CN110297908A - Diagnosis and treatment program prediction method and device - Google Patents
Diagnosis and treatment program prediction method and device Download PDFInfo
- Publication number
- CN110297908A CN110297908A CN201910584729.0A CN201910584729A CN110297908A CN 110297908 A CN110297908 A CN 110297908A CN 201910584729 A CN201910584729 A CN 201910584729A CN 110297908 A CN110297908 A CN 110297908A
- Authority
- CN
- China
- Prior art keywords
- text
- case history
- predicted
- vector
- diagnosis
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/284—Lexical analysis, e.g. tokenisation or collocates
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
- G16H10/60—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for patient-specific data, e.g. for electronic patient records
-
- G—PHYSICS
- 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
Abstract
The present invention provides diagnosis and treatment program prediction method and device, the feature vector for obtaining the medicine entity that case history text to be predicted is included obtains medicine entity vector, meanwhile, obtain the Text eigenvector of the case history text to be predicted.Then, the corresponding medicine entity vector sum Text eigenvector of case history text to be predicted is input in the diagnosis and treatment program prediction model that training obtains in advance, prediction obtains the diagnosis and treatment scheme to match with the case history text to be predicted, wherein, diagnosis and treatment scheme includes at least one of diagnosis and treatment drug, operation and check item.As shown in the above, the present invention is when carrying out feature extraction to case history text to be predicted, both it had been extracted the feature vector of medicine entity and its incidence relation, it is extracted the language ambience information of case history text simultaneously, reduce information loss, the accuracy of the feature vector of case history text to be predicted is improved, and then improves the predictablity rate of diagnosis and treatment scheme.
Description
Technical field
The invention belongs to medical diagnosis technical field more particularly to diagnosis and treatment program prediction method and devices.
Background technique
Currently, clinically, the diagnosis and treatment scheme that doctor provides places one's entire reliance upon the experience of doctor itself, in order to accelerate doctor
Clinical decision speed, propose diagnosis and treatment program prediction system, so that doctor be assisted quickly to provide effective diagnosis and treatment scheme, reduce
Patient's waiting time.
Current diagnosis and treatment program prediction system, which is mainly sorted out according to the clinical experience of medical expert with clinical guidelines, faces
Bed rule, then makes inferences by ontology inference machine, and then predicts corresponding diagnosis and treatment scheme.But this mode needs people
To extract clinical rules from clinical expertise and clinical guidelines, this process is time-consuming and laborious, moreover, Rulemaking is complete
Face property and logicality directly influence the accuracy of prediction result, while this mode the problem of there is also low efficiencys.
Summary of the invention
In view of this, the purpose of the present invention is to provide a kind of diagnosis and treatment program prediction method and device, to solve at present
The technical issues of diagnosis and treatment program prediction method accuracy rate is low and low efficiency, specific technical solution is as follows:
On the one hand, the present invention provides a kind of diagnosis and treatment program prediction methods, comprising:
Obtain case history text to be predicted；
The corresponding medicine entity vector of medicine entity that the case history text to be predicted includes is obtained, and is obtained described to pre-
Survey the corresponding Text eigenvector of case history text；
It is real using the corresponding medicine of case history text to be predicted described in the preparatory diagnosis and treatment program prediction model analysis trained and obtained
Body vector sum text vector obtains the diagnosis and treatment scheme to match with the case history text to be predicted, and the diagnosis and treatment scheme includes medicine
At least one of object, operation and check item.
Optionally, the diagnosis and treatment program prediction model includes the corresponding prediction model of pre-set categories；
It is described to utilize the corresponding medicine entity vector sum text of case history text to be predicted described in diagnosis and treatment program prediction model analysis
This vector obtains the diagnosis and treatment scheme to match with the case history text to be predicted, comprising:
By the corresponding Text eigenvector of the case history text to be predicted be input to that pre-selection training obtains based on depth
The textual classification model of habit obtains the corresponding text categories of the case history text to be predicted；
By the corresponding medicine entity vector sum text vector of the case history text to be predicted, it is input to and the disease to be predicted
It goes through in the corresponding prediction model of text categories of text, prediction obtains the diagnosis and treatment side to match with the case history text to be predicted
Case.
Optionally, the process of textual classification model of the training based on deep learning, comprising:
Obtain the case history training sample for having marked text categories；
The case history training sample is converted into text vector using term vector；
The text vector of the case history training sample is input to the multi-level attention model constructed in advance, extraction obtains
Text eigenvector, and the case history training sample is predicted based on the Text eigenvector to obtain prediction text class
Not；
Based on the corresponding prediction text categories of case history training sample described in each and text categories are marked, described in adjustment
Model parameter in multi-level attention model, until training sample to the case history using multi-level attention model adjusted
This prediction text type predicted meets the default condition of convergence.
Optionally, the corresponding Text eigenvector of the case history text to be predicted is obtained, comprising:
The case history text to be predicted is segmented using segmentation methods, obtains the text of the case history text to be predicted
Word segmentation result；
Each word in the text word segmentation result is mapped as corresponding vector using term vector, is obtained described to be predicted
The corresponding text vector of case history text；
It is extracted from the corresponding text vector of the case history text to be predicted using multi-level attention model and obtains text
Feature vector；
Wherein, the Text eigenvector includes semantic feature and text entirety semantic feature between semantic feature, sentence between word.
Optionally, the corresponding medicine entity vector of medicine entity for obtaining the case history text to be predicted and including, packet
It includes:
Obtain the medicine entity that the case history text to be predicted is included；
Learning model is indicated according to the medical knowledge map, and the medicine entity in the case history text to be predicted is mapped
For corresponding medicine entity vector.
Optionally, the process of the training diagnosis and treatment program prediction model, comprising:
Obtain the case history training sample for having marked diagnosis and treatment scheme；
The medicine entity vector of the included medicine entity of the case history training sample is obtained, and obtains the case history training sample
This corresponding Text eigenvector；
The corresponding medicine entity vector of the case history training sample and Text eigenvector be input to and is constructed in advance
Seq2Seq model, prediction obtain the corresponding prediction diagnosis and treatment scheme of the case history training sample；
Based on the corresponding prediction diagnosis and treatment scheme of case history training sample described in each and diagnosis and treatment scheme is marked, described in adjustment
Model parameter in Seq2Seq model, until being predicted using Seq2Seq model adjusted the case history training sample
Obtained prediction diagnosis and treatment scheme meets the default condition of convergence.
On the other hand, the present invention also provides a kind of diagnosis and treatment program prediction devices, comprising:
First obtains module, for obtaining case history text to be predicted；
Second obtains module, for obtain the corresponding medicine entity of medicine entity that the case history text to be predicted includes to
Amount；
Third obtains module, for obtaining the corresponding Text eigenvector of the case history text to be predicted；
Prediction module, for utilizing case history text to be predicted described in the diagnosis and treatment program prediction model analysis that training obtains in advance
Corresponding medicine entity vector sum text vector obtains the diagnosis and treatment scheme to match with the case history text to be predicted, described to examine
Treatment scheme includes at least one of drug, operation and check item.
Optionally, the diagnosis and treatment program prediction model includes the corresponding prediction model of pre-set categories；The prediction module packet
It includes:
Classify submodule, it is trained for the corresponding Text eigenvector of the case history text to be predicted to be input to pre-selection
The textual classification model based on deep learning arrived obtains the corresponding text categories of the case history text to be predicted；
Predict submodule, for will the corresponding medicine entity vector sum text vector of the case history text to be predicted, input
To in prediction model corresponding with the text categories of the case history text to be predicted, prediction is obtained and the case history text to be predicted
The diagnosis and treatment scheme to match.
Optionally, the third acquisition module includes:
Submodule is segmented, for being segmented using segmentation methods to the case history text to be predicted, is obtained described to pre-
Survey the text word segmentation result of case history text；
Mapping submodule, for using term vector by each word in the text word segmentation result be mapped as it is corresponding to
Amount, obtains the corresponding text vector of the case history text to be predicted；
Feature extraction submodule, for using multi-level attention model from the corresponding text of the case history text to be predicted
It is extracted in vector and obtains Text eigenvector；
Wherein, the Text eigenvector includes semantic feature and text entirety semantic feature between semantic feature, sentence between word.
Optionally, the second acquisition module includes:
First acquisition submodule, the medicine entity for being included for obtaining the case history text to be predicted；
Mapping submodule will be in the case history text to be predicted for indicating learning model according to medical knowledge map
Medicine entity is mapped as corresponding medicine entity vector.
Diagnosis and treatment program prediction method provided by the invention obtains the feature for the medicine entity that case history text to be predicted is included
Vector obtains medicine entity vector, meanwhile, obtain the Text eigenvector of the case history text to be predicted.Then, by disease to be predicted
The corresponding medicine entity vector sum Text eigenvector of text is gone through to be input in the diagnosis and treatment program prediction model that training obtains in advance,
Prediction obtains the diagnosis and treatment scheme to match with the case history text to be predicted, wherein diagnosis and treatment scheme includes diagnosis and treatment drug, operation and inspection
Look at least one of item.As shown in the above, this method was both extracted when carrying out feature extraction to case history text to be predicted
The feature vector of medicine entity and its incidence relation, while it being extracted the language ambience information of case history text, reduce information loss,
The accuracy of the feature vector of case history text to be predicted is improved, and then improves the predictablity rate of diagnosis and treatment scheme.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is the present invention
Some embodiments for those of ordinary skill in the art without creative efforts, can also basis
These attached drawings obtain other attached drawings.
Fig. 1 is a kind of flow chart of diagnosis and treatment program prediction method provided in an embodiment of the present invention；
Fig. 2 is the structural schematic diagram of diagnosis and treatment program prediction system provided in an embodiment of the present invention；
Fig. 3 is the flow chart of the training process of text classification prediction model provided in an embodiment of the present invention；
Fig. 4 is the flow chart of the training process of diagnosis and treatment program prediction model provided in an embodiment of the present invention；
Fig. 5 is a kind of block diagram of diagnosis and treatment program prediction device provided in an embodiment of the present invention；
Fig. 6 is the provided in an embodiment of the present invention a kind of second block diagram for obtaining module；
Fig. 7 is the block diagram that a kind of third provided in an embodiment of the present invention obtains module；
Fig. 8 is a kind of block diagram of prediction module provided in an embodiment of the present invention.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
Every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
Referring to Figure 1, a kind of flow chart of diagnosis and treatment program prediction method provided in an embodiment of the present invention, this method are shown
Applied in terminal or server, as shown in Figure 1, this method may include steps of:
S110 obtains case history text to be predicted.
Case history text to be predicted refers to the electronic health record text of patient to be treated, and many medical institutions can generate at present
Electronic health record text.
S120 obtains the corresponding medicine entity vector of medicine entity that case history text to be predicted includes, and obtains to be predicted
The corresponding Text eigenvector of case history text.
The each medicine entity for being included is extracted from case history text to be predicted, and obtains the pass between each medicine entity
Connection relationship, and the medicine entity and its incidence relation are converted to corresponding medicine entity vector.
Wherein, medicine entity refers to the entity of medical domain objective reality, for example, disease, illness, drug, operation etc..Example
Such as, it is treatment relationship that the incidence relation between medicine entity, which may include: between drug and disease, be between illness and disease because
Fruit relationship is treatment relationship between operation and disease.
In one embodiment of the invention,
Medical knowledge map is a kind of entity, concept and incidence relation between them for describing medical domain objective reality
Semantic network, using semantic technology form expression system, institutional, integrated medical domain knowledge.
Medical knowledge map indicates learning model it is intended that each medicine entity and incidence relation in knowledge mapping learn
Corresponding vector is obtained, meanwhile, retain original structure in medical knowledge map or gives information.Utilize medical knowledge map table
Show that learning model promotes computational efficiency during knowledge fusion and knowledge reasoning.
Learning model is indicated using medical knowledge map, obtains the vector for the medicine entity that case history text to be predicted is included
It indicates to get medicine entity vector is arrived.It can be more using the medicine entity vector that medical knowledge map indicates that learning model obtains
Accurately express its relationship between the semantic information of medical domain and the medicine entity and other medicine entities.
In one embodiment of the invention, medical knowledge map indicates that learning model can use the translation based on distance
Model, specifically, the medical knowledge map form of triple (e.g., head entity, relationship, tail entity) is indicated, then
After the medical knowledge map that triple form indicates is input to the translation model based on distance, obtain each in medical knowledge map
The vector of a medicine entity and its incidence relation.
There are the context of co-text information of medicine entity in case history text to be predicted, therefore, it is also desirable to by case history to be predicted
Text conversion is corresponding Text eigenvector.
In one embodiment of the invention, case history text to be predicted is segmented first with segmentation methods, is somebody's turn to do
The text word segmentation result of case history text to be predicted；Then, each word in text word segmentation result is mapped as using term vector
Corresponding vector obtains the corresponding text vector of the case history text to be predicted；Then, the text table based on deep learning is recycled
Representation model extracts from text vector and obtains Text eigenvector.
Text feature vector includes semantic feature and text entirety semantic feature between semantic feature, sentence between word.
Wherein, term vector is used to indicate that the vector of word, and effect is that each word in text is mapped as one
Vector realizes the conversion of text to vector.
S130 utilizes the corresponding doctor of case history text to be predicted described in the preparatory diagnosis and treatment program prediction model analysis trained and obtained
Entity vector sum text vector is learned, the diagnosis and treatment scheme to match with case history text to be predicted is obtained.
Wherein, the diagnosis and treatment scheme includes at least one of drug, operation and check item.
The medicine entity vector sum Text eigenvector that S120 is obtained is input in diagnosis and treatment program prediction model, obtain with
The diagnosis and treatment scheme that the case history text to be predicted matches, for example, check item, therapeutic agent, operation etc..
In one embodiment of the application, which uses Seq2Seq (sequence-to-
Sequence) model.Seq2Seq model is the network of an Encoder-Decoder structure, its input is a sequence,
Output is also a sequence, and Encoder expresses the vector that the signal sequence of a variable-length becomes regular length,
The vector of this regular length is become the signal sequence of the target of variable-length by Decoder.In brief, Seq2Seq model
It is exactly a translation model, a language sequence is translated into another language sequence, entire treatment process is by using depth
A sequence is mapped as another output sequence by neural network (e.g., LSTM network) or RNN network.
As shown in Fig. 2, indicating that learning model obtains patient medical record (case history text i.e. to be predicted) using medical knowledge map
Medicine entity vector；And learn to obtain the text feature of patient medical record using the textual classification model based on deep learning
Vector, finally, the medicine entity vector and Text eigenvector of patient medical record are input to Seq2Seq model, Seq2Seq model
Output phase answers prediction result.
Specifically, combining the medicine entity vector sum Text eigenvector in case history text to be predicted, specifically, medicine
Entity vector forms a sequence according to the sequencing that medicine entity occurs in case history text to be predicted, then, by text spy
Sign vector is added to the sequence least significant end, obtains list entries.After the list entries is input to Seq2Seq model, disease is exported
Sick sequence (such as a variety of diseases), alternatively, operation sequence (such as a variety of operations), alternatively, check item sequence (such as multiple inspections).
In addition, each single item in the sequence generated in Decoder because of Seq2Seq, before all relying on this in sequence
The item in face, therefore the correlation between each item i.e. prediction result is considered using Seq2Seq model,
In practical application scene, doctor by a patient prescribe in drug more than one, similarly, check item and
Operation may also more than one, so, when carrying out diagnosis and treatment program prediction, prediction model can predict to obtain a variety of drugs, a variety of
Check item, a variety of operations etc., i.e. prediction obtains multiple items.For drug prediction model, a kind of corresponding result label of drug,
And often there is correlation between prediction result, for example, have side effect relationship that cannot use simultaneously between drug A and drug B, this
When, it should the identical drug replacement drug A or drug B of another function is used instead, if prediction model considers this correlation
Prediction result can be made more reliable, more useful.
It for different types of diagnosis and treatment scheme, needs to be predicted to obtain using corresponding Seq2Seq model, that is, need
The corresponding Seq2Seq model of different type diagnosis and treatment scheme is trained in advance, for example, the Seq2Seq model of predictive disease, prediction hand
The Seq2Seq model of the Seq2Seq model of art, predicted treatment drug, predicts Seq2Seq model of check item etc..
In the Seq2Seq model of the different prediction result types of training, trained sample corresponding with prediction result type is utilized
This progress model training, for example, obtaining predicted treatment drug using the training sample training Seq2Seq model comprising therapeutic agent
Seq2Seq model.Therefore, it is necessary to classify to case history training sample.
In one embodiment of the application, the textual classification model based on deep learning can use to a large amount of case history
Training sample is classified.By in case history text input to the textual classification model based on deep learning, the case history text is obtained
Corresponding text categories.
In one possible implementation of the present invention, case history text is carried out based on the textual classification model of deep learning
The case history text can be first converted to corresponding Text eigenvector during classification, then, according to text feature vector
Classify to case history text.As it can be seen that extracting Text eigenvector is that the intermediate of the textual classification model based on deep learning produces
Object, therefore, S120, which extracts the corresponding Text eigenvector of case history text to be predicted, can also use the text based on deep learning
Disaggregated model is realized.
For example, the textual classification model based on deep learning can be realized using multi-level attention model (HAN model).
Certainly, it in other possible implementations of the invention, can be replaced using other models based on deep learning, herein not
It is described in detail again.
Diagnosis and treatment program prediction method provided in this embodiment obtains the spy for the medicine entity that case history text to be predicted is included
Sign vector obtains medicine entity vector, meanwhile, obtain the Text eigenvector of the case history text to be predicted.It then, will be to be predicted
The corresponding medicine entity vector sum Text eigenvector of case history text is input to the diagnosis and treatment program prediction model that training obtains in advance
In, prediction obtains the diagnosis and treatment scheme to match with the case history text to be predicted, wherein diagnosis and treatment scheme includes diagnosis and treatment drug, operation
At least one of with check item.As shown in the above, this method is when carrying out feature extraction to case history text to be predicted, both
It is extracted the feature vector of medicine entity and its incidence relation, while being extracted the language ambience information of case history text, reduces information
Loss, improves the accuracy of the feature vector of case history text to be predicted, and then improve the predictablity rate of diagnosis and treatment scheme.
The training process of HAN model is discussed in detail by taking HAN model as an example below, as shown in figure 3, the training process can wrap
Include following steps:
S210 obtains the case history training sample for having marked text categories.
Case history training sample in the present embodiment has been labelled with text categories belonging to the case history training sample in advance,
In, it can manually mark text categories belonging to existing electronic health record text, wherein text categories and prediction result classification phase
Together, including at least one in disease category, drug categories, Operative category.
Case history training sample is converted to text vector using term vector by S220.
Term vector is used to indicate that the vector of word, and effect is that each word in text is mapped as a vector,
Realize the conversion of text to vector.
But general term vector is obtained by general large-scale corpus training, this term vector since meaning is more extensive,
Meaning representated by most term vector does not have the word meaning of specific area.Therefore, certain domain term is had more in order to obtain
The term vector of language feature needs to be trained term vector model using the word corpus in the field, the word obtained using training
Vector model generates the term vector with the word meaning in the field；The term vector that training obtains is capable of the word conversion in the field
Preferably expressed the vector of the domanial words meaning.
For example, the corpus training using medical domain obtains the term vector in the field, turned using the term vector of medical domain
The vector got in return can preferably characterize the word of medical domain.
The text vector of case history training sample is input to the multi-level attention model constructed in advance, extracted by S230
To Text eigenvector, and case history training sample is predicted based on Text eigenvector to obtain prediction text categories.
Multi-level attention model (HAN model) includes three layers and is followed successively by lexis, sentence layer and output layer；Wherein, first
Case history text to be predicted is first divided into multiple sentences, then, with CNN (Convolutional Neural Networks, volume
Product neural network) vector that extracts in each sentence of/LSTM (long-short term memory, shot and long term memory) obtains sentence
Subcharacter vector, it is contemplated that each word there is different information content and each sentence to have difference to entire text sentence
Information content, therefore Attention mechanism can be added in the layer；Then, sentence characteristics vector encode using two-way GRU
To the whole semantic feature vector of text, while in view of information content of each sentence to entire text contributes different introducing attentions
Mechanism (that is, Attention mechanism).Finally, obtained Text eigenvector to be input to the full articulamentum with activation primitive
That is output layer finally obtains the classification confidence level of case history text to be predicted.
Wherein, GRU network is a kind of good variant of effect of LSTM network, the structure ratio LSTM network of GRU network
Structure is simpler, and effect is good, and LSTM network is capable of the dependence of Chief Learning Officer, CLO, remembers longer historical information, equally,
The variant GRU network of LSTM network can also learn longer dependence, for example, the position in text between each word is closed
System.
Natural language has positional relationship (that is, word order relationship), GRU model energy as a kind of sequence type, between word and word
The mutual alignment relation between this word and word is enough recorded, it is more preferable for analyzing text vector effect.Wherein, two-way GRU (Bi-
GRU) model can both carry out positive and reverse two-way analysis to text vector and obtain positive semantic and reverse semanteme, so as to
More accurately extract the semantic feature in text vector.
Attention mechanism, is called attention mechanism of doing, and this mechanism can make model during training pattern
Some important information are more concerned about, such as in medical domain, those words related with medicine entity is focused more on, mould can be made in this way
Type learning ability is more preferable, so that the accuracy rate that model is finally predicted is higher.For example, the semanteme obtained using Bi-GRU model extraction
The vector of each word dimension does not indicate the weight of importance in feature, after increasing Attention mechanism, so that the semanteme is special
Word dimension vector in sign has the weight of importance, so that the classification results of final case history training sample are more acurrate.
Parameter or customized parameter when the model parameter in HAN model constructed in advance is initialization, model training
Process is exactly the process of constantly Optimized model parameter.
S240 based on the corresponding prediction text categories of each case history training sample and has marked text categories, optimizes more
Model parameter in level attention model, the multi-level attention model after utilizing optimization is to the case history training sample
The prediction text type predicted meets the default condition of convergence.
Model training process is exactly the model parameter constantly optimized in the HAN model constructed in advance, default until meeting
The condition of convergence, wherein the condition of convergence may include that loss function minimizes, while accuracy rate rises to highest.In other words, mould
Shape parameter optimization process is exactly the inherence between the Text eigenvector by learning case history text and the text categories belonging to it
Relationship finally determines the model parameter combination that internal relation can be recognized accurately.
Specifically, obtaining the prediction knot of text categories belonging to all case history training samples using current HAN model prediction
Fruit；Since case history training sample is labeled with its text categories label, according to the true of the prediction result of case history training sample and mark
The penalty values and accuracy rate of current HAN model can be calculated in real text categories.Constantly according to penalty values and accuracy rate to working as
Model parameter in preceding HAN model optimizes, pre- until being obtained using the HAN model prediction case history training sample after optimization
The penalty values surveyed between result and its text categories result marked are less than penalty values threshold value, and the accuracy rate of the prediction result is high
In accuracy rate threshold value, determine that "current" model parameter is that optimal model parameters combine to get final HAN model is arrived at this time.
Fig. 4 is referred to, the flow chart of the training process of diagnosis and treatment program prediction model provided in an embodiment of the present invention is shown.
As shown in figure 4, the training process mainly comprises the steps that
S310 obtains the case history training sample for having marked diagnosis and treatment scheme.
In one embodiment of the invention, diagnosis and treatment side employed in electronic health record samples of text can manually be marked
Case, for example, therapeutic agent, operation, in check item at least one of.
S320 obtains the medicine entity vector of the included medicine entity of case history training sample.
The medicine entity that case history training sample is included is extracted, and obtains case history training sample institute according to medical knowledge map
The incidence relation between each medicine entity for including；Then, learning model is indicated according to medical knowledge map, by medicine entity
And its incidence relation is mapped as corresponding medicine entity vector.
S330, using the textual classification model based on deep learning, obtain the corresponding text feature of case history training sample to
Amount and text categories.
Diagnosis and treatment scheme type corresponding to different case history training samples may be different, and different diagnosis and treatment scheme types is corresponding
Feature vector may be different, therefore, it is necessary to respectively for the different corresponding case history training sample difference of diagnosis and treatment scheme type
The corresponding diagnosis and treatment program prediction model of training.
Wherein, the textual classification model based on deep learning is used to be classified to obtain case history instruction to case history training sample
Practice the text categories of sample.
In one embodiment provided by the invention, case history training sample is segmented first, and utilizes term vector will
It segments obtained word segmentation result and is mapped as corresponding vector, then each vector is encoded to obtain the text of case history training sample
This vector.The text vector of case history training sample is input in the textual classification model based on deep learning, text classification
Model first extracts from the text vector of case history training sample and obtains Text eigenvector, and according to text feature vector to disease
Training sample is gone through to be classified to obtain the corresponding text categories of case history training sample.As it can be seen that Text eigenvector is based on deep
Therefore the intermediate product for spending the textual classification model of study is obtaining case history using the textual classification model based on deep learning
While the text categories of training sample, moreover it is possible to obtain the Text eigenvector of the case history training sample.
For example, the textual classification model based on deep learning can be HAN model.
The corresponding medicine entity vector of case history training sample and Text eigenvector are input to and to construct in advance by S340
Seq2Seq model, prediction obtain the corresponding prediction diagnosis and treatment scheme of case history training sample.
It is constructed being input to after case history training sample corresponding medicine entity vector and Text eigenvector combination in advance
In Seq2Seq model, wherein form a vector sequence according to the sequencing that medicine entity occurs in case history training sample
Column, then, are added to the sequence least significant end for Text eigenvector and obtain list entries, which is input to
Seq2Seq model, the Seq2Seq model export the corresponding prediction diagnosis and treatment scheme of the case history training sample.
Different classes of case history training sample is separately input into different Seq2Seq models, for example, case history training sample
Text categories include 4 classes, then need respectively 4 Seq2Seq models of building in advance, same category of case history training sample difference
It is input in the same Seq2Seq model.
S350, for the diagnosis and treatment program prediction model of each classification, based on corresponding each case history training of the category
The corresponding prediction diagnosis and treatment scheme of sample and diagnosis and treatment scheme is marked, adjusted the model parameter in Seq2Seq model, until using adjusting
The prediction diagnosis and treatment scheme that Seq2Seq model after whole predicts case history training sample meets the default condition of convergence.
Model training process is exactly the model parameter constantly optimized in the Seq2Seq model constructed in advance, until meeting
The default condition of convergence, wherein the condition of convergence may include that loss function minimizes, while accuracy rate rises to highest.Change speech
It, Model Parameter Optimization process be exactly by learn case history training sample medical features vector sum Text eigenvector with it is corresponding
Diagnosis and treatment scheme between internal relation, finally determine can be recognized accurately internal relation model parameter combination.
Specifically, prediction diagnosis and treatment scheme is to be predicted to obtain to case history training sample using current Seq2Seq model
Prediction result.Each case history training sample is labeled with its corresponding diagnosis and treatment scheme, that is, has marked diagnosis and treatment scheme.Then, according to
The prediction medical treatment result of case history training sample and the loss letter that current Seq2Seq model is calculated between medical treatment result is marked
Several and accuracy rate.Constantly the model parameter in current Seq2Seq model is optimized according to loss function and accuracy rate, directly
The mark marked to the prediction diagnosis and treatment scheme for obtaining each case history training sample using the Seq2Seq model prediction after optimization with it
It infuses the loss function between diagnosis and treatment scheme and is less than penalty values threshold value, and accuracy rate is higher than accuracy rate threshold value, at this point, to be determined current
Model parameter is that optimal model parameters combine to get final Seq2Seq model is arrived.
Diagnosis and treatment program prediction model training process provided in this embodiment after obtaining case history training sample, extracts case history instruction
Practice the medicine entity in sample and obtain body characteristics vector corresponding to the incidence relation between each medicine entity, obtains medicine
Entity vector.Then, case history training sample is classified to obtain each case history using based on deep learning textual classification model
The corresponding text categories of training sample and Text eigenvector.By the corresponding medical features of same category of case history training sample to
Amount and Text eigenvector be input in Seq2Seq model, prediction obtain prediction diagnosis and treatment scheme, and according to prediction diagnosis and treatment scheme and
The model parameter for having marked the current Seq2Seq model of diagnosis and treatment scheme optimization, until meeting the default condition of convergence.Above-mentioned entire instruction
Practice process and does not need artificial constructed rule and feature, it is time saving and energy saving.
Corresponding to above-mentioned diagnosis and treatment program prediction embodiment of the method, the present invention also provides corresponding Installation practices.
Fig. 5 is referred to, a kind of block diagram of diagnosis and treatment program prediction device provided in an embodiment of the present invention is shown, which answers
For server-side, the device as shown in Figure 5 includes: that the first acquisition module 110, second obtains module 120, third obtains module
130 and prediction module 140.
First obtains module 110, for obtaining case history text to be predicted.
Second obtains module 120, for obtain the corresponding medicine entity of medicine entity that case history text to be predicted includes to
Amount.
In one embodiment of the invention, as shown in fig. 6, the second acquisition module 120 may include the first acquisition submodule
Block 121 and mapping submodule 122；
First acquisition submodule 121, the medicine entity for being included for obtaining case history text to be predicted.
The mapping submodule 122 will be in case history text to be predicted for indicating learning model according to medical knowledge map
Medicine entity is mapped as corresponding medicine entity vector.
Third obtains module 130, for obtaining the corresponding Text eigenvector of case history text to be predicted.
In one embodiment of the invention, as shown in fig. 7, the third obtain module 130 include: participle submodule 131,
Mapping submodule 132 and feature extraction submodule 133；
The participle submodule 131 obtains disease to be predicted for segmenting using segmentation methods to case history text to be predicted
Go through the text word segmentation result of text.
The mapping submodule 132, for using term vector by each word in text word segmentation result be mapped as it is corresponding to
Amount, obtains the corresponding text vector of case history text to be predicted.
This feature extracting sub-module 133, for utilizing multi-level attention model from the corresponding text of case history text to be predicted
It is extracted in this vector and obtains Text eigenvector.
Wherein, Text eigenvector includes semantic feature and text entirety semantic feature between semantic feature, sentence between word.
Prediction module 140, for utilizing the diagnosis and treatment program prediction model analysis case history text to be predicted that training obtains in advance
Corresponding medicine entity vector sum text vector obtains the diagnosis and treatment scheme to match with the case history text to be predicted.
Wherein, the diagnosis and treatment scheme includes at least one of drug, operation and check item.
In one embodiment of the invention, diagnosis and treatment program prediction model includes the corresponding prediction mould of multiple pre-set categories
Type, as shown in figure 8, the prediction module 140 includes: classification submodule 141 and prediction submodule 142.
The classification submodule 141, for the corresponding Text eigenvector of case history text to be predicted to be input to pre-selection training
The obtained textual classification model based on deep learning obtains the corresponding text categories of case history text to be predicted.
The prediction submodule 142, for inputting the corresponding medicine entity vector sum text vector of case history text to be predicted
To in prediction model corresponding with the text categories of case history text to be predicted, what prediction obtained matching with case history text to be predicted
Diagnosis and treatment scheme.
In one embodiment of the invention, the process of textual classification model of the training based on deep learning is as follows:
Obtain the case history training sample for having marked text categories；
The case history training sample is converted into text vector using term vector；
The text vector of the case history training sample is input to the multi-level attention model constructed in advance, extraction obtains
Text eigenvector, and the case history training sample is predicted based on the Text eigenvector to obtain prediction text class
Not；
Based on the corresponding prediction text categories of case history training sample described in each and text categories are marked, described in adjustment
Model parameter in multi-level attention model, until training sample to the case history using multi-level attention model adjusted
This prediction text type predicted meets the default condition of convergence.
In another embodiment of the present invention, the process of the training diagnosis and treatment program prediction model includes:
Obtain the case history training sample for having marked diagnosis and treatment scheme；
The medicine entity vector of the included medicine entity of the case history training sample is obtained, and obtains the case history training sample
This corresponding Text eigenvector；
The corresponding medicine entity vector of the case history training sample and Text eigenvector be input to and is constructed in advance
Seq2Seq model, prediction obtain the corresponding prediction diagnosis and treatment scheme of the case history training sample；
Based on the corresponding prediction diagnosis and treatment scheme of case history training sample described in each and diagnosis and treatment scheme is marked, described in adjustment
Model parameter in Seq2Seq model, until being predicted using Seq2Seq model adjusted the case history training sample
Obtained prediction diagnosis and treatment scheme meets the default condition of convergence.
Diagnosis and treatment program prediction device provided in this embodiment obtains the spy for the medicine entity that case history text to be predicted is included
Sign vector obtains medicine entity vector, meanwhile, obtain the Text eigenvector of the case history text to be predicted.It then, will be to be predicted
The corresponding medicine entity vector sum Text eigenvector of case history text is input to the diagnosis and treatment program prediction model that training obtains in advance
In, prediction obtains the diagnosis and treatment scheme to match with the case history text to be predicted, wherein diagnosis and treatment scheme includes diagnosis and treatment drug, operation
At least one of with check item.As shown in the above, the device is when carrying out feature extraction to case history text to be predicted, both
It is extracted the feature vector of medicine entity and its incidence relation, while being extracted the language ambience information of case history text, reduces information
Loss, improves the accuracy of the feature vector of case history text to be predicted, and then improve the predictablity rate of diagnosis and treatment scheme.
For the various method embodiments described above, for simple description, therefore, it is stated as a series of action combinations, but
Be those skilled in the art should understand that, the present invention is not limited by the sequence of acts described because according to the present invention, certain
A little steps can be performed in other orders or simultaneously.Secondly, those skilled in the art should also know that, it is retouched in specification
The embodiment stated belongs to preferred embodiment, and related actions and modules are not necessarily necessary for the present invention.
It should be noted that all the embodiments in this specification are described in a progressive manner, each embodiment weight
Point explanation is the difference from other embodiments, and the same or similar parts between the embodiments can be referred to each other.
For device class embodiment, since it is basically similar to the method embodiment, so being described relatively simple, related place ginseng
See the part explanation of embodiment of the method.
Step in each embodiment method of the application can be sequentially adjusted, merged and deleted according to actual needs.
Device in each embodiment of the application and the module in terminal and submodule can merge according to actual needs,
It divides and deletes.
In several embodiments provided herein, it should be understood that disclosed terminal, device and method, Ke Yitong
Other modes are crossed to realize.For example, terminal embodiment described above is only schematical, for example, module or submodule
Division, only a kind of logical function partition, there may be another division manner in actual implementation, for example, multiple submodule or
Module may be combined or can be integrated into another module, or some features can be ignored or not executed.Another point is shown
The mutual coupling, direct-coupling or communication connection shown or discussed can be through some interfaces, between device or module
Coupling or communication connection are connect, can be electrical property, mechanical or other forms.
Module or submodule may or may not be physically separated as illustrated by the separation member, as mould
The component of block or submodule may or may not be physical module or submodule, it can and it is in one place, or
It may be distributed on multiple network modules or submodule.Some or all of mould therein can be selected according to the actual needs
Block or submodule achieve the purpose of the solution of this embodiment.
In addition, each functional module or submodule in each embodiment of the application can integrate in a processing module
In, it is also possible to modules or submodule physically exists alone, it can also be integrated with two or more modules or submodule
In a module.Above-mentioned integrated module or submodule both can take the form of hardware realization, can also use software function
Energy module or the form of submodule are realized.
Finally, it is to be noted that, herein, relational terms such as first and second and the like be used merely to by
One entity or operation are distinguished with another entity or operation, without necessarily requiring or implying these entities or operation
Between there are any actual relationship or orders.Moreover, the terms "include", "comprise" or its any other variant meaning
Covering non-exclusive inclusion, so that the process, method, article or equipment for including a series of elements not only includes that
A little elements, but also including other elements that are not explicitly listed, or further include for this process, method, article or
The intrinsic element of equipment.In the absence of more restrictions, the element limited by sentence "including a ...", is not arranged
Except there is also other identical elements in the process, method, article or apparatus that includes the element.
The foregoing description of the disclosed embodiments can be realized those skilled in the art or using the present invention.To this
A variety of modifications of a little embodiments will be apparent for a person skilled in the art, and the general principles defined herein can
Without departing from the spirit or scope of the present invention, to realize in other embodiments.Therefore, the present invention will not be limited
It is formed on the embodiments shown herein, and is to fit to consistent with the principles and novel features disclosed in this article widest
Range.
The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also answered
It is considered as protection scope of the present invention.
Claims (10)
1. a kind of diagnosis and treatment program prediction method characterized by comprising
Obtain case history text to be predicted；
The corresponding medicine entity vector of medicine entity that the case history text to be predicted includes is obtained, and obtains the disease to be predicted
Go through the corresponding Text eigenvector of text；
The corresponding medicine entity of case history text to be predicted described in the diagnosis and treatment program prediction model analysis obtained using preparatory training to
Amount and text vector, obtain the diagnosis and treatment scheme to match with the case history text to be predicted, the diagnosis and treatment scheme includes drug, hand
At least one of art and check item.
2. the method according to claim 1, wherein the diagnosis and treatment program prediction model includes that pre-set categories are corresponding
Prediction model；
It is described using the corresponding medicine entity vector sum text of case history text to be predicted described in diagnosis and treatment program prediction model analysis to
Amount, obtains the diagnosis and treatment scheme to match with the case history text to be predicted, comprising:
By the corresponding Text eigenvector of the case history text to be predicted be input to that pre-selection training obtains based on deep learning
Textual classification model obtains the corresponding text categories of the case history text to be predicted；
By the corresponding medicine entity vector sum text vector of the case history text to be predicted, it is input to and the case history text to be predicted
In this corresponding prediction model of text categories, prediction obtains the diagnosis and treatment scheme to match with the case history text to be predicted.
3. according to the method described in claim 2, it is characterized in that, the mistake of textual classification model of the training based on deep learning
Journey, comprising:
Obtain the case history training sample for having marked text categories；
The case history training sample is converted into text vector using term vector；
The text vector of the case history training sample is input to the multi-level attention model constructed in advance, extraction obtains text
Feature vector, and the case history training sample is predicted based on the Text eigenvector to obtain prediction text categories；
Based on the corresponding prediction text categories of case history training sample described in each and text categories have been marked, have adjusted the multilayer
Model parameter in secondary attention model, until using multi-level attention model adjusted to the case history training sample into
The prediction text type that row prediction obtains meets the default condition of convergence.
4. method according to claim 1-3, which is characterized in that it is corresponding to obtain the case history text to be predicted
Text eigenvector, comprising:
The case history text to be predicted is segmented using segmentation methods, obtains the text participle of the case history text to be predicted
As a result；
Each word in the text word segmentation result is mapped as corresponding vector using term vector, obtains the case history to be predicted
The corresponding text vector of text；
It is extracted from the corresponding text vector of the case history text to be predicted using multi-level attention model and obtains text feature
Vector；
Wherein, the Text eigenvector includes semantic feature and text entirety semantic feature between semantic feature, sentence between word.
5. the method according to claim 1, wherein the medicine for obtaining the case history text to be predicted and including
The corresponding medicine entity vector of entity, comprising:
Obtain the medicine entity that the case history text to be predicted is included；
Learning model is indicated according to medical knowledge map, the medicine entity in the case history text to be predicted is mapped as corresponding
Medicine entity vector.
6. the method according to claim 1, wherein the process of the training diagnosis and treatment program prediction model, comprising:
Obtain the case history training sample for having marked diagnosis and treatment scheme；
The medicine entity vector of the included medicine entity of the case history training sample is obtained, and obtains the case history training sample pair
The Text eigenvector answered；
The corresponding medicine entity vector of the case history training sample and Text eigenvector are input to the Seq2Seq constructed in advance
Model, prediction obtain the corresponding prediction diagnosis and treatment scheme of the case history training sample；
Based on the corresponding prediction diagnosis and treatment scheme of case history training sample described in each and diagnosis and treatment scheme is marked, described in adjustment
Model parameter in Seq2Seq model, until being predicted using Seq2Seq model adjusted the case history training sample
Obtained prediction diagnosis and treatment scheme meets the default condition of convergence.
7. a kind of diagnosis and treatment program prediction device characterized by comprising
First obtains module, for obtaining case history text to be predicted；
Second obtains module, the corresponding medicine entity vector of medicine entity for including for obtaining the case history text to be predicted；
Third obtains module, for obtaining the corresponding Text eigenvector of the case history text to be predicted；
Prediction module, for corresponding using case history text to be predicted described in the diagnosis and treatment program prediction model analysis that training obtains in advance
Medicine entity vector sum text vector, obtain the diagnosis and treatment scheme to match with the case history text to be predicted, the diagnosis and treatment side
Case includes at least one of drug, operation and check item.
8. device according to claim 7, which is characterized in that the diagnosis and treatment program prediction model includes that pre-set categories are corresponding
Prediction model；The prediction module includes:
Classify submodule, for the corresponding Text eigenvector of the case history text to be predicted to be input to what pre-selection training obtained
Textual classification model based on deep learning obtains the corresponding text categories of the case history text to be predicted；
Predict submodule, for will the corresponding medicine entity vector sum text vector of the case history text to be predicted, be input to and
In the corresponding prediction model of text categories of the case history text to be predicted, prediction is obtained and the case history text phase to be predicted
The diagnosis and treatment scheme matched.
9. device according to claim 7 or 8, which is characterized in that the third obtains module and includes:
It segments submodule and obtains the disease to be predicted for segmenting using segmentation methods to the case history text to be predicted
Go through the text word segmentation result of text；
Mapping submodule is obtained for each word in the text word segmentation result to be mapped as corresponding vector using term vector
To the corresponding text vector of the case history text to be predicted；
Feature extraction submodule, for using multi-level attention model from the corresponding text vector of the case history text to be predicted
Middle extraction obtains Text eigenvector；
Wherein, the Text eigenvector includes semantic feature and text entirety semantic feature between semantic feature, sentence between word.
10. device according to claim 7, which is characterized in that described second, which obtains module, includes:
First acquisition submodule, the medicine entity for being included for obtaining the case history text to be predicted；
Mapping submodule, for indicating learning model according to medical knowledge map, by the medicine in the case history text to be predicted
Entity is mapped as corresponding medicine entity vector.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910584729.0A CN110297908B (en) | 2019-07-01 | 2019-07-01 | Diagnosis and treatment scheme prediction method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910584729.0A CN110297908B (en) | 2019-07-01 | 2019-07-01 | Diagnosis and treatment scheme prediction method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110297908A true CN110297908A (en) | 2019-10-01 |
| CN110297908B CN110297908B (en) | 2021-09-21 |
Family
ID=68029775
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910584729.0A Active CN110297908B (en) | 2019-07-01 | 2019-07-01 | Diagnosis and treatment scheme prediction method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110297908B (en) |
Cited By (27)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111128391A (en) * | 2019-12-24 | 2020-05-08 | 北京推想科技有限公司 | Information processing apparatus, method and storage medium |
| CN111191035A (en) * | 2019-12-30 | 2020-05-22 | 四川大学华西医院 | Method and device for recognizing lung cancer clinical database text entity |
| CN111259111A (en) * | 2020-01-13 | 2020-06-09 | 安徽科大讯飞医疗信息技术有限公司 | Medical record-based decision-making assisting method and device, electronic equipment and storage medium |
| CN111259112A (en) * | 2020-01-14 | 2020-06-09 | 北京百度网讯科技有限公司 | Medical fact verification method and device |
| CN111274397A (en) * | 2020-01-20 | 2020-06-12 | 北京百度网讯科技有限公司 | Method and device for establishing entity relationship detection model |
| CN111383770A (en) * | 2020-03-09 | 2020-07-07 | 北京大学 | Clinical research and follow-up visit fusion system and method |
| CN111507827A (en) * | 2020-04-20 | 2020-08-07 | 上海商涌网络科技有限公司 | Health risk assessment method, terminal and computer storage medium |
| CN111540468A (en) * | 2020-04-21 | 2020-08-14 | 重庆大学 | ICD automatic coding method and system for visualization of diagnosis reason |
| CN111724897A (en) * | 2020-06-12 | 2020-09-29 | 电子科技大学 | Motion function data processing method and system |
| CN111834017A (en) * | 2020-07-09 | 2020-10-27 | 上海市精神卫生中心（上海市心理咨询培训中心） | Method, system and device for predicting treatment effect of psychotropic drugs |
| CN112017776A (en) * | 2020-10-27 | 2020-12-01 | 平安科技（深圳）有限公司 | Disease prediction method based on dynamic graph and medical knowledge map and related equipment |
| CN112259246A (en) * | 2020-10-21 | 2021-01-22 | 平安科技（深圳）有限公司 | Disease prediction method fusing medical concept hierarchical structure and related equipment |
| CN112331332A (en) * | 2020-10-14 | 2021-02-05 | 北京工业大学 | Disease prediction method and system based on multi-granularity feature fusion |
| CN112435761A (en) * | 2020-12-04 | 2021-03-02 | 中国信息通信研究院 | Information recommendation method and device |
| CN112635050A (en) * | 2020-12-23 | 2021-04-09 | 安徽科大讯飞医疗信息技术有限公司 | Diagnosis recommendation method, electronic equipment and storage device |
| CN112786198A (en) * | 2021-02-08 | 2021-05-11 | 科大讯飞股份有限公司 | Diagnosis and treatment information recommendation model construction method, diagnosis and treatment information recommendation method and device |
| CN112820411A (en) * | 2021-01-27 | 2021-05-18 | 清华大学 | Medical relation extraction method and device |
| CN113094476A (en) * | 2021-06-08 | 2021-07-09 | 明品云(北京)数据科技有限公司 | Risk early warning method, system, equipment and medium based on natural language processing |
| WO2021151351A1 (en) * | 2020-09-04 | 2021-08-05 | 平安科技（深圳）有限公司 | Data processing method and apparatus, computer device, and storage medium |
| CN113270189A (en) * | 2021-05-19 | 2021-08-17 | 复旦大学附属肿瘤医院 | Tumor treatment aid decision-making method based on reinforcement learning |
| CN113345286A (en) * | 2021-08-03 | 2021-09-03 | 南京大经中医药信息技术有限公司 | Teacher-and-bearing teaching system and method integrating AI technology and video technology |
| CN113470827A (en) * | 2021-06-30 | 2021-10-01 | 上海商汤智能科技有限公司 | Classification method and device, electronic equipment and storage medium |
| CN113535974A (en) * | 2021-06-28 | 2021-10-22 | 科大讯飞华南人工智能研究院（广州）有限公司 | Diagnosis recommendation method and related device, electronic equipment and storage medium |
| CN114999599A (en) * | 2022-05-26 | 2022-09-02 | 中国科学技术信息研究所 | Data processing method and device, electronic equipment and computer readable storage medium |
| CN116306925A (en) * | 2023-03-14 | 2023-06-23 | 中国人民解放军总医院 | Method and system for generating end-to-end entity link |
| CN116504331A (en) * | 2023-04-28 | 2023-07-28 | 东北林业大学 | Frequency score prediction method for drug side effects based on multiple modes and multiple tasks |
| CN116306925B (en) * | 2023-03-14 | 2024-05-03 | 中国人民解放军总医院 | Method and system for generating end-to-end entity link |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8126736B2 (en) * | 2009-01-23 | 2012-02-28 | Warsaw Orthopedic, Inc. | Methods and systems for diagnosing, treating, or tracking spinal disorders |
| US20170147567A1 (en) * | 2003-08-20 | 2017-05-25 | Intellectual Ventures I Llc | File managing method for a digital apparatus |
| CN108305690A (en) * | 2018-02-07 | 2018-07-20 | 王博冉 | The nervous system disease point examines the system for distribution of out-patient department of method and the nervous system disease |
| CN108595708A (en) * | 2018-05-10 | 2018-09-28 | 北京航空航天大学 | A kind of exception information file classification method of knowledge based collection of illustrative plates |
| CN108984745A (en) * | 2018-07-16 | 2018-12-11 | 福州大学 | A kind of neural network file classification method merging more knowledge mappings |
| CN109087691A (en) * | 2018-08-02 | 2018-12-25 | 科大智能机器人技术有限公司 | A kind of OTC drugs recommender system and recommended method based on deep learning |
| CN109346169A (en) * | 2018-10-17 | 2019-02-15 | 长沙瀚云信息科技有限公司 | A kind of artificial intelligence assisting in diagnosis and treatment system and its construction method, equipment and storage medium |
| CN109378064A (en) * | 2018-10-29 | 2019-02-22 | 南京医基云医疗数据研究院有限公司 | Medical data processing method, device electronic equipment and computer-readable medium |
| CN109754886A (en) * | 2019-01-07 | 2019-05-14 | 广州达美智能科技有限公司 | Therapeutic scheme intelligent generating system, method and readable storage medium storing program for executing, electronic equipment |
| CN109785928A (en) * | 2018-12-25 | 2019-05-21 | 平安科技（深圳）有限公司 | Diagnosis and treatment proposal recommending method, device and storage medium |
| CN109920508A (en) * | 2018-12-28 | 2019-06-21 | 安徽省立医院 | prescription auditing method and system |
| CN109920501A (en) * | 2019-01-24 | 2019-06-21 | 西安交通大学 | Electronic health record classification method and system based on convolutional neural networks and Active Learning |
-
2019
- 2019-07-01 CN CN201910584729.0A patent/CN110297908B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170147567A1 (en) * | 2003-08-20 | 2017-05-25 | Intellectual Ventures I Llc | File managing method for a digital apparatus |
| US8126736B2 (en) * | 2009-01-23 | 2012-02-28 | Warsaw Orthopedic, Inc. | Methods and systems for diagnosing, treating, or tracking spinal disorders |
| CN108305690A (en) * | 2018-02-07 | 2018-07-20 | 王博冉 | The nervous system disease point examines the system for distribution of out-patient department of method and the nervous system disease |
| CN108595708A (en) * | 2018-05-10 | 2018-09-28 | 北京航空航天大学 | A kind of exception information file classification method of knowledge based collection of illustrative plates |
| CN108984745A (en) * | 2018-07-16 | 2018-12-11 | 福州大学 | A kind of neural network file classification method merging more knowledge mappings |
| CN109087691A (en) * | 2018-08-02 | 2018-12-25 | 科大智能机器人技术有限公司 | A kind of OTC drugs recommender system and recommended method based on deep learning |
| CN109346169A (en) * | 2018-10-17 | 2019-02-15 | 长沙瀚云信息科技有限公司 | A kind of artificial intelligence assisting in diagnosis and treatment system and its construction method, equipment and storage medium |
| CN109378064A (en) * | 2018-10-29 | 2019-02-22 | 南京医基云医疗数据研究院有限公司 | Medical data processing method, device electronic equipment and computer-readable medium |
| CN109785928A (en) * | 2018-12-25 | 2019-05-21 | 平安科技（深圳）有限公司 | Diagnosis and treatment proposal recommending method, device and storage medium |
| CN109920508A (en) * | 2018-12-28 | 2019-06-21 | 安徽省立医院 | prescription auditing method and system |
| CN109754886A (en) * | 2019-01-07 | 2019-05-14 | 广州达美智能科技有限公司 | Therapeutic scheme intelligent generating system, method and readable storage medium storing program for executing, electronic equipment |
| CN109920501A (en) * | 2019-01-24 | 2019-06-21 | 西安交通大学 | Electronic health record classification method and system based on convolutional neural networks and Active Learning |
Non-Patent Citations (1)
| Title |
|---|
| 桑盛田等: "融合知识图谱与深度学习的药物发现方法", 《模式识别与人工智能》 * |
Cited By (42)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111128391B (en) * | 2019-12-24 | 2021-01-12 | 推想医疗科技股份有限公司 | Information processing apparatus, method and storage medium |
| CN111128391A (en) * | 2019-12-24 | 2020-05-08 | 北京推想科技有限公司 | Information processing apparatus, method and storage medium |
| CN111191035A (en) * | 2019-12-30 | 2020-05-22 | 四川大学华西医院 | Method and device for recognizing lung cancer clinical database text entity |
| CN111191035B (en) * | 2019-12-30 | 2022-07-19 | 四川大学华西医院 | Method and device for recognizing lung cancer clinical database text entity |
| CN111259111A (en) * | 2020-01-13 | 2020-06-09 | 安徽科大讯飞医疗信息技术有限公司 | Medical record-based decision-making assisting method and device, electronic equipment and storage medium |
| CN111259111B (en) * | 2020-01-13 | 2022-02-18 | 安徽科大讯飞医疗信息技术有限公司 | Medical record-based decision-making assisting method and device, electronic equipment and storage medium |
| CN111259112A (en) * | 2020-01-14 | 2020-06-09 | 北京百度网讯科技有限公司 | Medical fact verification method and device |
| US11636936B2 (en) | 2020-01-14 | 2023-04-25 | Beijing Baidu Netcom Science And Technology Co., Ltd. | Method and apparatus for verifying medical fact |
| CN111259112B (en) * | 2020-01-14 | 2023-07-04 | 北京百度网讯科技有限公司 | Medical fact verification method and device |
| CN111274397A (en) * | 2020-01-20 | 2020-06-12 | 北京百度网讯科技有限公司 | Method and device for establishing entity relationship detection model |
| CN111274397B (en) * | 2020-01-20 | 2023-10-27 | 北京百度网讯科技有限公司 | Method and device for establishing entity relation detection model |
| CN111383770B (en) * | 2020-03-09 | 2022-08-30 | 北京大学 | Clinical research and follow-up visit fusion system and method |
| CN111383770A (en) * | 2020-03-09 | 2020-07-07 | 北京大学 | Clinical research and follow-up visit fusion system and method |
| CN111507827A (en) * | 2020-04-20 | 2020-08-07 | 上海商涌网络科技有限公司 | Health risk assessment method, terminal and computer storage medium |
| CN111540468A (en) * | 2020-04-21 | 2020-08-14 | 重庆大学 | ICD automatic coding method and system for visualization of diagnosis reason |
| CN111724897A (en) * | 2020-06-12 | 2020-09-29 | 电子科技大学 | Motion function data processing method and system |
| CN111724897B (en) * | 2020-06-12 | 2022-07-01 | 电子科技大学 | Motion function data processing method and system |
| CN111834017A (en) * | 2020-07-09 | 2020-10-27 | 上海市精神卫生中心（上海市心理咨询培训中心） | Method, system and device for predicting treatment effect of psychotropic drugs |
| WO2021151351A1 (en) * | 2020-09-04 | 2021-08-05 | 平安科技（深圳）有限公司 | Data processing method and apparatus, computer device, and storage medium |
| CN112331332A (en) * | 2020-10-14 | 2021-02-05 | 北京工业大学 | Disease prediction method and system based on multi-granularity feature fusion |
| CN112259246A (en) * | 2020-10-21 | 2021-01-22 | 平安科技（深圳）有限公司 | Disease prediction method fusing medical concept hierarchical structure and related equipment |
| CN112259246B (en) * | 2020-10-21 | 2023-07-25 | 平安科技（深圳）有限公司 | Disease prediction method integrating medical concept hierarchy structure and related equipment |
| CN112017776A (en) * | 2020-10-27 | 2020-12-01 | 平安科技（深圳）有限公司 | Disease prediction method based on dynamic graph and medical knowledge map and related equipment |
| CN112017776B (en) * | 2020-10-27 | 2021-01-15 | 平安科技（深圳）有限公司 | Disease prediction method based on dynamic graph and medical knowledge map and related equipment |
| CN112435761A (en) * | 2020-12-04 | 2021-03-02 | 中国信息通信研究院 | Information recommendation method and device |
| CN112635050A (en) * | 2020-12-23 | 2021-04-09 | 安徽科大讯飞医疗信息技术有限公司 | Diagnosis recommendation method, electronic equipment and storage device |
| CN112635050B (en) * | 2020-12-23 | 2022-12-06 | 安徽讯飞医疗股份有限公司 | Diagnosis recommendation method, electronic equipment and storage device |
| CN112820411B (en) * | 2021-01-27 | 2022-07-29 | 清华大学 | Medical relation extraction method and device |
| CN112820411A (en) * | 2021-01-27 | 2021-05-18 | 清华大学 | Medical relation extraction method and device |
| CN112786198A (en) * | 2021-02-08 | 2021-05-11 | 科大讯飞股份有限公司 | Diagnosis and treatment information recommendation model construction method, diagnosis and treatment information recommendation method and device |
| CN113270189B (en) * | 2021-05-19 | 2023-08-18 | 复旦大学附属肿瘤医院 | Tumor treatment auxiliary decision-making method based on reinforcement learning |
| CN113270189A (en) * | 2021-05-19 | 2021-08-17 | 复旦大学附属肿瘤医院 | Tumor treatment aid decision-making method based on reinforcement learning |
| CN113094476A (en) * | 2021-06-08 | 2021-07-09 | 明品云(北京)数据科技有限公司 | Risk early warning method, system, equipment and medium based on natural language processing |
| CN113535974A (en) * | 2021-06-28 | 2021-10-22 | 科大讯飞华南人工智能研究院（广州）有限公司 | Diagnosis recommendation method and related device, electronic equipment and storage medium |
| CN113535974B (en) * | 2021-06-28 | 2024-04-09 | 科大讯飞华南人工智能研究院（广州）有限公司 | Diagnostic recommendation method and related device, electronic equipment and storage medium |
| CN113470827A (en) * | 2021-06-30 | 2021-10-01 | 上海商汤智能科技有限公司 | Classification method and device, electronic equipment and storage medium |
| CN113345286B (en) * | 2021-08-03 | 2021-11-19 | 南京大经中医药信息技术有限公司 | Teacher-and-bearing teaching system and method integrating AI technology and video technology |
| CN113345286A (en) * | 2021-08-03 | 2021-09-03 | 南京大经中医药信息技术有限公司 | Teacher-and-bearing teaching system and method integrating AI technology and video technology |
| CN114999599A (en) * | 2022-05-26 | 2022-09-02 | 中国科学技术信息研究所 | Data processing method and device, electronic equipment and computer readable storage medium |
| CN116306925A (en) * | 2023-03-14 | 2023-06-23 | 中国人民解放军总医院 | Method and system for generating end-to-end entity link |
| CN116306925B (en) * | 2023-03-14 | 2024-05-03 | 中国人民解放军总医院 | Method and system for generating end-to-end entity link |
| CN116504331A (en) * | 2023-04-28 | 2023-07-28 | 东北林业大学 | Frequency score prediction method for drug side effects based on multiple modes and multiple tasks |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110297908B (en) | 2021-09-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110297908A (en) | Diagnosis and treatment program prediction method and device | |
| CN108363753B (en) | Comment text emotion classification model training and emotion classification method, device and equipment | |
| Mathews et al. | Semstyle: Learning to generate stylised image captions using unaligned text | |
| Zhou et al. | Deep learning based fusion approach for hate speech detection | |
| CN107516110B (en) | Medical question-answer semantic clustering method based on integrated convolutional coding | |
| CN109918644B (en) | Traditional Chinese medicine health consultation text named entity identification method based on transfer learning | |
| CN108416065B (en) | Hierarchical neural network-based image-sentence description generation system and method | |
| CN109871538A (en) | A kind of Chinese electronic health record name entity recognition method | |
| CN110032648A (en) | A kind of case history structuring analytic method based on medical domain entity | |
| CN109800411A (en) | Clinical treatment entity and its attribute extraction method | |
| CN109669994A (en) | A kind of construction method and system of health knowledge map | |
| CN111984766B (en) | Missing semantic completion method and device | |
| CN110825881A (en) | Method for establishing electric power knowledge graph | |
| CN110222163A (en) | A kind of intelligent answer method and system merging CNN and two-way LSTM | |
| CN110675944A (en) | Triage method and device, computer equipment and medium | |
| CN110807320A (en) | Short text emotion analysis method based on CNN bidirectional GRU attention mechanism | |
| Wen et al. | Dynamic interactive multiview memory network for emotion recognition in conversation | |
| Sharma et al. | A survey of methods, datasets and evaluation metrics for visual question answering | |
| WO2023029502A1 (en) | Method and apparatus for constructing user portrait on the basis of inquiry session, device, and medium | |
| CN111222049A (en) | Top-k similarity searching method on semantically enhanced heterogeneous information network | |
| CN108647191A (en) | It is a kind of based on have supervision emotion text and term vector sentiment dictionary construction method | |
| CN113707339B (en) | Method and system for concept alignment and content inter-translation among multi-source heterogeneous databases | |
| CN112309528B (en) | Medical image report generation method based on visual question-answering method | |
| CN112420191A (en) | Traditional Chinese medicine auxiliary decision making system and method | |
| CN111858940A (en) | Multi-head attention-based legal case similarity calculation method and system |
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 |