CN109935337A - A kind of medical record lookup method and system based on similarity measurement - Google Patents
A kind of medical record lookup method and system based on similarity measurement Download PDFInfo
- Publication number
- CN109935337A CN109935337A CN201910137294.5A CN201910137294A CN109935337A CN 109935337 A CN109935337 A CN 109935337A CN 201910137294 A CN201910137294 A CN 201910137294A CN 109935337 A CN109935337 A CN 109935337A
- Authority
- CN
- China
- Prior art keywords
- medical record
- medical
- record group
- group
- value
- 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.)
- Granted
Links
Landscapes
- Medical Treatment And Welfare Office Work (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
Abstract
The invention discloses a kind of medical record lookup method and system based on similarity measurement, step of the invention include obtaining medical record group set C for inquiry medical record set A building medical record group；Medical record group data set with similar tags is generated to medical record group set C and closes D；Construct machine learning model, and D is closed by medical record group data set and completes training, all medical records in target medical record, inquiry medical record set A are input to machine learning model together, obtain target medical record and inquire the similarity measure values in medical record set A between all medical records and the highest N number of medical record output of similarity measurement.The present invention takes full advantage of medical record self information and medical record correlation theories knowledge, it can be improved the precision of medical record similarity measurement, promote the accuracy of medical record sequence, the medical record Similarity Measures based on machine learning have better precision improvement potentiality simultaneously, have the advantages that high-precision, high applicability, robustness be good, sustainable optimization potentiality.
Description
Technical field
The present invention relates to the similar medical records of medical field to search technology, and in particular to a kind of medical record based on similarity measurement
Lookup method and system are, it can be achieved that the similar medical record under heterogeneous medical information is searched and similar medical record is sorted based on similitude.
Background technique
Medical record refer to by canonical record patient disease performance and diagnosis and treatment situation archives, specifically include that patient's essential information,
Patient history information, inspection information, doctor's advice information, diagnostic message, therapeutic scheme, state of an illness feedback etc..Medical record describes patient just
The complete state of an illness during examining, is saved patient's state of an illness by way of data information.Using medical record as the research of object point
Analysis is of great significance, and it is exactly in this way, for example when following hospital doctor at county level is for coming to go to a doctor that the medical record in the present invention, which is searched,
It, can be based on the present invention by searching for the current disease of therapeutic scheme auxiliary of expert in similar medical record when the state of an illness of patient can not be held
The diagnosing and treating of people.Therefore, the research that medical record is searched has great theory and practice meaning.
Similar medical record searches the similar sequence for being based primarily upon medical record.The similar sequence of medical record for understand medical record type,
The trend of the state of an illness has extremely important fundamental role in relationship and prediction medical record between identification medical record, is medical record application
Premise and basis.The similar sequence of medical record refers to for specified medical record, by medical record library all medical records and the medical record into
Row compares, and the size for being then based on similitude is ranked up medical record.Wherein, most important work is how to determine two diseases
Similar value size between case, the i.e. similarity measurement of medical record: two medical records are closer, their similarity measurement is also got over
Greatly, and two medical records are more become estranged, their similarity measurement also just it is smaller.
The measure of existing medical record similitude can be divided into two classes: machine learning model based on medical record data and be based on
The traditional theory model of theoretical knowledge.Traditional theory model from medicine domain knowledge, by pathological analysis judge medical record it
Between similitude size relation, it is explanatory good that the advantages of this model, is, the similarity measurements accuracy of measurement between a small amount of medical record is high,
Disadvantage is more demanding to professional knowledge, while being limited by professional domain knowledge, it is big that model accuracy promotes difficulty.It is based on
The machine learning model of medical record data is from medical record data itself, by dividing the medical record data for largely having formed relationship
Analysis study, and then learn to arrive similarity relationships therein, the advantages of this model, is that precision may be positively correlated with data volume, no
It is limited to domain knowledge, disadvantage is explanatory bad, has certain demand to medical record data similarity label.Therefore, it actually answers
There is model accuracy and promote that difficulty is big, medical record similitude label is difficult to obtain asks in the medical record method for measuring similarity in
Topic, which affects the precision improvement of medical record similarity measurement, and then impacts to the accuracy of the similar sequence of medical record.Cause
This, the medical record method for measuring similarity that can obtain medical record similitude label automatically is particularly important, and this method has important reason
By meaning and practice demand.
The Chinese patent literature of Publication No. CN104572675B discloses a kind of system and method for similar case history retrieval,
A kind of case history similarity calculating method based on pathology is devised in the technical solution, and then is realized and retrieved from case history library
The function of similar case history.The technical solution has ignored case history itself from the method that the angle of pathology carries out case history similarity calculation
Data information, and this Similarity Measures can not based on error carry out self-optimizing；Meanwhile the technical solution is with case history
The case where object carries out similar to search, and case history can not completely react patient, the part of similar only patient's state of an illness of case history
It is similar.Yang Hui et al. discloses a kind of similar case history retrieval system based on medical big data platform on " southeast national defence medicine "
It unites, natural language processing technique is based primarily upon in the technical solution, realize the similar disease in medical big data platform in case history library
Search function is gone through, the Chinese patent literature one of the method for the technical solution and previously described Publication No. CN104572675B
Sample has certain defect in the integrality of accuracy and case history the reaction state of an illness of similarity measurement.Therefore how to realize
The accuracy of medical record inquiry, integrity degree still have the space advanced optimized.
Summary of the invention
The technical problem to be solved in the present invention: it in view of the above problems in the prior art, provides a kind of based on similarity measurement
Medical record lookup method and system, the present invention can be good at the similarity measurement suitable for medical record, can be good at solving
The missings of medical record similar tags, the realistic problems such as not perfect of medical record theory of similarity knowledge, take full advantage of medical record self information
With medical record correlation theories knowledge, the precision of medical record similarity measurement can be improved, promote the accuracy of medical record sequence, be based on simultaneously
The medical record Similarity Measures of machine learning have better precision improvement potentiality, have high-precision, high applicability, robustness
The advantages of good, sustainable optimization potentiality.
In order to solve the above-mentioned technical problem, the technical solution adopted by the present invention are as follows:
A kind of medical record lookup method based on similarity measurement, implementation steps include:
1) medical record group set C is obtained for inquiry medical record set A building medical record group；
2) the medical record group data set with similar tags is obtained for the medical record group set C similar tags assignment for carrying out medical record group
Close D；
3) machine learning model is constructed, and D is closed by the medical record group data set with similar tags and is completed to machine learning mould
The training of type, the machine learning model establish the mapping relations between medical record group, medical record group similarity by training；
4) all medical records in target medical record, inquiry medical record set A are input to machine learning model together, obtain target disease
Similarity measure values in case and inquiry medical record set A between all medical records, select the highest N number of medical record of similarity measure values defeated
Out.
Preferably, the detailed step of step 1) includes:
1.1) it for all medical records in inquiry medical record set A, is carried out fully intermeshing two-by-two and combines to obtain medical record group collection
B is closed, the element in medical record group set B is medical record group and each medical record group is made of two medical records；
1.2) medical record group selection set C is obtained for medical record group set B random selection part medical record group0；For medical record group
Set B calculates separately similar value index value according to specified a variety of similar value indexs, and is directed to different similar value indexs, respectively
Descending is carried out based on similar value index value to arrange to obtain medical record group ordered set Bi；For all medical record group ordered set Bi, point
It Ji Yu not similar value index probability assignments selection generation medical record group selection set Ci, specified a variety of similar value indexs include
Euclidean distance, COS distance, Jie Kade distance, adjustment COS distance at least two；
1.3) by medical record group selection set C0, all medical record group selection set CiIt carries out set merging and obtains medical record group collection
Close C.
Preferably, described that medical record group selection set C is generated based on the selection of similar value index probability assignmentsiWhen, medical record group has
Ordered sets BiIn each medical record group randomly selected probability such as formula (1) shown in；
In formula (1), P (SMi) it is medical record group ordered set BiIn the probability that is selected of i-th of medical record group, SMiFor medical record group
Ordered set BiIn i-th of medical record group similar value index value, f (SMj) it is medical record group ordered set BiIn i-th medical record group
Similar value is through Regularization as a result, m is medical record group ordered set BiThe number of middle medical record group, f (SMj) function expression such as formula
(2) shown in；
In formula (2), SM1For medical record group ordered set BiIn the 1st medical record group similar value index value, SMmHave for medical record group
Ordered sets BiIn m-th of medical record group similar value index value, m be medical record group ordered set BiThe number of middle medical record group.
Preferably, the detailed step of step 2) includes:
2.1) medical record group set C is expressed as medical record group set matrix b, every a line table of the medical record group set matrix b
Show a medical record, the diagnosis that the preceding n column of medical record respectively indicate n feature of the medical record, last column feature s is the medical record is believed
Breath；
2.2) weighted value of each feature of medical record is determined according to the medical record group set matrix b；
2.3) pass through each in similitude and the corresponding weight calculation medical record group set C of feature between the feature of medical record
The similar value s of medical record group, and then obtain the medical record group set D with similar tags.
Preferably, the detailed step of step 2.2) includes: the original weight that each feature of medical record is calculated according to formula (4)
Value, is normalized the weighted value final as each feature for the original weighted value of all features；
In formula (4), yi' indicate medical record every i feature original weighted value,Indicate that medical record group set matrix b is all
The feature vector that i-th column feature of medical record is constituted；Indicate what the diagnostic message s of all medical records of medical record group set matrix b was constituted
Diagnostic message feature vector；σiIndicate the variance of the i-th column feature of all medical records of medical record group set matrix b.
Preferably, the function expression such as formula of the similar value s of each medical record group in medical record group set C is calculated in step 2.3)
(6) shown in；
In formula (6), sijIndicate that, by the similitude of medical record i and medical record j the medical record group constituted, n is characterized total quantity,For
The value of x-th of feature of medical record iFor the value of x-th of feature of medical record j, yxFor normalization after x-th of feature weight,For the maximum value of x-th of feature,For the minimum value of x-th of feature.
Preferably, the detailed step of building machine learning model includes: in step 3)
3.1) scoring loss function, sequence loss function, sequence three kinds of loss functions of probability loss function are separately designed；Its
In, the input for the loss function that scores is two medical record data, output is similarity score, and scoring loss function passes through prediction scoring
Absolute value representation between value and label score value；The input of sequence loss function is three medical record data, one of inquiry
Medical record, two sequence medical records export as similarity score, and the loss function that sorts passes through inquiry medical record respectively at two sequence medical records
Between predict score value height and label score value height between difference indicate；Sequence probability loss function input be
Three medical record data, one of inquiry medical record, two sequence medical records export and compare probability value, sequence probability loss for scoring
Function is predicted between the height of score value and the height of label score value between two sequence medical records by inquiry medical record
Disparity probability indicates；
3.2) neural network model is constructed, which is made of input layer, hidden layer and output layer, input layer
For all dimensions of medical record completely to be inputted network access network, hidden layer is to be fully connected layer network, the characteristic processing for medical record；It is defeated
Layer is used to export the similarity measure values between two medical records out；
3.3) respectively using loss function, sequence loss function, sequence three kinds of loss functions of probability loss function as nerve
The loss function of network model simultaneously chooses the wherein best loss function of effect；
3.4) according to the activation primitive of the type selection neural network model of selection loss function, loss function selection scoring
Activation primitive uses linear activation primitive when loss function, and activation primitive uses tanh when loss function selected and sorted loss function
Function is as activation primitive, and activation primitive is using sigmoid function as swashing when loss function selected and sorted probability loss function
Function living.
Preferably, three kinds of probability loss function design scoring loss function, sequence loss function, sequence damages in step 3.1)
When losing function, shown in the function expression for the loss function that scores such as formula (7), the function expression for the loss function that sorts such as formula (8)
It is shown, it sorts shown in the function expression such as formula (9) of probability loss function；
In formula (7), L (θ) is the penalty values of loss function, and t is the quantity of medical record group；For medical recordWith disease
CaseModel prediction score value,For medical recordWith medical recordLabel score value；
In formula (8), L (θ) is the penalty values of sequence loss function, and t is the quantity of medical record group,For medical record qiWith
Medical recordNeural Network model predictive score value,For medical record qiWith medical recordNeural Network model predictive scoring
Value,For medical record qiWith medical recordLabel score value,For medical record qiWith medical recordLabel score value,
Sign is sign function；
In formula (9), L (θ) indicates the penalty values of sequence probability loss function, and t indicates the quantity of medical record group,
Indicate medical record q under label score valueiAnd medical recordBetween similitude be greater than medical record qiAnd medical recordBetween similitude probability,Indicate medical record q under model prediction score valueiAnd medical recordBetween similitude be greater than medical record qiAnd medical recordBetween it is similar
The probability of property；WhereinFunction expression respectively as shown in formula (10) and formula (11)；
In formula (10) and formula (11),For medical record qiWith medical recordLabel score value,For medical record qiWith
Medical recordLabel score value,For medical record qiWith medical recordNeural Network model predictive score value,
For medical record qiWith medical recordNeural Network model predictive score value.
The present invention also provides a kind of, and the medical record based on similarity measurement searches system, including computer equipment, the calculating
Machine equipment is programmed to perform the step of aforementioned medical record lookup method based on similarity measurement of the invention or the computer
It is stored with and is programmed to perform based on the aforementioned medical record lookup method by similarity measurement of the present invention in the storage medium of equipment
Calculation machine program.
The present invention also provides a kind of computer readable storage medium, it is stored with and is compiled in the computer readable storage medium
Journey is to execute the computer program of the aforementioned medical record lookup method based on similarity measurement of the present invention.
Compared to the prior art, the present invention has an advantage that
1. the present invention is based on the lookups of the medical record of similarity measurement great demand and important meaning in practical applications
Justice.The even phenomenon of maldistribution of the resources is presented in existing medical field: a large amount of medical resource and Expert Resources concentrate on minority
Large hospital in, most of following hospital at county level only possesses a small number of medical resources and doctor's professional skill is integrally relatively low, but
Actually these following hospitals at county level medical object for being most patients, therefore, there have been most patients to obtain for this
To the state of high-level medical services.The present invention can alleviate the problem to a certain extent, include magnanimity in medical record library
Medical record data, these medical record data include diagnosis and treatment condition of the expert to patient, and essence is exactly a kind of medical resource, when
When following hospital doctor at county level can not hold the state of an illness for the outpatients that come, can by patient carry out trial inspection,
A primary disease is formed in the input systems such as information to check patient's essential information, patient symptom, patient history, patient
Then the primary medical record is input in medical record lookup system by case as a whole, the present invention can be based between medical record
Similitude similar medical record of output par, c from medical record library, in this way, following hospital doctor at county level by diagnosis to similar medical record and
The analysis of therapeutic scheme uses for reference expert doctor to the diagnosing and treating situation of similar patient, and then auxiliary diagnosis treatment is current sick
Expert Resources, can be carried out knowledge sharing by people by way of electronic data through the invention, and medical assistance preferably services
In medical field.
2, the present invention can be good at the similarity measurement suitable for medical record, can be good at solving medical record similar tags
Missing, the realistic problems such as not perfect of medical record theory of similarity knowledge, take full advantage of medical record self information reason related to medical record
By knowledge, the precision of medical record similarity measurement can be improved, promote the accuracy of medical record sequence, while the disease based on machine learning
Case Similarity Measures have better precision improvement potentiality, have that high-precision, high applicability, robustness be good, sustainable optimization
The advantages of potentiality.
3, the method for machine learning is more demanding to the distribution situation of training data, the present invention medical record distribution situation not
A kind of method that multi objective probability assignments are designed in the case where knowing carries out the selection of medical record group, and this method avoids to a certain extent
The distributions shift situation of medical record group data under single index.
4, the present invention is from the actual conditions of medical record data, comprehensive traditional theory model and machine learning model, tradition
Theoretical model completes the work of weak label, and machine learning model carries out the study of medical record similitude, makes full use of the excellent of each model
Gesture improves the accuracy of medical record similarity measurement.
5, for the method for the present invention compared to traditional theoretical model, the application of machine learning techniques is medical record similarity measurement essence
The promotion of degree is provided convenience, and the optimization of data, the adjustment of parameter, the improvement of learning method are all traditional medical record similarity measure sides
The optimization method that method cannot provide, method of the invention are that the Continuous optimization of medical record similarity measurement provides the foundation, Neng Gouzeng
Big medical record similarity measure precision improvement potentiality.
Detailed description of the invention
Fig. 1 is the basic principle schematic of present invention method.
Fig. 2 is the process schematic that medical record group set C is generated in the embodiment of the present invention.
Fig. 3 is the structural schematic diagram of the scoring loss function in present invention method.
Fig. 4 is the structural schematic diagram of the sequence loss function in present invention method.
Fig. 5 is the structural schematic diagram of the sequence probability loss function in present invention method.
Fig. 6 is the structural schematic diagram of the network model in present invention method.
Specific embodiment
As shown in Figure 1, the implementation steps of medical record lookup method of the present embodiment based on similarity measurement include:
1) medical record group set C is obtained for inquiry medical record set A building medical record group；
2) the medical record group data set with similar tags is obtained for the medical record group set C similar tags assignment for carrying out medical record group
Close D；
3) machine learning model is constructed, and D is closed by the medical record group data set with similar tags and is completed to machine learning mould
The training of type, machine learning model establish the mapping relations between medical record group, medical record group similarity by training；
4) all medical records in target medical record, inquiry medical record set A are input to machine learning model together, obtain target disease
Similarity measure values in case and inquiry medical record set A between all medical records, select the highest N number of medical record of similarity measure values defeated
Out.
In practical application, medical record refers to by the archives of canonical record patient disease performance and diagnosis and treatment situation, specifically includes that disease
People's essential information, patient history information, inspection information, doctor's advice information, diagnostic message, therapeutic scheme, state of an illness feedback etc..Medical record
Data format is also varied: key-value pair data, text data, image data, audio data of formatting etc..The present embodiment
The medical record data of middle application are the key-value pair medical record data of the formatting after arranging to original medical record data.
The method of machine learning is more demanding to the distribution situation of training data, and the present invention is unknown in the distribution situation of medical record
In the case where for inquiry medical record set A building medical record group obtain medical record group set C when, specifically be based on multi objective probability assignments
Mode construct medical record group and obtain medical record group set C, this method avoids medical record group data under single index to a certain extent
Distributions shift situation, so as to achieve the purpose that reduce data distribution error.
In the present embodiment, the detailed step of step 1) includes:
1.1) it for all medical records in inquiry medical record set A, is carried out fully intermeshing two-by-two and combines to obtain medical record group collection
B is closed, the element in medical record group set B is medical record group and each medical record group is made of two medical records；
1.2) medical record group selection set C is obtained for medical record group set B random selection part medical record group0；For medical record group
Set B calculates separately similar value index value according to specified a variety of similar value indexs, and is directed to different similar value indexs, respectively
Descending is carried out based on similar value index value to arrange to obtain medical record group ordered set Bi；For all medical record group ordered set Bi, point
It Ji Yu not similar value index probability assignments selection generation medical record group selection set Ci；
1.3) by medical record group selection set C0, all medical record group selection set CiIt carries out set merging and obtains medical record group collection
Close C.
In the present embodiment, medical record group selection set C is generated based on the selection of similar value index probability assignmentsiWhen, medical record group has
Ordered sets BiIn each medical record group randomly selected probability such as formula (1) shown in；
In formula (1), P (SMi) it is medical record group ordered set BiIn the probability that is selected of i-th of medical record group, SMiFor medical record group
Ordered set BiIn i-th of medical record group similar value index value, f (SMj) it is medical record group ordered set BiIn i-th medical record group
Similar value is through Regularization as a result, m is medical record group ordered set BiThe number of middle medical record group, f (SMj) function expression such as formula
(2) shown in；
In formula (2), SM1For medical record group ordered set BiIn the 1st medical record group similar value index value, SMmHave for medical record group
Ordered sets BiIn m-th of medical record group similar value index value, m be medical record group ordered set BiThe number of middle medical record group.
In the present embodiment, specified a variety of similar value indexs include Euclidean distance, COS distance, Jie Kade distance, adjustment
Furthermore COS distance also can according to need and increase other similar values using therein at least two or further progress extension
Index.
Euclidean distance: referring to fig. 2, Euclidean is calculated separately according to specified a variety of similar value indexs for medical record group set B
Distance is used as similar value index value, and is directed to different similar value indexs, is based respectively on similar value index value and carries out descending arrangement
Obtain the first medical record group ordered set B1；For the first all medical record group ordered set B1.First medical record group ordered set B1In
Element, the selection of medical record group is carried out based on similar value, the similar value of each element is denoted as SM respectively in ordered set1、SM2…
SMm, descending arrangement, the probability such as formula (1) of specific medical record group selection, (2) are shown, are based ultimately upon similar value index probability
Distribution selection generates the first medical record group selection set C1。
COS distance: referring to fig. 2, cosine is calculated separately according to specified a variety of similar value indexs for medical record group set B
Distance is used as similar value index value, and is directed to different similar value indexs, is based respectively on similar value index value and carries out descending arrangement
Obtain the second medical record group ordered set B2；For the second all medical record group ordered set B2.Second medical record group ordered set B2In
Element, the selection of medical record group is carried out based on similar value, the similar value of each element is denoted as SM respectively in ordered set1、SM2…
SMm, descending arrangement, the probability such as formula (1) of specific medical record group selection, (2) are shown, are based ultimately upon similar value index probability
Distribution selection generates the second medical record group selection set C2。
Jie Kade distance: referring to fig. 2, outstanding person is calculated separately according to specified a variety of similar value indexs for medical record group set B
Card moral distance is used as similar value index value, and is directed to different similar value indexs, is based respectively on similar value index value and carries out descending
Arrangement obtains third medical record group ordered set B3；For all third medical record group ordered set B3.Third medical record group ordered set
B3In element, the selection of medical record group is carried out based on similar value, the similar value of each element is denoted as SM respectively in ordered set1、
SM2…SMm, descending arrangement, the probability such as formula (1) of specific medical record group selection, (2) are shown, are based ultimately upon similar value index
Probability assignments selection generates third medical record group selection set C2。
It adjusts COS distance: referring to fig. 2, being calculated separately for medical record group set B according to specified a variety of similar value indexs
COS distance is adjusted as similar value index value, and is directed to different similar value indexs, is based respectively on the progress of similar value index value
Descending arranges to obtain filatow-Dukes disease case group ordered set B2；For all filatow-Dukes disease case group ordered set B4.Filatow-Dukes disease case group is orderly
Set B4In element, the selection of medical record group is carried out based on similar value, the similar value of each element is denoted as SM respectively in ordered set1、
SM2…SMm, descending arrangement, the probability such as formula (1) of specific medical record group selection, (2) are shown, are based ultimately upon similar value index
Probability assignments selection generates filatow-Dukes disease case group selection set C3。
Referring to fig. 2, finally by medical record group selection set C0, the first medical record group selection set C1, the second medical record group selection set
C2, third medical record group selection set C2, filatow-Dukes disease case group selection set C3It carries out set merging and obtains medical record group set C.Obtain disease
Case group set C is a kind of method for carrying out the building of medical record group to medical record based on multi objective probability assignments, can by the above method
The medical record group for being largely suitable for avoids the distributions shift situation of medical record group data under single index to a certain extent, from
And it can achieve the purpose that reduce data distribution error.
In the present embodiment, the detailed step of step 2) includes:
2.1) medical record group set C is expressed as medical record group set matrix b, every a line of medical record group set matrix b indicates one
The preceding n column of a medical record, medical record respectively indicate n feature of the medical record, last column feature s as the diagnostic message of the medical record；
2.2) weighted value of each feature of medical record is determined according to medical record group set matrix b；
2.3) pass through each in similitude and the corresponding weight calculation medical record group set C of feature between the feature of medical record
The similar value s of medical record group, and then obtain the medical record group set D with similar tags.
The basis that accurately description is the similar sequence of medical record is carried out to medical record, medical record is described as by belonging in the present embodiment
Property feature constitute vector:Wherein, biIndicate medical record i；Indicate the attributive character of medical record；siIt indicates
The specific characteristic of medical record --- diagnostic message.Therefore, shown in the function expression such as formula (3) of medical record group set matrix b；
Referring to formula (3), every a line of medical record group set matrix b indicates that a medical record, the preceding n column of medical record respectively indicate the disease
N feature of case, the diagnostic message that last column feature s is the medical record, by taking the medical record of the 1st row as an example, whereinIt indicates
N feature of the medical record, last column feature s1For the diagnostic message of the medical record.
In the present embodiment, the detailed step of step 2.2) include: according to formula (4) calculate medical record each feature it is original
The weighted value final as each feature, normalizing is normalized for the original weighted value of all features in weighted value
Shown in the function expression such as formula (5) for changing processing；
In formula (4), yi' indicate medical record every i feature original weighted value,Indicate that medical record group set matrix b is all
The feature vector that i-th column feature of medical record is constituted, Indicate all medical records of medical record group set matrix b
The diagnostic message feature vector that diagnostic message s is constituted,σiIndicate all medical records of medical record group set matrix b
I-th column featureVariance.The processing of polyfactorial tax power is one and is widely used but difficult problem, medical record phase
Like in property metrics process, the tax power of different data structure corresponding contents is handled since difficulty of its modeling makes in medical record data
It must be difficult to assign from the angle of model and weigh, and general subjective weighting method excessively relies on subjective factor and easily generates subjective error,
The original weighted value for calculating each feature of medical record in the present embodiment according to formula (4) is the visitor proposed based on data stability
Enabling legislation is seen, this method can reduce such error to a certain extent, assign power error so as to achieve the purpose that reduce.
In formula (4), yi' indicate medical record every i feature original weighted value, n indicate medical record group set matrix b it is ill
The total quantity of case, yj' indicate medical record every j feature original weighted value.
In the present embodiment, the function expression of the similar value s of each medical record group in medical record group set C is calculated in step 2.3)
As shown in formula (6)；
In formula (6), sijIndicate that, by the similitude of medical record i and medical record j the medical record group constituted, n is characterized total quantity,For disease
The value of x-th of feature of case iFor the value of x-th of feature of medical record j, yxFor normalization after x-th of feature weight,For the maximum value of x-th of feature,For the minimum value of x-th of feature.
For the medical record group data in medical record group set C={ (b1, b2), (b3, b4) ... } in the present embodiment, it is based on BM25
The similitude between medical record is characterized by the method for similitude and the corresponding weight of feature between feature in algorithm, specifically
Shown in calculation method such as formula (6), the similar value s of each medical record group in set C is obtained by calculation, and then obtains band similar tags
Medical record group set D={ (b1, b2, sm12)、(b3,b4,sm34)…},sm12Indicate the medical record being made of medical record b1 and medical record b2
The similitude of group, i.e. similitude between medical record b1 and medical record b2.
In the present embodiment, the detailed step of building machine learning model includes: in step 3)
3.1) scoring loss function, sequence loss function, sequence three kinds of loss functions of probability loss function are separately designed,
In:
As shown in figure 3, the input of scoring loss function is two medical record data, output is similarity score, scoring loss
Function passes through the absolute value representation between prediction score value and label score value；
As shown in figure 4, the input of sequence loss function is three medical record data, one of inquiry medical record, two sequences
Medical record exports as similarity score, and the loss function that sorts predicts scoring between two sequence medical records by inquiry medical record
Difference between the height of value and the height of label score value indicates；
As shown in figure 5, the input of sequence probability loss function is three medical record data, one of inquiry medical record, two
Sort medical record, exports and compares probability value for scoring, sequence probability loss function by inquiry medical record respectively at two sequence medical records it
Between predict score value height and label score value height between disparity probability indicate；
3.2) neural network model is constructed, as shown in fig. 6, the neural network model is by input layer, hidden layer and output layer
Composition, input layer are used to all dimensions of medical record completely inputting network access network, and hidden layer is to be fully connected layer network, for medical record
Characteristic processing；Output layer is used to export the similarity measure values between two medical records；
3.3) respectively using loss function, sequence loss function, sequence three kinds of loss functions of probability loss function as nerve
The loss function of network model simultaneously chooses the wherein best loss function of effect；
3.4) according to the activation primitive of the type selection neural network model of selection loss function, loss function selection scoring
Activation primitive uses linear activation primitive when loss function, and activation primitive uses tanh when loss function selected and sorted loss function
Function is as activation primitive, and activation primitive is using sigmoid function as swashing when loss function selected and sorted probability loss function
Function living.
In the present embodiment, design scoring loss function, sequence loss function, sequence probability loss function three in step 3.1)
When kind loss function, scoring loss function is used to regard the scoring of medical record similar value as loss function, the function for the loss function that scores
Shown in expression formula such as formula (7)；The loss function that sorts is used for using medical record similar value ordering relation as loss function, sequence loss letter
Shown in several function expressions such as formula (8)；The probability loss function that sorts is used for using medical record similar value sequence probability as loss letter
Number sorts shown in the function expression such as formula (9) of probability loss function；
In formula (7), L (θ) is the penalty values of loss function, and t is the quantity of medical record group；For medical recordWith medical recordModel prediction score value,For medical recordWith medical recordLabel score value；
In formula (8), L (θ) is the penalty values of sequence loss function, and t is the quantity of medical record group,For medical record qiWith
Medical recordNeural Network model predictive score value,For medical record qiWith medical recordNeural Network model predictive scoring
Value,For medical record qiWith medical recordLabel score value,For medical record qiWith medical recordLabel score value,
Sign is sign function；
In formula (9), L (θ) indicates the penalty values of sequence probability loss function, and t indicates the quantity of medical record group,
Indicate medical record q under label score valueiAnd medical recordBetween similitude be greater than medical record qiAnd medical recordBetween similitude probability,Indicate medical record q under model prediction score valueiAnd medical recordBetween similitude be greater than medical record qiAnd medical recordBetween it is similar
The probability of property；WhereinFunction expression respectively as shown in formula (10) and formula (11)；
In formula (10) and formula (11),For medical record qiWith medical recordLabel score value,For medical record qiWith
Medical recordLabel score value,For medical record qiWith medical recordNeural Network model predictive score value,
For medical record qiWith medical recordNeural Network model predictive score value.
For the medical record group set D with similar tags, enter data into neural network model type, according to designed
Model carries out the training of medical record similitude, finally obtains a kind of neural network mould that can measure two medical record similitude sizes
Type, that is, return device.The recurrence device is a kind of network structure that parameter determines, is a kind of side for judging two medical record similar values
Method, realize input two medical records, export medical record similar value function, such as input medical record b1 and b2, output medical record b1 and
The medical record similar value s (b1, b2) of b2.On this basis, for target medical record and inquiry medical record set A, then medical record will be inquired
All medical records are input to together with medical record return in device respectively in set A, obtain owning in target medical record and inquiry medical record set A
The similar value of medical record, before finding out wherein the big similar value of N to get arrived with the highest N number of medical record of target medical record similarity, N's
Value, which can according to need, is specified, and can be one or more.It should be noted that inquiry medical record set hereinbefore
A, medical record group set B, medical record group selection set C0, the first medical record group selection set C1, the second medical record group selection set C2, third
Medical record group selection set C2, filatow-Dukes disease case group selection set C3, medical record group set C, letter involved in medical record group set D only
For any specific restriction should not be constituted to above-mentioned data acquisition system itself immediately for distinguishing above-mentioned data acquisition system.
The present embodiment has fully considered the various actual conditions during medical record similarity measurement, tired to the reality of the process
Difficulty is described in detail, and has carried out comprehensive discussion for each stage of medical record similarity measurement, in the analysis in each stage
In propose technological difficulties and correspondence gives solution, finally integrate the analysis in each stage, give a kind of based on weak prison
Superintend and direct the medical record method for measuring similarity of machine learning, the present embodiment take full advantage of machine learning model based on medical record data and
The advantage of two class method of traditional theory model based on theoretical knowledge, the deficiency for avoiding two class methods.Weakly supervised machine learning
It is a kind of machine learning model for attempting to construct prediction by weaker supervision, the modeling provides a kind of synthesis to be based on theory
The method of the theoretical model of knowledge and the machine learning model based on data, the present embodiment method is using theoretical model without number
Create Weakly supervised label according to advantage is relied on, using machine learning model unrestricted domain knowledge the advantages of carry out Weakly supervised
It practises.This method can be good at the similarity measurement suitable for medical record, can be good at solving the missing of medical record similar tags,
The realistic problems such as not perfect of medical record theory of similarity knowledge.
In order to further be verified to the present embodiment based on the medical record lookup method of similarity measurement, JK doctor used below
Treat the practical medical record data and common data sets Robust04 of data center；Evaluation index uses MAP, P@20 and nDCG@20,
The mean accuracy of middle all retrieval medical records of MAP, P@20 indicate that the mean accuracy of preceding 20 medical records of retrieval, nDCG@20 are retrieval
Preceding 20 medical records aggregated rebates precision, i.e. each of which medical record precision respective weights are different, and forward medical record weight is larger；It is logical
Two class medical record data sets are crossed to be tested；The distribution of the MAP of each algorithm, P@20 and nDCG@20 such as 1 institute of table under different data collection
Show.
Table 1: medical record similarity measurement precision comparison result.
Referring to table 1 it is found that with existing based on theoretical model method (BM25), the Weakly supervised learning algorithm based on SVM
(RanksSVM) it compares, in the case where sorting probability loss function as loss function using use, the present embodiment is based on similar
Property measurement medical record lookup method (this method) have biggish advantage under each evaluation index；Wherein, under 20 index of nDCG@
This method advantage is maximum, and this method advantage is relatively small under MAP index, shows that the present embodiment is looked into based on the medical record of similarity measurement
The medical record group for looking for method to have larger similitude is sensitive.
It is compared with the Chinese patent literature of Publication No. CN104572675B, medical record of the present embodiment based on similarity measurement
Lookup method is based on machine learning method from the angle of data and carries out medical record similarity measurement, takes full advantage of medical record data letter
Breath, provides the self-optimization function based on error；It meanwhile with medical record being that retrieval object avoids disclosed above number and is
Case history can not completely react the problem of patient's state of an illness in the Chinese patent literature of CN104572675B.Yang Hui et al. is in " southeast state
Anti- medicine " on disclose a kind of similar case history searching system based on medical big data platform, the method for the technical solution and before
The Chinese patent literature of the Publication No. CN104572675B of face description is the same, anti-in the accuracy and case history of similarity measurement
Answering has certain defect in the integrality of the state of an illness, the method that the present embodiment is proposed based on the medical record lookup method of similarity measurement
Solves the problems, such as this to a certain extent.
In addition, the present embodiment also provides a kind of medical record lookup system based on similarity measurement, including computer equipment, it should
Computer equipment is programmed to perform the step of the present embodiment aforementioned medical record lookup method based on similarity measurement.The present embodiment
A kind of medical record lookup system based on similarity measurement, including computer equipment, the storage medium of the computer equipment are also provided
In be stored with the computer program for being programmed to perform the aforementioned medical record lookup method based on similarity measurement of the present embodiment.This reality
It applies example and a kind of computer readable storage medium is also provided, be stored in the computer readable storage medium and be programmed to perform this reality
Apply the computer program of the aforementioned medical record lookup method based on similarity measurement of example.
The above is only a preferred embodiment of the present invention, protection scope of the present invention is not limited merely to above-mentioned implementation
Example, all technical solutions belonged under thinking of the present invention all belong to the scope of protection of the present invention.It should be pointed out that for the art
Those of ordinary skill for, several improvements and modifications without departing from the principles of the present invention, these improvements and modifications
It should be regarded as protection scope of the present invention.
Claims (10)
1. a kind of medical record lookup method based on similarity measurement, it is characterised in that implementation steps include:
1) medical record group set C is obtained for inquiry medical record set A building medical record group；
2) the medical record group data set with similar tags is obtained for the medical record group set C similar tags assignment for carrying out medical record group and closes D；
3) machine learning model is constructed, and D is closed by the medical record group data set with similar tags and is completed to machine learning model
Training, the machine learning model establish the mapping relations between medical record group, medical record group similarity by training；
4) by target medical record, inquiry medical record set A in all medical records be input to machine learning model together, obtain target medical record and
The similarity measure values in medical record set A between all medical records are inquired, the highest N number of medical record output of similarity measure values is selected.
2. the medical record lookup method according to claim 1 based on similarity measurement, which is characterized in that step 1) it is detailed
Step includes:
1.1) it for all medical records in inquiry medical record set A, is carried out fully intermeshing two-by-two and combines to obtain medical record group set B,
Element in medical record group set B is medical record group and each medical record group is made of two medical records；
1.2) medical record group selection set C is obtained for medical record group set B random selection part medical record group0；For medical record group set B
Similar value index value is calculated separately according to specified a variety of similar value indexs, and is directed to different similar value indexs, is based respectively on
Similar value index value carries out descending and arranges to obtain medical record group ordered set Bi；For all medical record group ordered set Bi, difference base
It selects to generate medical record group selection set C in similar value index probability assignmentsi, specified a variety of similar value indexs include Euclidean
Distance, COS distance, Jie Kade distance, adjustment COS distance at least two；
1.3) by medical record group selection set C0, all medical record group selection set CiIt carries out set merging and obtains medical record group set C.
3. the medical record lookup method according to claim 2 based on similarity measurement, which is characterized in that described based on similar
It is worth the selection of index probability assignments and generates medical record group selection set CiWhen, medical record group ordered set BiIn each medical record group it is random
Shown in the probability of selection such as formula (1)；
In formula (1), P (SMi) it is medical record group ordered set BiIn the probability that is selected of i-th of medical record group, SMiIt is orderly for medical record group
Set BiIn i-th of medical record group similar value index value, f (SMj) it is medical record group ordered set BiIn i-th medical record group it is similar
It is worth through Regularization as a result, m is medical record group ordered set BiThe number of middle medical record group, f (SMj) function expression such as formula (2) institute
Show；
In formula (2), SM1For medical record group ordered set BiIn the 1st medical record group similar value index value, SMmFor medical record group ordered set
Close BiIn m-th of medical record group similar value index value, m be medical record group ordered set BiThe number of middle medical record group.
4. the medical record lookup method according to claim 1 based on similarity measurement, which is characterized in that step 2) it is detailed
Step includes:
2.1) medical record group set C is expressed as medical record group set matrix b, every a line of the medical record group set matrix b indicates one
The preceding n column of a medical record, medical record respectively indicate n feature of the medical record, last column feature s as the diagnostic message of the medical record；
2.2) weighted value of each feature of medical record is determined according to the medical record group set matrix b；
2.3) pass through each medical record in similitude and the corresponding weight calculation medical record group set C of feature between the feature of medical record
The similar value s of group, and then obtain the medical record group set D with similar tags.
5. the medical record lookup method according to claim 4 based on similarity measurement, which is characterized in that step 2.2) it is detailed
Thin step includes: the original weighted value that each feature of medical record is calculated according to formula (4), for the original weighted value of all features
The weighted value final as each feature is normalized；
In formula (4), yi' indicate medical record every i feature original weighted value,Indicate all medical records of medical record group set matrix b
The feature vector that i-th column feature is constituted；Indicate the diagnosis letter that the diagnostic message s of all medical records of medical record group set matrix b is constituted
Cease feature vector；σiIndicate the variance of the i-th column feature of all medical records of medical record group set matrix b.
6. the medical record lookup method according to claim 4 based on similarity measurement, which is characterized in that step 2.3) is fallen into a trap
It calculates in medical record group set C shown in the function expression such as formula (6) of the similar value s of each medical record group；
In formula (6), sijIndicate that, by the similitude of medical record i and medical record j the medical record group constituted, n is characterized total quantity,For medical record i
X-th of feature valueFor the value of x-th of feature of medical record j, yxFor normalization after x-th of feature weight,
For the maximum value of x-th of feature,For the minimum value of x-th of feature.
7. the medical record lookup method according to claim 1 based on similarity measurement, which is characterized in that building in step 3)
The detailed step of machine learning model includes:
3.1) scoring loss function, sequence loss function, sequence three kinds of loss functions of probability loss function are separately designed；Wherein,
The input of scoring loss function is two medical record data, output is similarity score, and scoring loss function passes through prediction score value
With the absolute value representation between label score value；The input of sequence loss function is three medical record data, one of inquiry disease
Case, two sequence medical records export as similarity score, sort loss function by inquiry medical record respectively at two sequence medical records it
Between predict score value height and label score value height between difference indicate；The input of sequence probability loss function is three
A medical record data, one of inquiry medical record, two sequence medical records export and compare probability value for scoring, and sequence probability loses letter
Number predicts the difference between the height of score value and the height of label score value by inquiry medical record between two sequence medical records
Different probability indicates；
3.2) neural network model is constructed, which is made of input layer, hidden layer and output layer, and input layer is used for
All dimensions of medical record are completely inputted into network access network, hidden layer is to be fully connected layer network, the characteristic processing for medical record；Output layer
For exporting the similarity measure values between two medical records；
3.3) respectively using loss function, sequence loss function, sequence three kinds of loss functions of probability loss function as neural network
The loss function of model simultaneously chooses the wherein best loss function of effect；
3.4) according to the activation primitive of the type selection neural network model of selection loss function, loss function selection scoring loss
Activation primitive uses linear activation primitive when function, and activation primitive uses tanh function when loss function selected and sorted loss function
As activation primitive, activation primitive is using sigmoid function as activation letter when loss function selected and sorted probability loss function
Number.
8. the medical record lookup method according to claim 7 based on similarity measurement, which is characterized in that set in step 3.1)
When meter scoring loss function, sequence loss function, sequence three kinds of loss functions of probability loss function, the function for the loss function that scores
Shown in expression formula such as formula (7), shown in the function expression for the loss function that sorts such as formula (8), the function for the probability loss function that sorts
Shown in expression formula such as formula (9)；
In formula (7), L (θ) is the penalty values of loss function, and t is the quantity of medical record group；For medical recordWith medical record's
Model prediction score value,For medical recordWith medical recordLabel score value；
In formula (8), L (θ) is the penalty values of sequence loss function, and t is the quantity of medical record group,For medical record qiWith medical recordNeural Network model predictive score value,For medical record qiWith medical recordNeural Network model predictive score value,For medical record qiWith medical recordLabel score value,For medical record qiWith medical recordLabel score value, sign is
Sign function；
In formula (9), L (θ) indicates the penalty values of sequence probability loss function, and t indicates the quantity of medical record group,It indicates
Medical record q under label score valueiAnd medical recordBetween similitude be greater than medical record qiAnd medical recordBetween similitude probability,Indicate medical record q under model prediction score valueiAnd medical recordBetween similitude be greater than medical record qiAnd medical recordBetween it is similar
The probability of property；WhereinFunction expression respectively as shown in formula (10) and formula (11)；
In formula (10) and formula (11),For medical record qiWith medical recordLabel score value,For medical record qiWith medical recordLabel score value,For medical record qiWith medical recordNeural Network model predictive score value,For disease
Case qiWith medical recordNeural Network model predictive score value.
9. a kind of medical record based on similarity measurement searches system, including computer equipment, it is characterised in that: the computer is set
The step of standby medical record lookup method being programmed to perform described in any one of claim 1~8 based on similarity measurement, or
It is stored with to be programmed to perform described in any one of claim 1~8 in the storage medium of computer equipment described in person and be based on
The computer program of the medical record lookup method of similarity measurement.
10. a kind of computer readable storage medium, it is characterised in that: be stored with and be programmed in the computer readable storage medium
With the computer program of the medical record lookup method based on similarity measurement described in any one of perform claim requirement 1~8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910137294.5A CN109935337B (en) | 2019-02-25 | 2019-02-25 | Medical record searching method and system based on similarity measurement |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910137294.5A CN109935337B (en) | 2019-02-25 | 2019-02-25 | Medical record searching method and system based on similarity measurement |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109935337A true CN109935337A (en) | 2019-06-25 |
| CN109935337B CN109935337B (en) | 2021-01-15 |
Family
ID=66985863
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910137294.5A Active CN109935337B (en) | 2019-02-25 | 2019-02-25 | Medical record searching method and system based on similarity measurement |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109935337B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110647912A (en) * | 2019-08-15 | 2020-01-03 | 深圳久凌软件技术有限公司 | Fine-grained image recognition method and device, computer equipment and storage medium |
| CN111091907A (en) * | 2019-11-15 | 2020-05-01 | 合肥工业大学 | Health medical knowledge retrieval method and system based on similar case library |
| CN112069783A (en) * | 2020-09-10 | 2020-12-11 | 卫宁健康科技集团股份有限公司 | Medical record input method and input system thereof |
| CN112652393A (en) * | 2020-12-31 | 2021-04-13 | 山东大学齐鲁医院 | ERCP quality control method, system, storage medium and equipment based on deep learning |
| CN112836012A (en) * | 2021-01-25 | 2021-05-25 | 中山大学 | Similar patient retrieval method based on sequencing learning |
| WO2024020238A1 (en) * | 2022-07-22 | 2024-01-25 | Opeeka, Inc. | Method and system utilizing machine learning to develop and improve care models for patients in an electronic patient system |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140089003A1 (en) * | 2012-09-27 | 2014-03-27 | University Of Utah Research Foundation | Patient health record similarity measure |
| CN104572675A (en) * | 2013-10-16 | 2015-04-29 | 中国人民解放军南京军区南京总医院 | Similar medical history searching system and method |
| CN106951719A (en) * | 2017-04-10 | 2017-07-14 | 荣科科技股份有限公司 | The construction method and constructing system of clinical diagnosis model, clinical diagnosing system |
| CN107180155A (en) * | 2017-04-17 | 2017-09-19 | 中国科学院计算技术研究所 | A kind of disease forecasting method and system based on Manufacturing resource model |
| CN107193919A (en) * | 2017-05-15 | 2017-09-22 | 清华大学深圳研究生院 | The search method and system of a kind of electronic health record |
| CN107656952A (en) * | 2016-12-30 | 2018-02-02 | 青岛中科慧康科技有限公司 | The modeling method of parallel intelligent case recommended models |
| CN108831559A (en) * | 2018-06-20 | 2018-11-16 | 清华大学 | A kind of Chinese electronic health record text analyzing method and system |
| CN109119134A (en) * | 2018-08-09 | 2019-01-01 | 脉景（杭州）健康管理有限公司 | Medical history data processing method, medical data recommender system, equipment and medium |
-
2019
- 2019-02-25 CN CN201910137294.5A patent/CN109935337B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140089003A1 (en) * | 2012-09-27 | 2014-03-27 | University Of Utah Research Foundation | Patient health record similarity measure |
| CN104572675A (en) * | 2013-10-16 | 2015-04-29 | 中国人民解放军南京军区南京总医院 | Similar medical history searching system and method |
| CN107656952A (en) * | 2016-12-30 | 2018-02-02 | 青岛中科慧康科技有限公司 | The modeling method of parallel intelligent case recommended models |
| CN106951719A (en) * | 2017-04-10 | 2017-07-14 | 荣科科技股份有限公司 | The construction method and constructing system of clinical diagnosis model, clinical diagnosing system |
| CN107180155A (en) * | 2017-04-17 | 2017-09-19 | 中国科学院计算技术研究所 | A kind of disease forecasting method and system based on Manufacturing resource model |
| CN107193919A (en) * | 2017-05-15 | 2017-09-22 | 清华大学深圳研究生院 | The search method and system of a kind of electronic health record |
| CN108831559A (en) * | 2018-06-20 | 2018-11-16 | 清华大学 | A kind of Chinese electronic health record text analyzing method and system |
| CN109119134A (en) * | 2018-08-09 | 2019-01-01 | 脉景（杭州）健康管理有限公司 | Medical history data processing method, medical data recommender system, equipment and medium |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110647912A (en) * | 2019-08-15 | 2020-01-03 | 深圳久凌软件技术有限公司 | Fine-grained image recognition method and device, computer equipment and storage medium |
| CN111091907A (en) * | 2019-11-15 | 2020-05-01 | 合肥工业大学 | Health medical knowledge retrieval method and system based on similar case library |
| CN112069783A (en) * | 2020-09-10 | 2020-12-11 | 卫宁健康科技集团股份有限公司 | Medical record input method and input system thereof |
| CN112652393A (en) * | 2020-12-31 | 2021-04-13 | 山东大学齐鲁医院 | ERCP quality control method, system, storage medium and equipment based on deep learning |
| CN112652393B (en) * | 2020-12-31 | 2021-09-07 | 山东大学齐鲁医院 | ERCP quality control method, system, storage medium and equipment based on deep learning |
| CN112836012A (en) * | 2021-01-25 | 2021-05-25 | 中山大学 | Similar patient retrieval method based on sequencing learning |
| WO2024020238A1 (en) * | 2022-07-22 | 2024-01-25 | Opeeka, Inc. | Method and system utilizing machine learning to develop and improve care models for patients in an electronic patient system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109935337B (en) | 2021-01-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109935337A (en) | A kind of medical record lookup method and system based on similarity measurement | |
| CN109635122A (en) | Intelligent disease inquiry method, apparatus, equipment and storage medium | |
| Feng et al. | Analysis and prediction of students’ academic performance based on educational data mining | |
| WO2016192612A1 (en) | Method for analysing medical treatment data based on deep learning, and intelligent analyser thereof | |
| CN106295186A (en) | A kind of method and system of aided disease diagnosis based on intelligent inference | |
| US20090287503A1 (en) | Analysis of individual and group healthcare data in order to provide real time healthcare recommendations | |
| CN109559300A (en) | Image processing method, electronic equipment and computer readable storage medium | |
| CN108804677A (en) | In conjunction with the deep learning question classification method and system of multi-layer attention mechanism | |
| Dabowsa et al. | A hybrid intelligent system for skin disease diagnosis | |
| WO2020224433A1 (en) | Target object attribute prediction method based on machine learning and related device | |
| CN109378066A (en) | A kind of control method and control device for realizing disease forecasting based on feature vector | |
| CN105825064A (en) | Daily traditional Chinese medicine therapy intelligent consultancy system | |
| Wu et al. | Hybrid TODIM method with crisp number and probability linguistic term set for urban epidemic situation evaluation | |
| KR20110098286A (en) | Self health diagnosis system of oriental medicine using fuzzy inference method | |
| CN114511759A (en) | Method and system for identifying categories and determining characteristics of skin state images | |
| Chakraborty et al. | Case based reasoning methodology for diagnosis of swine flu | |
| CN110473636A (en) | Intelligent doctor's advice recommended method and system based on deep learning | |
| Martsenyuk et al. | On development of machine learning models with aim of medical differential diagnostics of the comorbid states | |
| Song et al. | Research of medical aided diagnosis system based on temporal knowledge graph | |
| WO2021114626A1 (en) | Method for detecting quality of medical record data and related device | |
| Yin et al. | Stroke risk prediction: Comparing different sampling algorithms | |
| CN110033862A (en) | A kind of Chinese medicine Quantitative Diagnosis system and storage medium based on weighted digraph | |
| Lei et al. | Intelligent generation technology of sub-health diagnosis case based on case reasoning | |
| Vijayakumar et al. | An Intelligent stacking Ensemble-Based Machine Learning Model for Heart abnormality | |
| Wei et al. | Medical College Education Data Analysis Method Based on Improved Deep Learning Algorithm |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |