CN109637669B - Deep learning-based treatment scheme generation method, device and storage medium - Google Patents
Deep learning-based treatment scheme generation method, device and storage medium Download PDFInfo
- Publication number
- CN109637669B CN109637669B CN201811407145.8A CN201811407145A CN109637669B CN 109637669 B CN109637669 B CN 109637669B CN 201811407145 A CN201811407145 A CN 201811407145A CN 109637669 B CN109637669 B CN 109637669B
- Authority
- CN
- China
- Prior art keywords
- patient
- treatment scheme
- treatment
- sequence
- information
- 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.)
- Active
Links
- 238000011282 treatment Methods 0.000 title claims abstract description 190
- 238000013135 deep learning Methods 0.000 title claims abstract description 48
- 238000000034 method Methods 0.000 title claims abstract description 31
- 238000003745 diagnosis Methods 0.000 claims abstract description 79
- 238000012545 processing Methods 0.000 claims abstract description 43
- 238000003062 neural network model Methods 0.000 claims abstract description 41
- 239000013598 vector Substances 0.000 claims description 93
- 208000024891 symptom Diseases 0.000 claims description 31
- 230000015654 memory Effects 0.000 claims description 25
- 201000010099 disease Diseases 0.000 claims description 15
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 claims description 15
- 239000003814 drug Substances 0.000 claims description 5
- 229940079593 drug Drugs 0.000 claims description 5
- 238000002347 injection Methods 0.000 claims description 2
- 239000007924 injection Substances 0.000 claims description 2
- 238000013528 artificial neural network Methods 0.000 abstract description 8
- 230000000750 progressive effect Effects 0.000 abstract description 3
- 238000012549 training Methods 0.000 description 11
- 238000011269 treatment regimen Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 239000000243 solution Substances 0.000 description 3
- 206010068319 Oropharyngeal pain Diseases 0.000 description 2
- 201000007100 Pharyngitis Diseases 0.000 description 2
- 208000036071 Rhinorrhea Diseases 0.000 description 2
- 206010039101 Rhinorrhoea Diseases 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 208000002173 dizziness Diseases 0.000 description 2
- 230000003902 lesion Effects 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- 230000001225 therapeutic effect Effects 0.000 description 2
- 238000011285 therapeutic regimen Methods 0.000 description 2
- 206010008479 Chest Pain Diseases 0.000 description 1
- 206010037660 Pyrexia Diseases 0.000 description 1
- 208000009205 Tinnitus Diseases 0.000 description 1
- 230000006793 arrhythmia Effects 0.000 description 1
- 206010003119 arrhythmia Diseases 0.000 description 1
- 208000020670 canker sore Diseases 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 235000014510 cooky Nutrition 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 239000008267 milk Substances 0.000 description 1
- 210000004080 milk Anatomy 0.000 description 1
- 235000013336 milk Nutrition 0.000 description 1
- 230000000474 nursing effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000001356 surgical procedure Methods 0.000 description 1
- 231100000886 tinnitus Toxicity 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- 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
- G16H50/50—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for simulation or modelling of medical disorders
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Public Health (AREA)
- Medical Informatics (AREA)
- Pathology (AREA)
- Databases & Information Systems (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Medical Treatment And Welfare Office Work (AREA)
Abstract
The invention discloses a method for generating a treatment scheme based on deep learning, which comprises the following steps: acquiring diagnostic information of a patient to be treated; inputting the diagnosis information into a deep neural network model for processing to obtain a prediction result of the current treatment scheme of the patient to be processed; and inputting the predicted result of the current treatment scheme into a sequence model for processing to obtain the predicted result of the future treatment scheme of the patient to be processed. The invention also discloses a device for generating the treatment scheme based on deep learning and a computer readable storage medium. The present invention generates the current treatment scheme of the patient through the feedforward type deep neural network with time connection, predicts the future treatment scheme of the patient through the sequence-to-sequence model, and provides progressive treatment scheme reference suggestion for the patient so as to achieve the aim of better auxiliary treatment.
Description
Technical Field
The present invention relates to the field of artificial intelligence, and in particular, to a method for generating a treatment plan based on deep learning, a device for generating a treatment plan based on deep learning, and a computer readable storage medium.
Background
At present, a large amount of electronic medical record texts are generated every day in hospitals or medical institutions, and the electronic medical record is a medical record with very strong professionals, is an original record of the whole process of diagnosis and treatment of patients in hospitals, and contains information such as patient course records, examination and inspection results, medical advice, operation records, nursing records and the like.
In recent years, because of rapid development of big data and artificial intelligence technology, people begin to apply machine learning related technology to the field of auxiliary diagnosis or treatment so as to intelligently generate and recommend treatment methods for patients and help doctors to quickly formulate treatment schemes.
However, in the existing technology of intelligent generation of the treatment scheme, the relation between the diagnosis information of the patient in continuous time is not considered, and the treatment scheme of the patient in the current visit can be only predicted, so that the comprehensive prediction of the multiple treatment schemes of the patient in the whole treatment period can not be performed.
Disclosure of Invention
The invention mainly aims to provide a method for generating a treatment scheme based on deep learning, a device for generating the treatment scheme based on deep learning and a computer readable storage medium, which solve the technical problem that the comprehensive prediction of multiple treatment schemes of a patient in the whole treatment period cannot be realized in the prior art.
In order to achieve the above object, the present invention provides a method for generating a deep learning-based therapeutic scheme, the method comprising the steps of:
acquiring diagnostic information of a patient to be treated;
inputting the diagnosis information into a deep neural network model for processing to obtain a prediction result of the current treatment scheme of the patient to be processed;
and inputting the predicted result of the current treatment scheme into a sequence model for processing to obtain the predicted result of the future treatment scheme of the patient to be processed.
Preferably, the step of inputting the diagnostic information into a deep neural network model for processing to obtain a predicted result of the current treatment scheme of the patient to be processed includes:
inputting the diagnosis information into a deep neural network model for processing to obtain hidden vectors corresponding to the diagnosis information;
inputting hidden vectors corresponding to the diagnosis information into a self-attention mechanism layer for processing to obtain weights corresponding to the hidden vectors;
obtaining a weighted hidden vector according to the hidden vector and the weight;
and obtaining a prediction result of the current treatment scheme of the patient to be treated according to the weighted hidden vector.
Preferably, the step of inputting the hidden vector corresponding to the diagnostic information into a self-attention mechanism layer for processing, and obtaining the weight corresponding to the hidden vector includes:
inputting hidden vectors corresponding to the diagnosis information into a self-attention mechanism layer for processing;
and the self-attention mechanism layer learns the weight corresponding to the hidden vector according to the grade information of the diagnosis information.
Preferably, the grade information of the diagnostic information includes: primary diagnosis, other diagnosis, lesion diagnosis, and other diagnosis.
Preferably, the deep neural network model comprises a multi-layer long-short-term memory network or a multi-layer gating recursion unit network, and the deep neural network model and the sequence-to-sequence model are jointly trained according to the diagnosis information of a preset number of patients and the treatment scheme corresponding to the diagnosis information of the patients.
Preferably, the diagnostic information of the predetermined number of patients is diagnostic information of patients whose number of admissions is not less than a predetermined number.
Preferably, the step of inputting the predicted result of the current treatment plan into a sequence model for processing, and obtaining the predicted result of the future treatment plan of the patient to be processed includes:
inputting a predicted result of the current treatment scheme into a sequence model for processing to obtain a predicted result of the next treatment scheme of the patient to be processed;
acquiring whether the number of the predicted results of the treatment scheme of the patient to be treated reaches a preset number;
and if the number of the predicted results of the treatment scheme of the patient to be treated does not reach the preset number, taking the predicted result of the next treatment scheme as the predicted result of the current treatment scheme, and returning to execute the step of inputting the predicted result of the current treatment scheme into a sequence model for processing.
Preferably, the step of obtaining diagnostic information of the patient to be treated further comprises:
obtaining a coding vector corresponding to the diagnosis information of the patient according to international disease coding standards;
and inputting the coding vector corresponding to the diagnosis information of the patient into a deep neural network model for processing to obtain a prediction result of the current treatment scheme of the patient to be processed.
In order to achieve the above object, the present invention also provides a deep learning-based therapeutic scheme generating device, including:
the deep learning-based treatment plan generating device comprises a memory, a processor and a deep learning-based treatment plan generating program which is stored in the memory and can run on the processor, wherein the deep learning-based treatment plan generating program realizes the steps of the deep learning-based treatment plan generating method when being executed by the processor.
To achieve the above object, the present invention also provides a computer-readable storage medium having stored thereon a deep learning-based therapeutic regimen generation program which, when executed by a processor, implements the steps of the deep learning-based therapeutic regimen generation method described above.
The invention provides a method for generating a treatment scheme based on deep learning, a device for generating the treatment scheme based on deep learning and a computer readable storage medium, which are used for acquiring diagnosis information of a patient to be treated; inputting the diagnosis information into a deep neural network model for processing to obtain a prediction result of the current treatment scheme of the patient to be processed; and inputting the predicted result of the current treatment scheme into a sequence model for processing to obtain the predicted result of the future treatment scheme of the patient to be processed. In this way, the accuracy of the predicted patient treatment is improved by generating the patient treatment by the feedforward deep neural network with time linkage and predicting the future patient treatment by the sequence-to-sequence model.
Drawings
FIG. 1 is a schematic diagram of a hardware operating environment of an embodiment terminal according to an embodiment of the present invention;
FIG. 2 is a flow chart of a first embodiment of a method for generating a deep learning-based treatment plan according to the present invention;
fig. 3 is a flow chart of a second embodiment of a method for generating a deep learning-based treatment plan according to the present invention.
The achievement of the objects, functional features and advantages of the present invention will be further described with reference to the accompanying drawings, in conjunction with the embodiments.
Detailed Description
It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
The invention provides a method for generating a treatment scheme based on deep learning, which generates the treatment scheme of a patient through a feedforward type deep neural network with time connection, predicts the future treatment scheme of the patient through a sequence-to-sequence model, and improves the accuracy of the predicted treatment scheme of the patient.
As shown in fig. 1, fig. 1 is a schematic diagram of a hardware running environment of an embodiment terminal according to an embodiment of the present invention;
the terminal of the embodiment of the invention can be a deep learning-based treatment scheme generating device or a server.
As shown in fig. 1, the terminal may include: a processor 1001, such as a CPU, memory 1002, a communications bus 1003. The communication bus 1003 is used to implement connection communication between the constituent elements in the terminal. The memory 1002 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory. The memory 1002 may alternatively be a storage device separate from the processor 1001 described above.
It will be appreciated by those skilled in the art that the configuration of the terminal shown in fig. 1 is not limiting of the terminal of the embodiments of the present invention, and may include more or less components than those illustrated, or may combine certain components, or may be arranged in different components.
As shown in fig. 1, a generation program of a deep learning-based treatment plan may be included in a memory 1002 as a kind of computer storage medium.
In the terminal shown in fig. 1, the processor 1001 may be configured to call a generation program of a deep learning-based treatment plan stored in the memory 1002, and perform the following operations:
acquiring diagnostic information of a patient to be treated;
inputting the diagnosis information into a deep neural network model for processing to obtain a prediction result of the current treatment scheme of the patient to be processed;
and inputting the predicted result of the current treatment scheme into a sequence model for processing to obtain the predicted result of the future treatment scheme of the patient to be processed.
Further, the processor 1001 may call a generation program of the deep learning-based treatment plan stored in the memory 1002, and further perform the following operations:
inputting the diagnosis information into a deep neural network model for processing to obtain hidden vectors corresponding to the diagnosis information;
inputting hidden vectors corresponding to the diagnosis information into a self-attention mechanism layer for processing to obtain weights corresponding to the hidden vectors;
obtaining a weighted hidden vector according to the hidden vector and the weight;
and obtaining a prediction result of the current treatment scheme of the patient to be treated according to the weighted hidden vector.
Further, the processor 1001 may call a generation program of the deep learning-based treatment plan stored in the memory 1002, and further perform the following operations:
inputting hidden vectors corresponding to the diagnosis information into a self-attention mechanism layer for processing;
inputting hidden vectors corresponding to the diagnosis information into a self-attention mechanism layer for processing;
and the self-attention mechanism layer learns the weight corresponding to the hidden vector according to the grade information of the diagnosis information.
Further, the processor 1001 may call a generation program of the deep learning-based treatment plan stored in the memory 1002, and further perform the following operations:
the grade information of the diagnosis information includes: primary diagnosis, other diagnosis, lesion diagnosis, and other diagnosis.
Further, the processor 1001 may call a generation program of the deep learning-based treatment plan stored in the memory 1002, and further perform the following operations:
the deep neural network model comprises a plurality of layers of long-short-term memory networks or a plurality of layers of gating recursion unit networks, and the deep neural network model and the sequence-to-sequence model are jointly trained according to the diagnosis information of a preset number of patients and the treatment scheme corresponding to the diagnosis information of the patients.
Further, the processor 1001 may call a generation program of the deep learning-based treatment plan stored in the memory 1002, and further perform the following operations:
the diagnosis information of the patient with the preset number is the diagnosis information of the patient with the admission frequency not less than the preset number.
Further, the processor 1001 may call a generation program of the deep learning-based treatment plan stored in the memory 1002, and further perform the following operations:
inputting a predicted result of the current treatment scheme into a sequence model for processing to obtain a predicted result of the next treatment scheme of the patient to be processed;
acquiring whether the number of the predicted results of the treatment scheme of the patient to be treated reaches a preset number;
and if the number of the predicted results of the treatment scheme of the patient to be treated does not reach the preset number, taking the predicted result of the next treatment scheme as the predicted result of the current treatment scheme, and returning to execute the step of inputting the predicted result of the current treatment scheme into a sequence model for processing.
Further, the processor 1001 may call a generation program of the deep learning-based treatment plan stored in the memory 1002, and further perform the following operations:
obtaining a coding vector corresponding to the diagnosis information of the patient according to international disease coding standards;
and inputting the coding vector corresponding to the diagnosis information of the patient into a deep neural network model for processing to obtain a prediction result of the current treatment scheme of the patient to be processed.
Referring to fig. 2, in an embodiment, the method for generating a deep learning-based treatment plan includes:
step S10, obtaining diagnosis information of a patient to be treated.
The diagnosis information of the patient to be processed is derived from an electronic medical record of the patient, and the electronic medical record can be an outpatient medical record or an admission information record. It should be understood that in one piece of diagnostic information, only one piece of diagnostic data corresponding to the disease is recorded, and the diagnostic data may include the name of the disease and the disease related symptoms to be diagnosed, such as fever, sore throat, tinnitus, runny nose, canker sore, arrhythmia, chest distress, dizziness, and the like.
And step S20, inputting the diagnosis information into a deep neural network model for processing, and obtaining a prediction result of the current treatment scheme of the patient to be processed.
Firstly, word vectors corresponding to diagnostic information of a patient to be processed can be obtained according to word vectors trained in advance. Specifically, a pre-constructed continuous bag of words model (CBOW) may be utilized to extract word vectors corresponding to diagnostic information of the patient to be processed. The training input of the CBOW model is a word vector corresponding to a word that is contextually relevant to a particular feature word, and the output is the word vector for that feature word. For example, in the diagnostic information, there is a sub-data of "sore throat", "runny nose", "dizziness", and "cold" as a code vector that can be extracted by the CBOW model.
Specifically, when a CBOW model is constructed, for important model parameter initialization, the learning rate is set to 0.015, the iteration number is set to 5, the minimum word frequency is set to 5, the window size is set to 6, the feature vector dimension is set to 64, the batch size is set to 500, and the downsampling threshold is set to 1e-3.
Furthermore, because different word expressions exist for the same diagnosis in different diagnosis information, the diagnosis information can be converted into diagnosis codes according to the international disease classification standard (ICD-9) first, and then the diagnosis codes are converted into word vectors by using a one-hot coding (one-hot) algorithm or a CBOW model. ICD-9 is a system that classifies a disease according to rules based on certain characteristics of the disease, and is represented by a coded method.
Thus, by acquiring the diagnosis information or the word vector of the diagnosis code through the pre-constructed CBOW continuous word bag model, the original diagnosis record in the diagnosis information can be corresponding to the specific disease name and the disease related symptoms, and the corresponding disease name and the disease related symptoms can be converted into the numerical form acceptable by the deep neural network model, namely the word vector.
The deep neural network model in this embodiment comprises a multi-layered LSTM Long short term memory network (Long/short term memory) or comprises a multi-layered GRU gating recursion unit network (Gated recurrent units), preferably the deep neural network model comprises a multi-layered GRU gating recursion unit network. It should be noted that, both the LSTM network and the GRU network are feedforward neural networks with time links, and the order in which the input vectors are input to the LSTM network or the GRU network will affect the training result of the neural networks. For example, the neural network model may output different results than when "cookie" is input before "milk" is input.
Therefore, the sequence of word vectors in the word vector sequence generated according to one piece of diagnosis information is consistent with the sequence of word segmentation corresponding to the diagnosis information, the word vector sequence can be regarded as having certain time sequence data, the word vectors in the word vector sequence are sequentially input into an LSTM network or a GRU network for processing, the LSTM network or the GRU network sequentially outputs hidden vectors corresponding to the input word vectors one by one, and a full-connection layer of the final deep neural network model obtains a prediction result of a treatment scheme corresponding to the piece of diagnosis information according to the hidden vectors corresponding to the word vectors.
It should be noted that, the prediction result of the treatment scheme corresponding to the diagnostic information of the patient to be treated is a multi-label probability vector, and the dimension of the multi-label probability vector is the number of preset treatment schemes, where the treatment schemes include drug treatment, injection, transfusion, surgery or other treatment means. One label corresponds to one treatment scheme, and one label probability is a probability value of the treatment scheme obtained by the current visit of the patient to be treated, and the probability value can reflect the reliability or the acceptance of the treatment scheme.
And step S30, inputting the predicted result of the current treatment scheme into a sequence model for processing, and obtaining the predicted result of the future treatment scheme of the patient to be processed.
The prediction result of the current treatment plan obtained in step S20 may be input into a sequence-to-sequence model, so as to obtain the prediction result of the future treatment plan of the patient to be treated. The Sequence-to-Sequence model is an End-to-End (End-to-End) Sequence-to-Sequence model, the Sequence-to-Sequence model is generally implemented through an encoding-decoding (Encoder-Decoder) framework, the Encoder and the Decoder can be any text, voice, image and video data, and the Sequence-to-Sequence model can be constructed by adopting a CNN convolutional neural network, an RNN cyclic neural network, an LSTM network, a GRU network and other deep learning networks, and preferably, the Sequence-to-Sequence model is constructed by adopting the RNN cyclic neural network.
The obtained prediction result of the current treatment scheme of the patient to be treated, namely, the multi-label probability vector is input into an RNN network as a group of treatment sequences, the sequence information of the multi-label probability vector is used for extracting, so as to obtain a prediction coding vector with a fixed length, the RNN is called an encoder, the prediction coding vector is input into another RNN, the decoding is carried out to obtain a decoding vector, the RNN is generally called a decoder, thus, the direct mapping from one group of treatment sequences to the next group of treatment sequences is directly obtained, and a treatment sequence can be obtained through decoding the decoder, wherein the treatment sequence is the prediction result of the next treatment scheme of the patient and is also a probability vector.
Further, the predicted result of the next treatment plan of the patient can be input into the sequence model for processing, so as to obtain the predicted result of the next treatment plan of the patient to be processed.
It will be appreciated that the preset number of predicted results of the treatment plan of the patient to be treated may also be preset, and the number of predicted results of the treatment plan of the patient to be treated is calculated each time the sequence-to-sequence model obtains a predicted result of a new treatment plan, and if the number of predicted results of the treatment plan of the patient to be treated does not reach the preset number, the predicted result of the currently obtained treatment plan is input again to the sequence-to-sequence model to obtain the new treatment plan. For example, it is possible that the current treatment regimen and the next treatment regimen correspond to the same drug, but the next treatment regimen may be administered in a smaller amount than the current treatment regimen.
In this embodiment, the direct mapping from a set of treatment sequences (i.e., a treatment plan) to the next set of treatment sequences obtained by the sequence-to-sequence model is an innovative solution, and the implementation of the solution does not need to rely on external features and additional training processes, so that a treatment path of a patient in the whole treatment cycle can be obtained, cumbersome treatment on the similarity measurement of the treatment path of the patient can be avoided, and progressive treatment plan reference suggestions are provided for the patient, thereby achieving the purpose of better auxiliary treatment.
In this embodiment, the deep neural network model and the sequence-to-sequence model are jointly trained according to a preset number of diagnostic information and a treatment scheme corresponding to the diagnostic information.
Specifically, firstly determining the number of treatment schemes to be predicted from the sequence to the sequence model, and determining that each patient diagnosis information in the training data sample at least needs to contain several diagnosis records according to the number of treatment schemes to be predicted, for example, the number of treatment schemes to be predicted is three, wherein each patient diagnosis information in the training data sample at least needs to contain three diagnosis records; during training, first diagnosis information of a patient to be treated is input into a deep neural network model for treatment, a prediction result of a corresponding first treatment scheme is obtained, and error calculation and parameter updating in the deep neural network model are performed according to the prediction result of the first treatment scheme and a real first treatment scheme; continuously inputting the predicted result of the predicted first treatment scheme into a sequence-to-sequence model for processing to obtain the predicted result of the second treatment scheme, and performing error calculation and parameter updating in the sequence-to-sequence model according to the predicted result of the second treatment scheme and the actual second treatment scheme; judging whether the number of the predicted results of the treatment scheme reaches the preset number, if not, returning and inputting the predicted results of the second treatment scheme to the sequence-to-sequence model, and repeating the steps of prediction, error calculation and sequence-to-sequence model parameter updating.
In the pre-built neural network model, a multi-layer LSTM network or a GRU network is included, specifically, the number of network layers is set to be initialized to 20, and the number of hidden units of each layer is initialized to 64. And performing multiple iterations from the obtained training data to the sequence model through the neural network model and the sequence, so that training generation from the neural network model and the sequence to the sequence model can be realized.
It should be noted that, in order to improve the efficiency and accuracy of model training, when the treatment scheme corresponding to the diagnostic information in the training data is preprocessed, only the first 1500 most common drugs and treatment means are screened out, and each drug and treatment means is represented by a code to be used as a basic database for generating the treatment scheme in the neural network model.
In one embodiment, diagnostic information of a patient to be treated is obtained; inputting the diagnosis information into a deep neural network model for processing to obtain a prediction result of the current treatment scheme of the patient to be processed; and inputting the predicted result of the current treatment scheme into a sequence model for processing to obtain the predicted result of the future treatment scheme of the patient to be processed. In this way, the current treatment scheme of the patient is generated through the feedforward type deep neural network with time connection, the future treatment scheme of the patient is predicted through the sequence-to-sequence model, and progressive treatment scheme reference advice is provided for the patient, so that the aim of better auxiliary treatment is achieved.
In a second embodiment, as shown in fig. 3, on the basis of the embodiment shown in fig. 2, the step of inputting the diagnostic information into a deep neural network model for processing to obtain a predicted result of the current treatment plan of the patient to be processed includes:
and S40, inputting the diagnosis information into a deep neural network model for processing to obtain hidden vectors corresponding to the diagnosis information.
Step S41, the hidden vector corresponding to the diagnosis information is input into a self-attention mechanism layer for processing, and the weight corresponding to the hidden vector is obtained.
And step S42, obtaining the weighted hidden vector according to the hidden vector and the weight.
And step S43, obtaining a prediction result of the current treatment scheme of the patient to be treated according to the weighted hidden vector.
In this embodiment, a word vector corresponding to diagnostic information is used as an input vector of the deep neural network model through a pre-built deep neural network model or a training-completed deep neural network model, and based on the input vector, the deep neural network model correspondingly outputs an output vector corresponding to the input vector as a hidden vector corresponding to the diagnostic information, where the hidden vector is a feature representation corresponding to the diagnostic information.
Because the patient's symptoms in the diagnosis information have the light and heavy degree of different illness states, therefore, can divide the diagnosis information into four symptoms grades of first order, second order, third order and fourth order according to the symptoms in the diagnosis information, namely grade information, wherein, the fourth order of symptoms corresponds to the symptoms of main diagnosis, the third order of symptoms corresponds to the symptoms of other diagnoses, the second order of symptoms is the symptoms of damage class, the first order of symptoms is other common slight symptoms, the importance degree reduces in proper order.
Specifically, before a treatment plan is generated according to hidden vectors, a Self-attention mechanism (Self-Attention Mechanism) is introduced as a weight layer, the hidden vectors are input into the weight layer, the Self-attention mechanism performs autonomous learning according to the symptom level classified by each hidden vector, a weight corresponding to the symptom level is generated, and the weight is given to the corresponding hidden vector. Of course, when constructing the deep neural network model, it is also possible to construct the deep neural network model including the weight layer.
It should be noted that the self-attention mechanism is different from the conventional attention mechanism in that the conventional attention mechanism is essentially an alignment operation, that is, a sentence that needs to be introduced into the attention mechanism is aligned with external information, and the self-attention mechanism does not need to introduce external information to update parameters. The self-attention mechanism has a great promotion effect on a sequence learning task, and can effectively improve the system performance of the sequence on a sequence model by carrying out data weighted transformation on a source data sequence.
Specifically, the hidden vector sequence H with the length n output by the deep neural network model can be expressed as:
H＝(h 1 ,h 2 ,…,h n )
the weight formula of the self-attention mechanism is:
a＝softmax(W s2 tanh(W s1 H T ))
wherein a is a weight sequence corresponding to a hidden vector sequence H T For the array inverted and transposed hidden vector sequence H, W s1 And W is s2 Are model parameters in the self-attention mechanism model, and the optimization is continuously updated in training iteration.
Through the self-attention mechanism, the hidden vector sequence M to which the weight value is given can be expressed as:
M＝aH
the hidden vector with the weight value is processed by the full connection layer to obtain a treatment scheme prediction result corresponding to the diagnosis information of the patient to be treated, and when the prediction result is input into the sequence-to-sequence model to generate the treatment scheme of the patient, the hidden vector with the larger weight value naturally can attract more importance of the sequence-to-sequence model, so that the accuracy of the future treatment scheme of the patient to be treated generated by the sequence-to-sequence model can be improved.
In one embodiment, the diagnostic information is input into a deep neural network model for processing, so as to obtain hidden vectors corresponding to the diagnostic information; inputting hidden vectors corresponding to the diagnosis information into a self-attention mechanism layer for processing to obtain weights corresponding to the hidden vectors; obtaining a weighted hidden vector according to the hidden vector and the weight; and obtaining a prediction result of the current treatment scheme of the patient to be treated according to the weighted hidden vector. In this way, the treatment scheme is generated according to the degree of the patient symptoms by giving the hidden vectors weight values corresponding to the symptom levels according to the symptom levels of the diagnosis information corresponding to the hidden vectors in the process of generating the treatment scheme under the autonomous learning of the self-attention mechanism, so that the accuracy of the generated treatment scheme is improved.
In addition, the invention also provides a deep learning-based treatment scheme generating device, which comprises a memory, a processor and a deep learning-based treatment scheme generating program stored on the memory and capable of running on the processor, wherein the processor realizes the steps of the deep learning-based treatment scheme generating method when executing the deep learning-based treatment scheme generating program.
Furthermore, the present invention proposes a computer-readable storage medium, characterized in that the computer-readable storage medium comprises a deep-learning-based treatment plan generation program, which when executed by a processor implements the steps of the deep-learning-based treatment plan generation method as described in the above embodiments.
The foregoing embodiment numbers of the present invention are merely for the purpose of description, and do not represent the advantages or disadvantages of the embodiments.
From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a storage medium (e.g. ROM/RAM, magnetic disk, optical disk) as described above, comprising instructions for causing a terminal device (which may be a television, a mobile phone, a computer, a deep learning based treatment plan generating device, an air conditioner, or a network device, etc.) to perform the method according to the embodiments of the present invention.
The foregoing description is only of the preferred embodiments of the present invention, and is not intended to limit the scope of the invention, but rather is intended to cover any equivalents of the structures or equivalent processes disclosed herein or in the alternative, which may be employed directly or indirectly in other related arts.
Claims (6)
1. A method for generating a deep learning-based treatment plan, the method comprising the steps of:
obtaining diagnostic information of a patient to be processed, wherein in the diagnostic information, only one diagnostic record corresponds to diagnostic data, and the diagnostic data comprises a diagnostic disease name and a disease related sign;
inputting the diagnosis information into a deep neural network model for processing to obtain hidden vectors corresponding to the diagnosis information, wherein the hidden vectors are characteristic representations corresponding to the diagnosis information;
inputting the hidden vector corresponding to the diagnosis information into a self-attention mechanism layer for processing, and learning the weight corresponding to the hidden vector by the self-attention mechanism layer according to the grade information of the diagnosis information, wherein the grade information of the diagnosis information comprises: a fourth symptom, a third symptom, a second symptom and a first symptom, wherein the fourth symptom corresponds to a main diagnosis symptom, the third symptom corresponds to other diagnosis symptoms, the second symptom is a damage symptom, and the first symptom is other common mild symptoms;
obtaining a weighted hidden vector according to the hidden vector and the weight;
obtaining a prediction result of the current treatment scheme of the patient to be treated according to the weighted hidden vector, wherein the prediction result is a multi-label probability vector, the dimension of the multi-label probability vector is the number of preset treatment schemes, the treatment schemes comprise medication, injection, transfusion and operation, one label corresponds to one treatment scheme, one label probability is a probability value of the treatment scheme obtained by the current visit of the patient to be treated, and the probability value reflects the reliability or the acceptance of the treatment scheme;
inputting the predicted result of the current treatment scheme into a sequence model for processing to obtain the predicted result of the future treatment scheme of the patient to be processed; wherein,,
the step of inputting the predicted result of the current treatment scheme into a sequence model for processing to obtain the predicted result of the future treatment scheme of the patient to be processed comprises the following steps:
inputting a predicted result of the current treatment scheme into a sequence model for processing to obtain a predicted result of the next treatment scheme of the patient to be processed;
acquiring whether the number of the predicted results of the treatment scheme of the patient to be treated reaches a preset number;
and if the number of the predicted results of the treatment scheme of the patient to be treated does not reach the preset number, taking the predicted result of the next treatment scheme as the predicted result of the current treatment scheme, and returning to execute the step of inputting the predicted result of the current treatment scheme into a sequence model for processing.
2. The method of claim 1, wherein the deep neural network model comprises a multi-layered long-short-term memory network or a multi-layered gated recursive unit network, and the deep neural network model and the sequence-to-sequence model are jointly trained according to a predetermined number of patient diagnostic information and a treatment plan corresponding to the patient diagnostic information.
3. The method of generating a deep learning based treatment plan of claim 2, wherein the diagnostic information of the predetermined number of patients is diagnostic information of patients having a number of admissions not less than a predetermined number.
4. A method of generating a deep learning based treatment plan as claimed in any one of claims 1 to 3, wherein the step of obtaining diagnostic information of the patient to be treated further comprises, after:
obtaining a coding vector corresponding to the diagnosis information of the patient according to international disease coding standards;
and inputting the coding vector corresponding to the diagnosis information of the patient into a deep neural network model for processing to obtain a prediction result of the current treatment scheme of the patient to be processed.
5. A deep learning-based treatment plan generation apparatus, characterized in that it includes a memory, a processor, and a deep learning-based treatment plan generation program stored on the memory and executable on the processor, which when executed by the processor, implements the steps of the deep learning-based treatment plan generation method according to any one of claims 1 to 4.
6. A computer-readable storage medium, characterized in that the computer-readable storage medium has stored thereon a generation program of a deep-learning-based treatment plan, which when executed by a processor, implements the steps of the deep-learning-based treatment plan generation method according to any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811407145.8A CN109637669B (en) | 2018-11-22 | 2018-11-22 | Deep learning-based treatment scheme generation method, device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811407145.8A CN109637669B (en) | 2018-11-22 | 2018-11-22 | Deep learning-based treatment scheme generation method, device and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109637669A CN109637669A (en) | 2019-04-16 |
| CN109637669B true CN109637669B (en) | 2023-07-18 |
Family
ID=66068934
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811407145.8A Active CN109637669B (en) | 2018-11-22 | 2018-11-22 | Deep learning-based treatment scheme generation method, device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109637669B (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110176311A (en) * | 2019-05-17 | 2019-08-27 | 北京印刷学院 | A kind of automatic medical proposal recommending method and system based on confrontation neural network |
| CN110110059B (en) * | 2019-05-20 | 2021-06-29 | 挂号网（杭州）科技有限公司 | Medical dialogue system intention identification and classification method based on deep learning |
| CN110880362B (en) * | 2019-11-12 | 2022-10-11 | 南京航空航天大学 | Large-scale medical data knowledge mining and treatment scheme recommending system |
| CN111192693B (en) * | 2019-12-19 | 2021-07-27 | 山东大学 | Method and system for correcting diagnosis codes based on medicine combination |
| CN111341437B (en) * | 2020-02-21 | 2022-02-11 | 山东大学齐鲁医院 | Digestive tract disease judgment auxiliary system based on tongue image |
| CN111815487B (en) * | 2020-06-28 | 2024-02-27 | 珠海中科先进技术研究院有限公司 | Deep learning-based health education assessment method, device and medium |
| CN111701150B (en) * | 2020-07-02 | 2022-06-17 | 中国科学院苏州生物医学工程技术研究所 | Intelligent optical diagnosis and treatment equipment |
| TWI825467B (en) * | 2021-08-23 | 2023-12-11 | 緯創資通股份有限公司 | Data analysis system and data analysis method |
| CN115115620B (en) * | 2022-08-23 | 2022-12-13 | 安徽中医药大学 | Pneumonia lesion simulation method and system based on deep learning |
| CN116013503B (en) * | 2022-12-27 | 2024-02-20 | 北京大学长沙计算与数字经济研究院 | Dental treatment plan determining method, electronic equipment and storage medium |
| CN116798630B (en) * | 2023-07-05 | 2024-03-08 | 广州视景医疗软件有限公司 | Myopia physiotherapy compliance prediction method, device and medium based on machine learning |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106778014A (en) * | 2016-12-29 | 2017-05-31 | 浙江大学 | A kind of risk Forecasting Methodology based on Recognition with Recurrent Neural Network |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2005509218A (en) * | 2001-11-02 | 2005-04-07 | シーメンス メディカル ソリューションズ ユーエスエー インコーポレイテッド | Patient data mining to maintain quality |
| AU2011207344B2 (en) * | 2010-01-21 | 2015-05-21 | Asthma Signals, Inc. | Early warning method and system for chronic disease management |
| DE102014209627B3 (en) * | 2014-05-21 | 2015-10-29 | Cytocentrics Bioscience Gmbh | In-vitro diagnostics for model-based therapy planning |
| US10046177B2 (en) * | 2014-06-18 | 2018-08-14 | Elekta Ab | System and method for automatic treatment planning |
| US10299735B2 (en) * | 2014-08-22 | 2019-05-28 | Pulse Tectonics Llc | Automated diagnosis based at least in part on pulse waveforms |
| CN107145746A (en) * | 2017-05-09 | 2017-09-08 | 北京大数医达科技有限公司 | The intelligent analysis method and system of a kind of state of an illness description |
| CN108717866B (en) * | 2018-04-03 | 2022-10-11 | 中国医学科学院肿瘤医院 | Method, device, equipment and storage medium for predicting radiotherapy plan dose distribution |
| CN108766563A (en) * | 2018-05-25 | 2018-11-06 | 戴建荣 | Radiotherapy prediction of result method and system based on dosage group |
-
2018
- 2018-11-22 CN CN201811407145.8A patent/CN109637669B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106778014A (en) * | 2016-12-29 | 2017-05-31 | 浙江大学 | A kind of risk Forecasting Methodology based on Recognition with Recurrent Neural Network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109637669A (en) | 2019-04-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109637669B (en) | Deep learning-based treatment scheme generation method, device and storage medium | |
| US11810671B2 (en) | System and method for providing health information | |
| Shorten et al. | Deep Learning applications for COVID-19 | |
| Shorfuzzaman et al. | Artificial Neural Network‐Based Deep Learning Model for COVID‐19 Patient Detection Using X‐Ray Chest Images | |
| Lee et al. | Machine learning in relation to emergency medicine clinical and operational scenarios: an overview | |
| CN113241135A (en) | Disease risk prediction method and system based on multi-mode fusion | |
| Elzeki et al. | COVID-19: a new deep learning computer-aided model for classification | |
| CN110069779B (en) | Symptom entity identification method of medical text and related device | |
| CN111709233A (en) | Intelligent diagnosis guiding method and system based on multi-attention convolutional neural network | |
| US20200380339A1 (en) | Integrated neural networks for determining protocol configurations | |
| CN112420191A (en) | Traditional Chinese medicine auxiliary decision making system and method | |
| KR102429319B1 (en) | Method and system for predicting patients needs for hospital resources | |
| CN115497616A (en) | Method, system, equipment and storage medium for aid decision making of infectious diseases | |
| CN114628001B (en) | Prescription recommendation method, system, equipment and storage medium based on neural network | |
| Yuan et al. | Efficient symptom inquiring and diagnosis via adaptive alignment of reinforcement learning and classification | |
| Hartsock et al. | Vision-language models for medical report generation and visual question answering: A review | |
| CN112035627B (en) | Automatic question and answer method, device, equipment and storage medium | |
| CN115631852B (en) | Certificate type recommendation method and device, electronic equipment and nonvolatile storage medium | |
| CN116092668A (en) | Prediction method for heart failure patient readmission fused with multi-element heterogeneous data | |
| CN116994695A (en) | Training method, device, equipment and storage medium of report generation model | |
| CN116543853A (en) | Drug interaction prediction method, computer device, and medium | |
| CN114627993A (en) | Information prediction method, information prediction device, storage medium and computer equipment | |
| CN115240811A (en) | Construction method and application of implicit relation drug recommendation model based on graph neural network | |
| CN113658688A (en) | Clinical decision support method based on word segmentation-free deep learning | |
| CN113569999A (en) | Training method and device for medicine recommendation model, storage medium and computer equipment |
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 |