CN113192598A

CN113192598A - Efficacy prediction method and system for taVNS therapy

Info

Publication number: CN113192598A
Application number: CN202110142508.5A
Authority: CN
Inventors: 赵亚楠; 岳真锐; 贾叶奇; 荣培晶
Original assignee: INSTITUTE OF ACUPUNCTURE AND MOXIBUSTION CHINA ACADEMY OF CHINESE MEDICAL SCIENCES
Current assignee: INSTITUTE OF ACUPUNCTURE AND MOXIBUSTION CHINA ACADEMY OF CHINESE MEDICAL SCIENCES
Priority date: 2021-02-02
Filing date: 2021-02-02
Publication date: 2021-07-30

Abstract

The invention discloses a method and a system for predicting the curative effect of a taVNS therapy. The method comprises the following steps: preprocessing the acquired multi-dimensional data relating to the patient; extracting data features from the preprocessed data through convolution operation; inputting the extracted data characteristics into a prediction model for prediction; and processing the prediction result output by the prediction model through a multilayer perceptron to obtain a simplified prediction result. The method comprises the steps of establishing a curative effect prediction model according to multi-dimensional neuroelectrophysiological characterization, disease specific clinical characterization and quantifiable traditional Chinese medicine characteristic indexes of a patient as features, finally fitting a prediction model of people who are effective in taVNS therapy, helping doctors judge whether the patient is effective after treatment by the taVNS and predict the taVNS improvement disease degree in clinical decision, helping non-medical staff to predict the taVNS curative effect to a greater extent, and better achieving home intelligent self-care medical treatment.

Description

Efficacy prediction method and system for taVNS therapy

Technical Field

The invention relates to the technical field of big data and medicine, in particular to a method and a system for predicting the curative effect of a taVNS therapy.

Background

The auricular point therapy is an important component of the traditional Chinese medicine acupuncture, is closely related to meridians and viscera of the whole body, is called as the convergence of the ear as a zoned pulse, is developed on the basis of auricular points, and has remarkable curative effects on treating epilepsy, insomnia, depression, functional gastrointestinal diseases, Parkinson and cerebrovascular diseases by regulating the central-autonomic nerve function and restoring homeostasis balance in an organism through a percutaneous auricular vagus nerve stimulation technology (taVNS), thereby providing a novel therapy which is safe, noninvasive, comfortable, wearable and easy to operate and can realize household treatment for an intelligent, convenient and efficient medical mode under the modern age background, and being gradually popularized and used in clinic.

The appropriate population for any treatment cannot be considered in any case. Because the pathogenesis of diseases such as insomnia disorder and depression is complex and is related to a plurality of factors such as pressure, work and rest, behavior cognition and the like, although the taVNS treatment has undeniable clinical effect, the effect is quite different among individuals, which is probably caused by heterogeneity among individuals and different physiological and pathological mechanisms of different disease subtypes. Due to the ambiguity of clinical manifestations of different diseases, the subjectivity of the traditional four diagnostic methods and the empirical nature of doctor's syndrome differentiation, there may be a deviation in the doctor's grasp of the patient's body state and the choice of treatment methods, which further affects the clinical efficacy. This limits the precise treatment and further application and popularization of taVNS to a certain extent. Therefore, before treatment of taVNS, a characteristic capable of predicting response of the organism state to the taVNS from the individual level is found, and the method has positive significance for improving clinical curative effect and assisting precise treatment of the taVNS.

With the coming of the precise medical age, the artificial intelligence is utilized to predict the curative effect, thereby providing new possibility for medical treatment. In 2019, the research of Nature Medicine's artificial intelligence special journal series shows that the clinical work efficiency can be remarkably improved by assisting the diagnosis and treatment of clinicians through machine learning and artificial intelligence technology, and a synergistic effect of 1+1 > 2 is generated. Because the traditional curative effect prediction scheme depends on the subjective comprehensive cognition of medical personnel on the medical mechanism of the disease, the speciality and the subjectivity are strong, and the application range is limited. Therefore, it is important to objectively and accurately identify the body state and the degree of response to a therapeutic method to improve clinical efficacy. In order to solve the above problems, in recent years, disease prediction schemes based on artificial intelligence techniques such as machine learning and deep learning have been proposed in succession, so that disease efficacy prediction is more intelligent, accurate and efficient. Although the existing curative effect prediction scheme based on the artificial intelligence technology solves the limitation problem of the traditional scheme, the traditional scheme still has the problem of being not negligible. Firstly, the method mainly aims at the difficult and complicated diseases such as malignant tumor and the like to carry out curative effect prediction, and lacks of curative effect prediction on common diseases such as epilepsy, insomnia disorder, depression, functional gastrointestinal diseases, Parkinson, cerebrovascular diseases and the like. Secondly, data analysis is emphasized, so that critical medical information of diseases is easy to ignore, the information acquisition pertinence is not strong, or the information is only single-dimensional information, the treatment state cannot be reflected comprehensively, and the prediction precision is difficult to achieve expectation. Thirdly, the traditional machine learning has limited fitting ability to high-dimensional nonlinear data, relatively high computational power requirement on equipment and weak generalization ability, so that the cost is correspondingly increased, and reliable prediction is difficult to obtain, so that the traditional machine learning is difficult to widely apply. Fourthly, the focus is on historical disease record analysis, so that the real-time performance is not high.

Disclosure of Invention

The technical problem to be solved by the invention is to provide a method and a system for predicting the curative effect of a taVNS therapy, which fit a prediction model of a population with effectiveness in the taVNS therapy and can help a doctor to judge whether a patient is effective after treatment by the taVNS and predict the improvement degree of the taVNS in clinical decision.

In order to solve the technical problems, the invention adopts the following technical scheme:

in one aspect, the invention features a method for predicting efficacy of a taVNS therapy. The efficacy prediction method for taVNS therapy comprises: preprocessing the acquired multi-dimensional data relating to the patient; extracting data features from the preprocessed data through convolution operation; inputting the extracted data characteristics into a prediction model for prediction; and processing the prediction result output by the prediction model through a multilayer perceptron to obtain a simplified prediction result.

Optionally, for the method of predicting efficacy of the taVNS therapy, the prediction model comprises a two-layer bidirectional long-short term memory model.

Optionally, for the method for predicting efficacy of taVNS therapy, the inputting the extracted data features into a prediction model for prediction includes: respectively carrying out forward sequence propagation and reverse sequence propagation on the preprocessed data through two layers of bidirectional long and short term memory models; and connecting the data obtained after the forward sequence propagation with the data obtained after the reverse sequence propagation to obtain output data.

Optionally, for the method of predicting the efficacy of a therapy of taVNS, the long-short term memory model comprises an activation function Tanh and an activation function Sigmoid.

Optionally, for the method of predicting efficacy of the taVNS therapy, the multi-layered sensor comprises a fully connected layer.

Optionally, for the efficacy prediction method of the taVNS therapy, the multi-layered perceptron comprises an activation function Relu and an activation function Sigmoid.

Optionally, for the method for predicting efficacy of taVNS therapy, the prediction model is obtained by training a long-short term memory model, the model training process comprising: preprocessing the acquired multi-dimensional dataset relating to the patient; labeling the preprocessed multi-dimensional data set to obtain a sample set, and dividing the sample set into a training set and a verification set; extracting data features from the training set by convolution operation; inputting the extracted data characteristics into a two-layer bidirectional long and short term memory model for forward propagation; obtaining a result through a multilayer perceptron; and calculating the loss of the long-term and short-term memory model by using the output result and the sample label, obtaining an updating gradient through an error back propagation algorithm, updating the model through a self-adaptive moment estimation optimizer, and selecting an optimal model through the model prediction accuracy rate obtained after the verification set test in the training process.

Optionally, for the method of efficacy prediction for taVNS therapy, the patient-related multidimensional data comprises at least some of the following data: demographic information, disease type, disease severity, traditional Chinese medical constitution typing, traditional Chinese medical syndrome differentiation typing, traditional Chinese medical meridian syndrome differentiation, mental state, electroencephalogram, and heart rate variability information.

In another aspect, the invention features a system for predicting efficacy of a taVNS therapy. The efficacy prediction system for taVNS therapy comprises: a preprocessing unit for preprocessing the acquired multi-dimensional data relating to the patient; the characteristic extraction unit is used for extracting data characteristics from the preprocessed data through convolution operation; the prediction unit is used for inputting the extracted data characteristics into a prediction model for prediction; and the output simplifying unit is used for processing the prediction result output by the prediction model through the multilayer perceptron to obtain a simplified prediction result.

In yet another aspect, the present invention is directed to a computing device. The computing device includes: one or more processors, and a memory coupled with the one or more processors, the memory storing instructions that, when executed by the one or more processors, cause the one or more processors to perform the above-described method.

In yet another aspect, the present invention is directed to a machine-readable storage medium. The machine-readable storage medium stores executable instructions that, when executed, cause the machine to perform the above-described method.

Compared with the prior art, the technical scheme of the invention has the following main advantages:

according to the efficacy prediction method and system of the taVNS therapy, provided by the embodiment of the invention, the disease characteristics are extracted through a depth algorithm according to the multidimensional information of the patient, and the prediction model of the taVNS therapy effective population is finally fitted, so that a doctor can be helped to judge whether the patient is effective after treatment by the taVNS and predict the taVNS improvement disease degree in clinical decision, the efficacy prediction of non-medical personnel at home is helped to a greater extent, and the home intelligent self-care medical treatment is better realized. The classification of patients can realize the accurate treatment of taVNS, improve the clinical efficacy, indicate that other treatment schemes are likely to be more useful if the predicted response is low, save medical resources and avoid ineffective treatment.

The method and the system for predicting the curative effect of the taVNS therapy introduce a novel model based on two-layer bidirectional long-term and short-term memory (LSTM) to efficiently and accurately process various input sources to judge the curative effect, and the model has the characteristics of small quantity, high accuracy and good generalization degree. By using the deep learning method, an efficient and accurate characteristic analysis model for predicting the effectiveness of the taVNS based on multiple signal sources can be constructed, so that the disease prediction is more intelligent, and the accurate treatment of the taVNS is improved.

Drawings

Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the invention. Also, like reference numerals are used to refer to like parts throughout the drawings. In the drawings:

FIG. 1 is a flow chart of a method for predicting efficacy of a taVNS therapy provided by an embodiment of the present invention;

FIG. 2 is a schematic structural diagram of a complete deep learning model provided by an example of the present invention;

FIG. 3 is a flow chart of formula equivalence of a Long Short Term Memory (LSTM) model provided in accordance with an example of the present invention;

FIG. 4 is a flow chart of a model training process provided by one example of the present invention;

FIG. 5 is a schematic diagram of a efficacy prediction system for taVNS therapy according to another embodiment of the present invention;

fig. 6 is a block diagram of a computing device for efficacy prediction processing for taVNS therapy, according to an embodiment of the invention.

Detailed Description

Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

Fig. 1 is a flowchart of a method for predicting efficacy of a taVNS therapy according to an embodiment of the present invention. The efficacy prediction method of the taVNS therapy is to study common diseases, which have high incidence and are closely related to brain dysfunction, such as epilepsy, insomnia, depression, functional gastrointestinal diseases, Parkinson's disease and cerebrovascular disease.

As shown in fig. 2, the model for processing the input signal in this embodiment is a new deep learning model based on Convolution (Conv), Long Short-Term Memory (LSTM), and multi-layer Perceptron (MLP), and this structure has the characteristics of controllable volume, reliable maturity, high generalization degree, and high prediction accuracy.

As shown in fig. 1 and 2, the acquired multi-dimensional data relating to the patient is preprocessed in step S110.

As a specific example, in the preprocessing step, before all input ports enter the prediction model, the port data is firstly resampled from the initial 512Hz to 16Hz to increase the data throughput capacity of the model, and then the inputs of different ports are respectively normalized to make each signal end input conform to the Gaussian distribution of N (0,1), thereby facilitating the further processing and training of the model on the input data. After this, the measurement data of 16Hz is split into 480 length samples, 480 length being equivalent to 30 seconds of measurement data, i.e. individual samples lasting 30 seconds, since 16Hz is sampled 16 times per second, in order to increase the number of samples and avoid the disappearance of gradients and explosions caused by over-long sequences, a single sample may contain aligned normalized input data at that length.

In step S120, data features are extracted from the preprocessed data by a convolution (Conv) operation.

The preprocessed data can be subjected to a feature extraction operation, which is implemented by a convolutional layer, a one-dimensional convolutional layer with the same number of input ends and the same size as the measured data and with 128 output ends is initialized, so that the convolutional kernel size is 3, the step is 1 and 0 Padding (Padding), after the operation, the input signal is converted into intermediate data with the length of 478 bits and the number of ports of 128, and the purpose of the step is to promote a data feature channel and retain low-dimensional data features, so that the further feature conversion and data forward propagation are facilitated.

In step S130, the extracted data features are input into a prediction model for prediction.

As an alternative embodiment, the predictive model may include a two-layer two-way Long Short-Term Memory (LSTM) model, as shown in FIG. 3 and equation 1. The LSTM model may include an activation function Tanh and an activation function Sigmoid. After data preprocessing, an LSTM forward propagation step may be performed, in which a two-layer bidirectional LSTM model with an input and hidden size of 128 is initialized, and then each bit number in the preprocessed data is separately calculated twice step by step through equation 1 in a positive sequence and a reverse sequence, respectively, that is, the forward propagation of the two-layer bidirectional LSTM. Wherein, the hidden size refers to the size of the intermediate variable, i.e. the size of the internal variable of the model. In fig. 3 and equation 1, tanh and σ are activation functions; x is input data of each step; h and C are hidden variables, which are equivalent to the output of the previous step; i, f and o are intermediate variables, and W and U are learnable model parameters. The LSTM forward propagation includes forward propagation and reverse propagation, taking 478-bit input data as an example, the forward propagation refers to sequential calculation from bit 1 to bit 478, the reverse propagation refers to sequential calculation from bit 478 to bit 1, each bit calculation includes all steps in formula 1 in turn, and then the data obtained after forward propagation and the data obtained after reverse propagation are aligned and concatenated according to 478-bit input arrangement, so as to obtain 478-bit output data with 256 ports.

In step S140, the prediction result output by the prediction model is processed by a Multilayer Perceptron (MLP) to obtain a simplified prediction result.

As an alternative embodiment, the multiple layers of the multilayer sensor may be fully connected. The multilayer perceptron may include an activation function Relu and an activation function Sigmoid. The multi-layer perceptron is used for processing output data of the prediction model, and mapping a plurality of input data ports into a single output scalar so as to finally obtain a judgment whether the model is effective for treatment. Firstly, two full connection layers (Linear) are initialized, the data length 478 bits and the characteristic dimension 256 are respectively reduced to 1, the output data length 478 bits are firstly reduced to 1 bit and activated through a Relu function, then the characteristic dimension 256 is reduced to 1, and the output result is limited between (0,1) through a Sigmoid function, which is the final judgment of the model on the input data. If the output is greater than 0.5, the treatment is judged to be effective, otherwise, the treatment is not effective.

Taking a real-time electroencephalogram signal as an example, based on the real-time electroencephalogram signal, the effective probability calculated by the deep learning model can be directly combined with information of other dimensions, such as at least one of information of disease type, disease severity, traditional Chinese medicine constitution typing, traditional Chinese medicine syndrome differentiation typing, traditional Chinese medicine meridian syndrome differentiation, mental state, heart rate variability and the like, and an integrated linear classification model, such as a linear regression, a linear support vector machine or a random forest model, is quickly constructed by an integrated learning method, so that the reliability and accuracy of the model are further improved.

In the embodiment, Chinese medicine syndrome differentiation elements are organically fused, and information collection is carried out by combining subjective and objective indexes with multiple dimensions. The input may comprise at least some of the following data: demographic information, disease types, disease severity, traditional Chinese medicine constitution typing, traditional Chinese medicine syndrome differentiation typing, traditional Chinese medicine meridian syndrome differentiation, mental state, electroencephalogram, heart rate variability and other multi-dimensional clinical information. Demographic information may include gender and age. In the case of insomnia disorder, the insomnia type may include a difficulty-in-falling type, a difficulty-in-sleep maintenance type, a early wake type, a daytime dysfunction type, a dreaminess type, and a mixed type. Insomnia severity can include light, moderate, and heavy. The traditional Chinese medicine constitution types can comprise mild nature, qi deficiency nature, yang deficiency nature, yin deficiency nature, phlegm dampness nature, damp-heat nature, blood stasis nature, qi stagnation nature and specific endowment nature. The five zang organs spirit type in the traditional Chinese medicine can include heart failure to store spirit, liver failure to store spirit, kidney failure to store mind, lung failure to store spirit and spleen failure to store spirit. Differentiation of syndromes according to the meridians in TCM may include differentiation of syndromes according to the twelve main meridians. Mental states may include anxiety and depression. The brain electricity may include: delta wave with frequency of 1-3 Hz and amplitude of 20-200 μ V; theta wave with frequency of 4-7 Hz and amplitude of 5-20 μ V; alpha wave with frequency of 8-13 Hz (average 10Hz) and amplitude of 20-100 μ V; and beta wave with frequency of 14-30 Hz and amplitude of 100-150 μ V. The HRV information includes: average heart rate (mean HR), abnormal heart rate, RR interval Standard Deviation (SDNN), root mean square of adjacent RR interval difference (RMSSD), pNN50, standard deviation of difference between whole-stroke adjacent NN interval lengths (SDSD), total energy (TP), Very Low Frequency (VLF), Low Frequency (LF), High Frequency (HF), low frequency standard value (n.u.lf), high frequency standard value (n.u.hf), and low frequency/high frequency ratio (LF/HF). The information collected by this embodiment is real-time electroencephalography and/or HRV signals, which can better reflect the current status of the treatment.

The reason why the real-time electroencephalogram signal is used as the prediction index of the curative effect is that electroencephalograms (EEG) have the advantages of being non-invasive, economical and practical, and high in time resolution. EEG reflects spontaneous electrical brain activity in the cerebral cortex, where different brain regions have different electrical brain activity, which represents that the cerebral cortex is in different functional states. There is a substantial body of evidence supporting the use of EEG as a biomarker for the treatment of various neuropsychiatric diseases, which provides real-time neurophysiological correlations to guide the procedure. The technology predicts the response degree of the brain dysfunction diseases such as insomnia, depression and the like to taVNS treatment by monitoring the real-time EEG dynamic change of the cerebral cortex.

The output end is the effectiveness and the ineffectiveness of the taVNS treatment method for insomnia. And finally, constructing a curative effect prediction model by using deep learning.

The prediction model can be obtained by training a long-short term memory (LSTM) model, and the model training process mainly comprises the following steps: data preprocessing, feature extraction, LSTM forward propagation, and MLP post-processing. As a specific example, as shown in fig. 4, the model training process includes the following steps:

step S410, preprocessing the acquired multidimensional data set.

And step S420, labeling the preprocessed multi-dimensional data set to obtain a sample set, and dividing the sample set into a training set and a verification set. A single sample may contain aligned normalized input data at a preset length and a label for that sample, i.e., a label of whether the treatment is effective.

Step S430, extracting data features from the training set by convolution operation.

Step S440, inputting the extracted data characteristics into a two-layer bidirectional long-short term memory model for forward propagation.

And S450, obtaining a result through the multilayer perceptron.

Step S460, calculating loss of the long-term and short-term memory model by using the output result and the sample label, obtaining an updating gradient through a back propagation (backspace) algorithm, calculating model error through a loss function after the model is propagated forwards, calculating partial derivative values of all parameters layer by layer from back to front through a partial differential equation and a chain rule, updating the model through an adaptive moment estimation (Adam) optimizer, and selecting an optimal model through model prediction accuracy obtained after a verification set test in a training process.

The model is updated by using an Adam optimizer, which is an updating model method for calculating the adaptive learning rate of parameters, and the learning rate is calculated in a momentum-like mode so as to avoid the unsteady state problem of sparse gradient. Adam optimizers are well suited to solving problems with large scale data or parameters. When initializing the Adam optimizer, a learning rate of 0.0001 is used and a loss function is defined as Cross Entropy (Cross Entropy). And updating the model by using the training set during training and saving the optimal model by the verification set. In forward propagation, signal side features are first extracted using convolutional layers, then input into a two-layer bi-directional LSTM model, and then the results are obtained by a multi-layer perceptron. When the model is trained, the output and the sample label are used for calculating model loss, an updating gradient is obtained through a back propagation algorithm, the model is updated through an Adam optimizer, all data are integrated and then trained, for example, 2000 cycles are carried out, each cycle means that all training set data are used for training the model once, and the model with the highest prediction accuracy is stored through a verification set after the cycle. And testing on the verification set after each cycle is finished, covering the models stored before if the prediction accuracy of the current model obtained by the cycle after the test of the verification set is higher than that of the stored models, otherwise, not storing the current model, so that the model finally stored is the optimal model on the verification set in 2000 cycles.

As an alternative embodiment, after the model training process, a model testing process may also be included. Sample data is divided into a training set, a verification set and a test set. And updating the model by using the training set during model training, storing the optimal model by using the verification set, and finally analyzing the effect of the model in the test set. Only forward propagation is required in the course of testing the model to arrive at the final predicted outcome, thereby inferring the therapeutic effectiveness of the taVNS therapy.

Fig. 5 is a schematic structural diagram of a efficacy prediction system for taVNS therapy according to another embodiment of the present invention. As shown in fig. 5, the efficacy prediction system 500 of the taVNS therapy provided by this embodiment includes a preprocessing unit 510, a feature extraction unit 520, a prediction unit 530, and an output simplification unit 540.

The pre-processing unit 510 is used for pre-processing the acquired multi-dimensional data relating to the patient. The operation of the preprocessing unit 510 may refer to the operation of step 110 described above with reference to fig. 1.

The feature extraction unit 520 is configured to extract data features from the preprocessed data through a convolution operation. The operation of the feature extraction unit 520 may refer to the operation of step 120 described above with reference to fig. 1.

The prediction unit 530 is configured to input the extracted data features into a prediction model for prediction. The operation of the prediction unit 530 may refer to the operation of step 130 described above with reference to fig. 1.

The output simplifying unit 540 is configured to process the prediction result output by the long-term and short-term memory model through a multi-layer sensor to obtain a simplified prediction result. The operation of the output reduction unit 540 may refer to the operation of step 140 described above with reference to fig. 1.

According to the efficacy prediction method and system of the taVNS therapy, provided by the embodiment of the invention, the disease characteristics are extracted through a depth algorithm according to the multi-dimensional neuroelectrophysiological characterization, the disease specific clinical characterization and the quantifiable traditional Chinese medicine characteristic indexes of a patient, and a prediction model of people who are effective in the taVNS therapy is finally fitted, so that a doctor can be helped to judge whether the patient is effective after being treated by the taVNS and predict the taVNS to improve the disease degree in clinical decision, the efficacy prediction of non-medical personnel at home is helped to a greater extent, and the home intelligent self-care medical treatment is better realized. The classification of patients can realize the accurate treatment of taVNS, improve the clinical efficacy, indicate that other treatment schemes are likely to be more useful if the predicted response is low, save medical resources and avoid ineffective treatment.

Furthermore, the efficacy prediction method and system of the taVNS therapy of the embodiment of the invention introduce a novel LSTM model based on 2 layers of two-way, so as to efficiently and accurately process various input sources to judge the treatment effect, and the model has the characteristics of small volume, high accuracy and good generalization degree. By using the deep learning method, an efficient and accurate characteristic analysis model for predicting the effectiveness of the taVNS based on multiple signal sources can be constructed, so that the disease prediction is more intelligent, and the accurate treatment of the taVNS is improved.

As shown in fig. 6, computing device 600 may include at least one processor 610, storage 620, memory 630, communication interface 640, and internal bus 650, and at least one processor 610, storage 620, memory 630, and communication interface 640 are connected together via bus 650. The at least one processor 610 executes at least one computer-readable instruction (i.e., an element described above as being implemented in software) stored or encoded in a computer-readable storage medium (i.e., the memory 620).

In one embodiment, computer-executable instructions are stored in the memory 620 that, when executed, cause the at least one processor 610 to perform: preprocessing the acquired multi-dimensional data relating to the patient; extracting data features from the preprocessed data through convolution operation; inputting the extracted data characteristics into a prediction model for prediction; and processing the prediction result output by the prediction model through a multilayer perceptron to obtain a simplified prediction result.

It should be appreciated that the computer-executable instructions stored in the memory 620, when executed, cause the at least one processor 610 to perform the various operations and functions described above in connection with fig. 1-5 in the various embodiments of the present invention.

In the present invention, computing device 600 may include, but is not limited to: personal computers, server computers, workstations, desktop computers, laptop computers, notebook computers, mobile computing devices, smart phones, tablet computers, cellular phones, Personal Digital Assistants (PDAs), handheld devices, messaging devices, wearable computing devices, consumer electronics, and so forth.

According to one embodiment, a program product, such as a non-transitory machine-readable medium, is provided. A non-transitory machine-readable medium may have instructions (i.e., elements described above as being implemented in software) that, when executed by a machine, cause the machine to perform various operations and functions described above in connection with fig. 1-5 in various embodiments of the invention.

Specifically, a system or apparatus may be provided which is provided with a readable storage medium on which software program code implementing the functions of any of the above embodiments is stored, and causes a computer or processor of the system or apparatus to read out and execute instructions stored in the readable storage medium.

In this case, the program code itself read from the readable medium can realize the functions of any of the above-described embodiments, and thus the machine-readable code and the readable storage medium storing the machine-readable code form part of the present invention.

Examples of the readable storage medium include floppy disks, hard disks, magneto-optical disks, optical disks (e.g., CD-ROMs, CD-R, CD-RWs, DVD-ROMs, DVD-RAMs, DVD-RWs), magnetic tapes, nonvolatile memory cards, and ROMs. Alternatively, the program code may be downloaded from a server computer or from the cloud via a communications network.

The above description is only an embodiment of the present invention, and not intended to limit the scope of the claims, and all equivalent structures or equivalent processes that are transformed by the content of the specification and the drawings, or directly or indirectly applied to other related technical fields are included in the scope of the claims.

Claims

1. A method of predicting efficacy of a taVNS therapy, comprising:

preprocessing the acquired multi-dimensional data relating to the patient;

extracting data features from the preprocessed data through convolution operation;

inputting the extracted data characteristics into a prediction model for prediction;

and processing the prediction result output by the prediction model through a multilayer perceptron to obtain a simplified prediction result.

2. The method of predicting efficacy of a taVNS therapy according to claim 1, wherein the predictive model comprises two layers of bi-directional long-short term memory models.

3. The method of predicting efficacy of a taVNS therapy as claimed in claim 2, wherein the inputting the extracted data features into a prediction model for prediction comprises:

respectively carrying out forward sequence propagation and reverse sequence propagation on the preprocessed data through two layers of bidirectional long and short term memory models;

and connecting the data obtained after the forward sequence propagation with the data obtained after the reverse sequence propagation to obtain output data.

4. The method of predicting efficacy of a taVNS therapy according to claim 1 or claim 2, wherein the long-short term memory model includes an activation function Tanh and an activation function Sigmoid.

5. The method of predicting efficacy of a taVNS therapy according to claim 1, wherein the multi-layered sensor comprises a fully-connected layer.

6. The method of predicting efficacy of a taVNS therapy according to claim 1 or claim 5, wherein the multi-tier perceptron comprises an activation function Relu and an activation function Sigmoid.

7. The method of predicting efficacy of a taVNS therapy as claimed in claim 1 or claim 2, wherein the predictive model is derived by training a long-short term memory model, the model training process comprising:

preprocessing the acquired multi-dimensional dataset relating to the patient;

labeling the preprocessed multi-dimensional data set to obtain a sample set, and dividing the sample set into a training set and a verification set;

extracting data features from the training set by convolution operation;

inputting the extracted data characteristics into a two-layer bidirectional long-short term memory model for forward propagation;

obtaining a result through a multilayer perceptron;

and calculating the loss of the long-term and short-term memory model by using the output result and the sample label, obtaining an updating gradient through an error back propagation algorithm, updating the model through a self-adaptive moment estimation optimizer, and selecting an optimal model through the model prediction accuracy rate obtained after the verification set test in the training process.

8. The method of predicting efficacy of a taVNS therapy according to claim 1, wherein the multi-dimensional patient-related data includes at least some of: demographic information, disease type, disease severity, traditional Chinese medical constitution typing, traditional Chinese medical syndrome differentiation typing, traditional Chinese medical meridian syndrome differentiation, mental state, electroencephalogram, and heart rate variability information.

9. A system for predicting efficacy of a taVNS therapy, comprising:

a preprocessing unit for preprocessing the acquired multi-dimensional data relating to the patient;

the characteristic extraction unit is used for extracting data characteristics from the preprocessed data through convolution operation;

the prediction unit is used for inputting the extracted data characteristics into a prediction model for prediction;

and the output simplifying unit is used for processing the prediction result output by the prediction model through the multilayer perceptron to obtain a simplified prediction result.

10. A computing device, comprising:

one or more processors, and

a memory coupled with the one or more processors, the memory storing instructions that, when executed by the one or more processors, cause the one or more processors to perform the method of any of claims 1-8.

11. A machine-readable storage medium having stored thereon executable instructions that, when executed, cause the machine to perform the method of any one of claims 1 to 8.