CN113539412A - Chinese herbal medicine recommendation system based on deep learning - Google Patents
Chinese herbal medicine recommendation system based on deep learning Download PDFInfo
- Publication number
- CN113539412A CN113539412A CN202110811844.4A CN202110811844A CN113539412A CN 113539412 A CN113539412 A CN 113539412A CN 202110811844 A CN202110811844 A CN 202110811844A CN 113539412 A CN113539412 A CN 113539412A
- Authority
- CN
- China
- Prior art keywords
- layer
- neural network
- convolutional
- chinese herbal
- symptom
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 241000411851 herbal medicine Species 0.000 title claims abstract description 63
- 238000013135 deep learning Methods 0.000 title claims abstract description 19
- 208000024891 symptom Diseases 0.000 claims abstract description 59
- 239000003814 drug Substances 0.000 claims abstract description 58
- 238000003745 diagnosis Methods 0.000 claims abstract description 32
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 30
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 29
- 238000000034 method Methods 0.000 claims abstract description 26
- 208000037386 Typhoid Diseases 0.000 claims abstract description 19
- 201000008297 typhoid fever Diseases 0.000 claims abstract description 19
- 208000011580 syndromic disease Diseases 0.000 claims description 28
- 230000006870 function Effects 0.000 claims description 17
- 235000008216 herbs Nutrition 0.000 claims description 17
- 210000002569 neuron Anatomy 0.000 claims description 14
- 238000013528 artificial neural network Methods 0.000 claims description 13
- 230000006698 induction Effects 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 10
- 238000007781 pre-processing Methods 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 4
- 210000004205 output neuron Anatomy 0.000 claims description 4
- 238000009795 derivation Methods 0.000 claims description 3
- 210000002364 input neuron Anatomy 0.000 claims description 3
- 230000000644 propagated effect Effects 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 3
- 229940126680 traditional chinese medicines Drugs 0.000 claims description 3
- 238000011160 research Methods 0.000 abstract description 10
- 230000000694 effects Effects 0.000 abstract description 6
- 238000010801 machine learning Methods 0.000 abstract description 6
- 238000007477 logistic regression Methods 0.000 abstract description 5
- 238000012706 support-vector machine Methods 0.000 abstract description 5
- 238000013473 artificial intelligence Methods 0.000 abstract description 3
- 238000012797 qualification Methods 0.000 abstract description 2
- 238000011002 quantification Methods 0.000 abstract description 2
- 230000008569 process Effects 0.000 description 13
- 201000010099 disease Diseases 0.000 description 10
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 10
- 230000004069 differentiation Effects 0.000 description 9
- 238000011156 evaluation Methods 0.000 description 7
- 238000002474 experimental method Methods 0.000 description 6
- 238000012549 training Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 240000006688 Telosma cordata Species 0.000 description 2
- 235000017352 Telosma cordata Nutrition 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 208000017667 Chronic Disease Diseases 0.000 description 1
- 235000007516 Chrysanthemum Nutrition 0.000 description 1
- 244000189548 Chrysanthemum x morifolium Species 0.000 description 1
- 244000247747 Coptis groenlandica Species 0.000 description 1
- 235000002991 Coptis groenlandica Nutrition 0.000 description 1
- 241001445332 Coxiella <snail> Species 0.000 description 1
- 208000015220 Febrile disease Diseases 0.000 description 1
- 244000137773 Viola philippica Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000003759 clinical diagnosis Methods 0.000 description 1
- 238000010835 comparative analysis Methods 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 229940079593 drug Drugs 0.000 description 1
- 239000000796 flavoring agent Substances 0.000 description 1
- 238000009472 formulation Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000002329 infrared spectrum Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 101150117787 outL gene Proteins 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 241000894007 species Species 0.000 description 1
- 238000007619 statistical method 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
- G16H20/00—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
- G16H20/10—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to drugs or medications, e.g. for ensuring correct administration to patients
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- 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
- G16H20/00—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
- G16H20/90—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to alternative medicines, e.g. homeopathy or oriental medicines
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H70/00—ICT specially adapted for the handling or processing of medical references
- G16H70/40—ICT specially adapted for the handling or processing of medical references relating to drugs, e.g. their side effects or intended usage
-
- 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)
- General Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Medical Informatics (AREA)
- Data Mining & Analysis (AREA)
- Public Health (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Medicinal Chemistry (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Pharmacology & Pharmacy (AREA)
- Chemical & Material Sciences (AREA)
- Pathology (AREA)
- Databases & Information Systems (AREA)
- Alternative & Traditional Medicine (AREA)
- Toxicology (AREA)
- Medical Treatment And Welfare Office Work (AREA)
- Medicines Containing Plant Substances (AREA)
Abstract
The invention relates to a Chinese herbal medicine recommendation system based on deep learning. Firstly, an open data set of the typhoid treatise widely recognized in the field of traditional Chinese medicine is taken as a research object, and a standardized database is established through Multi-hot coding pretreatment of symptom and traditional Chinese medicine prescription information. Secondly, a convolutional neural network is used for simulating an internal matching rule between a clinical symptom group and Chinese herbal medicines, a clinical Chinese herbal medicine prescription recommendation algorithm is designed by utilizing a Pythrch frame and is respectively compared with three traditional machine learning models, namely a support vector machine, a Bayesian classifier, logistic regression and the like, and the effectiveness and the application value of the algorithm are described from two dimensions of quantification and qualification. The method can accurately recommend the Chinese herbal medicine prescription and has important promotion effect on the traditional Chinese medicine diagnosis and treatment research based on artificial intelligence.
Description
Technical Field
The invention belongs to the technical field of recommendation, and particularly relates to a Chinese herbal medicine recommendation system based on deep learning.
Background
Along with the leap-type development of Chinese economy, the influence of Chinese culture in the world is gradually increased, and Chinese medicine culture as the representative of Chinese culture is more and more emphasized by people[1]. The traditional Chinese medicine is a treasure of Chinese nationality, and has peculiar curative effect on the treatment of chronic diseases. The traditional Chinese medicine prescription is a group of herbal medicine sets which are prescribed for the disease treatment aiming at the symptoms of patients, the symptom sets in the traditional Chinese medicine prescription are called as 'symptom groups', and the herbal medicine sets are called as 'herbal medicine sets'. Literature reference[2]Note that prescription records are now over 10 thousand. In the field of traditional Chinese medicine, both clinical treatment and new prescriptions require the study of the intrinsic relationship between symptoms and herbs in the prescription[3]. This makes the study of the internal connection between the symptoms of traditional Chinese medicine and herbs a hot spot in the field of research in traditional Chinese medicine.
Most traditional Chinese medicine doctors and traditional Chinese medicine researchers have more researches and researches on traditional Chinese medicine diagnosis and treatment logics, but few researches on how to generate traditional Chinese medicine prescriptions aiming at the diseases of patients by applying the traditional Chinese medicine theory are carried out. One of the typical ideas in the logic of traditional Chinese medicine diagnosis and treatment is the concept of diagnosis and treatment based on syndrome differentiation[4]It is different from the diagnosis and treatment thinking of modern medicine, but carries out the process of thinking, reasoning, recognition and judgment on the input and output of the whole boundary effect of the medical object[5]. In recent years, Lispandong et al[6]Provides another diagnosis and treatment logic 'state identification', introduces state element information such as 'disease position', 'disease property', 'degree' and the like, combines a computer technology and a mathematical statistical method, and further simplifies the traditional Chinese medicine diagnosis and treatmentThe process. From the perspective of modern logical analysis, the unique diagnosis and treatment process of traditional Chinese medicine is to make the main contradiction judgment on the symptom group presented by the disease stage of the patient and recommend the corresponding herb medicine for treatment according to the symptoms[7]. Because the patient's symptoms do not correspond to the herbal medicine in the prescription in a one-to-one linear relationship, and different combinations of herbs may have the same efficacy[2]This presents a challenge to the recommendation of traditional Chinese medicine formulations.
In recent years, deep learning techniques have been widely used in the field of pharmaceutical research[8]. For example, Guo Yongkun[9]And the relation between the drug properties and the efficacies of the traditional Chinese medicine is explored by means of the fuzziness and the fitting of the neural network, so that the testing and the prediction of the efficacies of single formulas are realized. Aged chrysanthemum[10]And by means of the hierarchical feature extraction characteristics of the neural network, the traditional Chinese medicine syndrome differentiation artificial neural network aiming at dominant disease species is constructed. Tangyanfeng (decoction for treating rheumatism)[11]And (3) establishing a radial basis function neural network classification model for wild Chinese violet and cultivated Chinese violet, wherein the accuracy rate reaches 95.24%. Taste of plum[8]And a convolutional neural network model simulating nonlinear mapping between the medicine properties and the efficacies of the tonifying traditional Chinese medicine compound is constructed, and the accuracy rate reaches 92.50%. The machine learning algorithm is continuously tried to simulate the diagnosis and treatment process of doctors, and reference is provided for the doctors to develop Chinese herbal medicine prescriptions in clinic.
Disclosure of Invention
The invention aims to provide a Chinese herbal medicine recommendation system based on deep learning, which simulates diagnosis and treatment logics of syndrome differentiation and state identification of traditional Chinese medicine clinical diagnosis and treatment, realizes automatic recommendation of traditional Chinese medicine formulas from traditional Chinese medicine diseases to Chinese herbal medicines, solves the complex nonlinear problems of abstract diagnosis thinking, Chinese herbal medicine compatibility difference and the like unique to traditional Chinese medicine, can accurately recommend the Chinese herbal medicine formulas, and has an important promotion effect on traditional Chinese medicine diagnosis and treatment research based on artificial intelligence.
In order to achieve the purpose, the technical scheme of the invention is as follows: a Chinese herbal medicine recommendation system based on deep learning comprises:
the data acquisition module is used for acquiring the symptom information of the typhoid fever, the corresponding state elements, the syndrome types and the traditional Chinese medicine prescriptions and constructing a data set of diagnosis and treatment knowledge of the typhoid fever;
the data preprocessing module is used for performing Multi-hot coding preprocessing on symptom information and corresponding Chinese medicine prescription information based on a data set of diagnosis and treatment knowledge of 'typhoid treatise' and establishing a database;
the deep learning Chinese herbal medicine recommending module is used for simulating an internal matching rule between a clinical symptom group and Chinese herbal medicines by using a convolutional neural network and designing a clinical Chinese herbal medicine prescription recommending algorithm by using a Pythrch frame so as to realize the function of recommending Chinese herbal medicines according to symptoms.
In an embodiment of the present invention, in the database, a one-dimensional vector including 158 elements is used to represent a symptom group in a piece of diagnosis and treatment data, one element corresponds to a specific symptom, if one symptom appears in a piece of diagnosis and treatment data, the element corresponding to the symptom is set to "1", and if the symptom does not appear in the diagnosis and treatment data, the element corresponding to the symptom is set to "0"; correspondingly, a one-dimensional vector containing 77 elements is used for representing the traditional Chinese medicine in one piece of diagnosis and treatment data.
In an embodiment of the present invention, the deep learning chinese herbal medicine recommendation module specifically realizes the following functions:
(1) forward propagation
The input to the convolutional neural network is defined as X ═ X (X)1,x2,…,xn) And n is 158, representing 158 symptoms of input; taking 158-bit coded values of input symptoms as input of a convolutional layer, and converting the convolutional layer into a 24 × 77 × 1848 matrix by performing convolutional operation through 24 1 × 6 convolutional kernels, wherein weights of the convolutional kernels are shared; will QLDefining the neuron value output by the convolutional layer, W is weight, b is a bias unit, m is the dimension of the convolutional kernel, and the operation of the convolutional layer is defined as:
Relu(x)＝max(0,x)
where σ is a nonlinear activation function, the output of the convolutional layer corresponds to 48 state elementsAfter the over-rolling layer induces the state elements, the induced state elements are integrated by the first full-connection layer, the operation between the second full-connection layer corresponds to the induction classification of the syndrome type, and the operation of the third full-connection layer corresponds to the induction classification of the herbal medicine; is provided with YL+1For the output neuron value, k is the input neuron dimension, then the L +1 th fully-connected layer outputs the neuron value ZL+1Is defined as:
the probability of correctly outputting herbs is defined as:
wherein, f adopts Sigmoid function for normalizing the output prediction probability, which is specifically defined as:
(2) loss calculation
Because a group of traditional Chinese medicines is required to be obtained finally, namely a plurality of labels are required to be output by the convolutional neural network model, the method belongs to the category of multi-label classification; therefore, after the forward propagation of step (1), the loss value L is calculated using the multi-label cross-entropy loss function:
wherein H represents the number of neurons in the output layer, i.e. the number of Chinese medicinal herbs, th＝(thE {0,1}) and yh(0≤yhLess than or equal to 1) respectively representing the actual label and the predicted value of the convolutional neural network model;
(3) counter-propagating
By deltaL(x, y) denotes the L-th layer coordinate of (x, y), i.e. the partial derivative of the error function to the output value before activation of the L layer of the neural network, and the delta can be obtained according to the chain derivation rule assuming that the delta error of the L +1 layer is knownL(x, y) calculation formula:
wherein the content of the first and second substances,
from the delta error for the L-th layer, the derivative of that layer to the weight and bias can be found:
for each parameter in the network, the following formula is used for updating:
wherein eta is the learning rate of the neural network; the following equation can be obtained for the back propagation of the fully connected layer:
δL＝(WL+1)TδL+1⊙σ'(ZL)
the weight updating formula of the full-connection layer is the same as that of the convolution layer; current parameter W of network for forward propagationLAnd bLInput data is predicted, and loss calculation is carried outL carries out an evaluation on the current prediction result and calculates the actual output value thAnd the predicted output value y of the convolutional neural network modelhFinally, the loss value is propagated reversely so as to update the network parameters and continuously improve the performance of predicting the correct herbal medicine by the network.
In one embodiment of the invention, the convolutional neural network is a convolutional neural network consisting of a convolutional layer and three fully-connected layers, the symptoms and the Chinese herbal medicine prescriptions of the sample are respectively used as the input and the output of the network, and the state elements and the syndrome types are used as the fully-connected layers; the induction of syndrome type corresponding to the second full junction layer sets the number of neurons as syndrome type 416, and the prediction of 77 possible Chinese herbal medicines corresponding to the third full junction layer sets the number of neurons as 77.
Compared with the prior art, the invention has the following beneficial effects: the existing mainstream Chinese herbal medicine recommendation methods basically use a theme model and an expansion method thereof for recommendation, the methods focus on the co-occurrence relationship from using symptoms to Chinese herbal medicines, neglect the relationship between symptoms and only use the diagnosis and treatment concept of 'treatment by syndrome differentiation'. The method applies the convolutional neural network to a Chinese herbal medicine recommending scene, excavates the relation between the traditional Chinese medicine symptoms and the Chinese herbal medicines, realizes the quick recommendation of the traditional Chinese medicine symptoms to the Chinese herbal medicines by utilizing the good nonlinear fitting capacity of the convolutional neural network, and realizes the Chinese herbal medicine prediction accuracy of 71.42%, the recall rate of 86.87% and the F1 fraction of 79.39%. Experimental results show that the performance of the algorithm is superior to that of traditional machine learning models such as a support vector machine, a Bayesian classifier and logistic regression, and the algorithm can assist Chinese medical development of Chinese herbal medicine prescriptions.
Drawings
FIG. 1 is a diagram of a convolutional neural network model of the present invention.
Detailed Description
The technical scheme of the invention is specifically explained below with reference to the accompanying drawings.
The invention relates to a Chinese herbal medicine recommendation system based on deep learning, which comprises the following components:
the data acquisition module is used for acquiring the symptom information of the typhoid fever, the corresponding state elements, the syndrome types and the traditional Chinese medicine prescriptions and constructing a data set of diagnosis and treatment knowledge of the typhoid fever;
the data preprocessing module is used for performing Multi-hot coding preprocessing on symptom information and corresponding Chinese medicine prescription information based on a data set of diagnosis and treatment knowledge of 'typhoid treatise' and establishing a database;
the deep learning Chinese herbal medicine recommending module is used for simulating an internal matching rule between a clinical symptom group and Chinese herbal medicines by using a convolutional neural network and designing a clinical Chinese herbal medicine prescription recommending algorithm by using a Pythrch frame so as to realize the function of recommending Chinese herbal medicines according to symptoms.
The following is a specific embodiment of the present invention.
The invention firstly takes a public data set of the typhoid treatise widely accepted in the field of traditional Chinese medicine as a research object, and establishes a standardized database through the Multi-hot coding pretreatment of symptom and traditional Chinese medicine prescription information. Secondly, a convolutional neural network is used for simulating an internal matching rule between a clinical symptom group and Chinese herbal medicines, a clinical Chinese herbal medicine prescription recommendation algorithm is designed by utilizing a Pythrch frame and is respectively compared with three traditional machine learning models, namely a support vector machine, a Bayesian classifier, logistic regression and the like, and the effectiveness and the application value of the algorithm are described from two dimensions of quantification and qualification. The algorithm can accurately recommend the Chinese herbal medicine prescription and has an important promoting effect on the traditional Chinese medicine diagnosis and treatment research based on artificial intelligence.
1. Data source
In the present invention, the differentiation of exogenous febrile disease syndrome is based on[12]"Shanghai Lun Fang Xin Jie[13]"New treatise on Cold-induced diseases" on thesaurus[14]And "differentiation and treatment of symptoms of miscellaneous diseases due to typhoid fever[15]The related content of the medical knowledge comprises a plurality of symptom information of the typhoid fever, corresponding state elements, syndrome types and traditional Chinese medicine prescriptions, the information is summarized and sorted, finally 358 diagnosis and treatment records are collected, and a data set of diagnosis and treatment knowledge of the typhoid fever is constructed, wherein the data set comprises 1407 pieces of traditional Chinese medicine symptom information, 1823 pieces of state elements, 416 pieces of syndrome types and 1846 pieces of traditional Chinese medicine information, and after the repetition is removed, the data set comprises 158 pieces of symptom information, 48 pieces of state elements and syndrome types184 and 77 Chinese herbal medicines, and the data set part entries are shown in Table 1.
Table 1 dataset part entry
The collected typhoid data set, symptoms and herbal information have characteristics of short and multiple dimensions, and the constructed network classification performance is improved for the convenience of processing by a deep learning method, and the Multi-hot coding technology is adopted for preprocessing. Specifically, a one-dimensional vector including 158 elements is used to represent a symptom group in a piece of clinical data, one element corresponding to a specific symptom, and if a certain symptom appears in the clinical data, its corresponding position element is set to "1", and if a certain symptom does not appear in the piece of clinical data, its corresponding position element is set to "0". Accordingly, a one-dimensional vector containing 77 elements is used to represent a Chinese medicine group in a piece of diagnosis and treatment data, and the coded data is shown in table 2.
TABLE 2 coded data representation
2. Algorithm design
Compared with the traditional deep learning model, the convolutional neural network reduces complexity through three methods of receptive field, weight sharing and down sampling. The method generally comprises an input layer, a convolutional layer, a pooling layer, a full-link layer and an output layer, and a training process can be divided into three steps of forward propagation, loss calculation and backward propagation.
2.1, forward propagation
The input to the network may be defined as X ═ X (X)1,x2,…,xn) The 158-bit coded specific value of the input syndrome is used as the input of the convolution layer, and is converted into a matrix of 24 × 77 × 1848 by performing convolution operation through 24 convolution kernels (each convolution kernel weight is shared). Will QLDefined as a convolutionThe neuron value of layer output, W is weight, b is bias unit, m is the dimension of convolution kernel, and the operation definition of convolution layer is:
Relu(x)＝max(0,x)
wherein sigma is a nonlinear activation function, the output of the convolutional layer corresponds to 48 state elements, after the state elements are summarized by the convolutional layer, the summarized state elements are integrated by a first fully-connected layer, the operation between the second fully-connected layers corresponds to the induction classification of syndrome types, and the operation of the third fully-connected layer corresponds to the induction classification of herbal medicines; is provided with YL+1For the output neuron value, k is the input neuron dimension, then the L +1 th fully-connected layer outputs the neuron value ZL+1Is defined as:
the probability of correctly outputting herbs is defined as:
wherein, f adopts Sigmoid function for normalizing the output prediction probability, which is specifically defined as:
2.2 loss calculation
Because a group of traditional Chinese medicines is required to be obtained finally, namely a plurality of labels are required to be output by the convolutional neural network model, the method belongs to the category of multi-label classification; therefore, after forward propagation, the loss value L is calculated using the multi-label cross-entropy loss function:
wherein H represents the number of neurons in the output layer, i.e. the number of Chinese medicinal herbs, th＝(thE {0,1}) and yh(0≤yhLess than or equal to 1) respectively representing the actual label and the predicted value of the convolutional neural network model;
2.3, counter-propagating
In the counter-propagating process, delta is used for the convolution layerL(x, y) represents delta error at the L-th layer coordinate of (x, y), namely represents partial derivative of an error function to the output value of the neural network before the L-th layer activation, and delta can be obtained according to the chain derivation rule on the assumption that the delta error of the L +1 layer is knownL(x, y) calculation formula:
wherein the content of the first and second substances,
from the delta error for the L-th layer, the derivative of that layer to the weight and bias can be found:
for each parameter in the network, the following formula is used for updating:
wherein eta is the learning rate of the neural network; the following equation can be obtained for the back propagation of the fully connected layer:
δL＝(WL+1)TδL+1⊙σ'(ZL)
the weight updating formula of the full-connection layer is the same as that of the convolution layer; current parameter W of network for forward propagationLAnd bLInput data is predicted, a loss calculation L evaluates the current prediction result, and an actual output value t is calculatedhAnd the predicted output value y of the convolutional neural network modelhFinally, the loss value is propagated reversely so as to update the network parameters and continuously improve the performance of predicting the correct herbal medicine by the network.
3. Network architecture
According to the traditional Chinese medicine diagnosis and treatment logic and the convolutional neural network theory, the algorithm of the invention respectively uses the symptoms of the sample and the Chinese Herbal medicine prescription as the input and output of the network, uses the state elements and the syndrome types as the full connection layers, and designs a convolutional neural network Chinese Herbal medicine prediction model which is composed of a convolutional layer and three full connection layers, which is called CNN-HP (CNN Herbal descriptors) for short, and the structure of the model is shown in figure 1.
For neural network model training, the setting of the hyper-parameters is very important. In a convolutional network, the parameters are weights in the convolutional kernel and offsets. Similar to the fully-connected feedforward network, the convolutional network can also perform parameter learning through an error back-propagation algorithm. The algorithm of the invention firstly uses 24 convolution kernels with the size of 5 and the step length of 2 to carry out feature extraction on input data to obtain a feature matrix of 24 multiplied by 77 multiplied by 1848, and the first full-connection layer 1823 output neuron normalized data forms a judgment process of state elements, thereby accelerating network convergence and preventing gradient disappearance. The second fully connected layer corresponds to the induction of syndrome, setting the neuron number as syndrome number 416. The third full junction layer corresponds to 77 possible Chinese herbal predictions, so the number of neurons is set to 77.
The propagation process between the neural network layers corresponds to the Chinese medicine prescription, the state elements are judged according to symptoms, then the corresponding syndrome type is deduced, and finally the reasoning process of the herbal medicine is developed, and the network structure simulates the mapping relation. Specifically, the extraction process of the convolutional layer simulating the symptoms from the input layer features; the induction process from symptom groups to state elements is simulated from the convolution layer to the first full-connection layer; the second full-link layer transmission process simulates the induction process from state elements to syndrome types, and the final output layer simulates the process of prescription of Chinese herbal medicines. Therefore, the structural design of the convolutional neural network model better simulates the logical thinking process of 'syndrome differentiation and treatment' and 'state identification' in the traditional Chinese medicine.
4. Experiments and analyses
All experiments of the invention are based on Python3.7 environment, and are realized by programming by using a Pytroch 1.6 deep learning library and a Sklearn machine learning library. When experiments are carried out on a data set of 'typhoid treatise', different K values (the first K items with the highest probability in Chinese herbal medicine prediction) are set firstly, and all evaluation indexes are compared to obtain the optimal K value, and the optimal K value is compared with three traditional machine learning algorithms under the condition of the optimal K value: and (4) comparing the support vector machine, the naive Bayes classifier and the logistic regression, and quantitatively and qualitatively analyzing the Chinese herbal medicine recommendation effect of each algorithm.
4.1 recommendation results at different K values
The Chinese herbal medicine recommendation performance is evaluated by three evaluation indexes of accuracy (Precision), Recall (Recall) and F1 score (F1-score), and the evaluation indexes are defined as follows:
wherein, Top (P)preK) representsK-flavor herbs with the highest probability in the prediction results are gathered, and the K values are set to be 3, 4, 5, 6 and 10 respectively during experiments; plabelA label representing the collection of herbs in the actual prescription, i.e., the data. The results of the experiment are shown in table 3. As can be seen from the experimental data, the accuracy is highest when 3 herbs are recommended, but the recall rate is too low, and more herbs are not recommended; although the recall rate is as high as 96.17% when 10 herbs are recommended to be output, the accuracy rate and the F1 score are very low. In the trade-off, K is preferably 5, and in the subsequent experiments, K is 5.
TABLE 3 qualitative evaluation of the algorithm recommendation results of the present invention at different K values
4.2 quantitative analysis of data set
The data sets were randomly divided into training sets and test sets using different proportions, with the number of data samples and the recommended accuracy at different proportions as shown in table 4. Experimental results show that the algorithm obtains the best performance in the Group1 experimental Group with the most sufficient training data. This shows that the more sufficient the training set contains the symptoms and the Chinese medicine type information, the more sufficient the model is trained, and the better the recommendation performance is naturally.
TABLE 4 number of samples obtained by dividing data sets of different proportions and corresponding recommended accuracy rates thereof
4.3 qualitative evaluation and analysis
Table 5 shows the recommended examples of the chinese herbal medicines of the algorithm of the present invention when K is 5, wherein the underlined and bolded chinese herbal medicines are the correctly recommended chinese herbal medicines. In example 1, the algorithm of the present invention successfully recommended 4 herbs in the herb label, but more recommended one herb "coptis". In example 2, the algorithm of the present invention successfully recommended 3 herbs in the herb label, but more than 2 herbs "Guizhi" and "Fuling" were recommended. In the final example, the algorithm recommendation of the present invention is consistent with the herbal label. Therefore, the model of the invention can recommend Chinese herbal medicines more accurately, carry out intelligent composition and has good practical application value.
TABLE 5 example of Chinese medicine recommendation results for convolutional neural networks
4.4 comparative analysis with other algorithms
The experimental results of the algorithm (CNN-HP) and the support vector machine, the Bayesian classifier and the logistic regression on different experimental groups are shown in the table 6. Experimental data show that the accuracy of the algorithm is 71.42%, the recall rate is 86.87%, the F1 is 79.39%, and the three evaluation indexes are superior to those of the traditional algorithm, so that the algorithm has better recommendation performance.
TABLE 6 comparison of the results of the different methods on the different experimental groups
Reference documents:
[1] LIAO Juan, WANGJIANSUP, YUYE, WENZHAN, Dengxin, Linjiang, the application of artificial intelligence in the inheritance of famous and old Chinese medicine [ J ]. China traditional Chinese medicine journal, 2020,35(04): 1671-.
[2]L.Yao,Y.Zhang,B.Wei,W.Zhang and Z.Jin,"A Topic Modeling Approach for Traditional Chinese Medicine Prescriptions,"[C].in IEEE Transactions on Knowledge and Data Engineering,vol.30,no.6,pp.1007-1021,1June 2018,doi:10.1109/TKDE.2017.2787158.
[3] Zhao Li, Zhangjian, nan shu ling, etc. Chinese medicine specialty "prescriptions" course teaching design discussion [ J ] Shanxi college of traditional Chinese medicine, 2017(6): 134-.
[4] Zhao Wen, Linxuan Juan, Zhou Chang, Li Can east, the connotation and extension of the syndrome differentiation of TCM [ J/OL ] J.1-11 [2021-03-03]
[5] The general theory of syndrome and disease differentiation [ J ] Chinese medicine, 1990(2):3-9.
[6] Li Liandong, Ganjiu Juan, Lu Yuhui, etc. health status identification based on syndrome differentiation principle [ J ] Chinese medicine J2011 (04): 132-plus 135.
[7] Li Shu Dong, Yangxi plum, Ganjiu Juan, etc. discussion of the recognition model algorithm for health status [ J ] China journal of TCM, 2011,26(06):1351-1355.
[8] Plum flavor, Zhangxinyou, lean and Zhouyui, prediction research based on BP neural network Chinese medicine compound effect [ J ], Chinese medicine guide report (stage 16): 38-41.
[9] Guo Yongkun, Zhangxinyou, Liu Li Nu, etc. Chinese medicine formula efficacy prediction system based on neural network research [ J ]. Shizhen national medicine, 2019(2).
[10] Chenju, Yangzhining, Zhaozuding, Cao Yue, Ma Yi, Wenchuan, an artificial neural network algorithm model for traditional Chinese medicine syndrome differentiation [ C ]// fifth Chinese medicine information great meeting-big data standardization and wisdom Chinese medicine, 0.
[11] The infrared spectrum and radial basis neural network identification of Zihua Di Ding (Viola philippica, Hou Zhan Zhong, Wang Zhi Bao, etc. [ J ]. Hubei agricultural science 2014(01):132 plus 134).
[12] Coxiella. typhoid sorting [ J ]. shanghai science and technology press, 1996.
[13] Fan xinrong, typhoid argumentation element differentiation [ J ] chinese traditional medicine press, 2014.
[14] Fu Yan. Shang Han Lun research dictionary [ M ]. Jinan science and technology Press, 1994.
[15] Symptom identification and treatment of Wangfu, typhoid fever and miscellaneous diseases [ M ] Min health publishing agency, 2005.
The above are preferred embodiments of the present invention, and all changes made according to the technical scheme of the present invention that produce functional effects do not exceed the scope of the technical scheme of the present invention belong to the protection scope of the present invention.
Claims (4)
1. A Chinese herbal medicine recommendation system based on deep learning is characterized by comprising:
the data acquisition module is used for acquiring the symptom information of the typhoid fever, the corresponding state elements, the syndrome types and the traditional Chinese medicine prescriptions and constructing a data set of diagnosis and treatment knowledge of the typhoid fever;
the data preprocessing module is used for performing Multi-hot coding preprocessing on symptom information and corresponding Chinese medicine prescription information based on a data set of diagnosis and treatment knowledge of 'typhoid treatise' and establishing a database;
the deep learning Chinese herbal medicine recommending module is used for simulating an internal matching rule between a clinical symptom group and Chinese herbal medicines by using a convolutional neural network and designing a clinical Chinese herbal medicine prescription recommending algorithm by using a Pythrch frame so as to realize the function of recommending Chinese herbal medicines according to symptoms.
2. The deep learning-based herbal recommendation system of claim 1, wherein the database represents a symptom group in a diagnosis data by using a one-dimensional vector comprising 158 elements, one element corresponding to a specific symptom, and if one symptom appears in a diagnosis data, the element corresponding to the position is set to "1", and if the symptom does not appear in the diagnosis data, the element corresponding to the position is set to "0"; correspondingly, a one-dimensional vector containing 77 elements is used for representing the traditional Chinese medicine in one piece of diagnosis and treatment data.
3. The deep learning based herbal recommendation system of claim 2, wherein the deep learning herbal recommendation module is implemented as follows:
(1) forward propagation
The input to the convolutional neural network is defined as X ═ X (X)1,x2,…,xn) And n is 158, representing 158 symptoms of input; taking 158-bit coded values of input symptoms as input of a convolutional layer, and converting the convolutional layer into a 24 × 77 × 1848 matrix by performing convolutional operation through 24 1 × 6 convolutional kernels, wherein weights of the convolutional kernels are shared; will QLNeurons defined as convolutional layer outputThe value, W is the weight, b is the bias unit, m is the dimension of the convolution kernel, the operation of the convolution layer is defined as:
Relu(x)＝max(0,x)
wherein sigma is a nonlinear activation function, the output of the convolutional layer corresponds to 48 state elements, after the state elements are summarized by the convolutional layer, the summarized state elements are integrated by a first fully-connected layer, the operation between the second fully-connected layers corresponds to the induction classification of syndrome types, and the operation of the third fully-connected layer corresponds to the induction classification of herbal medicines; is provided with YL+1For the output neuron value, k is the input neuron dimension, then the L +1 th fully-connected layer outputs the neuron value ZL+1Is defined as:
the probability of correctly outputting herbs is defined as:
wherein, f adopts Sigmoid function for normalizing the output prediction probability, which is specifically defined as:
(2) loss calculation
Because a group of traditional Chinese medicines is required to be obtained finally, namely a plurality of labels are required to be output by the convolutional neural network model, the method belongs to the category of multi-label classification; therefore, after the forward propagation of step (1), the loss value L is calculated using the multi-label cross-entropy loss function:
wherein H represents the number of neurons in the output layer, i.e. the number of Chinese medicinal herbs, th＝(thE {0,1}) and yh(0≤yhLess than or equal to 1) respectively representing the actual label and the predicted value of the convolutional neural network model;
(3) counter-propagating
By deltaL(x, y) represents delta error at the L-th layer coordinate of (x, y), namely represents partial derivative of an error function to the output value of the neural network before the L-th layer activation, and delta can be obtained according to the chain derivation rule on the assumption that the delta error of the L +1 layer is knownL(x, y) calculation formula:
wherein the content of the first and second substances,
from the delta error for the L-th layer, the derivative of that layer to the weight and bias can be found:
for each parameter in the network, the following formula is used for updating:
wherein eta is the learning rate of the neural network; the following equation can be obtained for the back propagation of the fully connected layer:
δL＝(WL+1)TδL+1⊙σ'(ZL)
the weight updating formula of the full-connection layer is the same as that of the convolution layer; current parameter W of network for forward propagationLAnd bLInput data is predicted, a loss calculation L evaluates the current prediction result, and an actual output value t is calculatedhAnd the predicted output value y of the convolutional neural network modelhFinally, the loss value is propagated reversely so as to update the network parameters and continuously improve the performance of predicting the correct herbal medicine by the network.
4. The deep learning-based herbal recommendation system according to claim 3, wherein the convolutional neural network is a convolutional neural network consisting of a convolutional layer and three fully-connected layers, the symptoms and the herbal prescriptions of the sample are respectively used as the input and the output of the network, and the state elements and the syndrome types are used as the fully-connected layers; the induction of syndrome type corresponding to the second full junction layer sets the number of neurons as syndrome type 416, and the prediction of 77 possible Chinese herbal medicines corresponding to the third full junction layer sets the number of neurons as 77.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110811844.4A CN113539412B (en) | 2021-07-19 | 2021-07-19 | Deep learning-based Chinese herbal medicine recommendation system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110811844.4A CN113539412B (en) | 2021-07-19 | 2021-07-19 | Deep learning-based Chinese herbal medicine recommendation system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113539412A true CN113539412A (en) | 2021-10-22 |
| CN113539412B CN113539412B (en) | 2023-11-28 |
Family
ID=78128620
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110811844.4A Active CN113539412B (en) | 2021-07-19 | 2021-07-19 | Deep learning-based Chinese herbal medicine recommendation system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113539412B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114121212A (en) * | 2021-11-19 | 2022-03-01 | 东南大学 | Traditional Chinese medicine prescription generation method based on knowledge graph and group representation learning |
| CN115050481A (en) * | 2022-06-17 | 2022-09-13 | 湖南中医药大学 | Traditional Chinese medicine prescription efficacy prediction method based on graph convolution neural network |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110288041A (en) * | 2019-07-01 | 2019-09-27 | 齐鲁工业大学 | Chinese herbal medicine classification model construction method and system based on deep learning |
-
2021
- 2021-07-19 CN CN202110811844.4A patent/CN113539412B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110288041A (en) * | 2019-07-01 | 2019-09-27 | 齐鲁工业大学 | Chinese herbal medicine classification model construction method and system based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| 徐佳君 等: "基于人工智能算中医状态辨识规则", 中医杂志, pages 204 - 208 * |
| 赵文 等: "智能化中医辅助诊疗系统模型构建", 中华中医药杂志(原中国医药学报), pages 2421 - 2424 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114121212A (en) * | 2021-11-19 | 2022-03-01 | 东南大学 | Traditional Chinese medicine prescription generation method based on knowledge graph and group representation learning |
| CN114121212B (en) * | 2021-11-19 | 2024-04-02 | 东南大学 | Traditional Chinese medicine prescription generation method based on knowledge graph and group representation learning |
| CN115050481A (en) * | 2022-06-17 | 2022-09-13 | 湖南中医药大学 | Traditional Chinese medicine prescription efficacy prediction method based on graph convolution neural network |
| CN115050481B (en) * | 2022-06-17 | 2023-10-31 | 湖南中医药大学 | Traditional Chinese medicine prescription efficacy prediction method based on graph convolution neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113539412B (en) | 2023-11-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Kamadi et al. | A computational intelligence technique for the effective diagnosis of diabetic patients using principal component analysis (PCA) and modified fuzzy SLIQ decision tree approach | |
| WO2016192612A1 (en) | Method for analysing medical treatment data based on deep learning, and intelligent analyser thereof | |
| Liu et al. | Medical-vlbert: Medical visual language bert for covid-19 ct report generation with alternate learning | |
| CN108511056A (en) | Therapeutic scheme based on patients with cerebral apoplexy similarity analysis recommends method and system | |
| CN113539412A (en) | Chinese herbal medicine recommendation system based on deep learning | |
| CN106295187A (en) | Construction of knowledge base method and system towards intelligent clinical auxiliary decision-making support system | |
| CN112489769A (en) | Intelligent traditional Chinese medicine diagnosis and medicine recommendation system for chronic diseases based on deep neural network | |
| Guo et al. | A disease inference method based on symptom extraction and bidirectional Long Short Term Memory networks | |
| Huan et al. | Transfer learning with deep convolutional neural network for constitution classification with face image | |
| CN109273094A (en) | A kind of construction method and building system of the Kawasaki disease risk evaluation model based on Boosting algorithm | |
| Chowdary et al. | An effective approach for detecting diabetes using deep learning techniques based on convolutional LSTM networks | |
| Sivanesan et al. | A Review on diabetes mellitus diagnoses using classification on Pima Indian diabetes data set | |
| Huang et al. | A sorting fuzzy min-max model in an embedded system for atrial fibrillation detection | |
| Peng et al. | Artificial neural network application to the stroke prediction | |
| Zhang et al. | Transformer-and Generative Adversarial Network–Based Inpatient Traditional Chinese Medicine Prescription Recommendation: Development Study | |
| CN111986814A (en) | Modeling method of lupus nephritis prediction model of lupus erythematosus patient | |
| Gonçalves et al. | CNN optimization using surrogate evolutionary algorithm for breast cancer detection using infrared images | |
| CN110827990A (en) | Typhoid fever syndrome differentiation reasoning system based on knowledge graph | |
| CN116313141A (en) | Knowledge-graph-based intelligent inquiry method for unknown cause fever | |
| Shiying et al. | Research on syndrome classification prediction model of tibetan medicine diagnosis and treatment based on data mining | |
| CN111180045B (en) | Method for mining relation between drug pairs and efficacy from prescription information | |
| Xiao | Classification for Covid-19 Diseases Based on Ensembled Models | |
| Ruan et al. | Tpgen: Prescription generation using knowledge-guided translator | |
| CN112287665B (en) | Chronic disease data analysis method and system based on natural language processing and integrated training | |
| Yu | Analysis and Prediction of Heart Disease Based on Machine Learning Algorithms |
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 |