CN110473636A - Intelligent doctor's advice recommended method and system based on deep learning - Google Patents
Intelligent doctor's advice recommended method and system based on deep learning Download PDFInfo
- Publication number
- CN110473636A CN110473636A CN201910778555.1A CN201910778555A CN110473636A CN 110473636 A CN110473636 A CN 110473636A CN 201910778555 A CN201910778555 A CN 201910778555A CN 110473636 A CN110473636 A CN 110473636A
- Authority
- CN
- China
- Prior art keywords
- doctor
- advice
- information
- state
- illness
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
-
- 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
- G16H70/20—ICT specially adapted for the handling or processing of medical references relating to practices or guidelines
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Public Health (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Databases & Information Systems (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Pathology (AREA)
- Bioethics (AREA)
- Medical Treatment And Welfare Office Work (AREA)
Abstract
The present disclosure proposes intelligent doctor's advice recommended method and system based on deep learning is proposed, the intelligent doctor's advice recommended method based on deep learning includes the following steps: step (1): building conditions of patients information bank.Step (2): setting the patient of doctor's advice to be opened as current patents, acquire the state of an illness information of current patents, judge whether current patents have doctor's advice information, if so, executing step (4)；It is no to then follow the steps (3)；Step (3): it determines in conditions of patients information bank with the most similar patient of current patents' state of an illness information, the doctor's advice of most close patient is recommended as to the initial doctor's advice of current patents；Step (4): by the trained deep learning model M of the state of an illness information input of current patents, the output of deep learning model M is the recommendation doctor's advice of current patents.Using depth learning technology, the state of an illness information data of patient's complexity, the suitable doctor's advice of intelligent recommendation are analyzed.
Description
Technical field
This disclosure relates to medical information correlative technology field, in particular to a kind of intelligence based on deep learning
It can doctor's advice recommended method and system.
Background technique
The statement of this part only there is provided background technical information relevant to the disclosure, not necessarily constitutes first skill
Art.
Doctor's advice be exactly doctor according to the state of an illness and treatment need to patient diet, medication, in terms of instruction.Root
According to the state of an illness of patient, selects medical care precess appropriate and issue the groundwork that doctor's advice is doctor, it is a large amount of that this occupies doctor
Working time.For doctor, on the one hand, the number of patients faced daily is more, situation is complicated, and the work for issuing doctor's advice occupies
Their a large amount of times；On the other hand, due to thinking inertia or lack certain medical knowledges, the doctor's advice of doctor it sometimes appear that
Problem.
As information technology is in the application of medical field, using information technology, auxiliary doctor issues doctor's advice, cures for improving
Raw working efficiency, the quality for improving the issued doctor's advice of doctor, have very important significance.Doctor's advice recommender system master at this stage
Medical knowledge base (for example, setting up clinical path) is utilized, i.e., need the doctor's advice template used to be stored in knowledge every kind of disease
In library, in use, recommending the doctor's advice in knowledge base according to the medical diagnosis on disease of patient.Inventors have found that clinical path at this stage
The case where system needs doctor in advance by data inputs such as therapeutic scheme, doctor's advices into system, and when use judges patient by doctor,
The therapeutic scheme needed to be implemented is selected, and then selects doctor's advice.Existing system has following problem:
1, since the state of an illness of different patients is different, doctor, which can not classify all state of an illness, is all included in clinical path, this
It has resulted in causing doctor rapid by clinical path system due to the case where not covering patient in clinical path system
Doctor's advice is issued, therefore, clinical path system can do nothing to help doctor in many cases and improve the efficiency for issuing doctor's advice.
2, doctor needs the case where judging patient, voluntarily selects therapeutic scheme and doctor's advice, clinical path system can not be intelligent
Judge the state of an illness of patient and recommends doctor's advice.
3, doctor can only treat according to preset therapeutic scheme in clinical path system, not utilize medical treatment
The benefit analysis conditions of patients of big data provides different thinkings to doctor, and doctor is helped to select more advanced therapeutic scheme.
Therefore, doctor's advice recommender system at this stage does not play the role of that doctor is assisted to determine doctor's advice.
Summary of the invention
The disclosure to solve the above-mentioned problems, proposes intelligent doctor's advice recommended method and system based on deep learning, adopts
With depth learning technology, the state of an illness information data of patient's complexity, the suitable doctor's advice of intelligent recommendation are analyzed.
To achieve the goals above, the disclosure adopts the following technical scheme that
One or more embodiments provide the intelligent doctor's advice recommended method based on deep learning, include the following steps:
Step (1): building conditions of patients information bank.
Step (2): setting the patient of doctor's advice to be opened as current patents, acquires the state of an illness information of current patents, and judgement is current
Whether patient has doctor's advice information, if so, executing step (4)；It is no to then follow the steps (3)；
Step (3): determining with the most similar patient of current patents' state of an illness information in conditions of patients information bank, will most close trouble
The doctor's advice of person is recommended as the initial doctor's advice of current patents；
Step (4): by the trained deep learning model M of the state of an illness information input of current patents, deep learning model M is defeated
It is out the recommendation doctor's advice of current patents.
Further, conditions of patients information bank is constructed in the step (1), specific steps can be such that
It determines the disease name that include state of an illness information in conditions of patients information bank, there is the disease for every kind of disease acquisition
The state of an illness information of at least N number of patient of disease diagnosis；
Vector data is converted by the state of an illness information of each patient of acquisition；
The vector data of each patient is formed into a data set and is stored respectively in database, forms conditions of patients information
Library.
Further, the state of an illness information of the patient includes text data and numeric data, by each patient's of acquisition
The method that state of an illness information is converted into vector data, specifically:
The each single item of text information in state of an illness information time category is assigned a value of n-dimensional vector respectively, text information is turned
Turn to digital information；Wherein the n of n-dimensional vector refers to the classifiable classification number of each single item time of text information；
Numerical information in the digital information of conversion and state of an illness information is combined, vector data, the vector data are constituted
Including with doctor's advice information related data and with doctor's advice information extraneous data；
Further, it determines in conditions of patients information bank in the step 3 with the most similar patient's of current patents' state of an illness
Method, specifically:
Step 31: converting vector data for the text information each single item in the state of an illness information of the patient p of acquisition；
Step 32: by the vector data of conversion in conjunction with the numeric data in state of an illness information, generating the state of an illness of current patents
Information vector data set；
Step 33: calculating the state of an illness information vector data set of current patents with each patient's in conditions of patients information bank
The similarity of state of an illness message data set, the maximum patient of similarity value are the most close patient of current patents' state of an illness.
Further, the method and step of training deep learning model M includes:
It acquires the state of an illness information of n patient corresponding to each medical diagnosis on disease title and is converted into vector data as sample
Collection；
The deep learning model M based on two-way LSTM model is established, the state of an illness information before doctor's advice is generated with each patient
Vector data is input, to generate the state of an illness information vector data after doctor's advice for output, by the state of an illness of the patient in sample set
Information vector data input two-way LSTM model and are trained, and determine the parameter of two-way LSTM model.
Further, the state of an illness information vector data by the patient in sample set input two-way LSTM model and instruct
Practice, the method for determining the parameter of two-way LSTM model specifically:
Two-way LSTM model is constructed, patient is generated to the state of an illness information vector data before current doctor's adviceIt is defeated
Enter two-way LSTM model, two-way LSTM model output generates the output valve of the state of an illness information vector data after doctor's advice
Current patents in sample set data are generated to the true doctor's advice information in the data after current doctor's adviceWith it is double
Doctor's advice information generation value into the output valve of LSTM model output state of an illness information vector dataUtilize the side of linear regression
Method is separately converted to single vector-quantitiesWith
With true doctor's advice informationWith doctor's advice informationLoss function of the square distance as deep learning model M,
The parameter of the minimum corresponding two-way LSTM model of loss function is the parameter of the two-way LSTM model finally determined, final to determine ginseng
The corresponding two-way LSTM model of number is trained deep learning model M.
Further, the recommendation doctor's advice of current patents is obtained using trained deep learning model M in the step 4,
Further include the steps that carrying out data processing to the recommendation doctor's advice that deep learning model M exports, method particularly includes: extract deep learning
With doctor's advice relevant information data, including digital information and text information in the recommendation doctor's advice of model M output, text information is carried out
Binary conversion treatment, setting boundary numerical value will be greater than decomposing numerical value and is set as 1, remaining is set as 0, will treated text information
Recommendation doctor's advice with digital information as current patents.
Intelligent doctor's advice recommender system based on deep learning, comprising:
Database: for storing conditions of patients information bank.
Doctor's advice signal judgement module: it is configured as setting the patient of doctor's advice to be opened as current patents, acquires current patents'
State of an illness information, judges whether current patents have doctor's advice information, if so, going to the doctor's advice based on deep learning model M recommends mould
Block；Otherwise the doctor's advice recommending module based on database information is gone to；
Doctor's advice recommending module based on database information: it is configured to determine that in conditions of patients information bank and current patents' disease
The doctor's advice of most close patient, is recommended as the initial doctor's advice of current patents by the most similar patient of feelings information；
Doctor's advice recommending module based on deep learning model M: it is configured as the state of an illness information input training of current patents
Good deep learning model M, the output of deep learning model M are the recommendation doctor's advice of current patents.
A kind of electronic equipment, the meter run on a memory and on a processor including memory and processor and storage
The instruction of calculation machine when the computer instruction is run by processor, completes step described in the above method.
A kind of computer readable storage medium, for storing computer instruction, the computer instruction is executed by processor
When, complete step described in the above method.
Compared with prior art, the disclosure has the beneficial effect that
(1) there is initial doctor's advice the present disclosure contemplates current patents and without initial doctor's advice, the patient without initial doctor's advice is led to
The doctor's advice for searching the most similar patient of state of an illness information in state of an illness information bank D is crossed as doctor's advice is recommended, is considered in method implementation procedure
Many factors, improve the comprehensive of disclosure recommender system, not needing doctor, voluntarily medicine path is established in typing, ensure that
The feasibility of the present embodiment method.
(2) disclosure uses the deep learning model based on two-way LSTM model, utilizes LSTM model analysis time series
Ability, the change of illness state of intellectual analysis patient, improve recommend doctor's advice accuracy
(3) disclosure analyses in depth state of an illness information bank D using big data technology, and different thinkings is provided for doctor, helps
Doctor selects more advanced therapeutic scheme.
Detailed description of the invention
The Figure of description for constituting a part of this disclosure is used to provide further understanding of the disclosure, and the disclosure is shown
Meaning property embodiment and its explanation do not constitute the restriction to the disclosure for explaining the disclosure.
Fig. 1 is the flow chart according to the method for the embodiment of the present disclosure 1；
Fig. 2 is the method flow diagram of the parameter of the two-way LSTM model of determination of the embodiment of the present disclosure 1；
Fig. 3 is the system function module figure of the embodiment of the present disclosure 2.
Specific embodiment:
The disclosure is described further with embodiment with reference to the accompanying drawing.
It is noted that described further below be all exemplary, it is intended to provide further instruction to the disclosure.Unless another
It indicates, all technical and scientific terms used herein has usual with disclosure person of an ordinary skill in the technical field
The identical meanings of understanding.
It should be noted that term used herein above is merely to describe specific embodiment, and be not intended to restricted root
According to the illustrative embodiments of the disclosure.As used herein, unless the context clearly indicates otherwise, otherwise singular
Also it is intended to include plural form, additionally, it should be understood that, when in the present specification using term "comprising" and/or " packet
Include " when, indicate existing characteristics, step, operation, device, component and/or their combination.It should be noted that not conflicting
In the case where, each embodiment in the disclosure and the feature in embodiment can be combined with each other.Below in conjunction with attached drawing to reality
Example is applied to be described in detail.
Embodiment 1
In the technical solution disclosed in one or more embodiments, as shown in Figure 1, the intelligence based on deep learning is cured
Recommended method is advised, is included the following steps:
Step (1): building conditions of patients information bank D.
Step (2): the patient of doctor's advice to be opened is set as current patents p, acquires the state of an illness information of current patents p, judgement is worked as
Whether preceding patient has doctor's advice information, if so, executing step (4)；It is no to then follow the steps (3)；
Step (3): determining with the most similar patient of current patents' state of an illness information in conditions of patients information bank D, will be most close
The doctor's advice of patient is recommended as the initial doctor's advice of current patents；
Step (4): by the trained deep learning model M of the state of an illness information input of current patents p, deep learning model M
Output is the recommendation doctor's advice of current patents.
Conditions of patients information bank D is constructed in the step (1), specific steps can be such that
Step (101): the disease name that state of an illness information is included in conditions of patients information bank D is determined, for every kind of disease
Acquire the state of an illness information at least N number of patient of the medical diagnosis on disease；Preferably, selected patient is the state of an illness after treatment
The patient to take an evident turn for the better.
Establish conditions of patients information bank D to include the disease name of state of an illness information can be from international disease criterion coded system
It is chosen in ICD, or chooses all medical diagnosis on disease titles in international disease criterion coded system ICD；The state of an illness information can be with
It is patient from the state of an illness information for being diagnosed as the later institute having time t of the disease.
Step (102): vector data is converted by the state of an illness information of each patient of acquisition；
Step (103): the vector data of each patient is formed into a data set and is stored respectively in database, forms patient
State of an illness information bank D.
Certain patient p is collected in the state of an illness information of time t, state of an illness information is that structural data mainly includes two classes, Yi Leishi
Text information, title, inspection diagnosis of gender, medical diagnosis on disease title, allergies, prescription drug or operation including patient etc.；
Another kind of is numerical information, age, inspection numerical result, prescription drug dosage including patient etc..
The specific implementation step of step (102) can be such that
(i) each single item of the text information in state of an illness information time category is assigned a value of n-dimensional vector respectively, by text information
It is converted into digital information；Wherein the n of n-dimensional vector refers to the classifiable classification number of each single item time of text information.
(ii) numerical information in the digital information of conversion and state of an illness information is combined, constitutes vector data.
, can be according to the corresponding 0-1 vector of classification for text information, converting digital information for text information (can claim
For categorization vector), it can text information is indicated with multiple numerical value, is exemplified below:
The gender of patient has two classes altogether: male and female, and male's categorization vector is (1,0), and women categorization vector is (0,1), that
The gender information of patient is just converted into vector " (1,0) or (0,1) " from text information " male or female ".
Assuming that 30000 kinds of medical diagnosis on disease titles are shared in icd system, then class corresponding to the medical diagnosis on disease title of patient
Other vector 30000 is tieed up totally, and dimension numerical value corresponding to the medical diagnosis on disease that patient has is 1, remaining dimension numerical value is 0.
Assuming that all doctor's advices share 10000 kinds of drugs or operation, then corresponding to the prescription drug or action name of patient
Categorization vector have 10000 dimensions, dimension numerical value corresponding to the title of prescription drug or operation that patient has be 1, codimension
Degree value is 0.
Remaining text information is converted according to respective classification quantity, is converted into the vector data of respective dimensions.
Optionally, step 102 is specially to each patient p, and the state of an illness information by patient in time t is converted into vectorSpecific transformation rule is as follows:
(1) state of an illness information is divided into related to doctor's advice information and unrelated with doctor's advice information；
(2) corresponding numerical information related to doctor's advice information and unrelated with doctor's advice information is respectively as vectorOne
Point.
Assuming that the state of an illness information category obtained in step (102) is x total, wherein total x1 unrelated with doctor's advice information,
Total x relevant to doctor's advice information2；The result of each single item can be with a numerical value (if this state of an illness information is numerical information, directly
Connect using the single number in numerical information) or multiple numerical value (if this state of an illness information is text information, in use classes vector
Multiple numerical value) indicate；Assuming that numerical value corresponding to whole x item state of an illness information is stacked total m numerical value, wherein
The m unrelated with doctor's advice1, m relevant to doctor's advice2, i.e. m=m1+m2。
VectorTotal m dimension, one per one-dimensional (or multidimensional) in corresponding state of an illness information, preceding m1Item correspondence is unrelated with doctor's advice
State of an illness information project (total x1), rear m2Corresponding state of an illness information project (the total x relevant to doctor's advice of item2)；It is per one-dimensional numerical value
In the numerical value (if corresponding state of an illness information is digital information) or multiple numerical value of certain categorization vector of corresponding state of an illness information project
One (if corresponding state of an illness information be text information).We willIt is divided into two parts, preceding m1Item is denoted asCorresponding and doctor
Advise unrelated state of an illness information；M afterwards2Item is denoted asCorresponding state of an illness information relevant to doctor's advice, is denoted as
The state of an illness information of acquisition current patents p in step (2), judges whether current patents have doctor's advice information, and definition corresponds to
The state of an illness message data set of each patientIn, when t=1,Data information before first doctor's advice；When t=2,It is
Data information before second doctor's advice；When t=n,Be n-th of doctor's advice before data information；If patient p does not have
There is initial doctor's advice information (there is no doctor's advice in moment t=1), at this time vectorTwo parts in, first partIt is one
A m1Dimensional vector, corresponding x1The item state of an illness information unrelated with doctor's advice information,Middle correspondence i-th (i=1,2 ..., x1) state of an illness
The part of information is denoted asVectorSecond partIt is a m20 vector of dimension.
Method with the most similar patient of current patents' state of an illness in conditions of patients information bank D is determined in step 3, it can be with are as follows:
Step 31: converting vector data for the text information each single item in the state of an illness information of acquisition；
Step 32: the vector data that current patents p is converted generates current suffer from conjunction with the numeric data in state of an illness information
The state of an illness message data set of person p；
Step 33: calculating the state of an illness message data set of current patents p and the disease of each patient in conditions of patients information bank D
The similarity of feelings message data set, the maximum patient of similarity value are the most close patient of current patents' state of an illness.
As a specific implementation manner, the calculating of the similarity in step 33 can be specially to calculate current patents' disease
Data set before the initial doctor's advice of state of an illness message data set of feelings message data set and each patient in conditions of patients information bank D
Similarity.T=1 at this time, similarity are calculated by following formula:
Wherein, p indicates that current patents, q indicate the patient in conditions of patients information bank D,Believe for the state of an illness of current patents
Data unrelated with doctor's advice information in data set are ceased,Middle correspondence i-th (i=1,2 ..., x1) state of an illness information part note
Forx1For total item,It is concentrated and doctor's advice letter for the state of an illness information data of each patient in conditions of patients information bank D
Unrelated data are ceased,Middle correspondence i-th (i=1,2 ..., x1) part of state of an illness information is denoted aswiTo believe with doctor's advice
I-th in unrelated state of an illness information weight is ceased, is metThe numerical value of weight can be with is defined as:
The doctor's advice of most close patient is recommended as the initial doctor's advice of current patents in step 3, when the doctor's advice of most close patient is
When multiple, the most close initial doctor's advice of patient is recommended as to the initial doctor's advice of current patents.Found in conditions of patients information bank D
With the highest patient q of patient's p similarity, using doctor's advice of the patient q in moment t=1 as the initial doctor's advice for recommending patient p,
I.e. using the initial doctor's advice of the highest patient q of similarity as the initial doctor's advice of current patents p.
If current patents do not have initial doctor's advice, the information vector data relevant to doctor's advice of current patents are 0, then pass through
Output is still 0 under the action of being in relevant parameter after the transformation of deep learning model, i.e., can not pass through depth without initial doctor's advice
Learning model M exports next doctor's advice.In view of this problem, setting steps 3, by searching for state of an illness information bank D state of an illness information most phase
The doctor's advice of close patient considers many factors in method implementation procedure, ensure that the present embodiment method as doctor's advice is recommended
Feasibility.
Deep learning model M can choose two-way LSTM model in step (4), construct the depth based on two-way LSTM model
Learning model M, the English of shot and long term memory are Long Short Term Memory, and referred to as LSTM, two-way LSTM are two-way
The method and step of long short-term memory Recognition with Recurrent Neural Network, training deep learning model M can be with are as follows:
41) state of an illness information vector of n patient corresponding to each medical diagnosis on disease title in conditions of patients information bank D is acquired
For data as sample set, the state of an illness information vector data include information vector data related with doctor's advice and unrelated with doctor's advice
Information vector data；
42) the deep learning model M based on two-way LSTM model is established, the state of an illness letter before doctor's advice is generated with each patient
Ceasing vector data is input, to generate the state of an illness information vector data after doctor's advice for output, by the disease of the patient in sample set
Feelings information vector data input two-way LSTM model and are trained, and determine the parameter of two-way LSTM model.
In the step 42), establishing the deep learning model M based on two-way LSTM model includes two-way LSTM model and line
Property regression model, as shown in Fig. 2, the method for the parameter of the two-way LSTM model of determination in the step 42 specifically:
Two-way LSTM model 42-1) is constructed, patient is generated to the state of an illness information vector data before current doctor's adviceTwo-way LSTM model is inputted, two-way LSTM model output generates the defeated of the state of an illness information vector data after doctor's advice
It is worth out
Current patents in sample set data 42-2) are generated to the true doctor's advice information in the data after current doctor's advice
With the doctor's advice information generation value in the output valve of two-way LSTM model output state of an illness information vector dataUtilize linear regression
Method be separately converted to single vector-quantitiesWithLinear regression model (LRM) can be specially
Wherein V, W are the parameter vector and parameter matrix in linear regression relation model respectively.
42-3) with true doctor's advice informationWith doctor's advice informationLoss of the square distance as deep learning model M
Function, i.e. loss function areThe training objective of the minimum deep learning model M of loss function loses letter
The parameter of the minimum corresponding two-way LSTM model of number is the parameter of the two-way LSTM model finally determined, final to determine that parameter is corresponding
Two-way LSTM model be trained deep learning model M.
The step 42-1) in the two-way LSTM model of building be training acquisition multiple two-way LSTM models primary mould
Type constructs each primary mold method particularly includes:
421.1: collecting sample concentrates the state of an illness information vector data of setting quantity patient；
421.2: with each patient generate doctor's advice before state of an illness information vector data be input, after generating doctor's advice
State of an illness information vector data be output, the state of an illness information vector data of the setting quantity patient of acquisition are defeated as input and output
Enter two-way LSTM model, obtain the parameter of two-way LSTM model, to obtain two-way LSTM model primary mold；
Below specifically to illustrate entire training process:
Sample set is divided into training set, verifying collection and inspection set.By each patient p in training set all moment t's
State of an illness information vectorBy the whole process in step (42), loss function will be minimizedIt is trained as target.
Whole state of an illness information of a patient p in training set are once trained, are referred to as had trained on training set primary.
I, every training 1000 times on training set save the initial model once obtained；Multiple 1000 acquisitions of training are more
A initial model.
Ii, after saving an initial model, one-time authentication is carried out on verifying collection using the model: will be in step (42-1)
The doctor's advice information being inferred to is compared with true doctor's advice information, is verified the loss for concentrating all patient's whole state of an illness information
The summation of functionAnd it saves；
Iii, after verifying 100 times, the corresponding parameter of the verifying the smallest model of loss function summation is selected, as model M
Final argument；
It can also include the steps that testing trained deep learning model M:
Iv, it is once tested on test set: the doctor's advice information being inferred in step (42-1) and true doctor's advice is believed
Breath compares, and obtains the summation of the loss function of all patient's whole state of an illness information in test setAnd save, by the number
It is worth as the parameter for indicating that model M infers accuracy rate.
By the trained deep learning model M of the state of an illness information input of current patents p in step 4, deep learning model M is defeated
It is out the recommendation doctor's advice of current patents.
The recommendation doctor's advice of current patents is wherein obtained using trained deep learning model M, further includes to deep learning
The step of recommendation doctor's advice of model M output carries out data processing, method particularly includes: it extracts the state of an illness information of current patents p is defeated
Enter believing in the state of an illness information for the current patents that trained deep learning model M obtains with doctor's advice relevant information, including number
Text information is carried out binary conversion treatment by breath and text information, and setting boundary numerical value will be greater than decomposition numerical value and be set as 1, remaining
It is set as 0, using treated text information and digital information as the recommendation doctor's advice of current patents.
To inpatient p, from moment t=0 at the time of newest doctor's adviceBy its corresponding state of an illness information vectorInput model M；
For the momentInputTrained deep learning model M obtains vectorVectorCorresponding doctor's advice
Associated conditions information, including digital information and text information；For digital information, directly as model M at the momentPush away
Recommend doctor's advice information；For text information, boundary numerical value can be set as 0.5, by them in vectorIn corresponding number press
According to 0.5 for line of demarcation be approximately 0 or 1, even number less than 0.5, convert 0 for number, if number be more than or equal to 0.5, will count
Word is converted into 1；Then using the numerical value after conversion as model M at the momentRecommendation doctor's advice information.
Embodiment 2
The present embodiment additionally provides the intelligent doctor's advice recommender system based on deep learning, as shown in Figure 3, comprising:
Database: for storing conditions of patients information bank.
Doctor's advice signal judgement module: it is configured as setting the patient of doctor's advice to be opened as current patents, acquires current patents'
State of an illness information, judges whether current patents have doctor's advice information, if so, going to the doctor's advice based on deep learning model M recommends mould
Block；Otherwise the doctor's advice recommending module based on database information is gone to；
Doctor's advice recommending module based on database information: it is configured to determine that in conditions of patients information bank and current patents' disease
The doctor's advice of most close patient, is recommended as the initial doctor's advice of current patents by the most similar patient of feelings information；
Doctor's advice recommending module based on deep learning model M: it is configured as the state of an illness information input training of current patents
Good deep learning model M, the output of deep learning model M are the recommendation doctor's advice of current patents.
Embodiment 3
Present embodiments provide a kind of electronic equipment, including memory and processor and storage on a memory and are being located
The computer instruction run on reason device when the computer instruction is run by processor, completes step described in 1 method of embodiment
Suddenly.
Embodiment 4
A kind of computer readable storage medium is present embodiments provided, for storing computer instruction, the computer refers to
When order is executed by processor, step described in 1 method of embodiment is completed.
The foregoing is merely preferred embodiment of the present disclosure, are not limited to the disclosure, for the skill of this field
For art personnel, the disclosure can have various modifications and variations.It is all within the spirit and principle of the disclosure, it is made any to repair
Change, equivalent replacement, improvement etc., should be included within the protection scope of the disclosure.
Although above-mentioned be described in conjunction with specific embodiment of the attached drawing to the disclosure, model not is protected to the disclosure
The limitation enclosed, those skilled in the art should understand that, on the basis of the technical solution of the disclosure, those skilled in the art are not
Need to make the creative labor the various modifications or changes that can be made still within the protection scope of the disclosure.
Claims (10)
1. the intelligent doctor's advice recommended method based on deep learning, characterized in that include the following steps:
Step (1): building conditions of patients information bank.
Step (2): the patient of doctor's advice to be opened is set as current patents, the state of an illness information of current patents is acquired, judges current patents
Whether doctor's advice information is had, if so, executing step (4)；It is no to then follow the steps (3)；
Step (3): determine in conditions of patients information bank with the most similar patient of current patents' state of an illness information, by most close patient's
Doctor's advice is recommended as the initial doctor's advice of current patents；
Step (4): by the trained deep learning model M of the state of an illness information input of current patents, the output of deep learning model M is
The recommendation doctor's advice of current patents.
2. the intelligent doctor's advice recommended method based on deep learning as described in claim 1, it is characterized in that: in the step (1)
Conditions of patients information bank is constructed, specific steps can be such that
It determines the disease name that include state of an illness information in conditions of patients information bank, for every kind of disease acquisition there is the disease to examine
The state of an illness information of disconnected at least N number of patient；
Vector data is converted by the state of an illness information of each patient of acquisition；
The vector data of each patient is formed into a data set and is stored respectively in database, forms conditions of patients information bank.
3. the intelligent doctor's advice recommended method based on deep learning as claimed in claim 2, it is characterized in that: the state of an illness of the patient
Information includes text data and numeric data, the method for converting vector data for the state of an illness information of each patient of acquisition, tool
Body are as follows:
The each single item of text information in state of an illness information time category is assigned a value of n-dimensional vector respectively, converts text information to
Digital information；Wherein the n of n-dimensional vector refers to the classifiable classification number of each single item time of text information；
Numerical information in the digital information of conversion and state of an illness information is combined, vector data is constituted, the vector data includes
With doctor's advice information related data and with doctor's advice information extraneous data.
4. the intelligent doctor's advice recommended method based on deep learning as described in claim 1, it is characterized in that: in the step 3 really
Determine the method in conditions of patients information bank with the most similar patient of current patents' state of an illness, specifically:
Step 31: converting vector data for the text information each single item in the state of an illness information of the patient p of acquisition；
Step 32: by the vector data of conversion in conjunction with the numeric data in state of an illness information, generating the state of an illness information of current patents
Vector data collection；
Step 33: calculating the state of an illness information vector data set of current patents and the state of an illness of each patient in conditions of patients information bank
The similarity of message data set, the maximum patient of similarity value are the most close patient of current patents' state of an illness.
5. the intelligent doctor's advice recommended method based on deep learning as described in claim 1, it is characterized in that: training deep learning mould
The method and step of type M includes:
It acquires the state of an illness information of n patient corresponding to each medical diagnosis on disease title and is converted into vector data as sample set；
The deep learning model M based on two-way LSTM model is established, the state of an illness information vector before doctor's advice is generated with each patient
Data are input, to generate the state of an illness information vector data after doctor's advice for output, by the state of an illness information of the patient in sample set
Vector data inputs two-way LSTM model and is trained, and determines the parameter of two-way LSTM model.
6. the intelligent doctor's advice recommended method based on deep learning as claimed in claim 5, it is characterized in that: described will be in sample set
The state of an illness information vector data of patient input two-way LSTM model and be trained, the method for determining the parameter of two-way LSTM model
Specifically:
Two-way LSTM model is constructed, patient is generated to the state of an illness information vector data before current doctor's adviceIt inputs two-way
LSTM model, two-way LSTM model output generate the output valve of the state of an illness information vector data after doctor's advice
Current patents in sample set data are generated to the true doctor's advice information in the data after current doctor's adviceWith two-way LSTM
Model exports the doctor's advice information generation value in the output valve of state of an illness information vector dataDistinguished using the method for linear regression
It is converted into single vector-quantitiesWith
With true doctor's advice informationWith doctor's advice informationLoss function of the square distance as deep learning model M, lose letter
The parameter of the minimum corresponding two-way LSTM model of number is the parameter of the two-way LSTM model finally determined, final to determine that parameter is corresponding
Two-way LSTM model be trained deep learning model M.
7. the intelligent doctor's advice recommended method based on deep learning as described in claim 1, it is characterized in that: being adopted in the step 4
The recommendation doctor's advice of current patents is obtained with trained deep learning model M, further includes the recommendation to the output of deep learning model M
Doctor's advice carries out the step of data processing, method particularly includes: it extracts related to doctor's advice in the recommendation doctor's advice of deep learning model M output
Text information is carried out binary conversion treatment by information data, including digital information and text information, and setting boundary numerical value will be greater than
It decomposes numerical value and is set as 1, remaining is set as 0, using treated text information and digital information as the recommendation of current patents doctor
It advises.
8. the intelligent doctor's advice recommender system based on deep learning, characterized in that include:
Database: for storing conditions of patients information bank.
Doctor's advice signal judgement module: it is configured as setting the patient of doctor's advice to be opened as current patents, acquires the state of an illness of current patents
Information, judges whether current patents have doctor's advice information, if so, going to the doctor's advice recommending module based on deep learning model M；It is no
Then go to the doctor's advice recommending module based on database information；
Doctor's advice recommending module based on database information: it is configured to determine that in conditions of patients information bank and believes with current patents' state of an illness
Most similar patient is ceased, the doctor's advice of most close patient is recommended as to the initial doctor's advice of current patents；
Doctor's advice recommending module based on deep learning model M: it is configured as the state of an illness information input of current patents is trained
Deep learning model M, the output of deep learning model M are the recommendation doctor's advice of current patents.
9. a kind of electronic equipment, characterized in that on a memory and on a processor including memory and processor and storage
The computer instruction of operation when the computer instruction is run by processor, is completed described in any one of claim 1-7 method
Step.
10. a kind of computer readable storage medium, characterized in that for storing computer instruction, the computer instruction is located
When managing device execution, step described in any one of claim 1-7 method is completed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910778555.1A CN110473636B (en) | 2019-08-22 | 2019-08-22 | Intelligent medical advice recommendation method and system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910778555.1A CN110473636B (en) | 2019-08-22 | 2019-08-22 | Intelligent medical advice recommendation method and system based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110473636A true CN110473636A (en) | 2019-11-19 |
| CN110473636B CN110473636B (en) | 2020-09-22 |
Family
ID=68512843
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910778555.1A Active CN110473636B (en) | 2019-08-22 | 2019-08-22 | Intelligent medical advice recommendation method and system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110473636B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111046672A (en) * | 2019-12-11 | 2020-04-21 | 山东众阳健康科技集团有限公司 | Multi-scene text abstract generation method |
| CN114171173A (en) * | 2021-12-08 | 2022-03-11 | 山东众阳健康科技集团有限公司 | Blood inventory early warning method and system based on deep learning |
| CN114388093A (en) * | 2021-12-28 | 2022-04-22 | 山东众阳健康科技集团有限公司 | Outpatient medical record intelligent writing method and system based on deep learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107491638A (en) * | 2017-07-28 | 2017-12-19 | 深圳和而泰智能控制股份有限公司 | A kind of ICU user's prognosis method and terminal device based on deep learning model |
| CN108417272A (en) * | 2018-02-08 | 2018-08-17 | 合肥工业大学 | Similar case with temporal constraint recommends method and device |
| CN108806785A (en) * | 2018-05-29 | 2018-11-13 | 四川长虹电器股份有限公司 | A kind of diagnosis and treatment section office recommendation method and system based on convolutional neural networks |
| CN110111888A (en) * | 2019-05-16 | 2019-08-09 | 闻康集团股份有限公司 | A kind of XGBoost disease probability forecasting method, system and storage medium |
-
2019
- 2019-08-22 CN CN201910778555.1A patent/CN110473636B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107491638A (en) * | 2017-07-28 | 2017-12-19 | 深圳和而泰智能控制股份有限公司 | A kind of ICU user's prognosis method and terminal device based on deep learning model |
| CN108417272A (en) * | 2018-02-08 | 2018-08-17 | 合肥工业大学 | Similar case with temporal constraint recommends method and device |
| CN108806785A (en) * | 2018-05-29 | 2018-11-13 | 四川长虹电器股份有限公司 | A kind of diagnosis and treatment section office recommendation method and system based on convolutional neural networks |
| CN110111888A (en) * | 2019-05-16 | 2019-08-09 | 闻康集团股份有限公司 | A kind of XGBoost disease probability forecasting method, system and storage medium |
Non-Patent Citations (3)
| Title |
|---|
| DONG XISHUANG 等: "Transfer Bi-directional LSTM RNN for Named Entity Recognition in Chinese Electronic Medical Records", 《2017 IEEE 19TH INTERNATIONAL CONFERENCE ON E-HEALTH NETWORKING, APPLICATIONS AND SERVICES (HEALTHCOM)》 * |
| MA FENGLONG 等: "Dipole: Diagnosis Prediction in Healthcare via Aention-based Bidirectional Recurrent Neural Networks", 《KDD 2017 APPLIED DATA SCIENCE PAPER》 * |
| 刘文洋: "面向临床决策支持的贝叶斯网络医嘱推荐方法研究", 《中国优秀硕士学位论文全文数据库 医药卫生科技辑》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111046672A (en) * | 2019-12-11 | 2020-04-21 | 山东众阳健康科技集团有限公司 | Multi-scene text abstract generation method |
| CN111046672B (en) * | 2019-12-11 | 2020-07-14 | 山东众阳健康科技集团有限公司 | Multi-scene text abstract generation method |
| CN114171173A (en) * | 2021-12-08 | 2022-03-11 | 山东众阳健康科技集团有限公司 | Blood inventory early warning method and system based on deep learning |
| CN114388093A (en) * | 2021-12-28 | 2022-04-22 | 山东众阳健康科技集团有限公司 | Outpatient medical record intelligent writing method and system based on deep learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110473636B (en) | 2020-09-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109669994B (en) | Construction method and system of health knowledge map | |
| Liu et al. | Medical-vlbert: Medical visual language bert for covid-19 ct report generation with alternate learning | |
| Pan et al. | Characterization multimodal connectivity of brain network by hypergraph GAN for Alzheimer’s disease analysis | |
| Hu et al. | Automatic construction of Chinese herbal prescriptions from tongue images using CNNs and auxiliary latent therapy topics | |
| CN109830303A (en) | Clinical data mining analysis and aid decision-making method based on internet integration medical platform | |
| CN107247881A (en) | A kind of multi-modal intelligent analysis method and system | |
| Chowriappa et al. | Introduction to machine learning in healthcare informatics | |
| CN109949929A (en) | A kind of assistant diagnosis system based on the extensive case history of deep learning | |
| CN109785927A (en) | Clinical document structuring processing method based on internet integration medical platform | |
| Wang et al. | Attention-based multi-instance neural network for medical diagnosis from incomplete and low quality data | |
| CN110473636A (en) | Intelligent doctor's advice recommended method and system based on deep learning | |
| WO2020224433A1 (en) | Target object attribute prediction method based on machine learning and related device | |
| CN109994216A (en) | A kind of ICD intelligent diagnostics coding method based on machine learning | |
| Li et al. | Personalized intelligent syndrome differentiation guided by TCM consultation philosophy | |
| Pendyala et al. | Automated medical diagnosis from clinical data | |
| Li et al. | Meddm: Llm-executable clinical guidance tree for clinical decision-making | |
| CN110164519B (en) | Classification method for processing electronic medical record mixed data based on crowd-sourcing network | |
| CN109686453A (en) | Medical aid decision-making system based on big data | |
| Wang et al. | [Retracted] Student Physical Fitness Test System and Test Data Analysis System Based on Computer Vision | |
| Ning et al. | Research on a vehicle-mounted intelligent TCM syndrome differentiation system based on deep belief network | |
| Wang et al. | Research on named entity recognition of doctor-patient question answering community based on bilstm-crf model | |
| Zuo et al. | Fusing Structural and Functional Connectivities Using Disentangled VAE for Detecting MCI | |
| Wang et al. | Research on assistant diagnostic method of TCM based on BERT and BiGRU recurrent neural network | |
| Shi et al. | DMMAM: Deep multi-source multi-task attention model for intensive care unit diagnosis | |
| Chen et al. | Automatic surgery duration prediction using artificial neural networks |
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 | ||
| CP01 | Change in the name or title of a patent holder |
Address after: 12 / F, building 1, Aosheng building, 1166 Xinluo street, hi tech Zone, Jinan City, Shandong Province Patentee after: Zhongyang Health Technology Group Co.,Ltd. Address before: 12 / F, building 1, Aosheng building, 1166 Xinluo street, hi tech Zone, Jinan City, Shandong Province Patentee before: SHANDONG MSUNHEALTH TECHNOLOGY GROUP Co.,Ltd. |
|
| CP01 | Change in the name or title of a patent holder |