Traditional Chinese medicine syndrome type classification and probability prediction method based on artificial intelligence

Abstract

The invention relates to the technical field of traditional Chinese medicine syndrome type classification, in particular to a traditional Chinese medicine syndrome type classification and probability prediction method based on artificial intelligence, and the method comprises the steps: carrying out the normalization processing of scattered data, natural language text data and image data related to the syndromes of a patient, obtaining normalized data, and enabling a set of the normalized data to be a normalized data set; inputting the normalized data set into a trained syndrome type classification model based on Transform, and obtaining the probability of each syndrome type corresponding to the syndrome; the model sequentially comprises a patient information multi-mode embedding layer, a first self-attention layer and a first linear mapping + Softmax function calculation layer. After a normalized data set in the patient information multi-mode embedding layer is input into the first self-attention layer for attention calculation, the probability of each syndrome type is obtained through calculation of the function calculation layer; wherein the patient information multi-modal embedding layer is used for storing the normalized data set. And intelligent classification and probability prediction of the traditional Chinese medicine syndrome types are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of TCM syndrome classification, and in particular to a TCM syndrome classification and probability prediction method based on artificial intelligence. Background Art

[0002] In the field of traditional Chinese medicine, when diagnosing patient symptoms, doctors usually make a diagnosis based on their experience and then make a diagnosis. Purely relying on manual judgment of current symptoms is highly subjective and difficult to quantify and standardize the syndrome. In addition, doctors have different medical skills, so it is difficult to make consistent decisions even for the same syndrome of the same patient. Based on this, there is an urgent need for a method that can intelligently classify and probabilistically predict TCM syndromes. Summary of the invention

[0003] 1. Technical issues to be resolved

[0004] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a TCM syndrome classification and probability prediction method based on artificial intelligence, which solves the technical problems that when judging the syndrome type, it relies purely on manual judgment of the current symptoms, which is highly subjective and difficult to quantify and standardize the syndrome type. In addition, the medical skills of doctors are different, and it is difficult to make consistent decisions even for the same syndrome of the same patient.

[0005] (II) Technical solution

[0006] In order to achieve the above object, the main technical solutions adopted by the present invention include:

[0007] The embodiment of the present invention provides a method for TCM syndrome classification and probability prediction based on artificial intelligence, which is characterized by comprising:

[0008] S1. Normalizing scattered data, natural language text data, and image data related to patient symptoms to obtain normalized data, wherein a collection of each normalized data is a normalized data set;

[0009] S2, inputting the normalized data set into the trained Transformer-based syndrome classification model to obtain the probability of each syndrome corresponding to the syndrome;

[0010] The Transformer-based syndrome classification model includes: a patient information multimodal embedding layer, a first self-attention layer, and a first linear mapping + Softmax function calculation layer;

[0011] The normalized data set in the patient information multimodal embedding layer is input into the first self-attention layer for attention calculation, and then the probability of each syndrome type is calculated through the linear mapping + Softmax function calculation layer; wherein the patient information multimodal embedding layer is used to store the normalized data set.

[0012] Preferably, before S1, the method further includes:

[0013] S0. Using the historical normalized data to train the Transformer-based syndrome classification model to obtain a trained Transformer-based syndrome classification model.

[0014] Preferably, it includes:

[0015] The scattered data include: gender, age, examination index, and test index;

[0016] The natural language text data includes: medical history text and syndrome description text;

[0017] The image data includes: MRI image pictures, facial diagnosis pictures, and tongue diagnosis pictures.

[0018] Preferably, the S1 includes:

[0019] S10, performing vectorization processing on the scattered data related to the syndrome, the natural language text data, and the image data to obtain a vectorized data set;

[0020] S11. Normalize each vectorized data in the vectorized data set by adding type coding according to different types to form the normalized data.

[0021] Preferably, the S10 includes:

[0022] Performing linear mapping on the scattered data to obtain first vectorized data;

[0023] The natural language text data is converted into word embedding data by using word embedding, the word embedding data is position-encoded by a sine-cosine position encoding method in a Transformer, and feature extraction is performed by a second self-attention layer to obtain second vectorized data;

[0024] Performing convolution calculation on the image data through a convolution layer to obtain third vectorized data;

[0025] The first vectorized data, the second vectorized data, and the third vectorized data are combined into a vectorized data set.

[0026] Preferably, after S2, the method further includes:

[0027] S3, horizontally splicing the current or historical syndrome probability with the information vector in the current or historical patient information multimodal embedding layer as the patient information summary, and inputting it into the trained syndrome evolution prediction model to obtain the syndrome probability after or without specific treatment;

[0028] wherein the syndrome type evolution prediction model is trained by using the historical syndrome type probabilities and the information in the historical patient information multimodal embedding layer to obtain the trained syndrome type evolution prediction model;

[0029] The syndrome evolution prediction model includes: a patient information summary layer, a third self-attention layer, a fully connected layer, and a second linear mapping + Softmax function calculation layer;

[0030] The patient information aggregation layer is used to store current and historical patient information;

[0031] The information of the patient information summary layer is input into the third self-attention layer after time series encoding, processed by the fully connected layer, and then input into the second linear mapping + Softmax function calculation layer to calculate the probability of the syndrome type after or without specific treatment;

[0032] The fully connected layer includes: a first fully connected layer and a second fully connected layer; the first fully connected layer is used to process the historical patient information and output the probability of the syndrome type after a specific treatment;

[0033] The second fully connected layer is used to process the current patient information and output the probability of the syndrome type without specific treatment;

[0034] The first fully connected layer and the second fully connected layer extract the corresponding patient information by setting different extraction parameters.

[0035] Preferably, performing the timing encoding includes:

[0036] Take T as the length of a time period and establish a historical timeline based on the time window T;

[0037] Determine whether the patient's historical information exists at the current time point;

[0038] If it does not exist, the mask mechanism is used to mask the current time point, and the information here is not included in the patient's historical information summary;

[0039] If it exists, the current time is set to 0 and the time point of the last existing data is set to -T.

[0040] Preferably, it includes: the temporal coding adopts Transformer position coding method based on sine and cosine.

[0041] Preferably, after S3, the method further includes:

[0042] S4, inputting the probability of the syndrome type after or without specific treatment into the trained syndrome evolution prediction model to obtain the probability of the syndrome corresponding to the syndrome type;

[0043] The syndrome evolution prediction model is trained using the historical syndrome type probabilities after or without specific treatment to obtain the trained syndrome evolution prediction model;

[0044] The syndrome evolution prediction model comprises: a third fully connected layer, at least one third linear mapping + Softmax function calculation layer;

[0045] The probability of the syndrome type after or without specific treatment is processed by the third fully connected layer and then input into the third linear mapping + Softmax function calculation layer to calculate the probability of the syndrome corresponding to the syndrome type.

[0046] Preferably, it includes:

[0047] When the probability of the syndrome type after the specific treatment is input, the probability of the syndrome corresponding to the syndrome type after the specific treatment is output;

[0048] When the probability of the syndrome type without specific treatment is input, the probability of the syndrome corresponding to the syndrome type without specific treatment is output.

[0049] (III) Beneficial effects

[0050] The beneficial effects of the present invention are as follows: the artificial intelligence-based TCM syndrome classification and probability prediction method of the present invention uses a normalized data set to input into a trained Transformer-based syndrome classification model to obtain the probability of each syndrome corresponding to the syndrome. Compared with the prior art, it can perform intelligent classification and probability prediction of TCM syndromes.

[0051] The present invention adopts a trained Transformer-based syndrome classification model, a trained syndrome evolution prediction model, and a trained syndrome evolution prediction model. When used, information such as syndromes related to the patient is input into the trained Transformer-based syndrome classification model to output the probability of each syndrome, and each syndrome probability is input into the trained syndrome evolution prediction model to output the probability of a treated or untreated syndrome, and then the probability of a treated or untreated syndrome is input into the trained syndrome evolution prediction model to output the corresponding syndrome probability. The intelligent and trained model can realize the prediction of the evolution of syndromes and syndromes, get rid of the traditional Chinese medicine syndrome and syndrome prediction method, reduce subjective prediction errors, and increase accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a flow chart of the TCM syndrome classification and probability prediction method based on artificial intelligence of the present invention;

[0053] Figure 2 This is a structural diagram of the syndrome classification model of the present invention;

[0054] Figure 3 It is a structural diagram of the syndrome evolution prediction model of the present invention;

[0055] Figure 4 This is a structural diagram of the syndrome evolution prediction model of the present invention. DETAILED DESCRIPTION

[0056] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation modes in conjunction with the accompanying drawings.

[0057] The artificial intelligence-based TCM syndrome classification and probability prediction method proposed in the embodiment of the present invention uses a normalized data set to input into the trained Transformer-based syndrome classification model to obtain the probability of each syndrome corresponding to the syndrome. Compared with the existing technology, it can perform intelligent classification and probability prediction of TCM syndromes.

[0058] In order to better understand the above technical solution, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a clearer and more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0059] Traditional Chinese medicine is an important part of the traditional Chinese medical system. Its core theories include "syndrome type" and "symptom", which are comprehensive descriptions of diseases and patient conditions. However, the diagnosis and treatment of traditional Chinese medicine usually rely on experienced physicians, which are highly subjective and difficult to quantify and standardize. In recent years, with the development of electronic medical data and artificial intelligence technology, it has become possible to try to introduce data-driven methods into the research of traditional Chinese medicine.

[0060] At present, artificial intelligence technology has been widely used in disease diagnosis and treatment decision-making in the field of modern medicine, such as using machine learning models to perform pathological diagnosis on imaging data, predict treatment effects, etc. However, in the field of traditional Chinese medicine, especially in the classification of syndromes and the study of syndrome evolution, the exploration of the application of artificial intelligence technology is still in its infancy. Traditional Chinese medicine syndrome classification relies more on expert experience, lacks unified objective standards, and the evolution laws of syndromes and syndromes have not yet been fully revealed. This embodiment inputs the normalized data set into the trained Transformer-based syndrome classification model to obtain the probability of each syndrome corresponding to the syndrome. Compared with the prior art, it can intelligently classify and probabilistically predict Chinese medicine syndromes.

[0061] Embodiment 1

[0062] like Figure 1 As shown, the present invention provides a TCM syndrome classification and probability prediction method based on artificial intelligence, comprising:

[0063] S1. Normalizing scattered data, natural language text data, and image data related to patient symptoms to obtain normalized data, wherein the collection of each normalized data is a normalized data set;

[0064] S2, input the normalized data set into the trained Transformer-based syndrome classification model to obtain the probability of each syndrome corresponding to the syndrome;

[0065] The syndrome classification model based on Transformer includes: patient information multimodal embedding layer, the first self-attention layer, the first linear mapping + Softmax function calculation layer;

[0066] The normalized data set in the multimodal embedding layer of patient information is input into the first self-attention layer for attention calculation, and then the probability of each syndrome type is calculated through the linear mapping + Softmax function calculation layer; the multimodal embedding layer of patient information is used to store the normalized data set.

[0067] In this embodiment, since the TCM diagnosis and treatment process involves multimodal data such as patient basic information, symptoms and signs, inspection and test indicators, and medical images, these data are heterogeneous and complex, so the data related to the patient's syndrome is divided into three categories according to different structures: scattered data, natural language text data, and image data; in the actual modeling process, other data can also be added. For example, for the disease of post-stroke cognitive impairment, the patient's EEG brain wave signal, fNIRS near-infrared spectrum signal and other data will be collected, which can also be used as the input of the syndrome classification model. The user can input the syndrome classification model according to the actual situation, and this embodiment is not particularly limited. Before inputting the data, the data needs to be vectorized and then normalized. Generally, the vector data is converted into a numerical form by type encoding the vector data so that it can be processed by the machine learning algorithm. Machine learning models usually require numerical input, so categorical variables need to be encoded into numbers in some way.

[0068] Before inputting the syndrome classification model, the syndrome classification model needs to be trained. In actual use, the target data is obtained after the data is input into the trained syndrome classification model. In this embodiment, the training method and the use method of the syndrome classification model are the same, that is, the normalized data set in the multimodal embedding layer of the patient information is input into the first self-attention layer for attention calculation, and then the probability of each syndrome is calculated by the linear mapping + Softmax function calculation layer.

[0069] like Figure 2 As shown, Figure 2 The basic information of patients includes: gender, age, historical text; syndrome information such as symptoms and signs includes: syndrome description text; examination and test indicators; medical images and pictures include: MRI images, facial diagnosis pictures, and tongue diagnosis pictures; among them, gender, age, examination and test indicators are vectorized by linear mapping to obtain the first vectorized data; historical text and syndrome description text are linearly mapped by word embedding, and then vectorized by position encoding in the self-attention layer (second self-attention layer) to obtain the second vectorized data; MRI images, facial diagnosis pictures, and tongue diagnosis pictures are vectorized after convolution processing in the convolution layer to obtain the third vectorized data. Each vectorized data is type-encoded and input into the patient information multimodal embedding layer, and then after attention calculation in the first self-attention layer, the probability of each syndrome type is calculated by linear mapping + Softmax function calculation layer.

[0070] As a preferred embodiment, before S1, the method further includes:

[0071] S0. Use historical normalized data to train the Transformer-based syndrome classification model to obtain a trained Transformer-based syndrome classification model.

[0072] In this embodiment, before the syndrome classification model is actually used, the syndrome classification model must be trained. Training is an indispensable process, which gives the intelligent syndrome classification model the ability to understand and handle specific types of problems. An untrained syndrome classification model is like a blank sheet of paper, without any prior knowledge, and naturally cannot effectively complete the predetermined task. The training method of this embodiment is the same as the method for using the syndrome classification model described above, wherein the input during the training of the syndrome classification model is the historical normalized data accumulated through experience, and the historical normalized data set in the multimodal embedding layer of the patient information is input to the first self-attention layer for attention calculation, and then the probability of each syndrome is calculated by the linear mapping + Softmax function calculation layer, and the training of the syndrome classification model is completed once.

[0073] Since the syndrome classification model is trained in advance before being used, the accuracy of the syndrome classification model when actually used is improved.

[0074] As a preferred embodiment, it includes:

[0075] Scattered data include: gender, age, examination indicators, and test indicators;

[0076] Natural language text data includes: medical history text, syndrome description text;

[0077] Image data includes: MRI images, facial diagnosis images, and tongue diagnosis images.

[0078] This embodiment classifies different types of data so as to perform vector processing according to the data types later.

[0079] As a preferred embodiment, S1 includes:

[0080] S10, performing vectorization processing on scattered data, natural language text data, and image data related to syndromes to obtain a vectorized data set;

[0081] S11. Normalize each vectorized data in the vectorized data set by adding type coding according to different types to form normalized data.

[0082] In this embodiment, it is necessary to vectorize different types of data, and convert non-numeric data (such as text, images, audio, etc.) into a numerical vector process. This process is very necessary for the training and use of the syndrome classification model based on Transformer in this embodiment. The main purpose of vectorization is to enable these unstructured data to be effectively processed and analyzed by computers, and also to enable the data to be input into various machine learning algorithms for training and prediction. At the same time, since the self-attention layer of Transformer cannot distinguish different types of data, we need to add the type encoding of this type to all types of data vectors. Type encoding is similar to position encoding. For example, if the gender type is regarded as the first position and the syndrome description is regarded as the fourth position, the position encoding can be transformed into type encoding. We can use Transformer to transform based on sine and cosine position encoding or other position encoding methods, but it cannot have a large correlation with the position encoding function, otherwise it will be contaminated with the position encoding. After adding the type encoding, all data are normalized to obtain the multimodal embedding of patient information at the current time.

[0083] According to the actual needs of computer data processing, different types of data are processed differently according to the type of data and usage requirements, thus achieving data unification.

[0084] As a preferred embodiment, S10 includes:

[0085] The scattered data are linearly mapped to obtain the first vectorized data;

[0086] The natural language text data is converted into word embedding data by using word embedding, the word embedding data is position-encoded by the sine-cosine position encoding method in the Transformer, and the second self-attention layer feature extraction is performed to obtain the second vectorized data;

[0087] Performing convolution calculation on the image data through the convolution layer to obtain third vectorized data;

[0088] The first vectorized data, the second vectorized data, and the third vectorized data are combined into a vectorized data set.

[0089] In this embodiment, for scattered data values, linear mapping is used for embedding, and the parameters of the linear mapping are learned during the training of the syndrome classification model. For natural language text data, first use pre-trained word vectors, open source embedded syndrome classification models or other models to convert the text into vector data; then add position encoding to the vector data, and the position encoding can use Transformer's position encoding method based on sine and cosine or other position encoding methods; then pass through a layer of Transformer's self-attention layer, and the parameters of the self-attention layer are learned during the training of the syndrome classification model. For image data, after the image is preprocessed, it is subjected to feature extraction through the CNN convolution layer, and the parameters of the convolution layer are learned during the training of the syndrome classification model. Different convolution layers can be set for different image data, and the hyperparameters of the convolution layer can also be different. The hyperparameter setting shall be based on the best experimental effect.

[0090] Different vectorization methods are adopted according to different data to achieve uniformity on the data acquisition side.

[0091] As a preferred embodiment, after S2, the method further includes:

[0092] S3, horizontally splicing the current or historical syndrome probability with the information vector in the current or historical patient information multimodal embedding layer as the patient information summary, inputting it into the trained syndrome evolution prediction model, and obtaining the syndrome probability after or without specific treatment;

[0093] Among them, the trained syndrome type evolution prediction model is obtained by using the historical syndrome type probability and the information in the historical patient information multimodal embedding layer to train the syndrome type evolution prediction model;

[0094] The syndrome evolution prediction model includes: patient information summary layer, third self-attention layer, full connection layer, second linear mapping + Softmax function calculation layer;

[0095] The patient information summary layer is used to store current and historical patient information;

[0096] The information in the patient information summary layer is temporally encoded and input into the third self-attention layer, processed by the fully connected layer, and then input into the second linear mapping + Softmax function calculation layer to calculate the probability of syndrome type after or without specific treatment;

[0097] The fully connected layer includes: the first fully connected layer and the second fully connected layer; the first fully connected layer is used to process the historical patient information and output the probability of syndrome type after specific treatment;

[0098] The second fully connected layer is used to process the current patient information and output the probability of syndrome type without specific treatment;

[0099] The first fully connected layer and the second fully connected layer extract corresponding patient information by setting different extraction parameters.

[0100] In this embodiment, Figure 3 As shown in FIG. 1 , after obtaining the syndrome probability, the historical / current patient information is summarized and time-series encoded, and then input into the trained syndrome evolution prediction model to obtain the syndrome probability of having or not having been treated. It should be noted that the syndrome evolution prediction model in this embodiment also needs to be trained before use, and the training process is the same as the use process. The historical patient information is summarized and input into Figure 3 The probability of a syndrome type that has been treated or not is obtained from the syndrome type evolution prediction model, and a model training is completed. In actual use, after the current and historical patient information is summarized, the patient information with time series coding is summarized and input into the n1 self-attention layer and two n2 fully connected layers. The parameters of the self-attention layer and the fully connected layer are learned during the training of the syndrome type evolution prediction model, and the settings of the hyperparameters n1 and n2 are based on the best experimental results. The two fully connected layers correspond to the prediction of the probability of syndrome types that have been treated and those that have not been treated. After the data passes through the two fully connected layers, it will be calculated by linear mapping + Softmax respectively to obtain the probability of each syndrome type that has been treated and has not been treated after T time. The parameters of the linear mapping are learned during the training of the syndrome type classification model. The syndrome type is set during training, which is the same as the type setting of the syndrome type classification. The sample categories are annotated by experts during training.

[0101] As a preferred embodiment, performing timing coding includes:

[0102] Take T as the length of a time period and establish a historical timeline based on the time window T;

[0103] Determine whether there is patient history information at the current time point;

[0104] If it does not exist, the mask mechanism is used to cover the current time point, and the information here is not included in the patient's historical information summary;

[0105] If it exists, the current time is set to 0 and the time point of the last existing data is set to -T.

[0106] In this embodiment, a historical timeline is established based on the time window T, with the current time being 0, the previous time point being -T, and the previous time point being -2T... Then, the historical information of the same patient is summarized into the most recent time point of the historical information. If there is no historical information summary of the patient at a certain time point, the mask mechanism in the Transformer is used to mask it. For example, if the patient has historical information summary of the patient 3 months ago and 9 months ago, the historical information summary of the patient 3 months ago and 9 months ago is used as the input of the -T and -3T moments, and the -2T moment in June is masked.

[0107] By temporally encoding historical data, time information is converted into a format that can be understood by the syndrome evolution prediction model. This is very important in predictive analysis, pattern recognition, and other tasks that require consideration of time factors.

[0108] As a preferred embodiment, it includes: the temporal coding adopts the Transformer position coding method based on sine and cosine.

[0109] In this embodiment, after horizontally splicing the current or historical syndrome probability with the information vector in the current or historical patient information multimodal embedding layer, it is necessary to add time series coding to the input vector. Time series coding is also a position coding, which can adopt Transformer-based sine-cosine position coding or other position coding methods. Then, the patient information with time series coding added is summarized and input into the n1 self-attention layer and two n2 fully connected layers.

[0110] As a preferred embodiment, after S3, the method further includes:

[0111] S4, inputting the probability of syndrome type after or without specific treatment into the trained syndrome evolution prediction model to obtain the probability of syndrome corresponding to the syndrome type;

[0112] Among them, the syndrome evolution prediction model is trained by using the historical syndrome type probability after or without specific treatment to obtain the trained syndrome evolution prediction model;

[0113] The syndrome evolution prediction model includes: a third fully connected layer, at least one third linear mapping + Softmax function calculation layer;

[0114] The probability of the syndrome type after or without specific treatment is processed by the third fully connected layer and then input into the third linear mapping + Softmax function calculation layer to calculate the probability of the syndrome corresponding to the syndrome type.

[0115] By setting different linear mapping + Softmax function calculation layers, the probability of syndrome types after or without specific treatment can be distinguished, making the processing results clearer.

[0116] In this embodiment, Figure 4 As shown, the syndrome evolution prediction model of this embodiment also needs to be trained. First, the historical syndrome type probability is input into Figure 4 After the syndrome evolution prediction model is built, the corresponding syndrome probability is output, and the training of the syndrome evolution prediction model is completed. In the actual use of the syndrome evolution prediction model, the syndrome type probability output in the syndrome type evolution prediction is first used as input, and after passing through the fully connected layer and multiple linear mappings + Softmax, the probability of occurrence of each syndrome is obtained respectively. The parameters of the third fully connected layer and linear mapping are learned during the training of the syndrome evolution prediction model. The syndrome types are set during training. For example, hundreds of syndromes such as taciturnity, memory loss, red tongue, less moss, and slippery pulse can be set. They are obtained from the patient's syndrome information and correspond to the syndrome type after the expert marks the patient's syndrome type during training. When predicting, the probability of the syndrome type after a specific treatment is input, and the probability of each syndrome occurring after a specific treatment is output; the probability of the syndrome type after not having a specific treatment is input, and the probability of each syndrome occurring after not having a specific treatment is output.

[0117] As a preferred embodiment, it includes:

[0118] When the probability of a syndrome type after a specific treatment is input, the probability of the syndrome corresponding to the syndrome type after the specific treatment is output;

[0119] When the probability of a syndrome type without specific treatment is input, the probability of the syndrome corresponding to the syndrome type without specific treatment is output.

[0120] Embodiment 2

[0121] W0. Use the patient's basic information, symptom and sign information, examination and test indicators, medical images and pictures in the historical text data to input into the Transformer-based syndrome classification model for training, and obtain the trained Transformer-based syndrome classification model.

[0122] W1. The patient's basic information, symptom and sign information, examination and test indicators, medical images and pictures obtained at the current time point are processed and input into the trained Transformer-based syndrome classification model for data processing to obtain the syndrome probability corresponding to the current data.

[0123] Wherein W1 includes the following sub-steps:

[0124] W10, obtain the first vectorized data by linear mapping the scattered data;

[0125] The natural language text data is converted into word embedding data by using word embedding, the word embedding data is position-encoded by the sine-cosine position encoding method in the Transformer, and the second self-attention layer feature extraction is performed to obtain the second vectorized data;

[0126] Performing convolution calculation on the image data through the convolution layer to obtain third vectorized data;

[0127] The first vectorized data, the second vectorized data, and the third vectorized data are combined into a vectorized data set.

[0128] Scattered data include: gender, age, examination indicators, and test indicators;

[0129] Natural language text data includes: medical history text, syndrome description text;

[0130] Image data includes: MRI images, facial diagnosis images, and tongue diagnosis images.

[0131] For example, a 24-year-old male is tested for fat content in his blood, i.e., a blood lipid test. The historical text and syndrome description text here correspond to the historical blood lipid test results of the male and the current syndrome description. MRI images are generally not required here. If necessary, facial and tongue diagnosis image data can be added. After quantification of the above data, type coding is added, normalization is performed, and the trained Transformer-based syndrome classification model is input.

[0132] W11. Normalize the first, second and third vectorized data by adding type coding according to different types to form normalized data, and input the normalized data into the trained Transformer-based syndrome classification model.

[0133] After W1 obtains the syndrome type probability corresponding to the current data.

[0134] W2. Horizontally concatenate the syndrome type probabilities of historical data and historical patient information to form a summary of historical patient information, input it into the syndrome type evolution prediction model, and train the syndrome type evolution prediction model.

[0135] W21. Horizontally concatenate the syndrome probability of the current historical data and the current patient information into a summary of the current patient information. Input the current and historical patient information summaries into the trained syndrome evolution prediction model, and output the corresponding syndrome probability of having or not having been treated.

[0136] After W21 obtains the syndrome type probability corresponding to the current data.

[0137] W3. Input the historical syndrome type probability into the syndrome evolution prediction model and train the syndrome evolution prediction model.

[0138] W31. Input the current syndrome type probability into the trained syndrome evolution prediction model, and output the probability of the current syndrome corresponding to the current syndrome type.

[0139] In this embodiment, by using the trained Transformer-based syndrome classification model, the trained syndrome evolution prediction model, and the trained symptom evolution prediction model, when using, the information such as syndromes related to the patient is input into the trained Transformer-based syndrome classification model to output the probability of each syndrome, and after each syndrome probability is input into the trained syndrome evolution prediction model to output the probability of a treated or untreated syndrome, the probability of a treated or untreated syndrome is input into the trained syndrome evolution prediction model to output the corresponding syndrome probability. The intelligent and trained model can realize the prediction of the evolution of syndromes and syndromes, get rid of the traditional Chinese medicine prediction method of syndromes and symptoms, reduce subjective prediction errors, and increase accuracy.

[0140] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0141] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions.

[0142] It should be noted that in the claims, any reference numerals placed between brackets shall not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention may be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In the claims enumerating several means, several of these means may be embodied by the same hardware. The use of the words first, second, third, etc., is for convenience of expression only and does not indicate any order. These words may be understood as part of the component name.

[0143] In addition, it should be noted that, in the description of this specification, the description of the terms "one embodiment", "some embodiments", "embodiment", "example", "specific example" or "some examples" refers to the specific features, structures, materials or specific features described in conjunction with the embodiment or example included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or specific features described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0144] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments after knowing the basic creative concept. Therefore, the claims should be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0145] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention should also include these modifications and variations.

Claims

1. A method for classification and probability prediction of TCM syndromes based on artificial intelligence, characterized in that: include: S1. Normalizing scattered data, natural language text data, and image data related to patient symptoms to obtain normalized data, wherein a collection of each normalized data is a normalized data set; S2, inputting the normalized data set into the trained Transformer-based syndrome classification model to obtain the probability of each syndrome corresponding to the syndrome; The Transformer-based syndrome classification model includes: a patient information multimodal embedding layer, a first self-attention layer, and a first linear mapping + Softmax function calculation layer; The normalized data set in the patient information multimodal embedding layer is input into the first self-attention layer for attention calculation, and then the probability of each syndrome type is calculated through the linear mapping + Softmax function calculation layer; wherein the patient information multimodal embedding layer is used to store the normalized data set.

2. The method for TCM syndrome classification and probability prediction based on artificial intelligence according to claim 1, characterized in that: Before the S1, it also includes: S0. Using the historical normalized data to train the Transformer-based syndrome classification model to obtain a trained Transformer-based syndrome classification model.

3. The method for TCM syndrome classification and probability prediction based on artificial intelligence according to claim 1, characterized in that: include: The scattered data include: gender, age, examination index, and test index; The natural language text data includes: medical history text and syndrome description text; The image data includes: MRI image pictures, facial diagnosis pictures, and tongue diagnosis pictures.

4. The method for TCM syndrome classification and probability prediction based on artificial intelligence according to claim 3 is characterized in that: The S1 includes: S10, performing vectorization processing on the scattered data related to the syndrome, the natural language text data, and the image data to obtain a vectorized data set; S11. Normalize each vectorized data in the vectorized data set by adding type coding according to different types to form the normalized data.

5. The method for TCM syndrome classification and probability prediction based on artificial intelligence according to claim 4 is characterized in that: The S10 includes: Performing linear mapping on the scattered data to obtain first vectorized data; The natural language text data is converted into word embedding data by using word embedding, the word embedding data is position-encoded by a sine-cosine position encoding method in a Transformer, and feature extraction is performed by a second self-attention layer to obtain second vectorized data; Performing convolution calculation on the image data through a convolution layer to obtain third vectorized data; The first vectorized data, the second vectorized data, and the third vectorized data are combined into a vectorized data set.

6. The artificial intelligence-based TCM syndrome classification and probability prediction method according to claim 5 is characterized in that: After S2, it also includes: S3, horizontally splicing the current or historical syndrome probability with the information vector in the current or historical patient information multimodal embedding layer as the patient information summary, and inputting it into the trained syndrome evolution prediction model to obtain the syndrome probability after or without specific treatment; wherein the syndrome type evolution prediction model is trained by using the historical syndrome type probabilities and the information in the historical patient information multimodal embedding layer to obtain the trained syndrome type evolution prediction model; The syndrome evolution prediction model includes: a patient information summary layer, a third self-attention layer, a fully connected layer, and a second linear mapping + Softmax function calculation layer; The patient information aggregation layer is used to store current and historical patient information; The information of the patient information summary layer is input into the third self-attention layer after time series encoding, processed by the fully connected layer, and then input into the second linear mapping + Softmax function calculation layer to calculate the probability of the syndrome type after or without specific treatment; The fully connected layer includes: a first fully connected layer and a second fully connected layer; the first fully connected layer is used to process the historical patient information and output the probability of the syndrome type after a specific treatment; The second fully connected layer is used to process the current patient information and output the probability of the syndrome type without specific treatment; The first fully connected layer and the second fully connected layer extract the corresponding patient information by setting different extraction parameters.

7. The method for TCM syndrome classification and probability prediction based on artificial intelligence according to claim 6 is characterized in that: The timing coding comprises: Take T as the length of a time period and establish a historical timeline based on the time window T; Determine whether the patient's historical information exists at the current time point; If it does not exist, the mask mechanism is used to mask the current time point, and the information here is not included in the patient's historical information summary; If it exists, the current time is set to 0 and the time point of the last existing data is set to -T.

8. The method for TCM syndrome classification and probability prediction based on artificial intelligence according to claim 7, characterized in that: include: The temporal coding adopts the Transformer position coding method based on sine and cosine.

9. The method for TCM syndrome classification and probability prediction based on artificial intelligence according to claim 8, characterized in that: After S3, it also includes: S4, inputting the probability of the syndrome type after or without specific treatment into the trained syndrome evolution prediction model to obtain the probability of the syndrome corresponding to the syndrome type; The syndrome evolution prediction model is trained using the historical syndrome type probabilities after or without specific treatment to obtain the trained syndrome evolution prediction model; The syndrome evolution prediction model comprises: a third fully connected layer, at least one third linear mapping + Softmax function calculation layer; The probability of the syndrome type after or without specific treatment is processed by the third fully connected layer and then input into the third linear mapping + Softmax function calculation layer to calculate the probability of the syndrome corresponding to the syndrome type.

10. The artificial intelligence-based TCM syndrome classification and probability prediction method according to claim 9, characterized in that: include: When the probability of the syndrome type after the specific treatment is input, the probability of the syndrome corresponding to the syndrome type after the specific treatment is output; When the probability of the syndrome type without specific treatment is input, the probability of the syndrome corresponding to the syndrome type without specific treatment is output.

Images