WO2019024380A1 - 一种中医智能诊断方法、系统及中医医疗系统 - Google Patents

一种中医智能诊断方法、系统及中医医疗系统 Download PDF

Info

Publication number
WO2019024380A1
WO2019024380A1 PCT/CN2017/116065 CN2017116065W WO2019024380A1 WO 2019024380 A1 WO2019024380 A1 WO 2019024380A1 CN 2017116065 W CN2017116065 W CN 2017116065W WO 2019024380 A1 WO2019024380 A1 WO 2019024380A1
Authority
WO
WIPO (PCT)
Prior art keywords
training
model
training data
data
trained
Prior art date
Application number
PCT/CN2017/116065
Other languages
English (en)
French (fr)
Inventor
王峰
曹彬
李诗语
Original Assignee
广东工业大学
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 广东工业大学 filed Critical 广东工业大学
Priority to US16/634,135 priority Critical patent/US20200211706A1/en
Publication of WO2019024380A1 publication Critical patent/WO2019024380A1/zh

Links

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
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
    • G16H10/60—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for patient-specific data, e.g. for electronic patient records
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H20/00—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
    • G16H20/30—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to physical therapies or activities, e.g. physiotherapy, acupressure or exercising
    • 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
    • G16H50/50—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for simulation or modelling of medical disorders
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H70/00—ICT specially adapted for the handling or processing of medical references
    • G16H70/20—ICT specially adapted for the handling or processing of medical references relating to practices or guidelines
    • 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
    • G16H70/60—ICT specially adapted for the handling or processing of medical references relating to pathologies
    • 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
    • G06N3/047—Probabilistic or stochastic networks

Definitions

  • the invention relates to the technical field of disease diagnosis, in particular to a method and system for intelligent diagnosis of Chinese medicine and a medical system for Chinese medicine.
  • China's traditional Chinese medicine treatment method is the most distinctive method for treating diseases in clinical medicine.
  • various information about human diseases can be obtained.
  • the physician can give the patient's treatment plan.
  • TCM treatment is safer and the effect is more stable.
  • people are increasingly hoping to combine traditional Chinese medicine treatment methods with artificial intelligence, thereby reducing the workload of Chinese medicine practitioners.
  • many medical systems that can assist in the diagnosis of traditional Chinese medicine have been developed.
  • the diagnosis information of the patient is often given for the disease information of a certain aspect of the patient's disease, and the diagnosis result often has an accuracy rate. Low or even misdiagnosed, this is also an urgent problem to be solved in the field of intelligent diagnostic systems for Chinese medicine.
  • the object of the present invention is to provide a TCM intelligent diagnosis method and system for improving the accuracy of the TCM intelligent diagnosis system.
  • the specific plan is as follows:
  • a method for intelligent diagnosis of Chinese medicine comprising:
  • the server obtains the patient's medical training data, the medical training data, the medical training data, and the training data from the distributed client cluster;
  • the server uses the medical training data, the medical training data, the medical training data, and the medical training data to train a model to be trained based on a deep neural network algorithm, and obtain training model;
  • the server uses the post-training model to diagnose the disease data of the patient, and obtains a diagnosis result of the disease data.
  • the method specifically uses a distributed client server architecture or a cloud computing architecture.
  • the training model to be trained based on the deep neural network algorithm is trained and trained by using the medical training data, the medical training data, the medical training data and the medical training data.
  • the process of the post model including:
  • the training model to be trained based on the convolutional neural network algorithm is trained to obtain a post-training model
  • the training model to be trained based on the BP neural network algorithm is trained, and the post-training model is obtained.
  • the training model to be trained based on the BP neural network algorithm is trained, and the post-training model is obtained.
  • the training model based on the deep neural network algorithm is trained to obtain the model after the training.
  • the training model to be trained based on the deep neural network algorithm is trained and trained by using the medical training data, the medical training data, the medical training data and the medical training data.
  • the process of the post model including:
  • the training model to be trained based on the convolutional neural network algorithm is trained to obtain a post-training model
  • the training model to be trained based on the BP neural network algorithm is trained, and the post-training model is obtained.
  • the training model to be trained based on the BP neural network algorithm is trained, and the post-training model is obtained.
  • the training model based on the deep neural network algorithm is trained to obtain the model after the training.
  • the training based on the probabilistic neural network algorithm is to be trained.
  • the model is trained to get the model after training.
  • the method further includes:
  • the hope training data is denoised and/or smoothed.
  • the method further includes:
  • the hearing training data is filtered and/or framed.
  • the method further includes:
  • the post-training model is optimized to improve the accuracy of the model after the training
  • the new training data is disease data obtained after the patient diagnosis result is verified.
  • the invention also discloses a TCM intelligent diagnosis system, comprising:
  • a data acquisition module configured to obtain, by the server side, the patient's medical training data, the medical training data, the medical training data, and the medical training data from the distributed client cluster;
  • a model building module configured to use, by the server end, the waiting training data, the medical training data, the medical training data, and the medical training data to construct a model to be trained based on a deep neural network algorithm Train and get the model after training;
  • the diagnosis result obtaining module is configured to use the post-training model to diagnose the disease data of the patient, and obtain a diagnosis result of the disease data.
  • the present invention also discloses a Chinese medical system, including the aforementioned disclosed TCM intelligent diagnosis system, and further comprising:
  • a TCM intelligent treatment system for determining a corresponding treatment plan by using the diagnosis result obtained by the TCM intelligent diagnosis system
  • the TCM intelligent treatment system is a treatment system trained based on a deep neural network algorithm, and the corresponding training samples include historical diagnosis results and corresponding treatment plans.
  • the treatment plan determined by the TCM intelligent treatment system comprises a Chinese patent prescription and/or a physiotherapy plan.
  • the deep neural network algorithm for training the TCM intelligent treatment system comprises a convolutional neural network algorithm.
  • the server side obtains the patient's medical training data, the medical training data, the medical training data, and the medical training data from the distributed client cluster; the server uses the medical training data, the medical training data, The training data and the training data are analyzed, and the model to be trained based on the deep neural network algorithm is trained to obtain the model after training; the server uses the post-training model to diagnose the disease data of the patient, and obtains the diagnosis result of the disease data.
  • the data input to the model to be trained is the patient's observation training data, the hearing training data, the consultation training data, and the cut training data collected from the distributed client cluster. Obviously, in this way The patient will get a large number of disease data, and the disease data are obtained by different methods.
  • disease data are related to each other and restrict each other, which can more fully reflect the patient's disease status.
  • a type of disease data is input at the input of the model to be trained, and the model training accuracy is higher.
  • the method provided by the present invention is applied in a distributed client server architecture, so the model in the present invention can not only acquire more disease data of the patient, but also make the training precision of the model more accurate, and diagnose the disease data of the patient. A faster diagnostic speed can be obtained.
  • FIG. 1 is a flow chart of a TCM intelligent diagnosis method according to an embodiment of the present invention.
  • FIG. 2 is a flowchart of a TCM intelligent diagnosis method according to Embodiment 2 of the present invention.
  • Figure 3 is a training flow chart of the model to be trained
  • Figure 4 is a structural diagram of a model to be trained
  • Figure 5 is a basic structural diagram of the entire model to be trained
  • Figure 6 is a flow chart of the pre-processing of the hearing training data
  • Figure 7 is a structural diagram of a model to be heard and trained
  • Figure 8 is a basic structural diagram of the entire model to be trained
  • Figure 9 is a basic structural diagram of the questionnaire to be trained
  • Figure 10 is a basic structural diagram of the entire questionnaire to be trained
  • FIG. 11 is a flowchart of a TCM intelligent diagnosis method according to Embodiment 3 of the present invention.
  • Figure 12 is a basic structural diagram of a probabilistic neural network
  • 13 is a schematic structural diagram of an entire deep neural network algorithm
  • FIG. 14 is a schematic diagram of a terminal cloud server
  • FIG. 15 is a structural diagram of a TCM intelligent diagnosis system according to an embodiment of the present invention.
  • Embodiment 1 of the present invention discloses a TCM intelligent diagnosis method. Referring to FIG. 1, the method includes:
  • Step S11 The server obtains the patient's medical training data, the medical training data, the medical training data, and the medical training data from the distributed client cluster.
  • the patient's medical training data, the medical training data, the medical training data, and the medical training data are obtained from a distributed client cluster set in a plurality of hospitals and clinics.
  • Comprehensive information on the patient's disease is comprehensively obtained, and then the disease information of the obtained patient is comprehensively analyzed, and the disease diagnosis result corresponding to the patient can be obtained more accurately.
  • the embodiment of the present invention is applied in a distributed client server architecture.
  • the training precision of the model is more accurate, and when the disease data of the patient is diagnosed, Get faster diagnostic speeds.
  • the model can diagnose multiple patients at the same time, and the practical performance of the model is improved compared to the prior art.
  • observation training data, the hearing training data, and the cut training data can be preprocessed to obtain ideal training data to facilitate subsequent processing.
  • Step S12 The server uses the medical training data, the medical training data, the medical training data and the medical training data to train the model to be trained based on the deep neural network algorithm, and obtain the trained model.
  • a deep neural network algorithm is constructed based on a deep neural network algorithm, including, but not limited to, a common convolutional neural network algorithm, a BP neural network algorithm, and a probabilistic neural network algorithm; of course, at the above depth
  • a new neural network layer is added to optimize the established post-training model.
  • a neural network algorithm can be used in the established model to be trained, and several neural network algorithms can also be used.
  • several small neural network subsystems can be used in a large neural network system; At this point, the goal should be to solve the actual problem.
  • the method provided by the embodiment of the present application is more accurate than the patient's tongue image information processed by a single convolutional neural network. It can be understood that the patient's tongue image information can only represent a part of the patient's disease information, and the result is that the established model has less training data at the input end, and the resulting disease diagnosis result is not very accurate.
  • the diagnosis system will be more accurate based on the expert system to provide the patient's treatment plan.
  • the deep neural network algorithm has the ability to learn and relearn. Knowing the data to learn and summarize, so that the model training can be used to have higher efficiency of the existing training data; and based on the treatment system provided by the expert system, this effect is not achieved, because Experienced experts, the training data established is also limited, and through the database established by the experts, can not cover all the diagnostic results corresponding to the patient's disease information, so in comparison, through this method, the diagnosis of the patient's disease will More precise.
  • Step S13 The server uses the post-training model to diagnose the disease data of the patient, and obtains the diagnosis result of the disease data.
  • the training data includes disease information for the patient in the four aspects of diagnosis, diagnosis, consultation, and diagnosis, and it is understood that the patient's disease
  • the data may be disease information of a certain aspect of the patient's disease, or may be information of several aspects of the patient's disease.
  • the model established according to the deep neural network algorithm will give corresponding information according to the disease information provided by the patient. diagnostic result.
  • the diagnosis result in the embodiment is substantially based on the classification result obtained by the information processing device such as a computer, which is different from the diagnosis result obtained by the doctor using the medical theory.
  • the patient's diagnosis result can be used to verify the disease data corresponding to the passage, and the established post-training model can be optimized, so that the established post-training model has more accurate diagnosis results and can better diagnose the patient.
  • the disease may be the disease data corresponding to the rehabilitation of the patient included in the cloud server, or the disease data obtained by other methods to recover the disease of the patient;
  • the data to be trained of the model after optimization training is not limited.
  • the embodiment of the present invention is applied to a distributed client server architecture. It can be understood that the distributed client server architecture can optimize the problem of resource shortage and response bottleneck of the client, and can also solve the data operation in the centralized system. A slower problem.
  • the traditional Chinese medicine intelligent diagnosis method in this embodiment may specifically use a distributed client server architecture or a cloud computing architecture.
  • the server side obtains the patient's medical training data, the medical training data, the medical training data, and the medical training data from the distributed client cluster; the server uses the medical training data, the medical training data, The training data and the training data are analyzed, and the model to be trained based on the deep neural network algorithm is trained to obtain the model after training; the server uses the post-training model to diagnose the disease data of the patient, and obtains the diagnosis result of the disease data.
  • the data input to the model to be trained is the patient's observation training data, the hearing training data, the consultation training data, and the cut-off training data collected from the distributed cluster.
  • the second embodiment of the present invention discloses a specific TCM intelligent diagnosis method. Referring to FIG. 2, the technical solution is further illustrated and optimized in this embodiment. specific:
  • Step S21 The server obtains the patient's medical training data, the medical training data, the medical training data, and the medical training data from the distributed client cluster.
  • the patient's observation training data includes, but is not limited to, obtaining information of the patient's face image and the tongue image;
  • the patient's hearing training data includes, but is not limited to, obtaining the patient's voice, cough, and wheezing sound.
  • the patient's consultation training data including but not limited to, obtaining the patient's cause and history of the disease;
  • the patient's cut-off training data including but not limited to, obtaining the patient's pulse information.
  • the patient's disease data can obtain the patient's disease data from hospitals and clinics across the country, so that the disease data samples obtained by the model can be more comprehensive and then obtained. Comprehensive analysis of all the disease information obtained can more accurately obtain the disease diagnosis results corresponding to the patient.
  • Step S22 The server uses the medical training data, the medical training data, the medical training data, and the medical training data to train the model to be trained based on the deep neural network algorithm to obtain the trained model.
  • the step specifically includes the following steps S221, S222, S223, and S224.
  • Step S221 Using the hope training data, training the model to be trained based on the convolutional neural network algorithm, and obtaining the model after the training.
  • the Convolutional Neural Network Algorithm (CNN) is used to construct a look-and-see training model, which is a supervised learning method. It can be understood that the observation training data is tagged data. .
  • the training process for the model to be trained is shown in Figure 3.
  • the non-supervised (supervised) learning is performed for the look-ahead training model, and the untrained data (which may also be calibrated data) is used for stratified training.
  • Layer parameters and then use the uncalibrated data (which can also be calibrated data) to train the first layer of the model to be trained.
  • the expectation training model obtains the parameters of the n-1th layer by learning, and then the output of the n-1th layer is used as the input of the nth layer, and the nth layer of the training model is trained and looked at, thereby obtaining the diagnosis. Train the parameters of each layer in the model.
  • Top-down supervised learning firstly through the tagged data to train the model to be trained, the error is transmitted from top to bottom, and the neural network can be fine-tuned to obtain the training learning results of the model to be trained.
  • the overall architecture of the model to be trained is shown in Figure 4.
  • C1, C2, C3, and C4 are the convolutional layers of the model to be trained.
  • C1 has 96 convolution kernels of 11*11
  • C2 has 5*5 convolution kernels of 256
  • C3 has 3*3 convolution kernels 384
  • C4 has 3*3 convolution kernels 256; in the expectation training model, there are 4 layers in the max-pooling layer, and each layer of max-pooling cores is 2*2;
  • the output of the four-layer max-pooling layer is used as the input of the fully connected layer; the fully connected layer links the output of the fourth layer of max-pooling into a one-dimensional vector; the output of the fully connected layer is classified into the softmax layer.
  • the specific setting method of the parameter may also be other manners, so as to achieve the purpose of practical application, which is not limited herein.
  • the input training data that is, the facial image, the tongue image, and the body image of the patient are input, and the images are all 3 channels, that is, RGB map.
  • the deliberation training data may also be denoised and/or smoothed.
  • Step S222 Using the medical training data, the training model to be trained based on the BP neural network algorithm is trained, and the post-training model is obtained.
  • the method for acquiring the hearing training data includes, but is not limited to, collecting the patient's voice data, including the patient's voice, cough, and wheezing.
  • the training training data may also be filtered and/or framed.
  • the medical training data is more ideal.
  • the pre-emphasis of the medical training data is mainly to perform certain filtering and framing on the collected sound signal, because windowing framing
  • the processing is to segment the collected sound data to make the sound signal continuous and maintain a certain overlapping rate, which makes it easier to analyze and analyze the subsequent steps.
  • the pre-processed data is subjected to feature extraction, and then the extracted feature vector is used as the input data of the model to be trained to be trained, and the insufficient portion is zero-padded.
  • the process of pre-processing the training data is shown in Figure 6.
  • the structure of the model to be trained is to be trained.
  • the established model for the diagnosis and training is to use a BP neural network with two layers of hidden layers, and the number of input neurons is 600. There are 54 neurons in the hidden layer and 5 neurons in the output layer.
  • the learning process of BP neural network consists of two processes: forward propagation of signal and back propagation of error. In the case of forward propagation, the input signal is input from the input layer, processed layer by layer through each hidden layer, and then transmitted to the output layer. If the actual output of the output layer does not match the expected output (label), it goes to the backpropagation phase of the error.
  • Error back propagation is to pass the output error back to the input layer through the hidden layer in some form, and distribute the error to all the units of each layer, so as to obtain the error signal of each layer unit.
  • This error signal can be used as the correction of each unit weight. Basis. After the signal forward propagation and error back propagation are continuously corrected to adjust the weight between the layers in the model, the training accuracy of the model can be continuously improved.
  • the basic structure of the entire hearing-to-train model is shown in Figure 8.
  • Step S223 Using the training data of the consultation, training the training model to be trained based on the BP neural network algorithm, and obtaining the model after the consultation training.
  • the interview training data is obtained by the patient through the questions set by the answering system, for example, system setting problems, including but not limited to, the patient's age, gender, medical history, family and living environment.
  • the established questionable training model is a BP neural network with 3 hidden layers and 2 input layers and 2 output layers.
  • the basic framework of the established questionnaire for training is shown in Figure 9.
  • the training model to be trained is an 8-input and 9-output BP neural network, and each hidden layer node is set to 8; it is understandable that Setting a plurality of hidden layers compared to setting a single hidden layer can better ensure the accuracy of the model to be trained and the general data generalization ability.
  • the basic structure of the entire questionnaire to be trained is shown in Figure 10.
  • the specific setting method of the parameter may also be other manners, so as to achieve the purpose of practical application, which is not limited herein.
  • Step S224 training the cut-off training model based on the deep neural network algorithm by using the cut-off training data, and obtaining the model after the training.
  • the cut training data is acquired by the digital pulse sensor HK-2000C designed by the Huake Electronics Research Institute. Specifically, the cut training data, after smoothing filter preprocessing, will obtain more ideal cut training data. It can be understood that by such a processing method, the medical training data is made easier to process analysis of the subsequent steps.
  • the cut training data of the model to be trained is input, that is, the collected pulse picture, and the pulse picture is 3 channels, that is, the RGB picture; the model to be trained is according to the depth.
  • the neural network learning algorithm is constructed. It can be understood that the data of the input end of the model to be trained is the collected data to be trained, and the output layer is the disease diagnosis result corresponding to the data to be trained, the number of hidden layers and the specific The parameter settings can be adjusted according to the actual situation, which is not limited here.
  • Step S23 The server uses the post-training model to diagnose the disease data of the patient, and obtains the diagnosis result of the disease data.
  • the disease information of the patient is classified, and a deep neural network model corresponding to the data to be trained, the data to be diagnosed, the data to be diagnosed, and the data to be trained are established.
  • a deep neural network model corresponding to the data to be trained, the data to be diagnosed, the data to be diagnosed, and the data to be trained are established.
  • the disease data information of the patient with more comprehensive diseases can be obtained, and then the disease information of the obtained patient is comprehensively analyzed, and the disease diagnosis result corresponding to the patient can be obtained more accurately, so that The diagnosis of the patient's tongue image information by a single convolutional neural network algorithm will make the diagnosis more accurate.
  • the third embodiment of the present invention discloses a specific TCM intelligent diagnosis method. As shown in FIG. 11, the technical solution is further illustrated and optimized in this embodiment. specific:
  • Step S31 The server obtains the patient's medical training data, the medical training data, the medical training data, and the medical training data from the distributed client cluster.
  • obtaining the patient's medical training data, the medical training data, the medical training data, and the medical training data from the distributed client cluster refer to the method disclosed in the second disclosed embodiment, and no further description is provided herein. .
  • Step S32 The server uses the medical training data, the medical training data, the medical training data, and the training data to train the training model to be trained based on the deep neural network algorithm to obtain the trained model.
  • the step specifically includes the following steps S321, S322, S323, S324, and S325.
  • Step S321 Using the observation training data, training the training model to be trained based on the convolutional neural network algorithm, and obtaining the post-training model.
  • Step S322 Using the medical training data, the training model to be trained based on the BP neural network algorithm is trained, and the post-training model is obtained.
  • Step S323 Using the training data of the consultation, training the training model to be trained based on the BP neural network algorithm, and obtaining the model after the consultation training.
  • Step S324 Using the cut training data, training the cut training model to be constructed based on the deep neural network algorithm, and obtaining the model after the training.
  • step S321, the step S322, the step S323, and the step S324 in the present embodiment may refer to the corresponding method steps in the second implementation of the foregoing disclosure, and details are not described herein again.
  • Step S325 training the model to be trained based on the probabilistic neural network algorithm by using the extracted post-training model, the post-training model, the post-training model, and the output data of the post-training model. Model after training.
  • a model to be trained is constructed based on a probabilistic neural network (PNN), and a post-training model, a post-training model, a post-training model, and a post-training model are diagnosed. Disease outcome, as input data to the input of the model to be trained.
  • PNN probabilistic neural network
  • Figure 12 The basic structure of the probabilistic neural network is shown in Figure 12.
  • the diagnostic results are optimized through a deep neural network. This will make the diagnosis of the patient's disease more accurate.
  • the number of neurons in the input layer of the model to be trained according to the probabilistic neural network algorithm is equal to the post-training model, the post-training model, the post-training model, and the post-training model.
  • the sum of the vector dimensions of the output; the entire model based on the deep neural network algorithm is shown in Figure 13.
  • the number of the neurons and the connection relationship and the specific parameters of each layer in the model to be trained need to be adjusted according to the actual problem, which is The parameters in the model are not limited.
  • Step S33 Using the new training data, the model after training is optimized to improve the accuracy of the model after training.
  • the new training data is the disease data obtained after the patient diagnosis result is verified.
  • this training model has two working modes, one is the training mode of the model, that is, when the established model cannot independently diagnose the patient's disease, it will continuously train through a large number of tagged data sets.
  • the other is the model work and gradual optimization mode, that is, after the training, the model will give the corresponding diagnosis results of the patient during normal use, and then the disease data obtained after passing the test will be verified according to the patient diagnosis results. Accuracy is continuously optimized.
  • Step S34 The server uses the post-training model to diagnose the disease data of the patient, and obtains the diagnosis result of the disease data.
  • a corresponding model is established for the observation training data, the hearing training data, the consultation training data, and the cut training data, and then based on the probabilistic neural network.
  • the training model will extract the data of the post-training model, the post-training model, the post-training model and the post-training model output as the data of the input of the model to be trained, and once again the disease data of the patient. Train optimization.
  • FIG. 14 it is a schematic diagram of a terminal cloud server established by the model.
  • the terminal cloud server is connected to the input end of each clinic, wherein the training data, the medical training data, and the consultation are observed.
  • the training data and the training data are collected through the observation terminal of the cloud server, the sound collection end of the diagnosis, the information collection end of the consultation, and the pulse collection end.
  • the number of diseases of the patient in the embodiment of the present application is training data with data tags; that is, the physician sets the corresponding data tags by sorting the disease data of the patient.
  • observation training data, the hearing training data, the consultation training data, and the cut training data are stored in the cloud server.
  • the established model can have the ability to train data on a large scale, and in this way, the established model can simultaneously diagnose multiple patients, greatly improving the practical performance of the model.
  • the cloud server can first detect whether the patient has a case database in the cloud server through relevant settings, and if so, perform a normal disease data diagnosis process; if not, the system will automatically establish a complete patient disease. database. Moreover, the system can also print out the prescription of the patient's treatment plan and the precautions of the patient in daily life at the terminal of the system.
  • an embodiment of the present invention further discloses a TCM intelligent diagnosis system, which is specifically disposed on a cloud-based distributed client server architecture.
  • the system includes:
  • the data obtaining module 41 is configured to obtain the patient's medical training data, the medical training data, the medical training data, and the medical training data from the distributed client cluster.
  • the model building module 42 is configured to use the medical training data, the medical training data, the medical training data and the medical training data to train the model to be trained based on the deep neural network algorithm, and obtain the trained model.
  • the diagnosis result obtaining module 43 is configured to diagnose the disease data of the patient by using the post-training model on the server side, and obtain a diagnosis result of the disease data.
  • the model construction module 42 includes a look-and-feel construction unit, a diagnosis construction unit, a diagnosis construction unit, and a diagnosis construction unit;
  • the look-and-see construction unit is used to train the look-ahead training model based on the convolutional neural network algorithm to obtain the post-training model.
  • the diagnosis and diagnosis building unit is used to train the training model to be trained based on the BP neural network algorithm, and obtain the post-training model.
  • the questioning construction unit is used to train the training model to be trained based on the BP neural network algorithm, and obtain the post-training model.
  • the cut-off construction unit is used to train the cut-off training model based on the deep neural network algorithm to obtain the post-training model.
  • model building module 42 includes a look-and-feel building unit, a hearing building unit, a question building unit, a cut-off building unit, and a model building unit;
  • the look-and-see construction unit is used to train the look-ahead training model based on the convolutional neural network algorithm to obtain the post-training model.
  • the diagnosis and diagnosis building unit is used to train the training model to be trained based on the BP neural network algorithm, and obtain the post-training model.
  • the questioning construction unit is used to train the training model to be trained based on the BP neural network algorithm, and obtain the post-training model.
  • the cut-off construction unit is used to train the cut-off training model based on the deep neural network algorithm to obtain the post-training model.
  • the model building unit is configured to perform the model to be trained based on the probabilistic neural network algorithm by using the extracted post-training model, the post-training model, the post-training model, and the output data of the post-training model. Train and get the model after training.
  • the TCM intelligent diagnosis system disclosed in the embodiment of the present invention further includes: a diagnosis data preprocessing module and a diagnosis data preprocessing module; wherein
  • the diagnostic data preprocessing module is used for denoising and/or smoothing the medical training data.
  • the data processing pre-processing module is used for filtering and/or framing the sounding training data.
  • the TCM intelligent diagnosis system disclosed in the embodiment of the present invention further includes:
  • the model optimization module is configured to optimize the post-training model by using the new training data before the diagnosis result obtaining module 43 diagnoses the disease data of the patient by using the post-training model to improve the accuracy of the model after the training.
  • the new training data is the disease data obtained after the patient diagnosis result is verified.
  • the embodiment of the present invention further discloses a Chinese medical system, including the aforementioned TCM intelligent diagnosis system, and further comprising:
  • the TCM intelligent treatment system is used to determine the corresponding treatment plan by using the diagnosis results obtained by the TCM intelligent diagnosis system;
  • the TCM intelligent treatment system is a treatment system based on deep neural network algorithm training, and the corresponding training samples include historical diagnosis results and corresponding treatment plans.
  • the treatment plan in the present embodiment substantially refers to the classification result obtained by the information processing device such as a computer based on deep learning, which is different from the treatment plan obtained by the doctor based on medical theory.
  • the treatment system obtained by using the deep neural network algorithm is used to give the patient's diagnosis result, and the corresponding treatment plan of the patient is given.
  • the treatment plan obtained by using the deep neural network algorithm is used to give the patient's diagnosis result, and the corresponding treatment plan of the patient is given.
  • it can also be optimized with a new training sample, which is not limited herein.
  • the treatment plan determined by the TCM intelligent treatment system includes a proprietary Chinese medicine prescription and/or a physical therapy plan.
  • the treatment plan determined by the TCM intelligent treatment system includes, but is not limited to, a proprietary Chinese medicine prescription and/or a physical therapy plan. This not only reduces the workload of the doctor, but also provides a corresponding reference treatment plan for the patient's diagnosis results, which improves the patient's treatment experience.
  • the deep neural network algorithm for training the TCM intelligent treatment system includes a convolutional neural network algorithm.
  • the convolutional neural network algorithm has the advantages of simple structure, less training parameters and strong adaptability, and the TCM intelligent treatment system is obtained.
  • other deep neural network algorithms may also be used, which are not limited herein.
  • the TCM intelligent diagnosis system mainly determines the disease corresponding to the patient according to the disease information of the patient.
  • the TCM intelligent treatment system can provide a corresponding disease diagnosis plan to the patient according to the medical diagnosis system disclosed above; and the drug treatment plan provided by the TCM intelligent treatment system can also flexibly change the proportional weight of the drug according to different conditions of the patient. Further, the system can also print out the prescription of the patient's treatment plan and the precautions of the patient in daily life at the terminal of the system.

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • General Health & Medical Sciences (AREA)
  • Epidemiology (AREA)
  • Primary Health Care (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Pathology (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Mathematical Physics (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Alternative & Traditional Medicine (AREA)
  • Pharmacology & Pharmacy (AREA)
  • Physical Education & Sports Medicine (AREA)
  • Bioethics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Medical Treatment And Welfare Office Work (AREA)

Abstract

一种中医智能诊断方法、系统及中医医疗系统，该方法包括：服务器端从分布式客户端集群中获取患者的望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据；服务器端利用望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据，对基于深度神经网络算法构建的待训练模型进行训练，得到训练后模型；服务器端利用训练后模型对患者的疾病数据进行诊断，得出疾病数据的诊断结果。本申请公开的技术方案通过对患者的疾病信息进行全面的采集，有效提升了医疗诊断结果的准确度，而且，本方法可以快速地同时处理多个患者的疾病数据。

Description

一种中医智能诊断方法、系统及中医医疗系统
本申请要求于2017年07月31日提交中国专利局、申请号为201710639260.7、发明名称为“一种中医智能诊断方法、系统及中医医疗系统”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本发明涉及疾病诊断技术领域，特别涉及一种中医智能诊断方法、系统及中医医疗系统。
背景技术
我国的中医治疗方法是临床医学中最具特色的一种治疗疾病的方法，通过中医医师对患者进行望诊、闻诊、问诊和切诊，就可以得到关于人体疾病的各种信息，中医医师通过对所获得到的疾病信息进行综合的分析，就可以给出患者的治疗方案，相比于西医治疗，中医治疗更为安全，效果更加稳定。近年来，随着现代医学技术的不断发展，人们愈加希望将传统的中医治疗方法与人工智能结合起来，进而减轻中医医师的工作量。目前已经研究出很多可以辅助中医诊断的医疗系统，但是在目前常见的医疗诊断系统当中，往往是针对患者疾病某一方面的疾病信息来给出患者的诊断方案，诊断结果经常会出现准确率较低，甚至是误诊的现象，这也是在中医智能诊断系统领域中亟待解决的一个问题。
发明内容
有鉴于此，本发明的目的在于提供一种中医智能诊断方法及系统，用于提升中医智能诊断系统的准确率。其具体方案如下：
一种中医智能诊断方法，包括：
服务器端从分布式客户端集群中获取患者的望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据；
所述服务器端利用所述望诊训练数据、所述闻诊训练数据、所述问诊训练数据和所述切诊训练数据，对基于深度神经网络算法构建的待训练模型进行训练，得到训练后模型；
所述服务器端利用所述训练后模型对患者的疾病数据进行诊断，得出所述疾病数据的诊断结果。
优选的，所述方法具体使用分布式客户端服务器架构或云计算架构。
优选的，所述利用所述望诊训练数据、所述闻诊训练数据、所述问诊训练数据和所述切诊训练数据，对基于深度神经网络算法构建的待训练模型进行训练，得到训练后模型的过程，包括：
利用所述望诊训练数据，对基于卷积神经网络算法构建的望诊待训练模型进行训练，得到望诊训练后模型；
利用所述闻诊训练数据，对基于BP神经网络算法构建的闻诊待训练模型进行训练，得到闻诊训练后模型；
利用所述问诊训练数据，对基于BP神经网络算法构建的问诊待训练模型进行训练，得到问诊训练后模型；
利用所述切诊训练数据，对基于深度神经网络算法构建的切诊待训练模型进行训练，得到切诊训练后模型。
优选的，所述利用所述望诊训练数据、所述闻诊训练数据、所述问诊训练数据和所述切诊训练数据，对基于深度神经网络算法构建的待训练模型进行训练，得到训练后模型的过程，包括：
利用所述望诊训练数据，对基于卷积神经网络算法构建的望诊待训练模型进行训练，得到望诊训练后模型；
利用所述闻诊训练数据，对基于BP神经网络算法构建的闻诊待训练模型进行训练，得到闻诊训练后模型；
利用所述问诊训练数据，对基于BP神经网络算法构建的问诊待训练模型进行训练，得到问诊训练后模型；
利用所述切诊训练数据，对基于深度神经网络算法构建的切诊待训练模型进行训练，得到切诊训练后模型；
利用提取到的所述望诊训练后模型、所述闻诊训练后模型、所述问诊训练后模型和所述切诊训练后模型的输出端数据，对基于概率神经网络算法构建的待训练模型进行训练，得到训练后模型。
优选的，还包括：
对所述望诊训练数据进行去噪和/或平滑处理。
优选的，还包括：
对所述闻诊训练数据进行滤波和/或分帧处理。
优选的，所述利用所述训练后模型对患者的疾病数据进行诊断，得出所述疾病数据的诊断结果的过程之前，还包括：
利用新训练数据，对所述训练后模型进行优化，以提高所述训练后模型的精度；
其中，所述新训练数据为对患者诊断结果验证通过后得到的疾病数据。
本发明还公开了一种中医智能诊断系统，包括：
数据获取模块，用于服务器端从分布式客户端集群中获取患者的望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据；
模型构建模块，用于所述服务器端利用所述望诊训练数据、所述闻诊训练数据、所述问诊训练数据和所述切诊训练数据，对基于深度神经网络算法构建的待训练模型进行训练，得到训练后模型；
诊断结果获取模块，用于所述服务器端利用所述训练后模型对患者的疾病数据进行诊断，得出所述疾病数据的诊断结果。
进一步的，本发明还公开了一种中医医疗系统，包括前述公开的中医智能诊断系统，还包括：
中医智能治疗系统，用于利用所述中医智能诊断系统得到的诊断结果，确定出相应的治疗方案；
其中，所述中医智能治疗系统为基于深度神经网络算法训练得到的治疗系统，对应的训练样本包括历史诊断结果以及相应的治疗方案。
优选的，所述中医智能治疗系统确定出的治疗方案包括中成药处方和/或理疗方案。
优选的，训练所述中医智能治疗系统的深度神经网络算法包括卷积神经网络算法。
本发明中，首先服务器端从分布式客户端集群中获取患者的望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据；服务器端利用望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据，对基于深度神经网络算法构建的待训练模型进行训练，得到训练后模型；服务器端利用训练后模型对患者的疾病数据进行诊断，得出疾病数据的诊断结果。在本发明中，待训练模型输入端的数据为从分布式客户端集群中采集到的患者的望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据，显然，通过这样的方式，会得到患者大量的疾病数据，而且，这些疾病数据是利用不同的方法得到的，这些疾病数据，相互联系，又彼此制约，能够更加全面的体现患者的疾病状况，相比之下，会比在待训练模型输入端输入一种类型的疾病数据，模型训练精度更高。而且，本发明提供的方法，是应用在分布式客户端服务器架构中，所以本发明中的模型不仅可以获取更多的患者的疾病数据，使得模型的训练精度更加精准，而且诊断患者的疾病数据时，可以获得更加快速的诊断速度。
附图说明
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据提供的附图获得其他的附图。
图1为本发明实施例一公开的一种中医智能诊断方法流程图；
图2为本发明实施例二公开的一种中医智能诊断方法流程图；
图3为望诊待训练模型的训练流程图；
图4为望诊待训练模型的结构图；
图5为整个望诊待训练模型的基本结构图；
图6为闻诊训练数据预处理的流程图；
图7为闻诊待训练模型的结构图；
图8为整个闻诊待训练模型的基本结构图；
图9为问诊待训练模型的基本结构图；
图10为整个问诊待训练模型的基本结构图；
图11为本发明实施例三公开的一种中医智能诊断方法流程图；
图12为概率神经网络的基本结构图；
图13为整个深度神经网络算法的结构示意图；
图14为终端云服务器的示意图；
图15为本发明实施例公开的一种中医智能诊断系统结构图。
具体实施方式
下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
本发明实施例一公开了一种中医智能诊断方法，参见图1所示，该方法包括：
步骤S11：服务器端从分布式客户端集群中获取患者的望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据。
本实施例中，从设于多个医院、诊所的分布式客户端集群中获取患者的望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据，通过这样的方法，可以更加全面的得到患者疾病的全部信息，然后将获取得到的患者的疾病信息进行综合分析，可以更加精准的得到患者所对应的疾病诊断结果。
而且，本发明实施例是应用在分布式客户端服务器架构中，通过这样的方式，不仅可以获取更多的患者的疾病数据，使得模型的训练精度更加精准，而且诊断患者的疾病数据时，可以获得更加快速的诊断速度。而且，本模型可以同时诊断多个患者，相比于现有技术，提高了模型的实用性能。
进一步的，还可以对望诊训练数据、闻诊训练数据和切诊训练数据进 行预处理，得到较为理想的训练数据，以方便后续步骤的处理。
步骤S12：服务器端利用望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据，对基于深度神经网络算法构建的待训练模型进行训练，得到训练后模型。
在本实施例中，基于深度神经网络算法构建待训练模型，深度神经网络算法，包括但不限于，常见的卷积神经网络算法、BP神经网络算法和概率神经网络算法；当然可以在上述的深度神经网络的输入端和输出端，添加新的神经网络层，来优化所建立的训练后模型。
可以理解的是，建立的待训练模型中，可以运用一种神经网络算法，也可以运用几种神经网络算法，当然也可以在一个大的神经网络系统中运用几种小的神经网络子系统；此时，应该以所要达到解决实际问题为目的。
可以理解的是，通过本申请实施例提供的这种方法，会相比通过单一的卷积神经网络处理患者的舌像信息，诊断结果更加精准。可以理解的是，患者的舌像信息只能表征出患者的一部分疾病信息，这样导致的结果，就是所建立的模型，输入端的训练数据较少，由此得到的疾病诊断结果也不是很精确。
而且，通过这样的方式，会相比于传统方法，基于专家系统提供患者的治疗方案，诊断结果会更加精准，可以理解的是，深度神经网络算法具有学习以及再学习的能力，可以通过对已知的数据进行学习和归纳总结，所以可以保证模型训练时，对已有的训练数据有更高的使用效率；而基于专家系统提供的治疗方案，就达不到这样的效果，因为，再有经验的专家，所建立的训练数据，也是有限的，而且通过专家建立的数据库，不能覆盖患者疾病信息所对应的所有诊断结果，所以相比而言，通过这样的方法，患者疾病的诊断结果会更精准。
步骤S13：服务器端利用训练后模型对患者的疾病数据进行诊断，得出疾病数据的诊断结果。
在本实施例中，利用深度神经网络算法所建立的模型当中，训练数据包含了对患者进行望诊、闻诊、问诊和切诊四个方面的疾病信息，可以理解的是，患者的疾病数据可以是患者疾病的某一方面的疾病信息，也可以 是患者疾病的几个方面的信息，此时，根据深度神经网络算法所建立的模型会根据患者所提供的疾病信息，给出相应的诊断结果。可以理解的是，本实施例中的诊断结果实质上是基于深度学习由计算机等信息处理设备得到的分类结果，这与医生利用医学理论得到的诊断结论是不相同的。
当然，更进一步的，还可以利用患者诊断结果验证通过后所对应的疾病数据，来对所建立的训练后模型进行优化，使得建立的训练后模型的诊断结果更加精准，能够更好地诊断患者的疾病。需要说明的是，此处患者诊断结果验证通过后所对应的疾病数据可以是包含在云端服务器中患者康复所对应的疾病数据，也可以是通过其他方法获得的使得患者疾病康复的疾病数据；此处对优化训练后模型的待训练数据不作限定。
而且本发明实施例是应用于分布式客户端服务器架构中，可以理解的是，分布式客户端服务器架构可以优化客户端资源紧张与响应瓶颈的问题，而且也可以解决集中式系统中的数据运算速度较慢的问题。
进一步的，本实施例中的中医智能诊断方法具体可以使用分布式客户端服务器架构或云计算架构。
本发明中，首先服务器端从分布式客户端集群中获取患者的望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据；服务器端利用望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据，对基于深度神经网络算法构建的待训练模型进行训练，得到训练后模型；服务器端利用训练后模型对患者的疾病数据进行诊断，得出疾病数据的诊断结果。在本发明中，待训练模型输入端的数据为从分布式集群中采集到的患者的望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据，显然，通过这样的方式，会得到患者大量的疾病数据，而且，这些疾病数据是利用不同的方法得到的，这些疾病数据，相互联系，又彼此制约，能够更加全面的体现患者的疾病状况，相比之下，会比在待训练模型输入端输入一种类型的疾病数据，模型训练精度更高。而且，本发明提供的方法，是应用在分布式客户端服务器架构中，所以本发明中的模型不仅可以获取更多的患者的疾病数据，使得模型的训练精度更加精准，而且诊断患者的疾病数据时，可以获得更加快速的诊断速度。
本发明实施例二公开了一种具体的中医智能诊断方法，参见图2所示，相对于上一实施例，本实施例对技术方案作了进一步的说明和优化。具体的：
步骤S21：服务器端从分布式客户端集群中获取患者的望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据。
在本实施例中，患者的望诊训练数据，包括但不限于，获取患者面像和舌像的信息；患者的闻诊训练数据，包括但不限于，获取患者的说话声音、咳嗽和喘息声；患者的问诊训练数据，包括但不限于，获取患者的病症原因和患病史；患者的切诊训练数据，包括但不限于，获取患者的脉象信息。
从分布式客户端集群中获取患者的疾病数据，也即，患者的疾病数据可是从全国各地的医院及诊所当中获得患者的疾病数据，这样可以使得模型获得的疾病数据样本更加全面，然后将获取得到的全部疾病信息进行综合分析，可以更加精准的得到患者所对应的疾病诊断结果。
步骤S22：服务器端利用望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据，对基于深度神经网络算法构建的待训练模型进行训练，得到训练后模型；在本实施例中，该步骤具体包括下面步骤S221、步骤S222、步骤S223和步骤S224。
步骤S221：利用望诊训练数据，对基于卷积神经网络算法构建的望诊待训练模型进行训练，得到望诊训练后模型。
在本实施例中，基于卷积神经网络算法(Convolution Neural Network Algorithm，CNN)构建望诊待训练模型，采用的是有监督学习的方式，可以理解的是，望诊训练数据是有标签的数据。望诊待训练模型的训练流程如图3所示。
具体的，在本实施例中，对于望诊待训练模型采用的是自下向上非监督(监督)学习，首先采用无标定数据(也可以是有标定数据)分层训练望诊待训练模型各层参数，然后用无标定数据(也可以是有标定数据)训练望诊待训练模型的第一层，可以理解的是，训练望诊待训练模型时先学 习模型当中第一层的参数，之后望诊待训练模型通过学习得到第n-1层的参数，再将第n-1层的输出作为第n层的输入，训练望诊待训练模型的第n层，由此分别得到望诊待训练模型当中各层的参数。
自顶向下的监督学习，首先通过带标签的数据去训练望诊待训练模型，误差自顶向下传输，可以对神经网络进行微调，从而得到望诊待训练模型的训练学习结果。望诊待训练模型的整体架构如图4所示。
如图4所示，C1，C2，C3和C4为望诊待训练模型的卷积层，其中C1有11*11卷积核96个，C2有5*5卷积核为256个，C3有3*3卷积核384个，C4有3*3卷积核256个；望诊待训练模型当中max-pooling层有4层，每层max-pooling的核都为2*2；其中，第四层max-pooling层的输出作为全连接层的输入；全连接层将第四层的max-pooling的输出链接成为一个一维向量；全连接层的输出到softmax层进行分类。
当然，参数的具体设置方法也可以是其他的方式，以达到实际应用为目的，此处不作限定。
需要说明的是，在本实施例中，望诊待训练模型中，输入的是望诊训练数据，也即，患者的面部图像、舌头图像和形体图像，而且，上述图像均为3通道，即RGB图。
进一步的，在本步骤中，还可以对望诊训练数据进行去噪去噪和/或平滑处理。
可以理解的是，通过这样的处理方式，使得望诊训练数据更加理想，更易于后续步骤的处理分析。整个望诊待训练模型的基本结构如图5所示。
步骤S222：利用闻诊训练数据，对基于BP神经网络算法构建的闻诊待训练模型进行训练，得到闻诊训练后模型。
具体的，在本实施例中，闻诊训练数据的获取方法，包括但不限于，采集患者的声音数据，包括患者的说话声音、咳嗽和喘息声。
进一步的，在本步骤中，还可以对闻诊训练数据进行滤波和/或分帧处理。
可以理解的是，通过这样的处理方式，使得望诊训练数据更加理想，具体的，对闻诊训练数据预加重主要是对采集到的声音信号进行一定的滤 波和分帧，因为加窗分帧处理是将采集到的声音数据进行分段处理，使声音信号连续并保持一定的重叠率，这样更易于后续步骤的处理分析。而且，在本实施例中，是将预处理后的数据进行特征提取，然后将提取到的特征向量作为闻诊待训练模型的输入数据，不足部分补零。闻诊训练数据预处理的过程如图6所示。
如图7所示是闻诊待训练模型的结构图，在本实施例中，建立的闻诊待训练模型是采用含有两层隐含层的BP神经网络，输入神经元为600个，中间每个隐含层的神经元为54个，输出层的神经元为5个。BP神经网络的学习过程由信号的正向传播与误差的反向传播两个过程组成。正向传播时，输入信号从输人层输入，经各隐含层逐层处理后，传向输出层。若输出层的实际输出与期望的输出(标签)不符，则转入误差的反向传播阶段。误差反传是将输出误差以某种形式通过隐藏层向输入层反传，并将误差分摊给各层的所有单元，从而获得各层单元的误差信号，此误差信号可以作为修正各单元权值的依据。经过信号正向传播与误差反向传播不断的修正调整模型当中各层之间的权值，可以不断的提高模型的训练精度。整个闻诊待训练模型的基本结构如图8所示。
步骤S223：利用问诊训练数据，对基于BP神经网络算法构建的问诊待训练模型进行训练，得到问诊训练后模型。
在本实施例中，问诊训练数据是患者通过回答系统所设置的问题获取到的，比如，系统设置的问题，包括但不限于，患者的年龄、性别、病史、家庭及生活环境等问题。
在本实施例中，建立的问诊待训练模型是采用含有3个隐含层和2个输入层和2个输出层的BP神经网络。建立的问诊待训练模型的基本构架如图9所示；具体的，问诊待训练模型为8输入9输出的BP神经网络，每个隐含层的节点设置为8个；可以理解的是，设置多个隐含层相比设置单个隐藏层能更好地保证问诊待训练模型的精确度，以及较强的数据泛化能力。整个问诊待训练模型的基本结构如图10所示。
当然，参数的具体设置方法也可以是其他的方式，以达到实际应用为目的，此处不作限定。
步骤S224：利用切诊训练数据，对基于深度神经网络算法构建的切诊待训练模型进行训练，得到切诊训练后模型。
在本实施例中，切诊训练数据是由是由华科电子研究所设计的数字化脉象传感器HK-2000C采集得到的。具体的，该切诊训练数据，经过平滑滤波预处理后，会得到更加理想的切诊训练数据。可以理解的是，通过这样的处理方式，使得望诊训练数据更易于后续步骤的处理分析。
具体的，在本实施例中，输入切诊待训练模型的切诊训练数据，也即采集到的脉象图片，而且，脉象图片为3通道，也即RGB图；切诊待训练模型是根据深度神经网络学习算法构建的，可以理解的是，切诊待训练模型输入端的数据为采集到的切诊待训练数据，输出层为切诊待训练数据对应的疾病诊断结果，隐藏层层数及具体参数的设定可以根据实际情况进行相应的调整，此处不作限定。
步骤S23：服务器端利用训练后模型对患者的疾病数据进行诊断，得出疾病数据的诊断结果。
在本实施例中，是对患者的疾病信息进行了分类，分别建立了与望诊待训练数据、闻诊待训练数据、问诊待训练数据和切诊待训练数据相对应的深度神经网络模型。可以理解的是，通过这样的方法，可以获得患者疾病较为全面的疾病数据信息，然后将获取得到的患者的疾病信息进行综合分析，可以更加精准的得到患者所对应的疾病诊断结果，这样相比于单一的通过卷积神经网络算法来处理患者的舌像信息，诊断结果会更加精确。
本发明实施例三公开了一种具体的中医智能诊断方法，如图11所示，相对于上一实施例，本实施例对技术方案作了进一步的说明和优化。具体的：
步骤S31：服务器端从分布式客户端集群中获取患者的望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据。
在本实施中，从分布式客户端集群中获取患者的望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据，可以参考前述公开的实施二公开的方法，在此不作赘述。
步骤S32：服务器端利用望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据，对基于深度神经网络算法构建的待训练模型进行训练，得到训练后模型；在本实施例中，该步骤具体包括下面步骤S321、步骤S322、步骤S323、步骤S324和步骤S325。
步骤S321：利用望诊训练数据，对基于卷积神经网络算法构建的望诊待训练模型进行训练，得到望诊训练后模型。
步骤S322：利用闻诊训练数据，对基于BP神经网络算法构建的闻诊待训练模型进行训练，得到闻诊训练后模型。
步骤S323：利用问诊训练数据，对基于BP神经网络算法构建的问诊待训练模型进行训练，得到问诊训练后模型。
步骤S324：利用切诊训练数据，对基于深度神经网络算法构建的切诊待训练模型进行训练，得到切诊训练后模型。
需要说明的是，本实施中的步骤S321、步骤S322、步骤S323和步骤S324可参考前述公开的实施二中对应的方法步骤，此处不再赘述。
步骤S325：利用提取到的望诊训练后模型、闻诊训练后模型、问诊训练后模型和切诊训练后模型的输出端数据，对基于概率神经网络算法构建的待训练模型进行训练，得到训练后模型。
在本实施例中，基于概率神经网络算法(Probabilistic neural network，PNN)构建待训练模型，将望诊训练后模型、闻诊训练后模型、问诊训练后模型和切诊训练后模型诊断出来的疾病结果，作为待训练模型输入端的输入数据。概率神经网络的基本结构如图12所示。
可以理解的是，通过在望诊训练后模型、闻诊训练后模型、问诊训练后模型和切诊训练后模型的基础上，再通过一层深度神经网络对所建立的模型进行诊断结果的优化，这样会使得患者疾病的诊断结果更加的精确。
具体的，在本实施例中，根据概率神经网络算法所构建的待训练模型输入层的神经元数目等于望诊训练后模型、闻诊训练后模型、问诊训练后模型和切诊训练后模型输出端的矢量维数之和；整个基于深度神经网络算法构建的模型如图13所示。
需要说明的是，在此步骤中，所建立的待训练模型中各个层的神经元的个数和连接关系以及具体的参数，都需要根据实际问题所要达到的目的进行相应的调整，此处对模型中的参数不作限定。
步骤S33：利用新训练数据，对训练后模型进行优化，以提高训练后模型的精度。
其中，新训练数据为对患者诊断结果验证通过后得到的疾病数据。
可以理解的是，利用患者诊断结果验证通过后的疾病数据对训练后模型的参数进行优化，可以提高模型的诊断精度。
需要说明的是，此训练模型会有两种工作模式，一种是模型的训练模式，也即，建立的模型不能独立诊断患者疾病时，它会通过大量的带标签的数据集进行不断的训练；另外一种是模型的工作和逐渐优化模式，也即，训练后模型在正常使用过程中，会给出患者相应的诊断结果，然后会根据患者诊断结果验证通过后得到的疾病数据对模型的精度进行不断的优化。
步骤S34：服务器端利用训练后模型对患者的疾病数据进行诊断，得出疾病数据的诊断结果。
在本实施例中，首先是针对望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据分别建立相应的模型，然后在此基础之上，又建立一个基于概率神经网络的待训练模型，将提取到的望诊训练后模型、闻诊训练后模型、问诊训练后模型和切诊训练后模型输出端的数据，作为待训练模型输入端的数据，再一次的对患者的疾病数据进行训练优化。
可以理解的是，通过本实施例提供的这种方法，显著提高了模型的训练精度。
如图14所示，是该模型建立的终端云服务器示意图，具体的，在本实施例中，终端云服务器和各个诊所的输入端连接，其中，望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据通过云服务器中的望诊采集端、闻诊声音采集端、问诊信息采集端和脉搏采集端采集得到的。需要说明的是，本申请实施例中患者的疾病数都是带有数据标签的训练数据；也即，医师通过对患者的疾病数据进行整理，设置相应的数据标签。
可以理解的是，将望诊训练数据、闻诊训练数据、问诊训练数据和切 诊训练数据存储在云服务器中。可以使所建立的模型具备大规模训练数据的能力，而且通过这样的方式，还可以使建立的模型能够同时诊断多个患者，大大提高了模型的实用性能。
更进一步的，云服务器通过相关的设置，可以首先检测患者是否在云服务器中存在有病例库，如果有，则进行正常的疾病数据诊断流程；如果没有，则系统将会自动建立完整的患者疾病数据库。而且，本系统还可以在系统的终端打印出患者的治疗方案的处方以及患者在日常生活中的注意事项。
相应的，本发明实施例还公开了一种中医智能诊断系统，该系统具体设于基于云计算的分布式客户端服务器架构上，参见图15所示，该系统包括：
数据获取模块41，用于服务器端从分布式客户端集群中获取患者的望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据。
模型构建模块42，用于服务器端利用望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据，对基于深度神经网络算法构建的待训练模型进行训练，得到训练后模型。
诊断结果获取模块43，用于服务器端利用训练后模型对患者的疾病数据进行诊断，得出疾病数据的诊断结果。
具体的，模型构建模块42，包括望诊构建单元、闻诊构建单元、问诊构建单元和切诊构建单元；其中，
望诊构建单元，用于利用望诊训练数据，对基于卷积神经网络算法构建的望诊待训练模型进行训练，得到望诊训练后模型。
闻诊构建单元，用于利用闻诊训练数据，对基于BP神经网络算法构建的闻诊待训练模型进行训练，得到闻诊训练后模型。
问诊构建单元，用于利用问诊训练数据，对基于BP神经网络算法构建的问诊待训练模型进行训练，得到问诊训练后模型。
切诊构建单元，用于利用切诊训练数据，对基于深度神经网络算法构建的切诊待训练模型进行训练，得到切诊训练后模型。
更为具体的，模型构建模块42，包括望诊构建单元、闻诊构建单元、问诊构建单元、切诊构建单元和模型构建单元；其中，
望诊构建单元，用于利用望诊训练数据，对基于卷积神经网络算法构建的望诊待训练模型进行训练，得到望诊训练后模型。
闻诊构建单元，用于利用闻诊训练数据，对基于BP神经网络算法构建的闻诊待训练模型进行训练，得到闻诊训练后模型。
问诊构建单元，用于利用问诊训练数据，对基于BP神经网络算法构建的问诊待训练模型进行训练，得到问诊训练后模型。
切诊构建单元，用于利用切诊训练数据，对基于深度神经网络算法构建的切诊待训练模型进行训练，得到切诊训练后模型。
模型构建单元，用于利用提取到的望诊训练后模型、闻诊训练后模型、问诊训练后模型和切诊训练后模型的输出端数据，对基于概率神经网络算法构建的待训练模型进行训练，得到训练后模型。
进一步的，本发明实施例公开的中医智能诊断系统，还包括，望诊数据预处理模块和闻诊数据预处理模块；其中，
望诊数据预处理模块，用于对望诊训练数据进行去噪和/或平滑处理。
闻诊数据预处理模块，用于对闻诊训练数据进行滤波和/或分帧处理。
进一步的，本发明实施例公开的中医智能诊断系统，还包括：
模型优化模块，用于在诊断结果获取模块43利用训练后模型对患者的疾病数据进行诊断之前，利用新训练数据，对训练后模型进行优化，以提高所述训练后模型的精度。
其中，新训练数据为对患者诊断结果验证通过后得到的疾病数据。
关于上述各个模块和各个单元更加详细的工作过程可以参考前述实施例中公开的相应内容，在此不再进行赘述。
相应的，本发明实施例还公开了一种中医医疗系统，包括前述公开的中医智能诊断系统，还包括：
中医智能治疗系统，用于利用中医智能诊断系统得到的诊断结果，确定出相应的治疗方案；
其中，中医智能治疗系统为基于深度神经网络算法训练得到的治疗系统，对应的训练样本包括历史诊断结果以及相应的治疗方案。可以理解的是，本实施例中的治疗方案实质上是指基于深度学习由计算机等信息处理设备得到的分类结果，这与医生基于医学理论得到的治疗方案是不同的。
在本系统中，是将患者的诊断结果，利用深度神经网络算法训练得到的治疗系统，来给出患者相对应的治疗方案。当然，为了得到更好的治疗方案，还可以利用新的训练样本对其进行优化，此处不作限定。
具体的，中医智能治疗系统确定出的治疗方案包括中成药处方和/或理疗方案。
在本实施例中，由中医智能治疗系统确定出的治疗方案，包括但不限于中成药处方和/或理疗方案。这样不仅可以减轻医生的工作量，也可以给患者诊断结果提供相应的参考治疗方案，提升了患者的治疗体验。
具体的，训练中医智能治疗系统的深度神经网络算法包括卷积神经网络算法。
在本实施例中，利用卷积神经网络算法结构简单、训练参数少和适应性强的优点，来得到中医智能治疗系统。当然实际的应用当中，也可以使用其他的深度神经网络算法，此处不作限定。
在本发明实施例中，中医智能诊断系统主要是根据患者的疾病信息判断出患者所对应的疾病。中医智能治疗系统可以根据前述公开的医疗诊断系统，给患者提供相应的疾病诊断方案；而且，中医智能治疗系统给患者提供的药物治疗方案，还可以根据患者的不同状况灵活变换药物的比例权重。进一步的，本系统还可以在系统的终端打印出患者的治疗方案的处方以及患者在日常生活中的注意事项。
最后，还需要说明的是，在本文中，诸如第一和第二等之类的关系术语仅仅用来将一个实体或者操作与另一个实体或操作区分开来，而不一定要求或者暗示这些实体或操作之间存在任何这种实际的关系或者顺序。而且，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、物品或者设备不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、 方法、物品或者设备所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括所述要素的过程、方法、物品或者设备中还存在另外的相同要素。
以上对本发明所提供的一种中医智能诊断方法及系统进行了详细介绍，本文中应用了具体个例对本发明的原理及实施方式进行了阐述，以上实施例的说明只是用于帮助理解本发明的方法及其核心思想；同时，对于本领域的一般技术人员，依据本发明的思想，在具体实施方式及应用范围上均会有改变之处，综上所述，本说明书内容不应理解为对本发明的限制。

Claims (11)

  1. 一种中医智能诊断方法，其特征在于，包括：
    服务器端从分布式客户端集群中获取患者的望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据；
    所述服务器端利用所述望诊训练数据、所述闻诊训练数据、所述问诊训练数据和所述切诊训练数据，对基于深度神经网络算法构建的待训练模型进行训练，得到训练后模型；
    所述服务器端利用所述训练后模型对患者的疾病数据进行诊断，得出所述疾病数据的诊断结果。
  2. 根据权利要求1所述的方法，其特征在于，所述方法具体使用分布式客户端服务器架构或云计算架构。
  3. 根据权利要求1所述的方法，其特征在于，所述利用所述望诊训练数据、所述闻诊训练数据、所述问诊训练数据和所述切诊训练数据，对基于深度神经网络算法构建的待训练模型进行训练，得到训练后模型的过程，包括：
    利用所述望诊训练数据，对基于卷积神经网络算法构建的望诊待训练模型进行训练，得到望诊训练后模型；
    利用所述闻诊训练数据，对基于BP神经网络算法构建的闻诊待训练模型进行训练，得到闻诊训练后模型；
    利用所述问诊训练数据，对基于BP神经网络算法构建的问诊待训练模型进行训练，得到问诊训练后模型；
    利用所述切诊训练数据，对基于深度神经网络算法构建的切诊待训练模型进行训练，得到切诊训练后模型。
  4. 根据权利要求1所述的方法，其特征在于，所述利用所述望诊训练数据、所述闻诊训练数据、所述问诊训练数据和所述切诊训练数据，对基于深度神经网络算法构建的待训练模型进行训练，得到训练后模型的过程，包括：
    利用所述望诊训练数据，对基于卷积神经网络算法构建的望诊待训练模型进行训练，得到望诊训练后模型；
    利用所述闻诊训练数据，对基于BP神经网络算法构建的闻诊待训练模型进行训练，得到闻诊训练后模型；
    利用所述问诊训练数据，对基于BP神经网络算法构建的问诊待训练模型进行训练，得到问诊训练后模型；
    利用所述切诊训练数据，对基于深度神经网络算法构建的切诊待训练模型进行训练，得到切诊训练后模型；
    利用提取到的所述望诊训练后模型、所述闻诊训练后模型、所述问诊训练后模型和所述切诊训练后模型的输出端数据，对基于概率神经网络算法构建的待训练模型进行训练，得到训练后模型。
  5. 根据权利要求1所述的方法，其特征在于，还包括：
    对所述望诊训练数据进行去噪和/或平滑处理。
  6. 根据权利要求1所述的方法，其特征在于，还包括：
    对所述闻诊训练数据进行滤波和/或分帧处理。
  7. 根据权利要求1至6任一项所述的方法，其特征在于，所述利用所述训练后模型对患者的疾病数据进行诊断，得出所述疾病数据的诊断结果的过程之前，还包括：
    利用新训练数据，对所述训练后模型进行优化，以提高所述训练后模型的精度；
    其中，所述新训练数据为对患者诊断结果验证通过后得到的疾病数据。
  8. 一种中医智能诊断系统，其特征在于，包括：
    数据获取模块，用于服务器端从分布式客户端集群中获取患者的望诊训练数据、闻诊训练数据、问诊训练数据和切诊训练数据；
    模型构建模块，用于所述服务器端利用所述望诊训练数据、所述闻诊训练数据、所述问诊训练数据和所述切诊训练数据，对基于深度神经网络算法构建的待训练模型进行训练，得到训练后模型；
    诊断结果获取模块，用于所述服务器端利用所述训练后模型对患者的疾病数据进行诊断，得出所述疾病数据的诊断结果。
  9. 一种中医医疗系统，其特征在于，包括权利要求8所述的中医智能诊断系统，还包括：
    中医智能治疗系统，用于利用所述中医智能诊断系统得到的诊断结果，确定出相应的治疗方案；
    其中，所述中医智能治疗系统为基于深度神经网络算法训练得到的治疗系统，对应的训练样本包括历史诊断结果以及相应的治疗方案。
  10. 根据权利要求9所述的中医医疗系统，其特征在于，所述中医智能治疗系统确定出的治疗方案包括中成药处方和/或理疗方案。
  11. 根据权利要求9所述的中医医疗系统，其特征在于，训练所述中医智能治疗系统的深度神经网络算法包括卷积神经网络算法。
PCT/CN2017/116065 2017-07-31 2017-12-14 一种中医智能诊断方法、系统及中医医疗系统 WO2019024380A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/634,135 US20200211706A1 (en) 2017-07-31 2017-12-14 Intelligent traditional chinese medicine diagnosis method, system and traditional chinese medicine system

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710639260.7 2017-07-31
CN201710639260.7A CN107451407A (zh) 2017-07-31 2017-07-31 一种中医智能诊断方法、系统及中医医疗系统

Publications (1)

Publication Number Publication Date
WO2019024380A1 true WO2019024380A1 (zh) 2019-02-07

Family

ID=60489851

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/116065 WO2019024380A1 (zh) 2017-07-31 2017-12-14 一种中医智能诊断方法、系统及中医医疗系统

Country Status (3)

Country Link
US (1) US20200211706A1 (zh)
CN (1) CN107451407A (zh)
WO (1) WO2019024380A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111724895A (zh) * 2020-05-16 2020-09-29 张东 一种基于人工智能的个性化中医诊疗机器人系统
CN113555106A (zh) * 2020-04-23 2021-10-26 浙江远图互联科技股份有限公司 一种基于生成对抗网络的智能中医远程辅助诊疗平台
CN117786590A (zh) * 2023-12-01 2024-03-29 上海源庐加佳信息科技有限公司 以大语言模型为先验的智能中医系统

Families Citing this family (35)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107451407A (zh) * 2017-07-31 2017-12-08 广东工业大学 一种中医智能诊断方法、系统及中医医疗系统
CN108231189B (zh) * 2017-12-12 2022-12-09 华南师范大学 基于双深度神经学习网络的数据处理方法和疾病诊断装置
CN107887024A (zh) * 2017-12-14 2018-04-06 华子昂 基于中医机器人的智能诊断系统及中医机器人
CN108836261A (zh) * 2018-03-23 2018-11-20 北京松投科技有限公司 智能中医脉诊方法及系统
KR102637339B1 (ko) * 2018-08-31 2024-02-16 삼성전자주식회사 음성 인식 모델을 개인화하는 방법 및 장치
CN109330846B (zh) * 2018-09-28 2021-03-12 哈尔滨工业大学 一种基于深度学习算法的空气波压力按摩仪参数优化方法
CN109431463B (zh) * 2018-10-23 2022-08-02 南开大学 基于中西医样本标记的深度学习中医智能诊疗系统
CN109524110A (zh) * 2018-11-30 2019-03-26 广州华久信息科技有限公司 一种望诊和问诊合参的中医体质辨识方法及设备
CN109346179B (zh) * 2018-12-10 2022-08-26 山东管理学院 一种冠心病介入术后复发预测模型及其建模方法、装置
CN109599164A (zh) * 2018-12-10 2019-04-09 北京唐冠天朗科技开发有限公司 一种信息处理方法及系统
CN109770875A (zh) * 2019-03-27 2019-05-21 上海铀米机器人科技有限公司 一种基于神经网络分类器的人体体质辨识方法及系统
CN110277147A (zh) * 2019-04-25 2019-09-24 胡盛寿 一种实现病案诊断智能化编目的系统及方法
CN110136839B (zh) * 2019-05-14 2021-10-08 北京百度网讯科技有限公司 一种症状信息处理方法、装置和电子设备
CN110115563A (zh) * 2019-05-22 2019-08-13 成都中医药大学 一种中医证型预测系统
CN110327023B (zh) * 2019-07-11 2022-01-11 广东工业大学 一种压力传感器阵列及脉诊装置
CN110619956A (zh) * 2019-09-18 2019-12-27 广州西思数字科技有限公司 一种中医智能机器人
CN111445995A (zh) * 2020-03-30 2020-07-24 崔书克 基于中医大数据分析的人工智能辅助诊断方法及系统
CN111640496B (zh) * 2020-05-16 2024-03-12 张东 一种基于人工智能的中医远程会诊系统
CN111667917A (zh) * 2020-06-10 2020-09-15 北京小白世纪网络科技有限公司 一种基于神经网络实现中医诊疗的方法、系统及设备
CN112133385A (zh) * 2020-09-14 2020-12-25 黄军彩 一种问诊信息采集分析系统及采集分析方法
CN112185548B (zh) * 2020-09-25 2022-10-28 智慧中医科技(广东)有限公司 一种基于神经网络算法的智能中医诊断方法及装置
CN112652373B (zh) * 2020-12-09 2022-10-25 华南理工大学 一种基于舌头图像检索的处方推荐方法
CN112652404A (zh) * 2020-12-30 2021-04-13 华南师范大学 基于大数据人工智能的中医自助辩证监测方法和机器人
CN112750529A (zh) * 2020-12-31 2021-05-04 平安科技（深圳）有限公司 智能医疗问诊装置、设备及介质
CN112750531A (zh) * 2021-01-21 2021-05-04 广东工业大学 一种中医自动化望诊系统、方法、设备和介质
CN113017579A (zh) * 2021-03-18 2021-06-25 湖南翰坤实业有限公司 智能把脉问诊仪
CN113488126A (zh) * 2021-07-27 2021-10-08 心医国际数字医疗系统(大连)有限公司 信息处理方法、装置、电子设备及存储介质
CN113724898B (zh) * 2021-08-31 2024-05-07 平安科技（深圳）有限公司 智能问诊方法、装置、设备及存储介质
WO2023210219A1 (ja) * 2022-04-28 2023-11-02 Tdk株式会社 情報処理装置、情報処理方法、及びプログラム
WO2023210217A1 (ja) * 2022-04-28 2023-11-02 Tdk株式会社 情報処理装置、情報処理方法、及びプログラム
CN117012339A (zh) * 2023-06-20 2023-11-07 安徽中医药大学 一种基于人工智能的中医诊断数据分析系统
CN116821779B (zh) * 2023-08-31 2023-11-14 中南大学湘雅医院 一种针对胃肠健康的大数据识别方法
CN117612694B (zh) * 2023-12-04 2024-06-25 西安好博士医疗科技有限公司 一种基于数据反馈的热疗机数据识别方法和系统
CN117423451B (zh) * 2023-12-19 2024-05-03 菏泽德康医学检验所有限公司 一种基于大数据分析的智能分子诊断方法及系统
CN118335337A (zh) * 2024-04-18 2024-07-12 中日友好医院(中日友好临床医学研究所) 一种中医辅助问诊模型的构建方法及装置

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102298662A (zh) * 2010-06-24 2011-12-28 上海中医药大学 基于自适应谐振理论的神经网络中医证候诊断系统
CN105912846A (zh) * 2016-04-07 2016-08-31 南京小网科技有限责任公司 基于云计算和医学知识库技术的智能医疗辅助决策系统
CN106202943A (zh) * 2016-07-14 2016-12-07 京东方科技集团股份有限公司 一种健康分析设备、服务器及系统
CN106295139A (zh) * 2016-07-29 2017-01-04 姹ゅ钩 一种基于深度卷积神经网络的舌体自诊健康云服务系统
CN106951688A (zh) * 2017-03-01 2017-07-14 新绎健康科技有限公司 一种中医四诊分型辨证系统
CN107451407A (zh) * 2017-07-31 2017-12-08 广东工业大学 一种中医智能诊断方法、系统及中医医疗系统

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
RU2286711C2 (ru) * 2000-02-14 2006-11-10 Фёрст Опинион Корпорэйшн Система и способ автоматической диагностики
US20100280350A1 (en) * 2009-05-02 2010-11-04 Xinyu Zhang Chinese medicine tele-diagnostics and triage system
US9275327B2 (en) * 2010-07-28 2016-03-01 Herbminers Informatics Limited AI for relating herbal ingredients to illnesses classified in traditional chinese medicine/TCM using probabilities and a relevance index
US20170308981A1 (en) * 2016-04-22 2017-10-26 New York University Patient condition identification and treatment
CA3039182C (en) * 2016-11-07 2021-05-18 Equifax Inc. Optimizing automated modeling algorithms for risk assessment and generation of explanatory data

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102298662A (zh) * 2010-06-24 2011-12-28 上海中医药大学 基于自适应谐振理论的神经网络中医证候诊断系统
CN105912846A (zh) * 2016-04-07 2016-08-31 南京小网科技有限责任公司 基于云计算和医学知识库技术的智能医疗辅助决策系统
CN106202943A (zh) * 2016-07-14 2016-12-07 京东方科技集团股份有限公司 一种健康分析设备、服务器及系统
CN106295139A (zh) * 2016-07-29 2017-01-04 姹ゅ钩 一种基于深度卷积神经网络的舌体自诊健康云服务系统
CN106951688A (zh) * 2017-03-01 2017-07-14 新绎健康科技有限公司 一种中医四诊分型辨证系统
CN107451407A (zh) * 2017-07-31 2017-12-08 广东工业大学 一种中医智能诊断方法、系统及中医医疗系统

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113555106A (zh) * 2020-04-23 2021-10-26 浙江远图互联科技股份有限公司 一种基于生成对抗网络的智能中医远程辅助诊疗平台
CN111724895A (zh) * 2020-05-16 2020-09-29 张东 一种基于人工智能的个性化中医诊疗机器人系统
CN117786590A (zh) * 2023-12-01 2024-03-29 上海源庐加佳信息科技有限公司 以大语言模型为先验的智能中医系统

Also Published As

Publication number Publication date
CN107451407A (zh) 2017-12-08
US20200211706A1 (en) 2020-07-02

Similar Documents

Publication Publication Date Title
WO2019024380A1 (zh) 一种中医智能诊断方法、系统及中医医疗系统
EP4002271A1 (en) Image segmentation method and apparatus, and storage medium
Oh et al. Classification of heart sound signals using a novel deep WaveNet model
CN105117611B (zh) 基于卷积神经元网络的中医舌诊模型的确定方法及系统
CN106709254B (zh) 一种医疗诊断机器人系统
CN110459328B (zh) 临床监护设备
EP3783619A1 (en) Human body health assessment method and system based on sleep big data
CN107392109A (zh) 一种基于深度神经网络的新生儿疼痛表情识别方法
CN111316281A (zh) 基于机器学习的自然语言情境中数值数据的语义分类
CN110619322A (zh) 一种基于多流态卷积循环神经网络的多导联心电异常信号识别方法及系统
CN111009324B (zh) 脑网络多特征分析的轻度认知障碍辅助诊断系统及方法
CN109935336A (zh) 一种儿童呼吸科疾病的智能辅助诊断方法及诊断系统
Yang et al. Large-scale brain functional network integration for discrimination of autism using a 3-D deep learning model
CN116098621B (zh) 一种基于注意力机制的情绪面孔及生理反应识别方法
CN114748053A (zh) 一种基于fMRI高维时间序列的信号分类方法及装置
Deshpande et al. Automated detection of Diabetic Retinopathy using VGG-16 architecture
CN113990502B (zh) 一种基于异构图神经网络的icu心衰预测系统
CN111276242A (zh) 一种针对医院重症监护室患者疾病诊断与病情状态评估建模方法
Prasad et al. Detection of ckd from ct scan images using knn algorithm and using edge detection
CN117373595A (zh) 一种基于ai的内科患者个性化治疗方案生成系统
CN112712895A (zh) 一种针对2型糖尿病并发症的多模态大数据的数据分析方法
Kumar et al. Analysis of CNN model based classification of diabetic retinopathy diagnosis
CN116843036A (zh) 面向阿尔兹海默症辅助诊断的个性化联邦学习方法
WO2024108483A1 (zh) 多模态神经生物信号处理方法、装置、服务器及存储介质
Deepika et al. Design of hybrid classifier for prediction of diabetes through feature relevance analysis

Legal Events

Date Code Title Description
NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 17919695

Country of ref document: EP

Kind code of ref document: A1