar case retrieval device and similar case retrieval method | |
CN109509517A (en) | A kind of medical test Index for examination modified method automatically | |
CN111191456B (en) | Method for identifying text segments by using sequence labels | |
CN106845058A (en) | The standardized method of disease data and modular station | |
de Sousa Costa et al. | Classification of malignant and benign lung nodules using taxonomic diversity index and phylogenetic distance | |
Wang et al. | A ResNet‐based approach for accurate radiographic diagnosis of knee osteoarthritis | |
US20230021868A1 (en) | Data-sharding for efficient record search | |
CN111680506A (en) | External key mapping method and device of database table, electronic equipment and storage medium | |
CN107169264B (en) | complex disease diagnosis system | |
CN111028940B (en) | Multi-scale lung nodule detection method, device, equipment and medium | |
CN111524600A (en) | Liver cancer postoperative recurrence risk prediction system based on neighbor2vec | |
CN111581969A (en) | Medical term vector representation method, device, storage medium and electronic equipment | |
CN118312816A (en) | Cluster weighted clustering integrated medical data processing method and system based on member selection | |
Lonij et al. | Open-world visual recognition using knowledge graphs | |
CN109885712A (en) | Lung neoplasm image search method and system based on content | |
CN109783483A (en) | A kind of method, apparatus of data preparation, computer storage medium and terminal | |
CN111640517B (en) | Medical record coding method and device, storage medium and electronic equipment | |
CN116737945B (en) | Mapping method for EMR knowledge map of patient | |
CN116110594B (en) | Knowledge evaluation method and system of medical knowledge graph based on associated literature | |
Wang et al. | Improved V-Net lung nodule segmentation method based on selective kernel |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WW01 | Invention patent application withdrawn after publication | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190322 |