CN112216379A - Disease diagnosis system based on intelligent joint learning - Google Patents
Disease diagnosis system based on intelligent joint learning Download PDFInfo
- Publication number
- CN112216379A CN112216379A CN201910630056.8A CN201910630056A CN112216379A CN 112216379 A CN112216379 A CN 112216379A CN 201910630056 A CN201910630056 A CN 201910630056A CN 112216379 A CN112216379 A CN 112216379A
- Authority
- CN
- China
- Prior art keywords
- medical
- layer
- model
- data
- feature
- 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.)
- Pending
Links
- 238000003745 diagnosis Methods 0.000 title claims abstract description 28
- 201000010099 disease Diseases 0.000 title claims abstract description 25
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 title claims abstract description 25
- 238000002474 experimental method Methods 0.000 claims abstract description 22
- 238000013528 artificial neural network Methods 0.000 claims abstract description 21
- 238000000605 extraction Methods 0.000 claims abstract description 8
- 238000012549 training Methods 0.000 claims description 44
- 238000013527 convolutional neural network Methods 0.000 claims description 30
- 239000013598 vector Substances 0.000 claims description 22
- 230000006870 function Effects 0.000 claims description 10
- 239000000284 extract Substances 0.000 claims description 8
- 230000002457 bidirectional effect Effects 0.000 claims description 7
- 238000000034 method Methods 0.000 claims description 7
- 230000000306 recurrent effect Effects 0.000 claims description 6
- 238000007477 logistic regression Methods 0.000 claims description 4
- 230000007787 long-term memory Effects 0.000 claims description 4
- 238000003062 neural network model Methods 0.000 claims description 4
- 230000006403 short-term memory Effects 0.000 claims description 4
- 230000007246 mechanism Effects 0.000 claims description 3
- 238000013519 translation Methods 0.000 claims description 3
- 230000004913 activation Effects 0.000 claims description 2
- 238000005457 optimization Methods 0.000 claims description 2
- 230000007170 pathology Effects 0.000 claims 1
- 238000013135 deep learning Methods 0.000 abstract description 16
- 238000013136 deep learning model Methods 0.000 abstract description 4
- 230000002708 enhancing effect Effects 0.000 abstract description 3
- 238000006243 chemical reaction Methods 0.000 description 5
- 238000013473 artificial intelligence Methods 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000013461 design Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000002068 genetic effect Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000001575 pathological effect Effects 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000007634 remodeling Methods 0.000 description 1
Images
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
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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 invention relates to a disease diagnosis system based on intelligent joint learning. The system comprises: the system comprises a context feature embedded representation layer, a medical image embedded representation layer, a medical physical examination experiment data embedded representation layer, a medical knowledge coding layer, a neural network hiding layer and a joint learning output layer. The system adopts a combined learning model of a traditional linear model and a deep learning model for effective disease diagnosis, can realize automatic extraction of high-level information of medical images, electronic medical records and physical examination experiment results by utilizing deep learning, avoids complex characteristic engineering, and can also integrate the knowledge and doctor experience of medical diagnostics by utilizing rules. The invention can comprehensively cover various modal information of patients, and effectively combine the field knowledge of medical diagnostics while keeping the strong generalization ability of the deep learning algorithm, thereby enhancing the interpretability and the accuracy of the whole disease diagnosis algorithm model.
Description
Technical Field
The invention belongs to the field of artificial intelligence, and relates to a disease diagnosis system based on intelligent joint learning, which is realized by effectively fusing medical knowledge and a machine learning algorithm by utilizing the joint learning of a traditional linear model and a deep learning model.
Background
Machine learning is the fundamental way for computers to have intelligence, and is the core of artificial intelligence. The essence of machine learning is the use of algorithms to mine from large amounts of data the rules underlying them and to use them for prediction or classification. The combination of emerging artificial intelligence technology and traditional medical treatment is remodeling diagnosis and treatment modes and industrial modes in the medical health field. With the increasing accumulation of medical data and the continuous breakthrough of artificial intelligence technology, machine learning algorithms represented by deep learning are increasingly applied in the medical field in recent years, and the applications mainly focus on disease screening and prediction based on the deep learning.
The existing deep learning-based disease screening and prediction applications can be mainly divided into the following three major categories. (1) Image analysis class based on deep learning. (2) And analyzing the class based on the deep learning sign data. (3) And E-medical record analysis class based on deep learning.
These methods suffer from the following disadvantages: the adopted deep learning algorithm only depends on data of a single source mode, and the training result of the model has one-sidedness. Due to the design limitation of the deep learning algorithm on the neural network structure, medical diagnosis knowledge accumulated and verified through long-term medical practice cannot be used in the neural network. The deep learning algorithm lacking medical knowledge guidance is completely limited by training data of the algorithm, and the result of algorithm analysis is not interpretable and cannot be effectively popularized.
A technical solution is needed to solve the above problems, which can process heterogeneous data of multiple modalities and enrich the expression dimension of patient information. Meanwhile, the domain knowledge of medical diagnostics can be effectively combined, the strong generalization capability of the deep learning algorithm is kept, and the interpretability and the accuracy of the whole disease diagnosis algorithm model are enhanced.
Disclosure of Invention
Aiming at the problems, the invention aims to design and realize a disease diagnosis system based on intelligent combined learning, adopts a combined learning model of a traditional linear model and a deep learning model for effective disease diagnosis, can realize automatic extraction of high-level information of medical images, electronic medical records and physical examination experiment results by utilizing deep learning, avoids complex feature engineering, and can also integrate the knowledge and doctor experience of medical diagnostics by utilizing rules. The combined learning model can comprehensively cover various modal information of a patient, and can effectively combine the field knowledge of medical diagnostics while keeping the strong generalization capability of a deep learning algorithm, thereby enhancing the interpretability and the accuracy of the whole disease diagnosis algorithm model.
In order to achieve the purpose, the invention adopts the following technical scheme:
a disease diagnostic system based on intelligent joint learning, comprising (one module for each of the following layers):
the context characteristic embedding layer is used for automatically extracting context characteristics in the electronic medical record of the text type and coding semantic information of the context;
the medical image embedding layer is used for automatically extracting the characteristics of the medical image of the patient;
the medical physical examination experiment data embedding presentation layer is used for automatically extracting the characteristics of result information generated by a patient physical examination experiment;
the medical knowledge coding layer is connected with the context feature embedding representation layer, the medical image embedding representation layer and the medical physical examination experiment data embedding representation layer and is used for coding the multi-mode feature data of the patient in a linear feature vector form according to medical diagnosis knowledge by adopting a rule-based extraction method;
the neural network hiding layer is connected with the context characteristic embedding representation layer, the medical image embedding representation layer and the medical physical examination experiment data embedding representation layer and is used for training the embedded representation of the text type electronic medical record data, the medical image data and the physical examination experiment data according to the type difference of the input patient multi-model data;
and the joint learning output layer is used for realizing the joint learning of the training result of the neural network hidden layer and the linear model of the medical knowledge coding layer and outputting the final disease diagnosis result.
Further, the context feature embedding representation layer adopts a bidirectional long and short term memory network (BLSTM) to automatically extract the context features in the electronic medical record of the text class. In addition, the context feature embedding representation layer can also adopt a unidirectional LSTM and a recurrent neural network RNN to replace a bidirectional LSTM, and the same implementation effect can be obtained; to enhance the local learning ability of the neural network, an LSTM model with attention mechanism may also be used.
Further, the medical image embedding representation layer automatically extracts the features of the medical image of the patient by using a Convolutional Neural Network (CNN).
Further, the medical physical examination experiment data embedding representation layer adopts a Convolutional Neural Network (CNN) to automatically extract the characteristics of the result information generated by the patient physical examination experiment.
Further, the encoding in the form of linear feature vectors performed by the medical knowledge encoding layer is one-hot encoding.
Further, the neural network hidden layer comprises a BLSTM network and a CNN network which are respectively used for training the embedded representation of the electronic medical record data, the medical image data and the physical examination experiment data of the text type.
Compared with the prior art, the invention has the following positive effects:
the invention designs and realizes a combined learning model of a traditional linear model and a deep learning model, which is used for effective disease diagnosis. The method utilizes the advantage of deep learning for automatically extracting the features, avoids manual feature engineering which has high time cost and high labor cost and is difficult to expand to a large data set, and realizes real data driving; meanwhile, medical professional knowledge and experience are fully exerted. The combined learning model can comprehensively cover various modal information of a patient, and can effectively combine the field knowledge of medical diagnostics while keeping the strong generalization capability of a deep learning algorithm, thereby enhancing the interpretability and the accuracy of the whole disease diagnosis algorithm model. The invention has the advantages of low overhead, high expression and multiple applications.
Drawings
Fig. 1 is a block diagram of a disease diagnosis system based on intelligent joint learning.
Fig. 2 is a workflow diagram of an intelligent joint learning based disease diagnosis system.
Detailed Description
The present invention will be described in detail below with reference to specific embodiments and accompanying drawings.
Fig. 1 is a block diagram of a disease diagnosis system based on intelligent joint learning according to an embodiment of the present invention. The functions and implementations of the modules shown in fig. 1 are described as follows:
(1) the context feature embedding presentation layer facing the bidirectional long and short term memory network (BLSTM) automatically extracts the context features in the electronic medical record of text type and encodes the semantic information of the context.
The electronic medical record is used as a text representation form, and input coding from text to a neural network training model is realized by embedding a representation layer. The invention adopts the prior Skip-Gram model to obtain word vectors based on a large amount of unmarked Chinese corpus training as the embedded expression of the electronic medical record. Meanwhile, labels need to be prepared for the electronic medical record in order to train various parameters of the algorithm. The label preparation has two means, and for the electronic medical record with the diagnosis conclusion, the diagnosis conclusion is directly extracted from the text to be used as the label of the current electronic medical record. For an electronic medical record containing only patient complaints, a professional doctor needs to be asked in advance to fill out the label. The training of the general case algorithm selects the former electronic medical record data.
(2) Embedding the medical image into a presentation layer facing a Convolutional Neural Network (CNN), and automatically extracting the characteristics of the medical image of the patient.
Since the medical image data is two-dimensional image data, which includes information of focus, background, occlusion, illumination, etc., the data needs to be preprocessed. Firstly, an automatic extraction method in the radiology is adopted, and each medical image is labeled by a pathological label. The image also needs to be normalized for mean and standard deviation before being input to the convolutional neural network. Meanwhile, in order to meet the requirement of convolutional neural network training on a large amount of data, a training data set is enhanced through changes such as rotation, translation and deformation.
(3) The Convolution Neural Network (CNN) -oriented medical physical examination experiment data is embedded into a presentation layer, and the characteristics of result information generated by a patient physical examination experiment are automatically extracted.
The medical physical examination experiment data embedding presentation layer is mainly used for processing laboratory detection data of various numerical types and simultaneously comprises sequence data received by sensor equipment such as electrocardiogram monitoring equipment. The data is represented in the form of a matrix as input to a convolutional neural network.
(4) And the medical knowledge coding layer is used for performing one-hot coding in a linear characteristic vector form on the multi-modal data of the patient according to the medical diagnostics knowledge by adopting a rule-based extraction method.
According to medical diagnostics knowledge, multi-modal feature data of a patient is defined in the form of a set of linear feature vectors, a set of features X ═<x1,x2,……,xn>And the number of the features in the feature set is n. The feature set comprises both the original medical features and the cross-transformation features. And carrying out one-hot coding on the value of the original medical characteristic. For example, the patient is eligible for xiInformation on the item characteristics, then xiIs 1, otherwise is 0. In order to add description on nonlinear characteristics on the basis of the original linear model, cross conversion characteristics are added. The cross-over feature is defined as follows:
wherein c iskiIs a Boolean variable, xiAnd d represents the number of the features, if the ith original medical feature is one part of the kth cross-over feature, the value is 1, and if not, the value is 0. For example, the value of the cross-over feature for two features (genetic feature dominant and weight feature obese) is 1, and if and only if the constituent features (genetic feature dominant and weight feature obese) are both 1, otherwise it is 0.
(5) The neural network hiding layer comprises an LSTM network and a CNN network according to the type difference of input patient multi-model data, and is used for training the embedded representation of text electronic medical record data, medical image data and physical examination experiment data respectively.
The recurrent neural network preserves the state of the context by periodically connecting hidden layer nodes. The cyclic structure part of this patent adopts two-way long-and-short-term memory network (BLSTM), and BLSTM has powerful ability to long-range dependence modeling, and it has inherited the structural feature of most recurrent neural network model to the gradient disappearance problem that probably appears in the recurrent neural network has been avoided. BLSTM includes both forward and backward LSTM structures. Wherein the forward LSTM and backward LSTM are used to capture past and future different feature information, respectively. The BLSTM is more suitable for processing longer texts, can extract more characteristic information, and is very suitable for analyzing long text data such as electronic medical records.
And performing algorithm training on the medical image training data set which is subjected to class labeling and preprocessing by adopting a convolutional neural network. Because the convolutional neural network needs a huge amount of training data, the medical image training data set often cannot meet the requirements of CNN parameter training. According to the method, a pre-training mode is adopted, firstly, large-scale image data of ImageNet is utilized to pre-train model parameters of the CNN, and the obtained parameters of the CNN are used as initial values of the model. And then, taking the medical image training data as the input of the CNN model, and adjusting and optimizing the parameters of the model.
(6) And the joint learning output layer realizes joint learning of the neural network hidden layer training result and the linear model of the medical knowledge coding layer and inputs a final disease diagnosis result.
A set of linear feature vectors obtained via the medical knowledge coding layer, by means of a linear model y ═ wTX + b constitutes the linear part of the joint learning model, where X is the feature vector, w is the weight vector, and b is the offset. And the neural network model constitutes the deep part of the joint learning model. The outputs of the linear part model and the depth part model are combined through a weighted sum form, and the joint learning training of the models is completed through a logistic regression mode. The logistic regression model is defined as follows:
wherein Y represents a label of the classification and x is texThe eigenvector, σ (·), represents the sigmod function, and θ (x) represents those cross-conversion features. w is a1Is a weight vector of the linear model, wdeepIs the final activation function valueThe vector of (2).
The training of the combined model adopts a back propagation method to simultaneously train a linear part and a depth part and adopts small-batch random optimization.
Fig. 2 is a flowchart of the work flow of the disease diagnosis system based on intelligent joint learning according to the embodiment of the present invention. The steps are specifically described as follows:
step 1.1 prepare the data.
Preparing marked data, segmenting the marked data into a training data set, developing the data set and a testing data set, wherein the training data set and the developing data set are used in a training stage, and the testing data set is used in a testing stage.
Step 1.2 batch sample input
Training of the training data set is input into the system in batches according to a small batch principle and the set size of the batch size.
Step 2.1 electronic medical record text participle
And segmenting each sentence according to the sentence unit by using the text in the electronic medical record.
Step 2.2 word vector embedding representation
For each character of each input sample, word embedding (English is a word, Chinese is a character level) is carried out, and the word is converted into a 300-dimensional vector according to a character index table and a linear layer.
Step 2.3 bidirectional LSTM network
The input sample sequence represented by the word vector is sent to a bidirectional LSTM network to extract deep feature information of the context.
Step 3.1 medical knowledge feature extraction
And constructing a one-dimensional original feature set by using medical diagnostics knowledge, and extracting one-hot codes of original features from multi-modal data of the patient.
Step 3.2 Cross-conversion feature Generation
And calculating the code of the cross conversion characteristics from the original characteristic code according to the definition of the cross conversion characteristics.
Step 4.1 image data normalization
And normalizing the mean value and the standard deviation of the image.
Step 4.2 dataset enhancement
The training data set is enhanced by changes in rotation, translation, and deformation.
Step 4.3CNN model pretraining
And pre-training model parameters of the CNN by using the large-scale image data of ImageNet, and taking the obtained parameters of the CNN as initial values of the model.
Step 5.1 numerical data serialization
And expressing the numerical data in a matrix form as an input of the convolutional neural network.
Step 5.2CNN model
And inputting the numerical data matrix into the CNN network, and training the hidden layer characteristics of the network.
Step 6 of connection
And splicing the calculation results of the linear model and the depth model to establish a target function.
Step 7 calculating a cost function
In the training process, the objective function is to maximize the log-likelihood of the correct label sequence of the training set.
The cost function is the negative of the objective function.
Step 8 back propagation algorithm
And training the model by using a back propagation gradient descent algorithm, and adaptively adjusting the learning rate according to the training speed. If the effect of the model on the test set is reduced, the overfitting is indicated, the training is stopped immediately, and otherwise, the training is continued.
The context feature embedding expression layer can also adopt a unidirectional LSTM and a recurrent neural network RNN to replace a bidirectional LSTM, and the same realization effect can be obtained. To enhance the local learning ability of the neural network, an LSTM model with attention mechanism may also be used.
Another embodiment of the present invention provides a computer/server, which includes the above disease diagnosis system based on intelligent joint learning, that is, the above disease diagnosis system based on intelligent joint learning is deployed as a software product on the computer/server.
The above embodiments are only intended to illustrate the technical solution of the present invention and not to limit the same, and a person skilled in the art can modify the technical solution of the present invention or substitute the same without departing from the principle and scope of the present invention, and the scope of the present invention should be determined by the claims.
Claims (10)
1. An intelligent joint learning system for disease diagnosis, comprising:
the context characteristic embedding layer is used for automatically extracting context characteristics in the electronic medical record of the text type and coding semantic information of the context;
the medical image embedding layer is used for automatically extracting the characteristics of the medical image of the patient;
the medical physical examination experiment data embedding presentation layer is used for automatically extracting the characteristics of result information generated by a patient physical examination experiment;
the medical knowledge coding layer is connected with the context feature embedding representation layer, the medical image embedding representation layer and the medical physical examination experiment data embedding representation layer and is used for coding the multi-mode feature data of the patient in a linear feature vector form according to medical diagnosis knowledge by adopting a rule-based extraction method;
the neural network hiding layer is connected with the context characteristic embedding representation layer, the medical image embedding representation layer and the medical physical examination experiment data embedding representation layer and is used for training the embedding representation of the electronic medical record data, the medical image data and the physical examination experiment data of the text type according to the type difference of the input multi-model data of the patient;
and the joint learning output layer is used for performing joint learning on the training result of the neural network hidden layer and the linear model of the medical knowledge coding layer and outputting a final disease diagnosis result.
2. The system of claim 1, wherein the contextual feature embedding representation layer automatically extracts contextual features in the electronic medical record of the textual class using one of the following neural network models: a bidirectional long and short term memory network BLSTM, a unidirectional long and short term memory network LSTM, a recurrent neural network RNN, and an LSTM with attention mechanism.
3. The system of claim 1, wherein the medical image embedded representation layer automatically extracts features of the patient medical image using a convolutional neural network.
4. The system of claim 3, wherein the medical image embedded representation layer firstly labels each medical image with a pathology label using an automatic extraction method in radiology; before the image is input into the convolutional neural network, the image is normalized by mean value and standard deviation, and the training data set is enhanced by rotation, translation, deformation and other changes, so as to meet the requirement of convolutional neural network training on a large amount of data.
5. The system of claim 1, wherein the medical physical examination experiment data embedding representation layer automatically extracts features of result information generated by the patient physical examination experiment using a convolutional neural network.
6. The system according to claim 1, wherein the encoding in the form of linear feature vectors by the medical knowledge encoding layer is a one-hot encoding.
7. The system according to claim 6, wherein the medical knowledge encoding layer defines the multi-modal feature data of the patient as a set of linear feature vectors according to the medical diagnostics knowledge, the feature set including both the original medical features and the cross-transformation features; carrying out one-hot coding on the value of the original medical characteristic; the cross-over feature is defined as follows:
wherein c iskiIs a Boolean variable, xiAnd d represents the number of the features, if the ith original medical feature is one part of the kth cross-over feature, the value is 1, and if not, the value is 0.
8. The system of claim 1, wherein the neural network hidden layer comprises a BLSTM network and a CNN network for training embedded representations of electronic medical record data, medical image data, and physical examination data of text type, respectively; the neural network hidden layer adopts a pre-training mode, firstly pre-trains model parameters of the CNN by using the existing large-scale image data, the obtained CNN parameters are used as initial values of the model, then medical image training data are used as input of the CNN model, and the parameters of the model are adjusted and optimized.
9. The system of claim 1, wherein the joint learning model of the joint learning output layer comprises a linear portion and a depth portion; a set of linear feature vectors obtained via a medical knowledge coding layer, by a linear model y ═ wTX + b constitutes the linear part, where X is a feature vector, w is a weight vector, and b is an offset; a neural network model constitutes the depth component; the outputs of the linear part and the depth part are combined in a weighted sum mode, and the joint learning training of the model is completed through a logistic regression model; the training adopts a back propagation method to train a linear part and a depth part simultaneously and adopts a small batchAnd (4) random optimization.
10. The system of claim 9, wherein the logistic regression model is defined as follows:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910630056.8A CN112216379A (en) | 2019-07-12 | 2019-07-12 | Disease diagnosis system based on intelligent joint learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910630056.8A CN112216379A (en) | 2019-07-12 | 2019-07-12 | Disease diagnosis system based on intelligent joint learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112216379A true CN112216379A (en) | 2021-01-12 |
Family
ID=74048563
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910630056.8A Pending CN112216379A (en) | 2019-07-12 | 2019-07-12 | Disease diagnosis system based on intelligent joint learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112216379A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113658683A (en) * | 2021-08-05 | 2021-11-16 | 重庆金山医疗技术研究院有限公司 | Disease diagnosis system and data recommendation method |
CN113743414A (en) * | 2021-08-02 | 2021-12-03 | 清华大学 | Method, device and system for identifying focus based on semantic coding |
CN116682551A (en) * | 2023-07-27 | 2023-09-01 | 腾讯科技(深圳)有限公司 | Disease prediction method, disease prediction model training method and device |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107247881A (en) * | 2017-06-20 | 2017-10-13 | 北京大数医达科技有限公司 | A kind of multi-modal intelligent analysis method and system |
US20180158552A1 (en) * | 2016-12-01 | 2018-06-07 | University Of Southern California | Interpretable deep learning framework for mining and predictive modeling of health care data |
CN108830853A (en) * | 2018-07-20 | 2018-11-16 | 东北大学 | A kind of melanoma aided diagnosis method based on artificial intelligence |
WO2018212710A1 (en) * | 2017-05-19 | 2018-11-22 | National University Of Singapore | Predictive analysis methods and systems |
CN109063712A (en) * | 2018-06-22 | 2018-12-21 | 哈尔滨工业大学 | A kind of multi-model Hepatic diffused lesion intelligent diagnosing method and system based on ultrasound image |
CN109528196A (en) * | 2018-11-14 | 2019-03-29 | 北京工业大学 | A kind of hepatic vein pressure gradient Noninvasive assessmet method based on multi-modal image and Heuristics |
CN109920540A (en) * | 2019-03-14 | 2019-06-21 | 宁波中云创科信息技术有限公司 | Construction method, device and the computer equipment of assisting in diagnosis and treatment decision system |
-
2019
- 2019-07-12 CN CN201910630056.8A patent/CN112216379A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180158552A1 (en) * | 2016-12-01 | 2018-06-07 | University Of Southern California | Interpretable deep learning framework for mining and predictive modeling of health care data |
WO2018212710A1 (en) * | 2017-05-19 | 2018-11-22 | National University Of Singapore | Predictive analysis methods and systems |
CN107247881A (en) * | 2017-06-20 | 2017-10-13 | 北京大数医达科技有限公司 | A kind of multi-modal intelligent analysis method and system |
CN109063712A (en) * | 2018-06-22 | 2018-12-21 | 哈尔滨工业大学 | A kind of multi-model Hepatic diffused lesion intelligent diagnosing method and system based on ultrasound image |
CN108830853A (en) * | 2018-07-20 | 2018-11-16 | 东北大学 | A kind of melanoma aided diagnosis method based on artificial intelligence |
CN109528196A (en) * | 2018-11-14 | 2019-03-29 | 北京工业大学 | A kind of hepatic vein pressure gradient Noninvasive assessmet method based on multi-modal image and Heuristics |
CN109920540A (en) * | 2019-03-14 | 2019-06-21 | 宁波中云创科信息技术有限公司 | Construction method, device and the computer equipment of assisting in diagnosis and treatment decision system |
Non-Patent Citations (1)
Title |
---|
ABDULLAH AL IMRAN ET AL: ""Classification of Chronic Kidney Disease using Logistic Regression, Feedforward Neural Networkand Wide & Deep Learning"", 《INTERNATIONAL CONFERENCE ON INNOVATION IN ENGINEERING AND TECHNOLOGY (ICIET)》, pages 1 - 3 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113743414A (en) * | 2021-08-02 | 2021-12-03 | 清华大学 | Method, device and system for identifying focus based on semantic coding |
CN113743414B (en) * | 2021-08-02 | 2022-08-05 | 清华大学 | Method, device and system for identifying focus based on semantic coding |
CN113658683A (en) * | 2021-08-05 | 2021-11-16 | 重庆金山医疗技术研究院有限公司 | Disease diagnosis system and data recommendation method |
CN116682551A (en) * | 2023-07-27 | 2023-09-01 | 腾讯科技(深圳)有限公司 | Disease prediction method, disease prediction model training method and device |
CN116682551B (en) * | 2023-07-27 | 2023-12-22 | 腾讯科技(深圳)有限公司 | Disease prediction method, disease prediction model training method and device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107977361B (en) | Chinese clinical medical entity identification method based on deep semantic information representation | |
CN109545302B (en) | Semantic-based medical image report template generation method | |
US20210034813A1 (en) | Neural network model with evidence extraction | |
GB2571825A (en) | Semantic class localization digital environment | |
CN111126024B (en) | Statement generation | |
CN112216379A (en) | Disease diagnosis system based on intelligent joint learning | |
Halvardsson et al. | Interpretation of swedish sign language using convolutional neural networks and transfer learning | |
Xu et al. | Intelligent emotion detection method based on deep learning in medical and health data | |
CN113657105A (en) | Medical entity extraction method, device, equipment and medium based on vocabulary enhancement | |
Yang et al. | Automatic ultrasound image report generation with adaptive multimodal attention mechanism | |
Zhao et al. | A comparison review of transfer learning and self-supervised learning: Definitions, applications, advantages and limitations | |
CN117391092B (en) | Electronic medical record multi-mode medical semantic alignment method based on contrast learning | |
Rajesh et al. | Machine learning for online automatic prediction of common disease attributes using never-ending image learner | |
EP4094198A1 (en) | Classifying out-of-distribution data using a contrastive loss | |
Lin et al. | Contrastive pre-training and linear interaction attention-based transformer for universal medical reports generation | |
CN117012370A (en) | Multi-mode disease auxiliary reasoning system, method, terminal and storage medium | |
Zaghir et al. | Real-world patient trajectory prediction from clinical notes using artificial neural networks and UMLS-based extraction of concepts | |
CN116109980A (en) | Action recognition method based on video text matching | |
CN114492386A (en) | Combined detection method for drug name and adverse drug reaction in web text | |
Li et al. | Classification of COVID-19 images based on transfer learning and feature fusion | |
Nainwal et al. | A comprehending deep learning approach for disease classification | |
Ma et al. | A pathology image diagnosis network with visual interpretability and structured diagnostic report | |
CN112115719A (en) | Chinese medicine medical record named entity recognition method and system based on multi-head attention mechanism | |
Mao et al. | Facial expression recognition based on deep convolutional neural network | |
Štajduhar et al. | Analysing large repositories of medical images |
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 |