A training method and device of a syndrome diagnosis model, and a syndrome prediction method and device

Abstract

The application discloses a syndrome diagnosis model training method, a syndrome prediction method and device. The syndrome diagnosis model training method comprises the following steps: unsupervised training of an initial model through a basic traditional Chinese medicine corpus to obtain a first diagnosis model; supervised training of the first diagnosis model through a general famous old Chinese medicine case database to obtain a second diagnosis model, wherein the general famous old Chinese medicine case database contains syndrome element information corresponding to patient symptoms; determining syndrome element pseudo-labels corresponding to a specific famous old Chinese medicine case database and correcting the syndrome element pseudo-labels to obtain target pseudo-labels; and optimizing training of the second diagnosis model through the specific famous old Chinese medicine case database and the corresponding target pseudo-labels to obtain a syndrome diagnosis model. The application solves the technical problems of limited accuracy and insufficient interpretability of an intelligent syndrome diagnosis model caused by the lack of syndrome element annotation in famous old Chinese medicine case data and the difficulty of annotation.

Description

Technical Field

[0001] This application relates to the field of TCM diagnostic technology, specifically to a training method for a syndrome diagnosis model, a syndrome prediction method, and a device. Background Technology

[0002] Traditional Chinese medicine (TCM), as an ancient and unique medical system, centers on "syndrome differentiation and treatment," which involves analyzing a patient's specific symptoms, signs, and other clinical manifestations to determine the essence of the illness—the syndrome—and then formulating a personalized treatment plan. The syndrome differentiation process relies on the doctor's rich clinical experience and profound theoretical foundation. Renowned veteran TCM doctors, with their lifelong accumulated diagnostic and treatment wisdom, have become shining pearls in this treasure trove of traditional medicine. Their experience is not only reflected in their profound insight into the nature of diseases but also includes a series of specific diagnostic and treatment techniques and insights into medication. These are exemplary models of the integration of TCM theory and practice, playing an irreplaceable role in promoting the high-quality development of TCM services and the academic inheritance of TCM.

[0003] In recent years, with the rapid development of artificial intelligence technology, especially deep learning algorithms, many studies have begun to explore how to apply these modern technologies to the field of traditional Chinese medicine (TCM) in order to improve the accuracy and efficiency of diagnosis and treatment, while also realizing the digital inheritance of the experience of renowned veteran TCM doctors. Natural language processing (NLP) technology has shown great potential in interpreting ancient TCM texts, compiling medical records of renowned veteran TCM doctors, and even assisting in diagnosis. However, these applications also face multiple challenges, especially in the area of ​​data.

[0004] First, the limited amount of data from a single renowned practitioner restricts the model's generalization ability and accuracy. Traditional Chinese medicine (TCM) diagnosis and treatment is a highly personalized decision-making process; each renowned TCM practitioner has their unique diagnostic style and preferences, which are often only reflected in their individual medical records. Therefore, how to enable the model to learn the essence of renowned TCM practitioners even with limited data is a pressing issue. Second, the inconsistent terminology used by renowned TCM practitioners increases the difficulty of model understanding and learning. Terms used in TCM diagnosis, such as syndrome elements and symptoms, have multiple expressions due to historical, regional, and personal differences. This inconsistency necessitates a strong semantic understanding capability in model construction to overcome these superficial differences and grasp their underlying logical connections. Furthermore, annotating syndrome elements in medical records is a time-consuming and complex task. Syndrome elements are key indicators in the process of syndrome differentiation and treatment. However, traditional medical records do not always clearly mark syndrome elements. This means that before training with machine learning methods, a lot of preprocessing of medical records is required, including the extraction and labeling of syndrome elements. This undoubtedly increases the development cost and cycle of intelligent syndrome differentiation and diagnosis models.

[0005] Currently, although some studies have attempted to use deep learning technology to analyze and learn from the medical records of renowned traditional Chinese medicine practitioners to assist in syndrome differentiation and diagnosis, most methods remain at a relatively rudimentary stage, particularly in terms of model training strategies and interpretability. Most models are trained solely on the medical record texts, failing to effectively integrate the crucial intermediate variable of syndrome elements, resulting in poor performance in syndrome differentiation and an inability to reveal the diagnostic and treatment logic of these renowned practitioners.

[0006] There is currently no effective solution to the above problems. Summary of the Invention

[0007] This application provides a training method, a syndrome prediction method, and an apparatus for a syndrome diagnosis model, in order to at least solve the technical problems of limited accuracy and insufficient interpretability of intelligent syndrome diagnosis models caused by the lack of syndrome element annotation and the difficulty of annotation in the medical case data of famous veteran TCM doctors.

[0008] According to one aspect of the embodiments of this application, a training method for a syndrome diagnosis model is provided, comprising: unsupervised training of an initial model using a basic TCM corpus to obtain a first diagnostic model, wherein the basic TCM corpus contains multiple TCM basic knowledge texts; supervised training of the first diagnostic model using a general database of medical records of renowned TCM doctors to obtain a second diagnostic model, wherein the general database of medical records of renowned TCM doctors contains general medical records of multiple renowned TCM doctors and syndrome element information corresponding to patient symptoms, the syndrome element information being used to guide the first diagnostic model in learning the general medical logic of renowned TCM doctors; determining pseudo-labels of syndrome elements corresponding to a specific database of medical records of renowned TCM doctors, and correcting the pseudo-labels of syndrome elements to obtain target pseudo-labels, wherein the specific database of medical records of renowned TCM doctors contains personal medical records of preset renowned TCM doctors; and optimizing the training of the second diagnostic model using the specific database of medical records of renowned TCM doctors and the corresponding target pseudo-labels to obtain a syndrome diagnosis model.

[0009] Optionally, an unsupervised training process is performed on the initial model using a basic TCM corpus to obtain a first diagnostic model. This includes: acquiring first symptom information and a first syndrome diagnosis result, wherein the first symptom information is the symptom information of any patient in the basic TCM corpus, and the first syndrome diagnosis result is the syndrome diagnosis result of any patient in the basic TCM corpus; predicting the first symptom information and the first syndrome diagnosis result using the initial model to obtain a first prediction result, wherein the first prediction result represents the probability that the first syndrome diagnosis result is the true diagnosis result of the first symptom information; determining a first loss value between the first prediction result and the first true label using a first loss function, wherein the first true label represents the true diagnosis result of the first symptom information; and optimizing the model parameters of the initial model based on the first loss value to obtain the first diagnostic model.

[0010] Optionally, the first diagnostic model is trained in a supervised manner using a database of medical records from renowned traditional Chinese medicine practitioners. This includes: acquiring second symptom information and determining a first feature vector corresponding to the second symptom information, wherein the second symptom information is the symptom information of any patient in the database of medical records from renowned traditional Chinese medicine practitioners; predicting the first feature vector using the first diagnostic model to obtain a second prediction result, wherein the second prediction result is used to represent the syndrome diagnosis result corresponding to the second symptom information; concatenating the first feature vector and the second prediction result to obtain a second feature vector; predicting the second feature vector using the first diagnostic model to obtain a third prediction result, wherein the third prediction result is used to represent the syndrome diagnosis result corresponding to the second symptom information; and training the first diagnostic model based on the second prediction result and the third prediction result.

[0011] Optionally, the first diagnostic model is trained based on the second and third prediction results, including: determining a second loss value between the second prediction result and the second true label using a second loss function, wherein the second true label is used to represent the true syndrome information corresponding to the second symptom information; determining a third loss value between the third prediction result and the third true label using a third loss function, wherein the third true label is used to represent the true syndrome information corresponding to the second symptom information; and optimizing the model parameters of the first diagnostic model based on the second loss value, the third loss value, and preset hyperparameters to obtain the second diagnostic model.

[0012] Optionally, determining the pseudo-label of the syndrome element corresponding to a specific database of medical records of renowned traditional Chinese medicine practitioners includes: obtaining third symptom information and determining a third feature vector corresponding to the third symptom information, wherein the third symptom information is the symptom information of any patient in the database of medical records of renowned traditional Chinese medicine practitioners; predicting the third feature vector through a second diagnostic model to obtain a fourth prediction result, wherein the fourth prediction result is used to represent the syndrome element diagnosis result corresponding to the third symptom information; determining the pseudo-label of the syndrome element corresponding to the third symptom information based on the fourth prediction result and a preset threshold; or, obtaining a list of syndrome elements corresponding to the third symptom information and determining the pseudo-label of the syndrome element corresponding to the third symptom information based on the list of syndrome elements.

[0013] Optionally, the pseudo-labels of the syndrome elements are corrected to obtain the target pseudo-label, including: determining a first deviation probability and a second deviation probability corresponding to the third symptom information, wherein the first deviation probability is used to represent the probability that the pseudo-label of the syndrome element is 0 when the fourth true label is 1, the second deviation probability is used to represent the probability that the pseudo-label of the syndrome element is 1 when the fourth true label is 0, and the fourth true label is used to represent the true syndrome element information corresponding to the third symptom information; and the pseudo-labels of the syndrome elements are corrected based on the first deviation probability and the second deviation probability to obtain the target pseudo-label.

[0014] Optionally, determining the first deviation probability corresponding to the third symptom information includes: determining the anchor sample corresponding to the third symptom information, and determining the cumulative density function and corresponding inverse function of the anchor sample based on the standard normal distribution, wherein the anchor sample is used to represent the sample in the third symptom information whose predicted probability obtained by the second diagnostic model is lower than a preset probability threshold; determining the correlation coefficient matrix between the symptom elements corresponding to the anchor sample, and determining the log-likelihood function corresponding to the anchor sample based on the correlation coefficient matrix; determining the normal distribution parameter corresponding to the anchor sample based on the log-likelihood function; and determining the first deviation probability based on the normal distribution parameter, the cumulative density function, and the inverse function.

[0015] Optionally, the method further includes: optimizing the fourth prediction result based on the target pseudo-label to obtain the target prediction result; determining the fourth loss value between the target prediction result and the syndrome pseudo-label through the fourth loss function; concatenating the third feature vector and the target prediction result to obtain the fourth feature vector; predicting the fourth feature vector through the second diagnostic model to obtain the fifth prediction result, wherein the fifth prediction result is used to represent the syndrome diagnosis result corresponding to the third symptom information; determining the fifth loss value between the fifth prediction result and the fifth true label through the third loss function, wherein the fifth true label is used to represent the true syndrome information corresponding to the third symptom information; and optimizing the model parameters of the second diagnostic model based on the fourth loss value and the fifth loss value to obtain the syndrome diagnosis model.

[0016] Optionally, the method further includes: predicting the third feature vector using a syndrome diagnosis model to obtain the target syndrome element diagnosis result; concatenating the third feature vector and the target syndrome element diagnosis result to obtain the target feature vector, and predicting the target feature vector using a syndrome diagnosis model to obtain the target syndrome diagnosis result; and determining the first syndrome element list and the second syndrome element list corresponding to the first target syndrome diagnosis result and the second target syndrome diagnosis result, respectively, wherein the first target syndrome diagnosis result and the second target syndrome diagnosis result are any two identical syndrome diagnosis results in the target syndrome diagnosis results, and the first syndrome element list and the second syndrome element list are... The list represents a predetermined number of the most probable syndrome elements in the target syndrome diagnosis results. Based on the first and second syndrome element lists, the syndrome element ratios for the first and second syndromes are determined. The first syndrome is any syndrome in the first target syndrome diagnosis results, and the second syndrome is any syndrome in the second target syndrome diagnosis results. The syndrome element ratios represent the relative proportions of the frequencies of each syndrome element in the first syndrome element list under the first syndrome, and the relative proportions of the frequencies of each syndrome element in the second syndrome element list under the second syndrome. Based on the syndrome element ratios, the pre-defined diagnostic and treatment focus of renowned traditional Chinese medicine practitioners is determined.

[0017] According to another aspect of the embodiments of this application, a syndrome prediction method is also provided, comprising: acquiring symptom information of a patient and determining feature vectors corresponding to the symptom information; predicting the feature vectors through a syndrome diagnosis model to obtain a syndrome element prediction result corresponding to the symptom information, wherein the syndrome diagnosis model is trained in the following manner: performing unsupervised training on an initial model through a basic TCM corpus to obtain a first diagnostic model, wherein the basic TCM corpus contains multiple TCM basic knowledge texts; and performing supervised training on the first diagnostic model through a general database of medical records of renowned TCM doctors to obtain a second diagnostic model, wherein the general database of medical records of renowned TCM doctors contains multiple general diagnosis and treatment records of renowned TCM doctors, and a prediction result corresponding to the patient's symptoms. The corresponding syndrome element information is used to guide the first diagnostic model in learning the general diagnostic and treatment logic of renowned traditional Chinese medicine practitioners. Syndrome element pseudo-labels corresponding to a specific database of medical records of renowned traditional Chinese medicine practitioners are determined, and these pseudo-labels are corrected to obtain target pseudo-labels. The database of medical records of renowned traditional Chinese medicine practitioners contains the personal medical records of preset renowned traditional Chinese medicine practitioners. The second diagnostic model is optimized and trained using the database of medical records of renowned traditional Chinese medicine practitioners and the corresponding target pseudo-labels to obtain a syndrome diagnosis model. Syndrome element prediction results and feature vectors are concatenated to obtain a target feature vector. The syndrome diagnosis model predicts the target feature vector to obtain a syndrome prediction result corresponding to the symptom information. The syndrome prediction result is used to provide a basis for the patient's syndrome differentiation diagnosis.

[0018] According to another aspect of the embodiments of this application, a training device for a syndrome diagnosis model is also provided, comprising: a first training module, used for unsupervised training of an initial model using a basic TCM corpus to obtain a first diagnostic model, wherein the basic TCM corpus contains multiple TCM basic knowledge texts; a second training module, used for supervised training of the first diagnostic model using a general database of medical records of renowned TCM doctors to obtain a second diagnostic model, wherein the general database of medical records of renowned TCM doctors contains general medical records of multiple renowned TCM doctors and syndrome element information corresponding to patient symptoms, the syndrome element information being used to guide the first diagnostic model to learn the general medical logic of renowned TCM doctors; a correction module, used for determining pseudo-labels of syndrome elements corresponding to a specific database of medical records of renowned TCM doctors and correcting the pseudo-labels of syndrome elements to obtain target pseudo-labels, wherein the specific database of medical records of renowned TCM doctors contains personal medical records of preset renowned TCM doctors; and a third training module, used for optimized training of the second diagnostic model using the specific database of medical records of renowned TCM doctors and the corresponding target pseudo-labels to obtain a syndrome diagnosis model.

[0019] According to another aspect of the embodiments of this application, a syndrome prediction device is also provided, comprising: an acquisition module, configured to acquire symptom information of a patient and determine a feature vector corresponding to the symptom information; and a first prediction module, configured to predict the feature vector using a syndrome diagnosis model to obtain a syndrome element prediction result corresponding to the symptom information, wherein the syndrome diagnosis model is trained in the following manner: an initial model is unsupervised trained using a basic TCM corpus to obtain a first diagnostic model, wherein the basic TCM corpus contains multiple TCM basic knowledge texts; the first diagnostic model is supervised trained using a general database of medical records of renowned TCM doctors to obtain a second diagnostic model, wherein the general database of medical records of renowned TCM doctors contains multiple general diagnosis and treatment records of renowned TCM doctors, and a prediction result corresponding to the patient's symptoms. The system includes: 1) Syndrome element information, used to guide the first diagnostic model in learning the general diagnostic and treatment logic of renowned traditional Chinese medicine (TCM) doctors; 2) Identifying pseudo-labels of syndrome elements corresponding to a specific database of medical records of renowned TCM doctors, and correcting these pseudo-labels to obtain target pseudo-labels, wherein the specific database of medical records of renowned TCM doctors contains the personal medical records of preset renowned TCM doctors; 3) Optimizing and training the second diagnostic model using the specific database of medical records of renowned TCM doctors and the corresponding target pseudo-labels to obtain a syndrome diagnosis model; 4) A splicing module, used to splice the syndrome element prediction results and feature vectors to obtain a target feature vector; and 5) A second prediction module, used to predict the target feature vector using the syndrome diagnosis model to obtain a syndrome prediction result corresponding to the symptom information, wherein the syndrome prediction result is used to provide a basis for the patient's syndrome differentiation diagnosis.

[0020] According to another aspect of the embodiments of this application, an electronic device is also provided, including: a memory and a processor, wherein the memory is used to store program instructions; the processor is connected to the memory and is used to execute the training method or the syndrome prediction method that implements the above-mentioned syndrome diagnosis model.

[0021] According to another aspect of the embodiments of this application, a non-volatile storage medium is also provided, the non-volatile storage medium including a stored computer program, wherein the device where the non-volatile storage medium is located executes the above-mentioned training method or syndrome prediction method of syndrome diagnosis model by running the computer program.

[0022] According to another aspect of the embodiments of this application, a computer program product is also provided, including computer instructions, which, when executed by a processor, implement the training method or the syndrome prediction method of the above-mentioned syndrome diagnosis model.

[0023] In this embodiment, an initial model is trained unsupervised using a basic TCM corpus to obtain a first diagnostic model. The basic TCM corpus contains multiple TCM basic knowledge texts. A second diagnostic model is trained in a supervised manner using a database of medical records from renowned TCM doctors. This database contains general treatment records from multiple renowned TCM doctors, as well as syndrome element information corresponding to patient symptoms. This syndrome element information guides the first diagnostic model in learning the general treatment logic of renowned TCM doctors. Pseudo-labels for syndrome elements corresponding to specific databases of medical records from renowned TCM doctors are identified, and these pseudo-labels are corrected. The process involves partial processing to obtain target pseudo-labels. A database of medical records from renowned traditional Chinese medicine (TCM) practitioners contains their individual clinical records. The second diagnostic model is then optimized and trained using this database and the corresponding target pseudo-labels to obtain a syndrome diagnosis model. This achieves a deep understanding and accurate simulation of the diagnostic and treatment logic of renowned TCM practitioners, thereby improving the accuracy and interpretability of syndrome diagnosis while reducing the complexity of data annotation. This solves the technical problem of limited accuracy and insufficient interpretability of intelligent syndrome diagnosis models caused by the lack of syndrome element annotations and the high difficulty of annotation in the medical record data of renowned TCM practitioners. Attached Figure Description

[0024] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0025] Figure 1 This is a hardware structure diagram of a computer terminal for implementing a training method for a syndrome diagnosis model according to an embodiment of this application;

[0026] Figure 2 This is a flowchart of a training method for a syndrome diagnosis model according to an embodiment of this application;

[0027] Figure 3 This is an overall framework diagram of a training method for a syndrome diagnosis model according to an embodiment of this application;

[0028] Figure 4 This is a flowchart of a syndrome prediction method according to an embodiment of this application;

[0029] Figure 5 This is a structural diagram of a training device for a syndrome diagnosis model according to an embodiment of this application;

[0030] Figure 6 This is a structural diagram of a syndrome prediction device according to an embodiment of this application. Detailed Implementation

[0031] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0032] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0033] First, some nouns or terms that appear in the explanation of the embodiments of this application shall be interpreted as follows:

[0034] Syndrome: In Traditional Chinese Medicine (TCM) theory, a syndrome is the nature and stage of a disease determined by analyzing and synthesizing all the information obtained from the four diagnostic methods of observation, auscultation and olfaction, inquiry, and palpation. It forms the basis for TCM's syndrome differentiation and treatment. For example, "liver and kidney yin deficiency" and "spleen and stomach damp-heat" represent different etiologies, pathogenesis, and stages of the disease, respectively.

[0035] Syndrome elements: These are the basic elements in traditional Chinese medicine syndrome differentiation, reflecting the specific manifestations and pathological characteristics of diseases in different organs, meridians, qi and blood. For example, "heart fire rising" and "liver qi stagnation" are specific elements that constitute syndromes and have guiding significance for syndrome differentiation and treatment.

[0036] Unsupervised and weakly supervised pre-training: Unsupervised pre-training trains the model to learn the inherent structure and features of the data without labeled data; weakly supervised pre-training uses partially labeled or indirectly labeled data to guide the model's learning, and has some guiding information compared to unsupervised pre-training.

[0037] Supervised training continues: After the model has initially learned some basic knowledge, it is further trained using a labeled dataset to enable it to learn more specific task knowledge or skills.

[0038] To address the issue of poor accuracy in detection algorithms in related technologies, this application provides a training method and a syndrome prediction method for a syndrome diagnosis model, which can be run on... Figure 1 The computer terminal shown is described below.

[0039] The training method or syndrome prediction method of the syndrome diagnosis model provided in this application can be executed on a mobile terminal, computer terminal or similar computing device. Figure 1 A hardware block diagram of a computer terminal for implementing a training method for a syndrome diagnosis model or a syndrome prediction method is shown. Figure 1 As shown, the computer terminal 10 may include one or more processors (shown as 102a, 102b, ..., 102n in the figure) (the processor may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, and a transmission module 106 for communication functions connected via wired and / or wireless networks. In addition, it may also include: a display, a keyboard, a cursor control device, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, and a BUS bus. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0040] It should be noted that the aforementioned one or more processors and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10. As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0041] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the training method or syndrome prediction method of the syndrome diagnosis model in this embodiment of the application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the above-mentioned training method or syndrome prediction method of the syndrome diagnosis model. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor, and these remote memories can be connected to the computer terminal 10 via a network. Examples of the above-mentioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0042] The transmission module 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission module 106 includes a network interface controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission module 106 may be a radio frequency (RF) module, used for wireless communication with the Internet.

[0043] The display can be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 10.

[0044] It should be noted here that, in some optional embodiments, the above... Figure 1 The computer terminal shown may include hardware components (including circuitry), software components (including computer code stored on a computer-readable medium), or a combination of both hardware and software components. It should be noted that... Figure 1 This is only one instance of a specific particular instance, and is intended to illustrate the types of components that may exist in the aforementioned computer terminal.

[0045] In the above operating environment, the embodiments of this application provide a training method for a syndrome diagnosis model and an embodiment of a syndrome prediction method. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Also, although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than that shown here.

[0046] Figure 2This is a flowchart of a training method for a syndrome diagnosis model according to an embodiment of this application, such as... Figure 2 As shown, the method includes the following steps:

[0047] Step S202: The initial model is trained unsupervised using a basic TCM corpus to obtain the first diagnostic model. The basic TCM corpus contains multiple TCM basic knowledge texts.

[0048] Step S204: Supervised training of the first diagnostic model is performed using a general database of medical records of renowned traditional Chinese medicine practitioners to obtain the second diagnostic model. The database contains general medical records of multiple renowned traditional Chinese medicine practitioners, as well as syndrome element information corresponding to the patient's symptoms. The syndrome element information is used to guide the first diagnostic model to learn the general diagnostic and treatment logic of renowned traditional Chinese medicine practitioners.

[0049] Step S206: Determine the pseudo-label of syndrome element corresponding to the specific famous traditional Chinese medicine doctor's medical record database, and perform correction processing on the pseudo-label of syndrome element to obtain the target pseudo-label. The specific famous traditional Chinese medicine doctor's medical record database contains the personal diagnosis and treatment records of the preset famous traditional Chinese medicine doctors.

[0050] Step S208: Optimize and train the second diagnostic model using a database of medical records from renowned traditional Chinese medicine practitioners and corresponding target pseudo-labels to obtain a syndrome diagnosis model.

[0051] Through the above steps S202 to S208, the goal of deeply understanding and accurately simulating the syndrome differentiation and treatment logic of renowned veteran TCM doctors is achieved, thereby improving the accuracy and interpretability of syndrome differentiation and diagnosis, while reducing the complexity of data annotation. This solves the technical problem of limited accuracy and insufficient interpretability of intelligent syndrome diagnosis models caused by the lack of syndrome element annotation and the high difficulty of annotation in the medical case data of renowned veteran TCM doctors.

[0052] In this embodiment, a multi-layered training and fine-tuning method is employed to construct the syndrome diagnosis model, mainly including three stages: pre-training, continued training, and fine-tuning optimization. The overall framework is as follows: Figure 3As shown. Specifically, in the pre-training stage, the BERT model is trained under self-supervised or weakly supervised conditions based on a basic TCM corpus (including key information such as symptoms, syndrome diagnosis, and symptom analysis), enabling the model to initially grasp TCM terminology and basic diagnostic and treatment concepts. In the continued training stage, the pre-trained BERT model is trained under supervised conditions based on a database of medical records from renowned TCM doctors (including key information such as syndrome elements, symptoms, and standard syndromes), achieving syndrome element prediction and syndrome prediction, allowing the model to better learn the common diagnostic and treatment ideas of renowned TCM doctors. In the fine-tuning stage, the continued-trained BERT model is trained under supervised conditions based on a specific dataset of medical records from renowned TCM doctors (including key information such as symptoms and standard syndromes, as well as generated pseudo-labels for syndrome elements and corrective measures), achieving syndrome element prediction and syndrome prediction, enabling the model to accurately learn the personalized diagnostic and treatment experience of a pre-defined renowned TCM doctor (such as a professor), forming a syndrome diagnosis model with a unique diagnostic perspective. The following combines... Figure 3 The training logic described above provides a detailed explanation of steps S202 to S208.

[0053] In step S202 above, the aim is to conduct unsupervised training on the initial model (such as a pre-trained language model like BERT) using a basic TCM corpus containing a large amount of TCM literature and basic knowledge. This enables the model to have basic TCM vocabulary and contextual understanding capabilities, laying the groundwork for subsequent supervised training and ensuring that the model can understand the professional knowledge and expressions in the TCM corpus.

[0054] Optionally, an unsupervised training process is performed on the initial model using a basic TCM corpus to obtain a first diagnostic model. This includes: acquiring first symptom information and a first syndrome diagnosis result, wherein the first symptom information is the symptom information of any patient in the basic TCM corpus, and the first syndrome diagnosis result is the syndrome diagnosis result of any patient in the basic TCM corpus; predicting the first symptom information and the first syndrome diagnosis result using the initial model to obtain a first prediction result, wherein the first prediction result represents the probability that the first syndrome diagnosis result is the true diagnosis result of the first symptom information; determining a first loss value between the first prediction result and the first true label using a first loss function, wherein the first true label represents the true diagnosis result of the first symptom information; and optimizing the model parameters of the initial model based on the first loss value to obtain the first diagnostic model. A detailed analysis follows:

[0055] In this embodiment, in addition to the masked language model (MLM) and NSP (Next Sentence Prediction) tasks used in the BERT model (i.e., the initial model), the specific characteristics of the medical case data are fully considered, and an additional SP (Syndrome Prediction) task is added. In the SP task, the symptom information (i.e., the first symptom information) of any patient in the basic TCM corpus and any syndrome diagnosis result (i.e., the first syndrome diagnosis result) are simultaneously input into the BERT model for prediction. The BERT model is then asked to determine whether the diagnosis result is correct, and the first prediction result is obtained.

[0056] It should be noted that the first prediction result is artificially set to have a 50% probability that the diagnosis result of the first syndrome is the actual diagnosis result of the patient corresponding to the first symptom information, and a 50% probability that the diagnosis result of the first syndrome is the diagnosis result of a random patient.

[0057] More specifically, the first symptom information and the first syndrome diagnosis result can be combined, represented as "[CLS] + Symptom + [SEP] + Diagnosis Result", where [CLS] and [SEP] are the classification and segmentation tokens of the BERT model. Subsequently, the combined result is input into the BERT model, and the first token output by the BERT model is processed using an MLP. Convert it to a real value, and then use the sigmoid function to convert it into a predicted probability. The specific expression is as follows:

[0058]

[0059] In the formula, This indicates the first prediction result. This indicates that the diagnosis of the first syndrome in this sample is the true diagnosis of the patient corresponding to the information of the first symptom. This indicates that the diagnosis result of the first syndrome in this sample is the diagnosis result of a randomly selected patient.

[0060] Furthermore, through the first loss function The loss for this SP task is calculated using the following expression:

[0061]

[0062] In the formula, This represents the real label corresponding to this task, namely the first real label mentioned above.

[0063] Total loss function during pre-training phase as follows:

[0064]

[0065] In the formula, This represents the loss function corresponding to the MLM task. This represents the loss function corresponding to the NSP task. This represents the loss function (first loss function) corresponding to the SP task.

[0066] Finally, the model weight parameters are optimized based on the loss function to obtain the pre-trained BERT model, i.e., the first diagnostic model, denoted as . .

[0067] It is worth noting that in the above pre-training stage, all diagnostic results are the original texts in the basic TCM corpus and do not require standardization. The MLM and NSP tasks are unsupervised training methods and do not require additional labeled data. The SP task is a weakly supervised training method and can directly extract diagnostic results from the TCM corpus without additional manual processing.

[0068] In step S204 above, based on the first diagnostic model obtained through unsupervised training, supervised training can be continued using a general database of medical records from renowned traditional Chinese medicine practitioners. This database contains abundant records of general medical cases from renowned practitioners and clear information on syndrome elements. Through supervised learning, the model can learn the general diagnostic and treatment logic of these practitioners from actual medical cases, including the mapping rules from symptoms to syndrome elements and the association patterns from syndrome elements to syndromes, thereby improving the model's basic accuracy and universality in syndrome differentiation and diagnosis.

[0069] Optionally, the first diagnostic model is trained in a supervised manner using a database of medical records from renowned traditional Chinese medicine practitioners. This includes: acquiring second symptom information and determining a first feature vector corresponding to the second symptom information, where the second symptom information is the symptom information of any patient in the database of medical records from renowned traditional Chinese medicine practitioners; predicting the first feature vector using the first diagnostic model to obtain a second prediction result, where the second prediction result represents the syndrome diagnosis result corresponding to the second symptom information; concatenating the first feature vector and the second prediction result to obtain a second feature vector; predicting the second feature vector using the first diagnostic model to obtain a third prediction result, where the third prediction result represents the syndrome diagnosis result corresponding to the second symptom information; and training the first diagnostic model based on the second and third prediction results. A detailed analysis follows:

[0070] In this embodiment, the syndrome element can be added as a mediating variable to the prediction process of the model, and the pre-trained model can be analyzed using a general database of medical records from renowned traditional Chinese medicine practitioners that contains syndrome element information. The model continues training.

[0071] First, the symptom information (i.e., the second symptom information) of any patient (e.g., patient i) from the database of medical records of renowned traditional Chinese medicine practitioners is used as input, and the feature vector (i.e., the first feature vector) corresponding to this symptom text is determined. Then, through... The model predicts the first feature vector to obtain the corresponding syndrome diagnosis result (i.e., the second prediction result). The specific expression is as follows:

[0072]

[0073] In the formula, This indicates the second prediction result. This represents the sigmoid function. Represents the first eigenvector. This represents the weight matrix in a neural network.

[0074] Finally, the diagnostic results of the syndrome elements and the corresponding feature vectors are concatenated, and then... The model then predicts the concatenated feature vector (i.e., the second feature vector) again to obtain the syndrome diagnosis result (i.e., the third prediction result). The specific expression is as follows:

[0075]

[0076]

[0077] In the formula, This represents the second feature vector after concatenation. This indicates the third prediction result. This represents the softmax normalization function. This represents the weight matrix in a neural network.

[0078] In the above process, the first diagnostic model is trained based on the second and third prediction results, including: determining a second loss value between the second prediction result and the second true label using a second loss function, wherein the second true label represents the true syndrome information corresponding to the second symptom information; determining a third loss value between the third prediction result and the third true label using a third loss function, wherein the third true label represents the true syndrome information corresponding to the second symptom information; and optimizing the model parameters of the first diagnostic model based on the second loss value, the third loss value, and preset hyperparameters to obtain the second diagnostic model. A detailed analysis follows:

[0079] In this embodiment, based on the second prediction result obtained from the above-mentioned element prediction and the third prediction result obtained from the syndrome prediction, combined with the corresponding loss function, optimization can be achieved. The model's weight parameters are obtained after secondary training. The model is the second diagnostic model mentioned above.

[0080] Among them, the second loss function corresponding to the evidence element prediction task as follows:

[0081]

[0082] In the formula, , indicating the quantity of evidence elements; , indicating the number of patients; This represents the predicted probability of the i-th patient regarding the k-th factor, i.e., the second prediction result; This represents the true label of the i-th patient regarding the k-th evidence element, i.e., the second true label.

[0083] The third loss function corresponding to the syndrome prediction task as follows:

[0084]

[0085] In the formula, , indicating the number of symptoms; This represents the predicted probability of the i-th patient regarding the j-th syndrome, i.e., the third prediction result; This represents the true label of the i-th patient regarding the j-th symptom, i.e., the third true label.

[0086] The total loss function during the continued training phase as follows:

[0087]

[0088] In the formula, These are preset hyperparameters used to control the weights of the two loss functions.

[0089] In step S206 above, the aim is to generate pseudo-labels of syndrome elements based on a database of medical records of famous traditional Chinese medicine doctors, and to perform correction processing on these pseudo-labels of syndrome elements to reduce the bias caused by insufficient data or incomplete labeling. This allows the model to learn and simulate the diagnostic logic and treatment preferences of the famous traditional Chinese medicine doctor even in the absence of complete syndrome element information, thereby improving the model's personalized adaptability.

[0090] Optionally, determining the pseudo-label of the syndrome element corresponding to a specific database of medical records of renowned traditional Chinese medicine practitioners includes: obtaining third symptom information and determining a third feature vector corresponding to the third symptom information, wherein the third symptom information is the symptom information of any patient in the database of medical records of renowned traditional Chinese medicine practitioners; predicting the third feature vector using a second diagnostic model to obtain a fourth prediction result, wherein the fourth prediction result is used to represent the syndrome element diagnosis result corresponding to the third symptom information; determining the pseudo-label of the syndrome element corresponding to the third symptom information based on the fourth prediction result and a preset threshold; or, obtaining a list of syndrome elements corresponding to the third symptom information and determining the pseudo-label of the syndrome element corresponding to the third symptom information based on the list of syndrome elements. Specific analysis is as follows:

[0091] In this embodiment of the application, when fine-tuning the model, it is necessary to generate pseudo-labels of syndrome elements based on a specific database of medical records of a famous traditional Chinese medicine doctor, and to correct the pseudo-labels so that the prediction effect of the model is close to the prediction effect obtained by training with real syndrome element labels.

[0092] Suppose a database of medical records of renowned traditional Chinese medicine practitioners contains N patients. For each patient, the following two methods are provided for generating pseudo-labels for syndrome elements.

[0093] Method 1: Determine the pseudo-label of syndrome elements based on the syndrome name.

[0094] Specifically, each syndrome corresponds to a list of syndrome elements. For example, for the syndrome of qi stagnation and blood stasis, the list of syndrome elements is {qi stagnation, blood stasis}. For any patient i in a specific database of medical records from renowned traditional Chinese medicine practitioners, The symptom information (i.e., the third symptom information) is determined, and the pseudo-label of the syndrome element in the syndrome element list corresponding to the third symptom information is determined to be 1, and the pseudo-label of the remaining syndrome elements is determined to be 0.

[0095] For example, if patient i is diagnosed with Qi stagnation and blood stasis syndrome, then the pseudo-labels "Qi stagnation" and "blood stasis" in the corresponding syndrome element list are set to 1, and the rest are set to 0.

[0096] Method 2: Based on The model identifies pseudo-labels for evidence elements.

[0097] The feature vector corresponding to the third symptom information of patient i (i.e., the third feature vector) is input into the model after further training. In this process, the corresponding diagnostic result (i.e., the fourth prediction result) is obtained, denoted as... , Subsequently, based on the preset threshold... Determine the final pseudo-tags, if Then the corresponding evidence element pseudo-label will be... Set it to 1, otherwise set it to 0.

[0098] Among them, the fourth loss function corresponds to the evidence element prediction and evidence element pseudo-label correction stage. as follows:

[0099]

[0100] Further, the pseudo-labels of the syndrome elements are corrected to obtain the target pseudo-label, including: determining the first deviation probability and the second deviation probability corresponding to the third symptom information, wherein the first deviation probability is used to represent the probability that the pseudo-label of the syndrome element is 0 when the fourth true label is 1, the second deviation probability is used to represent the probability that the pseudo-label of the syndrome element is 1 when the fourth true label is 0, and the fourth true label is used to represent the true syndrome element information corresponding to the third symptom information; the pseudo-labels of the syndrome elements are corrected based on the first deviation probability and the second deviation probability to obtain the target pseudo-label. The specific analysis is as follows:

[0101] In this embodiment, the obtained false labels of evidence elements need to be corrected to make them closer to the true labels. Specifically, a first deviation probability is first defined. Second deviation probability ,in, This represents the pseudo-label of the syndrome element corresponding to the third symptom information of patient i. This represents the true evidence element label corresponding to the third symptom information of patient i. Subsequently, based on the first bias probability... Second deviation probability Corrective measures are taken to address false labels on evidence.

[0102] More specifically, the first bias probability Second deviation probability It can be estimated using the copula tool, with the first bias probability. For example, the determination method is as follows: Determine the anchor sample corresponding to the third symptom information, and determine the cumulative density function and corresponding inverse function of the anchor sample based on the standard normal distribution. The anchor sample represents the sample in the third symptom information whose predicted probability obtained through the second diagnostic model is lower than a preset probability threshold. Determine the correlation coefficient matrix between the symptom elements corresponding to the anchor sample, and determine the log-likelihood function corresponding to the anchor sample based on the correlation coefficient matrix. Determine the normal distribution parameters corresponding to the anchor sample based on the log-likelihood function. Determine the first bias probability based on the normal distribution parameters, cumulative density function, and inverse function. Specific analysis is as follows:

[0103] First, a separate BERT model can be trained (denoted as BERT). ), used to predict false labels for evidence elements. For the first Individual evidence was selected from all patients. The lowest probability of prediction We take n samples as anchor samples, and assume that these anchor samples follow a standard normal distribution, i.e. For each patient This allows us to find the predicted value of the corresponding anchor sample. ,in, Indicates the patient The corresponding number of anchor point samples, This indicates that the predicted value corresponds to the first... Individual evidence.

[0104] Secondly, the copula tool is used to model and estimate the Beta distribution parameters and correlation coefficients corresponding to the anchor samples. For any Define a new variable ,in, express The cumulative density function of the distribution, It represents the inverse function of the cumulative density function, and it is easy to know that... It follows a standard normal distribution. (Hypothetical variable) With variables The correlation coefficient between them is Then the patient All anchor point samples have corresponding correlation coefficients between each pair of evidence elements, and these correlation coefficients can form a correlation coefficient matrix. .

[0105] Then for First, it is converted into Then the corresponding log-likelihood function is obtained. for:

[0106]

[0107] in, This indicates that the mean is 0 and the covariance matrix is... The joint probability density function of the multivariate normal distribution. The parameter is The probability density function of the standard normal distribution. Then, for all patients, the total log-likelihood function. for:

[0108]

[0109] Subsequently, by maximizing The normal distribution parameters can then be obtained. and correlation coefficient .

[0110] Finally, the first bias probability can be obtained by using the following estimation steps. .

[0111] S1: Based on all correlation coefficients Sampling from a multivariate normal distribution Next, record the first The result of the second sampling is ;

[0112] S2: Transform each element in the sample using the following method: ,in, This represents the standard normal cumulative density function. The parameter is The inverse function of the cumulative density function of the standard normal distribution;

[0113] S3: Will The estimated value is denoted as ,but .

[0114] For the second deviation probability The estimate, selection The highest probability of prediction Take one sample, denoted as the anchor sample, and repeat the above steps to obtain the anchor sample. .

[0115] In step S208 above, based on the second diagnostic model, further fine-tuning and training are performed using specific case data of renowned traditional Chinese medicine (TCM) doctors combined with corresponding target pseudo-labels. This yields a syndrome diagnosis model that accurately reflects the diagnostic experience of that renowned TCM doctor. This step primarily focuses on the diagnostic style and medication characteristics of a specific renowned TCM doctor. Through fine-tuning, the model can learn the unique diagnostic and treatment ideas of this renowned TCM doctor in a refined manner, achieving the effect of both inheriting general diagnostic wisdom and taking into account personalized diagnostic experience. This makes the model more targeted and practical in real-world applications.

[0116] Optionally, the second diagnostic model is optimized and trained using a database of medical records from renowned traditional Chinese medicine practitioners and corresponding target pseudo-labels to obtain a syndrome diagnosis model. This includes: optimizing the fourth prediction result based on the target pseudo-labels to obtain the target prediction result; determining the fourth loss value between the target prediction result and the syndrome pseudo-label using a fourth loss function; concatenating the third feature vector and the target prediction result to obtain the fourth feature vector; predicting the fourth feature vector using the second diagnostic model to obtain the fifth prediction result, where the fifth prediction result represents the syndrome diagnosis result corresponding to the third symptom information; determining the fifth loss value between the fifth prediction result and the fifth true label using a third loss function, where the fifth true label represents the true syndrome information corresponding to the third symptom information; and optimizing the model parameters of the second diagnostic model based on the fourth and fifth loss values ​​to obtain the syndrome diagnosis model.

[0117] In this embodiment of the application, based on the first deviation probability Second deviation probability It can be deduced that And further inferences .

[0118] Based on this, the concatenated vector containing the feature vector corresponding to the third symptom information and the syndrome diagnosis result will be... Change to:

[0119]

[0120] at this time, This represents the feature vector corresponding to the third symptom information, i.e., the third feature vector; This indicates the diagnostic result of the syndrome element corresponding to the third symptom information, i.e., the fourth predictive result; This represents the target prediction result after optimizing the fourth prediction result based on the target pseudo-label.

[0121] Among them, the total loss function during the model fine-tuning stage as follows:

[0122]

[0123] In the formula, This represents the fourth loss function corresponding to the element prediction in the fine-tuning phase. For the corresponding weight parameters, This represents the fifth loss function corresponding to the fine-tuning stage and syndrome prediction.

[0124] By minimizing the total loss function This allows us to obtain a syndrome diagnosis model that inherits the diagnostic experience of the veteran traditional Chinese medicine doctor. .

[0125] It is worth noting that, due to the use of pseudo-labels for syndrome elements, when applied to the inheritance work of any renowned veteran TCM doctor, it is not necessary to label the syndrome element information in the medical record database of a specific renowned veteran TCM doctor. This greatly reduces the difficulty of data preparation in the inheritance work of renowned veteran TCM doctors and broadens the scope of application.

[0126] In this embodiment of the application, in addition to using a trained syndrome diagnosis model In addition to direct prediction, this approach also allows for in-depth analysis and inheritance of the diagnostic experience of renowned veteran TCM doctors, particularly by revealing key information in the diagnostic process through the intermediate variable of syndrome elements. The implementation can be as follows: Predict the third feature vector using a syndrome diagnosis model to obtain the target syndrome element diagnosis result; concatenate the third feature vector and the target syndrome element diagnosis result to obtain the target feature vector, and predict the target feature vector using the syndrome diagnosis model to obtain the target syndrome diagnosis result; determine the first syndrome element list and the second syndrome element list corresponding to the first and second target syndrome diagnosis results, respectively. The first and second target syndrome diagnosis results are any two identical syndrome diagnosis results from the target syndrome diagnosis results. The list represents a predetermined number of the most probable syndrome elements in the target syndrome diagnosis results. Based on the first and second syndrome element lists, the syndrome element ratios for the first and second syndromes are determined. The first syndrome is any syndrome in the first target syndrome diagnosis results, and the second syndrome is any syndrome in the second target syndrome diagnosis results. The syndrome element ratios represent the relative proportions of the frequencies of each syndrome element in the first syndrome element list under the first syndrome, and the relative proportions of the frequencies of each syndrome element in the second syndrome element list under the second syndrome. Based on the syndrome element ratios, the pre-defined diagnostic and treatment focus of renowned traditional Chinese medicine practitioners is determined.

[0127] Specifically, firstly, utilize the previously trained... The model analyzes the input symptom information (here, we still use the symptom information of any patient from a database of medical records of renowned traditional Chinese medicine practitioners as an example, i.e., the third symptom information) to predict the optimal combination of syndrome elements, i.e., the target syndrome element diagnosis result. Then, the third feature vector corresponding to the third symptom information is concatenated with the target syndrome element diagnosis result to generate a target feature vector that integrates intuitive symptom information with a deep understanding of syndrome elements. The model makes predictions again, obtaining the target syndrome diagnosis results. Next, the model-predicted target syndrome diagnoses are analyzed, particularly for recurring syndrome diagnoses (i.e., the same syndrome in the first and second target syndrome diagnoses). The top N most probable syndrome elements associated with these recurring syndromes are extracted to form a syndrome element list (i.e., the first syndrome element list and the second syndrome element list). Subsequently, the relative frequencies of each syndrome element in the first and second syndrome element lists are compared and analyzed to obtain the pre-set syndrome element ratios under different syndrome diagnoses by renowned traditional Chinese medicine practitioners. Finally, based on the syndrome element ratio results, the diagnostic and treatment focus of the renowned traditional Chinese medicine practitioners is determined, such as the syndrome element combinations that they particularly focus on during the syndrome differentiation process.

[0128] The above methods not only help young doctors understand the diagnostic reasoning of renowned veteran TCM doctors, but also provide a precise quantitative tool for TCM teaching and research, transforming the inheritance of experience from abstract theoretical learning into concrete and intuitive practical guidance. By imitating the diagnostic process of renowned veteran TCM doctors, young doctors can more effectively grasp the essence of syndrome differentiation and treatment. Simultaneously, the interpretability of the model provides a theoretical basis for clinical decision-making, enhancing the model's credibility and applicability in practical applications.

[0129] In this embodiment, unsupervised learning is first used to enable the model to master TCM terminology and basic concepts in a basic TCM corpus. Then, supervised learning is used to learn widely accepted diagnostic principles and treatment procedures in a general database of medical records from renowned veteran TCM doctors. Subsequently, on a specific database of medical records from renowned veteran TCM doctors, pseudo-labels of syndrome elements are generated and corrected, allowing the model to specifically learn the unique diagnostic ideas and habits of that veteran TCM doctor. The resulting syndrome diagnosis model can not only make accurate diagnostic diagnoses based on patients' unstructured symptom texts, but also reveal and inherit the essence of the veteran TCM doctors' experience. This method not only significantly improves the accuracy and interpretability of syndrome differentiation and treatment, but also cleverly solves the problems of limited data on renowned veteran TCM doctors and complex syndrome element labeling, reducing the labeling workload for doctors, providing a convenient learning platform for young doctors, promoting the modernization of TCM diagnosis and treatment, and truly realizing the digital inheritance and application of the valuable experience of renowned veteran TCM doctors.

[0130] Figure 4 This is a flowchart of a symptom prediction method according to an embodiment of this application, such as... Figure 4 As shown, the method includes the following steps:

[0131] Step S402: Obtain the patient's symptom information and determine the feature vector corresponding to the symptom information.

[0132] In step S402 above, it is first necessary to collect various symptom information reported by the patient, including but not limited to subjective feelings, lifestyle and dietary habits, sleep patterns, bowel and bladder function, tongue and pulse examination, etc. Subsequently, through preprocessing and natural language processing techniques, this unstructured symptom text information is converted into machine-readable feature vectors, providing information input for subsequent intelligent diagnosis.

[0133] Step S404: The feature vector is predicted using the syndrome diagnosis model to obtain the syndrome element prediction results corresponding to the symptom information. The syndrome diagnosis model is trained in the following ways: Unsupervised training is performed on the initial model using a basic TCM corpus to obtain the first diagnostic model. The basic TCM corpus contains multiple TCM basic knowledge texts. Supervised training is performed on the first diagnostic model using a database of medical records from renowned TCM doctors to obtain the second diagnostic model. This database contains general medical records from multiple renowned TCM doctors and syndrome element information corresponding to the patient's symptoms. The syndrome element information is used to guide the first diagnostic model in learning the general diagnostic logic of renowned TCM doctors. Pseudo-labels for syndrome elements corresponding to a specific database of medical records from renowned TCM doctors are determined, and the pseudo-labels are corrected to obtain target pseudo-labels. This specific database contains personal medical records from preset renowned TCM doctors. The second diagnostic model is optimized and trained using the specific database of medical records from renowned TCM doctors and the corresponding target pseudo-labels to obtain the syndrome diagnosis model.

[0134] In step S404 above, a pre-trained, multi-layered syndrome diagnosis model can be used to perform in-depth analysis of the feature vectors corresponding to the patient's symptoms, obtaining the syndrome element prediction result that best matches the symptom information, i.e., the specific manifestations of the disease in different organs, meridians, qi and blood. This syndrome diagnosis model understands TCM terminology through unsupervised learning, learns the general diagnostic and treatment logic of renowned TCM doctors through supervised training, and finally inherits personalized treatment ideas through fine-tuning based on specific medical cases of renowned TCM doctors.

[0135] Step S406: Concatenate the evidence prediction results and feature vectors to obtain the target feature vector.

[0136] In step S406 above, after obtaining the preliminary syndrome element prediction results, they can be merged with the original feature vector to create a target feature vector containing symptom text information and syndrome element information. This enhances the input information of the model and ensures that the model not only relies on symptom descriptions when making syndrome predictions, but also takes into account the intermediate results obtained through syndrome element diagnosis, thereby improving the accuracy and depth of syndrome differentiation and diagnosis.

[0137] Step S408: The target feature vector is predicted by the syndrome diagnosis model to obtain the syndrome prediction result corresponding to the symptom information, wherein the syndrome prediction result is used to provide a basis for the patient's syndrome differentiation diagnosis.

[0138] In step S408 above, the target feature vector needs to be input into the syndrome diagnosis model again. Based on the integrated information, the model can more accurately identify the patient's pathogenesis state and obtain the syndrome prediction results corresponding to the patient's symptom information, thereby providing a basis for subsequent treatment and realizing the formulation of scientific personalized treatment plans.

[0139] Through the above steps S402 to S408, the goal of deep learning and accurate simulation of the diagnostic and treatment logic of renowned traditional Chinese medicine practitioners is achieved, thereby improving the accuracy and interpretability of diagnostic methods, reducing the need for data annotation, and accelerating the inheritance of traditional Chinese medicine experience and the training of young doctors.

[0140] It should be noted that the syndrome diagnosis model obtained from the above training is used to implement the syndrome prediction method. Therefore, the relevant explanations about model training in the training method of the above syndrome diagnosis model also apply to the syndrome prediction method, and will not be repeated here.

[0141] According to embodiments of this application, a training apparatus for a syndrome diagnosis model is provided. It should be noted that the training apparatus for the syndrome diagnosis model of this application can be used to execute the training method for the syndrome diagnosis model provided in the embodiments of this application. The training apparatus for the syndrome diagnosis model provided in the embodiments of this application will be described below.

[0142] Figure 5 This is a structural diagram of a training device for a syndrome diagnosis model provided according to an embodiment of this application. Figure 5 As shown, the device includes:

[0143] The first training module 50 is used to perform unsupervised training on the initial model using a basic TCM corpus to obtain the first diagnostic model. The basic TCM corpus contains multiple TCM basic knowledge texts.

[0144] The second training module 52 is used to conduct supervised training on the first diagnostic model through a general database of medical records of famous traditional Chinese medicine doctors to obtain a second diagnostic model. The general database of medical records of famous traditional Chinese medicine doctors contains general diagnosis and treatment records of multiple famous traditional Chinese medicine doctors, as well as syndrome element information corresponding to the patient's symptoms. The syndrome element information is used to guide the first diagnostic model to learn the general diagnosis and treatment logic of famous traditional Chinese medicine doctors.

[0145] The correction module 54 is used to determine the pseudo-label of syndrome element corresponding to the medical record database of a specific famous traditional Chinese medicine doctor, and to perform correction processing on the pseudo-label of syndrome element to obtain the target pseudo-label. The medical record database of a specific famous traditional Chinese medicine doctor contains the personal diagnosis and treatment records of the preset famous traditional Chinese medicine doctor.

[0146] The third training module 56 is used to optimize and train the second diagnostic model using a database of medical records from specific renowned traditional Chinese medicine practitioners and corresponding target pseudo-labels, thereby obtaining a syndrome diagnosis model.

[0147] Through the first training module, second training module, correction module, and third training module in the training device of the above-mentioned syndrome diagnosis model, the goal of deeply understanding and accurately simulating the syndrome differentiation and treatment logic of famous veteran TCM doctors is achieved. This realizes the technical effect of improving the accuracy and interpretability of syndrome diagnosis, while reducing the complexity of data annotation. In turn, it solves the technical problem of limited accuracy and insufficient interpretability of intelligent syndrome diagnosis model caused by the lack of syndrome element annotation and the high difficulty of annotation in the medical case data of famous veteran TCM doctors.

[0148] In the training device for the syndrome diagnosis model provided in this application embodiment, the first training module is further used to acquire first symptom information and first syndrome diagnosis result, wherein the first symptom information is the symptom information of any patient in the basic TCM corpus, and the first syndrome diagnosis result is the syndrome diagnosis result of any patient in the basic TCM corpus; the first symptom information and the first syndrome diagnosis result are predicted by an initial model to obtain a first prediction result, wherein the first prediction result is used to represent the probability that the first syndrome diagnosis result is the true diagnosis result of the first symptom information; a first loss value is determined by a first loss function between the first prediction result and the first true label, wherein the first true label is used to represent the true diagnosis result of the first symptom information; and the model parameters of the initial model are optimized based on the first loss value to obtain a first diagnosis model.

[0149] In the training device for the syndrome diagnosis model provided in this application embodiment, the second training module is further used to acquire second symptom information and determine a first feature vector corresponding to the second symptom information, wherein the second symptom information is the symptom information of any patient in the general database of medical records of famous traditional Chinese medicine practitioners; predict the first feature vector through the first diagnostic model to obtain a second prediction result, wherein the second prediction result is used to represent the syndrome diagnosis result corresponding to the second symptom information; concatenate the first feature vector and the second prediction result to obtain a second feature vector; predict the second feature vector through the first diagnostic model to obtain a third prediction result, wherein the third prediction result is used to represent the syndrome diagnosis result corresponding to the second symptom information; and train the first diagnostic model based on the second prediction result and the third prediction result.

[0150] In the training device for the syndrome diagnosis model provided in this application embodiment, the second training module is further configured to determine a second loss value between the second prediction result and the second true label through a second loss function, wherein the second true label is used to represent the true syndrome information corresponding to the second symptom information; determine a third loss value between the third prediction result and the third true label through a third loss function, wherein the third true label is used to represent the true syndrome information corresponding to the second symptom information; and optimize the model parameters of the first diagnostic model based on the second loss value, the third loss value, and preset hyperparameters to obtain the second diagnostic model.

[0151] In the training device of the syndrome diagnosis model provided in this application embodiment, the correction module is further used to acquire third symptom information and determine the third feature vector corresponding to the third symptom information, wherein the third symptom information is the symptom information of any patient in a specific database of medical records of famous traditional Chinese medicine doctors; predict the third feature vector through the second diagnostic model to obtain a fourth prediction result, wherein the fourth prediction result is used to represent the syndrome element diagnosis result corresponding to the third symptom information; determine the syndrome element pseudo-label corresponding to the third symptom information based on the fourth prediction result and a preset threshold; or, acquire a list of syndrome elements corresponding to the third symptom information and determine the syndrome element pseudo-label corresponding to the third symptom information based on the list of syndrome elements.

[0152] In the training device of the syndrome diagnosis model provided in this application embodiment, the correction module is further used to determine the first deviation probability and the second deviation probability corresponding to the third symptom information. The first deviation probability is used to represent the probability that the pseudo-label of the syndrome element is 0 when the fourth true label is 1, and the second deviation probability is used to represent the probability that the pseudo-label of the syndrome element is 1 when the fourth true label is 0. The fourth true label is used to represent the true syndrome element information corresponding to the third symptom information. The pseudo-label of the syndrome element is corrected according to the first deviation probability and the second deviation probability to obtain the target pseudo-label.

[0153] In the training device of the syndrome diagnosis model provided in this application embodiment, the correction module is further used to determine the anchor sample corresponding to the third symptom information, and to determine the cumulative density function and the corresponding inverse function of the anchor sample based on the standard normal distribution. The anchor sample is used to represent the sample in the third symptom information whose predicted probability obtained by the second diagnosis model is lower than a preset probability threshold; to determine the correlation coefficient matrix between the syndrome elements corresponding to the anchor sample, and to determine the log-likelihood function corresponding to the anchor sample based on the correlation coefficient matrix; to determine the normal distribution parameter corresponding to the anchor sample based on the log-likelihood function; and to determine the first deviation probability based on the normal distribution parameter, the cumulative density function and the inverse function.

[0154] In the training apparatus of the syndrome diagnosis model provided in this application embodiment, the third training module is further used to optimize the fourth prediction result based on the target pseudo-label to obtain the target prediction result; determine the fourth loss value between the target prediction result and the syndrome pseudo-label through the fourth loss function; concatenate the third feature vector and the target prediction result to obtain the fourth feature vector; predict the fourth feature vector through the second diagnosis model to obtain the fifth prediction result, wherein the fifth prediction result is used to represent the syndrome diagnosis result corresponding to the third symptom information; determine the fifth loss value between the fifth prediction result and the fifth true label through the third loss function, wherein the fifth true label is used to represent the true syndrome information corresponding to the third symptom information; optimize the model parameters of the second diagnosis model based on the fourth loss value and the fifth loss value to obtain the syndrome diagnosis model.

[0155] In the training apparatus for the syndrome diagnosis model provided in this application embodiment, the third training module is further configured to predict the third feature vector using the syndrome diagnosis model to obtain the target syndrome element diagnosis result; concatenate the third feature vector and the target syndrome element diagnosis result to obtain the target feature vector, and predict the target feature vector using the syndrome diagnosis model to obtain the target syndrome diagnosis result; and determine the first syndrome element list and the second syndrome element list corresponding to the first target syndrome diagnosis result and the second target syndrome diagnosis result, respectively, wherein the first target syndrome diagnosis result and the second target syndrome diagnosis result are any two identical syndrome diagnosis results in the target syndrome diagnosis results. The first and second syndrome element lists are a preset set of the most probable syndrome elements in the target syndrome element diagnosis results. The syndrome element ratio results for the first and second syndromes are determined based on the first and second syndrome element lists. The first syndrome is any syndrome in the first target syndrome diagnosis results, and the second syndrome is any syndrome in the second target syndrome diagnosis results. The syndrome element ratio results are used to determine the relative proportion of the frequency of occurrence of each syndrome element in the first syndrome element list under the first syndrome, and the relative proportion of the frequency of occurrence of each syndrome element in the second syndrome element list under the second syndrome. The diagnostic and treatment focus of the preset renowned traditional Chinese medicine practitioners is determined based on the syndrome element ratio results.

[0156] According to embodiments of this application, a syndrome diagnosis device is also provided. It should be noted that the syndrome diagnosis device of this application can be used to execute the syndrome prediction method provided in the embodiments of this application. The syndrome diagnosis device provided in the embodiments of this application will be described below.

[0157] Figure 6 This is a structural diagram of a syndrome prediction device provided according to an embodiment of this application. Figure 6 As shown, the device includes:

[0158] The acquisition module 60 is used to acquire the patient's symptom information and determine the feature vector corresponding to the symptom information;

[0159] The first prediction module 62 is used to predict feature vectors using a syndrome diagnosis model to obtain syndrome element prediction results corresponding to symptom information. The syndrome diagnosis model is trained as follows: An initial model is unsupervised-trained using a basic TCM corpus to obtain a first diagnostic model, where the basic TCM corpus contains multiple TCM basic knowledge texts; the first diagnostic model is supervised-trained using a database of medical records from renowned TCM doctors to obtain a second diagnostic model, where the database contains general treatment records from multiple renowned TCM doctors and syndrome element information corresponding to patient symptoms. This syndrome element information guides the first diagnostic model in learning the general treatment logic of renowned TCM doctors; pseudo-labels for syndrome elements are determined corresponding to a specific database of medical records from renowned TCM doctors, and these pseudo-labels are corrected to obtain target pseudo-labels, where the specific database contains personal treatment records from preset renowned TCM doctors; the second diagnostic model is optimized and trained using the specific database of medical records from renowned TCM doctors and the corresponding target pseudo-labels to obtain the syndrome diagnosis model.

[0160] The splicing module 64 is used to splice the evidence prediction results and feature vectors to obtain the target feature vector;

[0161] The second prediction module 66 is used to predict the target feature vector through the syndrome diagnosis model to obtain the syndrome prediction result corresponding to the symptom information, wherein the syndrome prediction result is used to provide a basis for the patient's syndrome diagnosis.

[0162] Through the acquisition module, first prediction module, splicing module and second prediction module in the training device of the above syndrome diagnosis model, the goal of deep learning and accurate simulation of the syndrome differentiation and treatment logic of famous veteran TCM doctors is achieved. This realizes the technical effect of improving the accuracy and interpretability of syndrome differentiation and diagnosis, reducing the data annotation requirements, and accelerating the inheritance of TCM experience and the training of young doctors.

[0163] This application also provides an electronic device, including: a memory and a processor, wherein the memory is used to store program instructions; the processor is connected to the memory and is used to execute the training method or the syndrome prediction method that implements the above-mentioned syndrome diagnosis model.

[0164] It should be noted that the aforementioned electronic equipment is used to perform Figure 2 The training method or the syndrome diagnosis model shown Figure 4 The syndrome prediction method shown above, therefore, the training method of the above syndrome diagnosis model or the relevant explanations in the syndrome prediction method are also applicable to this electronic device, and will not be repeated here.

[0165] This application embodiment also provides a non-volatile storage medium, which includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the above-mentioned training method or syndrome prediction method of syndrome diagnosis model by running the computer program.

[0166] It should be noted that the aforementioned non-volatile storage media is used for execution. Figure 2 The training method or the syndrome diagnosis model shown Figure 4 The syndrome prediction method shown above, therefore, the training method of the above syndrome diagnosis model or the relevant explanations in the syndrome prediction method also apply to this non-volatile storage medium, and will not be repeated here.

[0167] This application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the training method or the syndrome prediction method of the above-mentioned syndrome diagnosis model.

[0168] It should be noted that the above-mentioned computer program product is used to execute Figure 2 The training method or the syndrome diagnosis model shown Figure 4 The syndrome prediction method shown above, therefore, the training method of the above syndrome diagnosis model or the relevant explanations in the syndrome prediction method also apply to this computer program product, and will not be repeated here.

[0169] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0170] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0171] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0172] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0173] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0174] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0175] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A training method for a syndrome diagnosis model, characterized in that, include: The initial model was trained unsupervised using a basic TCM corpus to obtain the first diagnostic model. The basic TCM corpus contains multiple TCM basic knowledge texts. The first diagnostic model is trained in a supervised manner using a database of medical records of renowned traditional Chinese medicine practitioners to obtain a second diagnostic model. The database contains general medical records of multiple renowned traditional Chinese medicine practitioners, as well as syndrome element information corresponding to the patient's symptoms. The syndrome element information is used to guide the first diagnostic model to learn the general diagnostic logic of renowned traditional Chinese medicine practitioners. Identify the pseudo-labels of syndrome elements corresponding to a specific database of medical records of renowned traditional Chinese medicine practitioners, and perform correction processing on the pseudo-labels of syndrome elements to obtain the target pseudo-labels. The specific database of medical records of renowned traditional Chinese medicine practitioners contains the personal diagnosis and treatment records of preset renowned traditional Chinese medicine practitioners. The second diagnostic model is optimized and trained using the specific database of medical records of renowned traditional Chinese medicine practitioners and the corresponding target pseudo-labels to obtain a syndrome diagnosis model.

2. The method according to claim 1, characterized in that, The initial model was trained unsupervised using a basic TCM corpus to obtain the first diagnostic model, which includes: Obtain first symptom information and first syndrome diagnosis result, wherein the first symptom information is the symptom information of any patient in the basic TCM corpus, and the first syndrome diagnosis result is the syndrome diagnosis result of any patient in the basic TCM corpus; The first symptom information and the first syndrome diagnosis result are predicted by the initial model to obtain a first prediction result, wherein the first prediction result is used to represent the probability that the first syndrome diagnosis result is the true diagnosis result of the first symptom information; A first loss value is determined by a first loss function between the first prediction result and the first true label, wherein the first true label is used to represent the true diagnosis result of the first symptom information; The model parameters of the initial model are optimized based on the first loss value to obtain the first diagnostic model.

3. The method according to claim 1, characterized in that, The first diagnostic model was trained in a supervised manner using a database of medical records from renowned traditional Chinese medicine practitioners, including: Obtain second symptom information and determine a first feature vector corresponding to the second symptom information, wherein the second symptom information is the symptom information of any patient in the general database of medical records of famous traditional Chinese medicine doctors; The first feature vector is predicted by the first diagnostic model to obtain a second prediction result, wherein the second prediction result is used to represent the syndrome diagnosis result corresponding to the second symptom information; The first feature vector and the second prediction result are concatenated to obtain the second feature vector; The second feature vector is predicted by the first diagnostic model to obtain a third prediction result, wherein the third prediction result is used to represent the syndrome diagnosis result corresponding to the second symptom information; The first diagnostic model is trained based on the second prediction result and the third prediction result.

4. The method according to claim 3, characterized in that, Training the first diagnostic model based on the second prediction result and the third prediction result includes: A second loss value is determined by a second loss function between the second prediction result and the second true label, wherein the second true label is used to represent the true syndrome information corresponding to the second symptom information; A third loss value is determined between the third prediction result and the third true label using a third loss function, wherein the third true label is used to represent the true syndrome information corresponding to the second symptom information; The model parameters of the first diagnostic model are optimized based on the second loss value, the third loss value, and preset hyperparameters to obtain the second diagnostic model.

5. The method according to claim 4, characterized in that, Identify pseudo-labels of syndrome elements corresponding to specific databases of medical records of renowned veteran TCM doctors, including: Obtain third symptom information and determine the third feature vector corresponding to the third symptom information, wherein the third symptom information is the symptom information of any patient in the specific famous traditional Chinese medicine medical case database; The third feature vector is predicted by the second diagnostic model to obtain a fourth prediction result, wherein the fourth prediction result is used to represent the syndrome diagnosis result corresponding to the third symptom information; Based on the fourth prediction result and the preset threshold, a pseudo-label of the syndrome element corresponding to the third symptom information is determined; Alternatively, obtain a list of syndrome elements corresponding to the third symptom information, and determine pseudo-labels of syndrome elements corresponding to the third symptom information based on the list of syndrome elements.

6. The method according to claim 5, characterized in that, The pseudo-labels of the evidence elements are subjected to bias correction processing to obtain the target pseudo-labels, including: Determine a first deviation probability and a second deviation probability corresponding to the third symptom information, wherein the first deviation probability is used to represent the probability that the pseudo label of the syndrome element is 0 when the fourth true label is 1, and the second deviation probability is used to represent the probability that the pseudo label of the syndrome element is 1 when the fourth true label is 0, wherein the fourth true label is used to represent the true syndrome element information corresponding to the third symptom information. The pseudo-label of the evidence element is corrected based on the first deviation probability and the second deviation probability to obtain the target pseudo-label.

7. The method according to claim 6, characterized in that, Determining the first deviation probability corresponding to the third symptom information includes: Anchor samples corresponding to the third symptom information are determined, and the cumulative density function and corresponding inverse function of the anchor samples are determined based on the standard normal distribution. The anchor samples are used to represent samples in the third symptom information whose predicted probability obtained by the second diagnostic model is lower than a preset probability threshold. Determine the correlation coefficient matrix between the evidence elements corresponding to the anchor sample, and determine the log-likelihood function corresponding to the anchor sample based on the correlation coefficient matrix; The normal distribution parameters corresponding to the anchor point sample are determined based on the log-likelihood function. The first deviation probability is determined based on the normal distribution parameters, the cumulative density function, and the inverse function.

8. The method according to claim 6, characterized in that, The method further includes: The fourth prediction result is optimized based on the target pseudo-label to obtain the target prediction result; The fourth loss value between the target prediction result and the evidence pseudo-label is determined by the fourth loss function; The fourth feature vector is obtained by concatenating the third feature vector and the target prediction result. The fourth feature vector is predicted by the second diagnostic model to obtain a fifth prediction result, wherein the fifth prediction result is used to represent the syndrome diagnosis result corresponding to the third symptom information; The fifth loss value between the fifth prediction result and the fifth true label is determined by the third loss function, wherein the fifth true label is used to represent the true syndrome information corresponding to the third symptom information; The model parameters of the second diagnostic model are optimized based on the fourth loss value and the fifth loss value to obtain the syndrome diagnosis model.

9. The method according to claim 7, characterized in that, The method further includes: The third feature vector is predicted using the syndrome diagnosis model to obtain the diagnostic result of the target syndrome element; By concatenating the third feature vector and the diagnostic result of the target syndrome element, a target feature vector is obtained, and the target feature vector is predicted by the syndrome diagnosis model to obtain the diagnostic result of the target syndrome. A first syndrome element list and a second syndrome element list are respectively determined to correspond to the first target syndrome diagnosis result and the second target syndrome diagnosis result, wherein the first target syndrome diagnosis result and the second target syndrome diagnosis result are any two identical syndrome diagnosis results in the target syndrome diagnosis results, and the first syndrome element list and the second syndrome element list are a set of the most probable syndrome elements in the target syndrome element diagnosis results. The syndrome element ratio results of the first syndrome and the second syndrome are determined based on the first syndrome element list and the second syndrome element list, wherein the first syndrome is any syndrome in the first target syndrome diagnosis result, and the second syndrome is any syndrome in the second target syndrome diagnosis result. The syndrome element ratio results are used to include the relative proportion of the occurrence frequency of each syndrome element in the first syndrome element list under the first syndrome, and the relative proportion of the occurrence frequency of each syndrome element in the second syndrome element list under the second syndrome. The focus of diagnosis and treatment of the pre-set renowned traditional Chinese medicine doctors is determined based on the results of the syndrome element ratio.

10. A method for predicting syndromes, characterized in that, include: Obtain the patient's symptom information and determine the feature vector corresponding to the symptom information; The feature vector is predicted using a syndrome diagnosis model to obtain the syndrome element prediction result corresponding to the symptom information. The syndrome diagnosis model is trained as follows: An initial model is unsupervised-trained using a basic TCM corpus to obtain a first diagnostic model, where the basic TCM corpus contains multiple TCM basic knowledge texts; the first diagnostic model is supervised-trained using a database of medical records from renowned TCM doctors to obtain a second diagnostic model, where the database contains general treatment records from multiple renowned TCM doctors and syndrome element information corresponding to the patient's symptoms. This syndrome element information guides the first diagnostic model in learning the general treatment logic of renowned TCM doctors; pseudo-labels for syndrome elements are determined corresponding to a specific database of medical records from renowned TCM doctors, and these pseudo-labels are corrected to obtain target pseudo-labels, where the specific database contains personal treatment records from preset renowned TCM doctors; the second diagnostic model is optimized and trained using the specific database of medical records from renowned TCM doctors and the corresponding target pseudo-labels to obtain the syndrome diagnosis model. By concatenating the evidence element prediction result and the feature vector, the target feature vector is obtained; The target feature vector is predicted by the syndrome diagnosis model to obtain the syndrome prediction result corresponding to the symptom information, wherein the syndrome prediction result is used to provide a basis for the dialectical diagnosis of the patient.

11. A training device for a syndrome diagnosis model, characterized in that, include: The first training module is used to perform unsupervised training on the initial model using a basic TCM corpus to obtain the first diagnostic model. The basic TCM corpus contains multiple TCM basic knowledge texts. The second training module is used to conduct supervised training on the first diagnostic model through a general database of medical records of renowned traditional Chinese medicine doctors to obtain a second diagnostic model. The general database of medical records of renowned traditional Chinese medicine doctors contains general diagnosis and treatment records of multiple renowned traditional Chinese medicine doctors, as well as syndrome element information corresponding to the patient's symptoms. The syndrome element information is used to guide the first diagnostic model to learn the general diagnosis and treatment logic of renowned traditional Chinese medicine doctors. The correction module is used to determine the pseudo-label of syndrome element corresponding to the medical record database of a specific famous traditional Chinese medicine doctor, and to perform correction processing on the pseudo-label of syndrome element to obtain the target pseudo-label. The specific medical record database of famous traditional Chinese medicine doctor contains the personal diagnosis and treatment records of the preset famous traditional Chinese medicine doctor. The third training module is used to optimize and train the second diagnostic model using the specific database of medical records of famous traditional Chinese medicine doctors and the corresponding target pseudo-labels, so as to obtain the syndrome diagnosis model.

12. A syndrome prediction device, characterized in that, include: The acquisition module is used to acquire the patient's symptom information and determine the feature vector corresponding to the symptom information; The first prediction module is used to predict the feature vector using a syndrome diagnosis model to obtain the syndrome element prediction result corresponding to the symptom information. The syndrome diagnosis model is trained as follows: an initial model is unsupervised-trained using a basic TCM corpus to obtain a first diagnostic model, wherein the basic TCM corpus contains multiple TCM basic knowledge texts; the first diagnostic model is supervised-trained using a database of medical records from renowned TCM doctors to obtain a second diagnostic model, wherein the database contains general treatment records from multiple renowned TCM doctors and syndrome element information corresponding to the patient's symptoms, the syndrome element information being used to guide the first diagnostic model in learning the general treatment logic of renowned TCM doctors; pseudo-labels of syndrome elements are determined corresponding to a specific database of medical records from renowned TCM doctors, and the pseudo-labels are corrected to obtain target pseudo-labels, wherein the specific database contains personal treatment records from preset renowned TCM doctors; the second diagnostic model is optimized and trained using the specific database of medical records from renowned TCM doctors and the corresponding target pseudo-labels to obtain the syndrome diagnosis model. The splicing module is used to splice the evidence element prediction result and the feature vector to obtain the target feature vector; The second prediction module is used to predict the target feature vector through the syndrome diagnosis model to obtain the syndrome prediction result corresponding to the symptom information, wherein the syndrome prediction result is used to provide a basis for the dialectical diagnosis of the patient.

13. An electronic device, characterized in that, include: A memory and a processor, wherein the memory is used to store program instructions; the processor is connected to the memory and is used to execute a training method for implementing the syndrome diagnosis model according to any one of claims 1 to 9 or a syndrome prediction method according to claim 10.

14. A non-volatile storage medium, characterized in that, The non-volatile storage medium includes a stored computer program, wherein the device containing the non-volatile storage medium executes the training method of the syndrome diagnosis model according to any one of claims 1 to 9 or the syndrome prediction method according to claim 10 by running the computer program.

15. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the training method of the syndrome diagnosis model according to any one of claims 1 to 9 or the syndrome prediction method according to claim 10.

Images