Traditional Chinese medicine gynecological auxiliary diagnosis and treatment system based on attention mechanism neural network

Abstract

The invention discloses a traditional Chinese medicine gynecological auxiliary diagnosis and treatment system based on an attention mechanism neural network, and relates to the field of intelligent traditional Chinese medicine gynecological diagnosis and treatment, and the system comprises the following steps: obtaining traditional Chinese medicine gynecological clinical manifestation multi-dimensional data of a patient, extracting multi-modal features, and carrying out cross-modal feature alignment and dynamic weight distribution based on multiple attention mechanisms. According to the traditional Chinese medicine gynecological auxiliary diagnosis and treatment system based on the attention mechanism neural network, symptoms-syndromes of traditional Chinese medicine gynecological diseases are described based on a composite syndrome type library and authoritative medical books through multi-modal features, and syndromes to which multiple attention mechanisms belong are dynamically identified; performing prescription-symptom attention matching according to the syndrome differentiation result, screening candidate prescriptions through multiple attention, and recommending dynamic addition and reduction of medicinal materials based on the physique bias of the patient; continuous monitoring data of a patient is collected, future pathological feature evolution is captured using multiple attention, future syndrome changes are predicted, and a health management scheme is generated.

Description

Technical Field

[0001] The present invention relates to the field of intelligent TCM gynecological diagnosis and treatment, and specifically to a TCM gynecological auxiliary diagnosis and treatment system based on an attention mechanism neural network. Background Art

[0002] With the rapid development of modern society, women are paying more and more attention to their own health. Traditional Chinese Medicine (TCM) gynecology, with its holistic view of "harmony between man and nature" and dynamic regulation, as well as its philosophy of syndrome differentiation and treatment and natural therapy, has outstanding advantages in the management of chronic gynecological diseases, functional regulation, prevention of diseases, and safety. Due to the relatively unique theoretical system and clinical experience of TCM gynecology, the integration of TCM gynecology with modern technology is somewhat difficult. Therefore, using artificial intelligence and big data technology to mine data on ancient TCM gynecology prescriptions and the experience of famous doctors, the inheritance of TCM gynecology experience and implicit knowledge can be made explicit and standardized, and a dynamic relationship can be constructed from the patient's clinical manifestations to the syndrome differentiation model and then to the generation of personalized treatment prescriptions. It is crucial to develop an interactive dynamic syndrome differentiation auxiliary diagnosis and treatment system by combining the four diagnostic mechanisms of "looking, listening, asking and feeling". The invention patent with publication number CN118942637A discloses a TCM gynecology treatment plan optimization method and system based on reinforcement learning. It generates and optimizes treatment plans through a multi-objective reinforcement learning model, providing an innovative TCM gynecology treatment plan optimization method. The invention patent with publication number CN118398201A discloses a TCM gynecological disease diagnosis auxiliary system based on a clinical knowledge graph. By constructing a clinical knowledge graph and performing coordinate processing, the clinical knowledge graph is visualized, and targeted auxiliary diagnosis information is obtained according to the information range. However, in the process of implementing the technical solutions of the invention in the embodiments of the present application, the present application discovered that the above-mentioned technology has the following technical problems: First, although there are design methods for auxiliary diagnosis and treatment systems for TCM gynecology in the existing technology, most of them adopt the method of clinical knowledge graph, do not distinguish the pathogenesis, clinical symptoms and treatment plans of different syndromes, simply coordinate the patient's clinical manifestations and provide auxiliary information based on the amount of coordinate information and the range of information, ignore the valuable experience of TCM syndrome differentiation and treatment, and reduce the accuracy of dynamic identification of complex syndromes; Second, the relatively basic reinforcement learning method is usually adopted, which ignores the impact of the dynamic development of TCM gynecological diseases and cannot cope with the multi-modal long-distance dependence in the complex pathogenesis, resulting in inconsistent matching between clinical manifestations and data, poor adaptive and dynamic analysis capabilities for complex syndromes, and even serious problems of deviation in diagnosis and treatment plans; Third, most existing auxiliary diagnosis and treatment systems for TCM gynecology focus only on the diagnosis and treatment process, ignore the advantages of TCM in preventive medicine, and lack the combination of TCM treatment and intervention methods in preventing diseases and predicting menstrual cycle changes or pregnancy-related risks. Due to the holistic view of TCM gynecology, disease prevention, early intervention and rehabilitation consolidation are also important. Therefore, it is necessary to propose a TCM gynecology auxiliary diagnosis and treatment system based on attention mechanism neural network. Summary of the Invention

[0003] (1) Technical problems solved

[0004] In view of the shortcomings of the existing technology, the present invention provides a TCM gynecology auxiliary diagnosis and treatment system based on an attention mechanism neural network. In order to solve the above technical problems, the present invention provides a TCM gynecology auxiliary diagnosis and treatment system based on an attention mechanism neural network. According to the patient's clinical manifestations and other relevant data, a data-driven TCM gynecology syndrome dynamic identification model is established to dynamically identify the nonlinear combination relationship between symptoms under complex syndromes. A Transformer network with a self-attention mechanism is used to dynamically capture the global dependency relationship of clinical manifestations-TCM syndrome type-treatment plan, and the network's ability to capture different semantic relationships of symptoms is enhanced based on a multi-head attention mechanism. It assists doctors in diagnosing patients' symptoms and recommending treatment plans, predicting future symptom changes of the disease, and realizing the multimodal, interpretable, and dynamic nature of the TCM gynecology auxiliary diagnosis and treatment system.

[0005] (2) Technical solution

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a TCM gynecological auxiliary diagnosis and treatment method based on an attention mechanism neural network, comprising the following steps:

[0007] Step S1: Obtain the patient's TCM gynecological four-diagnosis combined syndrome differentiation results and multi-dimensional data of clinical manifestations, then input the architecture to extract multimodal features, perform cross-modal feature alignment and dynamic weight allocation based on the multi-head attention mechanism, and meet the "four-diagnosis combined" holistic view constraints of "diagnosis in traditional Chinese medicine";

[0008] Step S2: Using the above multimodal features, combined with the symptom-syndrome descriptions in the "Traditional Chinese Medicine Gynecology" compound syndrome database and authoritative medical books such as "Gynecological Syndrome and Treatment Standards" and "Fu Qingzhu's Gynecology", the nonlinear combination relationship between symptoms is captured, and the multi-head attention mechanism dynamically identifies the corresponding syndromes to obtain the identified syndrome results;

[0009] Step S3: Based on the syndrome identification results, perform prescription-symptom attention matching, calculate the similarity between the patient's syndrome and the prescription indication, screen candidate prescriptions through multi-headed attention, and recommend dynamic addition or subtraction of medicinal materials based on the patient's physical constitution bias;

[0010] Step S4: Collect continuous monitoring data of patients, use multi-head attention to capture the evolution of pathological characteristics at different time points, extract periodic patterns, predict future syndrome changes and generate health management plans;

[0011] Step S5: The case diagnosis and treatment process is digitized, and dynamic knowledge evolution is achieved through the Transformer network. This is combined with a closed-loop clinical feedback loop and knowledge updates, focusing on the adaptive integration of regional schools of thought, and iteratively optimizing the auxiliary diagnosis and treatment system.

[0012] Furthermore, in step S1, the patient's multi-dimensional data and features are preprocessed. The multimodal input features are text, and the multimodal input features include text, images, and time series signals. The text includes the patient's chief complaint and medical history, and the image includes tongue image and complexion; the time series signals include pulse condition and physical condition monitoring data. The multi-head attention mechanism splits the input feature data into multiple subspaces, with each attention head focusing on different modal associations, such as the association between "thick yellow tongue coating" and "slippery and rapid pulse" for damp-heat syndrome. Through self-attention, the global dependency is dynamically captured, the importance of symptoms is automatically determined, and the impact weight of different syndromes on the patient's health is dynamically adjusted. For example, "thirst and desire to drink" contributes more to "yin deficiency syndrome" than "mild fatigue." When the tongue diagnosis is "stasis syndrome" but the pulse does not show "wiry pulse", cross-modal contradiction detection is triggered, and features with higher correlation with the chief complaint are prioritized and the impact weight is dynamically adjusted.

[0013] Among them, the contradiction detection mechanism includes a confidence scoring module: when the tongue diagnosis is "blood stasis syndrome" but the pulse is not "astringent pulse", the confidence scores of the two modalities are calculated; if the score difference exceeds the threshold of 0.3, the priority weighting of the chief complaint text is triggered, and the weight is increased by 50%.

[0014] Furthermore, in step S2, a symptom-syndrome mapping matrix is ​​constructed based on the compound syndrome type library of "Chinese Medicine Gynecology" and authoritative medical books such as "Gynecological Syndrome and Treatment Standards" and "Fu Qingzhu's Gynecology";

[0015] Furthermore, in step S3, when recommending and adding or subtracting prescriptions, classic prescriptions need to be flexibly adapted to individual differences. Therefore, it is necessary to add or subtract medicinal materials during the process. At the same time, the prescription composition, efficacy, and contraindications are structured as key-value pairs of the attention mechanism, and a penalty term is introduced in the attention weight. For example, if peach kernel and safflower are prohibited during pregnancy, the weight is forced to zero to ensure the safety of the recommendation. The multi-head attention matching stage converts the prescription into a structured vector as follows:

[0016]

[0017] Prescriptions are initially screened based on the menstrual cycle stage, and different dimensions are focused on in the cross-attention calculation: medicinal material-symptom matching, pathogenesis fit, pregnancy safety control, cycle synchronization optimization, and the addition and subtraction recommendation stages are ranked according to the importance of the medicinal materials.

[0018] The rules for adding or subtracting medicinal materials include:

[0019] Core herbs, such as Angelica sinensis and Ligusticum chuanxiong in Siwu Decoction, have an importance weight of ≥ 0.8 and cannot be deleted.

[0020] Adjustment of auxiliary medicines: Dynamic replacement based on physical constitution bias, such as adding Astragalus for those with Qi deficiency, weight = 0.6.

[0021] Furthermore, in step S4, TCM gynecology's prevention of disease may involve prevention and early intervention in aspects such as irregular menstruation, infertility, and postpartum conditioning. In combination with the multi-head attention mechanism, a prevention and disease prediction module is designed, including data input, model architecture, and health management. Patient historical data slices are converted into multi-source heterogeneous data input, including menstrual cycle logs, basal body temperature curves, tongue dynamic changes, pulse trends, life behavior data, and modern medical indicators; a three-level prevention prediction model is constructed to prevent disease before it occurs, prevent the onset of disease before it occurs, and prevent the progression of disease after it has occurred; based on the patient's possible future health problems, different health management plans and personalized health care suggestions are generated that take into account the patient's physical characteristics and the current solar terms.

[0022] Among them, the three-level prevention prediction model includes the following:

[0023]

[0024] Furthermore, in step S5, TCM gynecology relies on accumulated experience, and regional differences in schools lead to inconsistent diagnosis and treatment, so it is necessary to introduce an iterative optimization module. It includes steps such as data construction, incremental learning, and quality control. The data collection specification requires structured four-diagnosis information and quantified efficacy feedback; in the incremental learning process, the symptom-syndrome attention mapping is updated in stages, the attention association between new cases and existing knowledge is calculated and enhanced with regional adaptation, and the experience of schools is integrated; in the quality control process, a dual-channel verification mechanism is set up, a model recommendation plan is recommended, random expert inspections are carried out, and rule engine verification is carried out to determine whether to enter the learning queue, a feedback closed-loop design is designed, and an efficacy score is obtained based on clinical results and implementation plans. If the score exceeds the set threshold, a difference analysis is performed on the predicted features and actual features, and the database adds revised cases to trigger model retraining.

[0025] Furthermore, the TCM gynecology auxiliary diagnosis and treatment system includes:

[0026] A TCM gynecology multimodal clinical manifestation fusion module is used to obtain patient information, extract and align multimodal features, and dynamically assign weights;

[0027] A dynamic identification module for TCM gynecological syndromes, which is used to dynamically identify syndromes based on the symptom-syndrome relationship using the patient's multimodal characteristics;

[0028] A TCM gynecological prescription recommendation and addition / subtraction module is used to recommend prescriptions and dynamically add / subtract prescriptions based on the syndrome identification results;

[0029] A TCM gynecology disease prevention prediction module, used to predict future symptom changes in patients and generate health management plans;

[0030] The iterative optimization module of TCM gynecological diagnosis and treatment experience is used to digitize the case diagnosis and treatment process and iteratively optimize the auxiliary diagnosis and treatment system.

[0031] Furthermore, the multimodal feature fusion extracts the patient's clinical manifestation features as input features through different encoders, namely text (self.text_encoder), image (self.image_encoder), and time series signal (self.signal_encoder); the dimension of the attention head is set to 8, namely menstrual disorders, leucorrhea, pregnancy disorders, postpartum diseases, gynecological diseases, overall syndrome differentiation and individualized analysis, complex syndrome type-treatment method reasoning, and symptom contribution evaluation; 8 linear transformations are performed on the three groups of input features to generate 8 independent input feature matrices, namely: Where, is the text feature linear projection matrix, is the image feature linear projection matrix, is the linear projection matrix of the time series signal feature. Cross-modal multi-head attention fusion, parallel calculation of the scaled dot product attention of each head: Where, s is the scaling factor to prevent the dot product result from being too large, which will cause the softmax gradient to disappear. To concatenate multimodal features, through multi-head attention interaction, the outputs of the eight attention heads are concatenated and linearly transformed to obtain the corresponding output data of the patient: Where, For the output projection matrix, ensure the dimensions align with the input.

[0032] Furthermore, the three-level prevention prediction model has a parallel triggering mechanism, focusing on the prediction of menstrual cycle regularity or pregnancy risk assessment, taking into account gynecological diseases such as irregular menstruation, leucorrhea, infertility, etc., as well as key elements in gynecological diagnosis, such as symptoms related to menstruation, leucorrhea, pregnancy, and childbirth. The first level is to prevent illness before it occurs and warn of health risks. The attention mechanism uses solar term sensitive position coding, which is dynamically adjusted according to the 24 solar terms. Based on temporal dependence and cross-modal interaction, it focuses on predicting the probability of gynecological diseases in the next 6 months and identifying sub-health conditions; the second level is to prevent illness before it occurs, intervening in the early stages of the disease. When the patient's menstrual cycle begins to become irregular but does not meet the diagnostic criteria, and imaging examinations show changes in gynecological appendages but no clinical manifestations, the patient's various clinical manifestations are monitored, and dynamic thresholds are used to warn of disease occurrence; the third level is to prevent changes after illness occurs, manage the patient's overall disease course, and focus on monitoring sudden abnormal indicators based on the patient's existing symptoms, which can easily lead to physiological indicators that worsen the disease.

[0033] (3) Beneficial effects

[0034] The present invention provides a TCM gynecology auxiliary diagnosis and treatment system based on an attention mechanism neural network. It has the following beneficial effects:

[0035] (1) This TCM gynecology auxiliary diagnosis and treatment system based on the attention mechanism neural network aligns multi-dimensional data according to the patient's cross-modal features, dynamically captures the global dependency of multimodal input features through self-attention, dynamically adjusts the weight of the impact of different syndromes on the patient's health, and improves the accuracy of the auxiliary diagnosis and treatment system in identifying complex TCM gynecology syndromes.

[0036] (2) This TCM gynecology auxiliary diagnosis and treatment system based on the attention mechanism neural network constructs a mapping from symptoms to syndromes through the Transformer network multi-head attention mechanism, dynamically identifies patient syndromes, preliminarily screens prescriptions according to the menstrual cycle stage, focuses on different dimensional indicators in the cross-attention calculation, recommends prescriptions based on complex syndrome types and adds or subtracts medicinal materials, and performs reasonable dynamic analysis to accurately and dialectically treat the syndromes.

[0037] (3) The TCM gynecology auxiliary diagnosis and treatment system based on the attention mechanism neural network is based on the holistic concept of TCM gynecology to prevent diseases before they occur, and performs prevention and early intervention before the disease occurs. The multi-head attention mechanism processes data at different time points and combines it with the TCM solar term theory to make predictions, predict future symptom changes and generate health management plans, effectively solving the problem that the TCM gynecology auxiliary diagnosis and treatment system lacks disease prevention, early intervention, and rehabilitation consolidation functions. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a schematic diagram of the overall structure of a TCM gynecology auxiliary diagnosis and treatment system based on an attention mechanism neural network according to the present invention;

[0039] Figure 2 This is a flowchart of the application of a TCM gynecology auxiliary diagnosis and treatment system based on an attention mechanism neural network according to the present invention;

[0040] Figure 3 Schematic diagram of the cross-modal multi-head attention fusion framework of the present invention. DETAILED DESCRIPTION

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0042] See also Figures 1 to 3 The present invention provides a technical solution: a method for using a TCM gynecology auxiliary diagnosis and treatment system based on an attention mechanism neural network comprises the following steps:

[0043] Step S1: Obtain the patient's TCM gynecological four-diagnosis combined syndrome differentiation results and multi-dimensional data of clinical manifestations, then input the architecture to extract multimodal features, perform cross-modal feature alignment and dynamic weight allocation based on the multi-head attention mechanism, and meet the "four-diagnosis combined" holistic view constraints of "diagnosis in traditional Chinese medicine";

[0044] It should be further explained that, during the specific implementation process, the patient's multi-dimensional data and features are preprocessed. The multi-modal input features are text, and the multi-modal input features are text, images, and time series signals. The multi-head attention mechanism splits the input feature data into multiple subspaces, with each attention head focusing on different modal associations, such as the association between "thick yellow tongue coating" and "slippery and rapid pulse" for damp-heat syndrome. Through self-attention, global dependencies are dynamically captured, the importance of symptoms is automatically determined, and the impact weight of different syndromes on patient health is dynamically adjusted. For example, "thirst and desire to drink" contributes more to "yin deficiency syndrome" than "mild fatigue." When the tongue diagnosis of "stasis syndrome" is confirmed but the pulse does not show "wiry pulse", cross-modal contradiction detection is triggered, and features with higher relevance to the main complaint are prioritized and the impact weight is dynamically adjusted.

[0045] It should be further explained that, in the specific implementation process, the multimodal feature fusion extracts the patient's clinical manifestation features as input features through different encoders, namely text (self.text_encoder), image (self.image_encoder), and time series signal (self.signal_encoder); the dimension of the attention head is set to 8, namely menstrual disorders, leucorrhea, pregnancy disorders, postpartum diseases, gynecological diseases, overall syndrome differentiation and individualized analysis, complex syndrome type-treatment method reasoning, and symptom contribution evaluation; 8 linear transformations are performed on the three groups of input features to generate 8 independent input feature matrices, namely: Where, is the text feature linear projection matrix, is the image feature linear projection matrix, is the linear projection matrix of the time series signal features.

[0046] Cross-modal multi-head attention fusion, parallel calculation of the scaled dot product attention of each head: Where, s is the scaling factor to prevent the dot product result from being too large, which will cause the softmax gradient to disappear. To concatenate multimodal features, through multi-head attention interaction, the outputs of the eight attention heads are concatenated and linearly transformed to obtain the corresponding output data of the patient: Where, For the output projection matrix, ensure the dimensions align with the input.

[0047] Step S2: Using the above multimodal features, combined with the symptom-syndrome descriptions in the "Traditional Chinese Medicine Gynecology" compound syndrome database and authoritative medical books such as "Gynecological Syndrome and Treatment Standards" and "Fu Qingzhu's Gynecology", the nonlinear combination relationship between symptoms is captured, and the multi-head attention mechanism dynamically identifies the corresponding syndromes to obtain the identified syndrome results;

[0048] It should be further explained that in the specific implementation process, based on the compound syndrome database of "Chinese Medicine Gynecology" and authoritative medical books such as "Gynecological Syndrome Treatment Standards" and "Fu Qingzhu's Gynecology", a symptom-syndrome mapping matrix was constructed, and the division of labor of each attention head is as follows:

[0049]

[0050] Step S3: Based on the syndrome identification results, perform prescription-symptom attention matching, calculate the similarity between the patient's syndrome and the prescription indication, screen candidate prescriptions through multi-headed attention, and recommend dynamic addition or subtraction of medicinal materials based on the patient's physical constitution bias;

[0051] It should be further explained that in the specific implementation process, when recommending and adding or subtracting prescriptions, classic prescriptions need to be flexibly adapted to individual differences. Therefore, it is necessary to add or subtract medicinal materials during the process. At the same time, the prescription composition, efficacy, and contraindications are structured as key-value pairs of the attention mechanism, and a penalty term is introduced in the attention weight. For example, if peach kernel and safflower are prohibited during pregnancy, the weight is forced to zero to ensure the safety of the recommendation. The multi-head attention matching stage converts the prescription into a structured vector as follows: Prescriptions are initially screened based on the menstrual cycle stage, and different dimensions are focused on in the cross-attention calculation: medicinal material-symptom matching, pathogenesis fit, pregnancy safety control, cycle synchronization optimization, and the addition and subtraction recommendation stages are ranked according to the importance of medicinal materials.

[0052] Step S4: Collect continuous monitoring data of patients, including menstrual cycle, tongue changes, and pulse trends, use multi-head attention to capture the evolution of pathological characteristics at different time points, extract cycle patterns, predict future syndrome changes, and generate health management plans;

[0053] It should be further explained that, in the specific implementation process, TCM gynecology's prevention of disease may involve prevention and early intervention in areas such as irregular menstruation, infertility, and postpartum conditioning. Incorporating a multi-headed attention mechanism, a disease prevention prediction module is designed, including data input, model architecture, and health management. Patient historical data slices are converted into multi-source heterogeneous data input, including menstrual cycle logs, basal body temperature curves, tongue dynamics, pulse trends, lifestyle data, and modern medical indicators. A three-level prevention prediction model is constructed to prevent illness before it occurs, prevent the onset of illness before it occurs, and prevent the progression of existing illness. Based on the patient's possible future health problems, different health management plans and personalized health recommendations are generated that take into account the patient's physical characteristics and the current solar term.

[0054] It should be further explained that, in the specific implementation process, the three-level prevention prediction model uses a parallel trigger mechanism, focusing on predicting the regularity of the menstrual cycle or assessing pregnancy risks, taking into account gynecological diseases such as irregular menstruation, leucorrhea, infertility, etc., as well as key elements in gynecological diagnosis, such as symptoms related to menstruation, leucorrhea, pregnancy, and childbirth. The first level is to prevent illness before it occurs and warn of health risks. The attention mechanism uses solar term sensitive position coding, which is dynamically adjusted according to the 24 solar terms. Based on temporal dependence and cross-modal interaction, it focuses on predicting the probability of gynecological diseases in the next 6 months and identifying sub-health conditions. The second level is to prevent illness before it occurs, intervening in the early stages of the disease. When the patient's menstrual cycle begins to become irregular but does not meet the diagnostic criteria, and imaging examinations show changes in gynecological appendages but no clinical manifestations, the patient's various clinical manifestations are monitored and dynamic thresholds are used to warn of disease occurrence. The third level is to prevent changes after illness occurs, managing the patient's overall disease course. Based on the patient's existing symptoms, it focuses on monitoring sudden abnormal indicators and physiological indicators that are likely to lead to worsening of the disease.

[0055] Among them, the dynamic warning threshold and periodic law extraction method include:

[0056] Time series feature extraction: Use LSTM+multi-head attention to capture the regularity of the menstrual cycle, input the basal body temperature curve of the past 6 months, and output the probability of abnormalities in the future cycle.

[0057] Solar term sensitive coding: Convert the 24 solar terms into periodic position coding. The formula is as follows:

[0058] ;

[0059] Where t is the solar term number, Lichun = 1, Yushui = 2, ..., Dahan = 24;

[0060] Step S5: digitize the aforementioned diagnosis and treatment process cases and implement dynamic knowledge evolution through a Transformer network. Combined with a closed-loop clinical feedback loop and knowledge updates, the system is optimized through adaptive integration of regional schools of thought and iterative optimization of the auxiliary diagnosis and treatment system. This adaptive integration of regional schools of thought involves encoding schools of thought, such as the "Lingnan School" and the "Shanghai School," into vectors; calculating the similarity between new cases and representative cases from each school of thought; and weighting and integrating the recommended results. When the efficacy score S is < 0.7 (with a maximum score of 1.0), compare the KL divergence of the predicted symptoms to the actual symptoms. If the KL divergence is > 0.5, incremental model training is triggered.

[0061] It should be further explained that, in the specific implementation process, TCM gynecology relies on accumulated experience, and regional differences in schools lead to inconsistent diagnosis and treatment, which requires the introduction of an iterative optimization module. This includes steps such as data construction, incremental learning, and quality control. Data collection specifications require structured information on the four diagnoses and quantified feedback on efficacy; in the incremental learning process, the symptom-syndrome attention mapping is updated in stages, the attention association between new cases and existing knowledge is calculated and enhanced with regional adaptation, and the experience of schools is integrated; in the quality control process, a dual-channel verification mechanism is set up, a model recommendation plan is recommended, random expert inspections are carried out, and rule engine verification is carried out to determine whether to enter the learning queue, a feedback closed-loop design is designed, and an efficacy score is derived based on clinical results and implementation plans. If the score exceeds the set threshold, a difference analysis is conducted on the predicted features and actual features, and the database is amended to add cases to trigger model retraining.

[0062] To better implement the above method, see Figure 2 The present invention further describes an application process of a TCM gynecology auxiliary diagnosis and treatment system based on an attention mechanism neural network as follows:

[0063] (1) The TCM gynecology multimodal clinical manifestation fusion module obtains multidimensional data of patients’ TCM gynecology clinical manifestations, extracts modal features, and performs cross-modal feature alignment and dynamic weight allocation based on the multi-head attention mechanism.

[0064] (2) The dynamic identification module of TCM gynecological syndromes is based on the description of TCM gynecological disease symptoms and syndromes in the complex syndrome database and authoritative medical books. The multi-headed attention mechanism dynamically identifies the corresponding syndromes and obtains the identification syndrome results.

[0065] (3) The TCM gynecological prescription recommendation and addition / subtraction module performs prescription-symptom attention matching, screens candidate prescriptions through multi-headed attention, and recommends dynamic addition / subtraction of medicinal materials based on the patient's physical constitution bias.

[0066] (4) The TCM gynecological disease prevention prediction module collects continuous monitoring data of patients, uses multi-head attention to capture the evolution of future pathological characteristics, predicts future syndrome changes and generates health management plans.

[0067] (5) Iterative optimization module of TCM gynecological diagnosis and treatment experience, digitizing the case diagnosis and treatment process, combining clinical feedback closed loop and knowledge update, focusing on adaptive integration of regional schools, iterative optimization of auxiliary diagnosis and treatment system,

[0068] An auxiliary diagnosis and treatment system for TCM gynecology based on an attention mechanism neural network aligns multi-dimensional data according to the patient's cross-modal features, dynamically captures the global dependency of multi-modal input features through self-attention, dynamically adjusts the weight of the impact of different syndromes on the patient's health, and improves the accuracy of the auxiliary diagnosis and treatment system in identifying complex TCM gynecological syndromes; constructs a symptom-to-syndrome mapping through the Transformer network multi-head attention mechanism, dynamically identifies patient syndromes, preliminarily screens prescriptions according to the menstrual cycle stage, pays attention to different dimensional indicators in the cross-attention calculation, recommends prescriptions based on complex syndromes and adds or subtracts medicinal materials, and rationally and dynamically analyzes and accurately treats syndromes.

[0069] Based on the holistic concept of preventing diseases before they occur in Traditional Chinese Medicine (TCM) gynecology, prevention and early intervention are carried out before the disease occurs. The multi-headed attention mechanism processes data at different time points and combines it with TCM solar term theory to make predictions, predict future symptom changes and generate health management plans, effectively solving the problem that the TCM gynecology auxiliary diagnosis and treatment system lacks disease prevention, early intervention, and rehabilitation consolidation functions.

[0070] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0071] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A TCM gynecology auxiliary diagnosis and treatment system based on attention mechanism neural network, characterized by: The steps include: Step S1: Obtain the patient's TCM gynecological four-diagnosis combined syndrome differentiation results and multi-dimensional data of clinical manifestations, then input the architecture to extract multimodal features, and perform cross-modal feature alignment and dynamic weight allocation based on the multi-head attention mechanism to meet the holistic view constraints of "four-diagnosis combined" in "TCM Diagnosis"; Step S2: Using multimodal features, combining the compound syndrome database and authoritative medical books on the description of symptoms and syndromes of gynecological diseases in Traditional Chinese Medicine, the nonlinear combination relationship between symptoms is captured, and the multi-head attention mechanism dynamically identifies the corresponding syndromes to obtain the identified syndrome results; Step S3: Based on the syndrome identification results, perform prescription-symptom attention matching, calculate the similarity between the patient's syndrome and the prescription indication, screen candidate prescriptions through multi-headed attention, and recommend dynamic addition or subtraction of medicinal materials based on the patient's physical constitution bias; Step S4: Collect continuous monitoring data of patients, use multi-head attention to capture the evolution of pathological characteristics at different time points, extract periodic patterns, predict future syndrome changes and generate health management plans; Step S5: The case diagnosis and treatment process is digitized, and dynamic knowledge evolution is achieved through the Transformer network. This is combined with a closed-loop clinical feedback loop and knowledge updates, focusing on the adaptive integration of regional schools of thought, and iteratively optimizing the auxiliary diagnosis and treatment system.

2. The TCM gynecology auxiliary diagnosis and treatment system based on an attention mechanism neural network according to claim 1 is characterized by: In step S1, the multi-dimensional data and features of the patient are pre-processed, wherein the multi-modal input features are text, and the multi-modal input features include text, images, and time series signals, wherein the text includes the patient's chief complaint and medical history, the image includes the tongue image and complexion, and the time series signals include the pulse condition and physical condition monitoring data; The multi-head attention mechanism splits the input feature data into multiple subspaces, with each attention head focusing on different modal associations. It dynamically captures global dependencies through self-attention, automatically determines the importance of symptoms, and dynamically adjusts the impact weights of different syndromes on patient health. When the tongue diagnosis indicates "stasis syndrome" but the pulse does not show "wiry pulse", cross-modal contradiction detection is triggered, and features with higher relevance to the main complaint are prioritized and the impact weights are dynamically adjusted.

3. The TCM gynecology auxiliary diagnosis and treatment system based on an attention mechanism neural network according to claim 1 is characterized by: In step S2, a symptom-syndrome mapping matrix is ​​constructed based on the compound syndrome type library of "Traditional Chinese Medicine Gynecology" and the authoritative medical books "Gynecological Syndrome and Treatment Standards" and "Fu Qingzhu's Gynecology".

4. The TCM gynecology auxiliary diagnosis and treatment system based on an attention mechanism neural network according to claim 1, characterized in that: In step S3, when recommending and adding or subtracting prescriptions, classic prescriptions need to be flexibly modified according to individual differences by adding or subtracting medicinal materials. At the same time, the prescription composition, efficacy, and contraindications are structured as key-value pairs of the attention mechanism.

5. The TCM gynecology auxiliary diagnosis and treatment system based on an attention mechanism neural network according to claim 1 is characterized by: In the multi-head attention matching stage in step S3, the formula for converting the prescription into a structured vector is: ; Prescriptions are initially screened based on the menstrual cycle stage, and different dimensions are focused on in the cross-attention calculation: medicinal material-symptom matching, pathogenesis fit, pregnancy safety control, cycle synchronization optimization, and the addition and subtraction recommendation stages are ranked according to the importance of the medicinal materials.

6. The TCM gynecology auxiliary diagnosis and treatment system based on an attention mechanism neural network according to claim 1 is characterized by: In step S4, a multi-head attention mechanism is combined to design a disease prevention prediction module, including data input, model architecture, and health management; Patient historical data slices are converted into multi-source heterogeneous data input, including menstrual cycle logs, basal body temperature curves, tongue dynamic changes, pulse trends, life behavior data, and modern medical indicators; Construct a three-level prevention prediction model to prevent illness before it occurs, prevent illness before it strikes, and prevent illness from progressing. Based on the patient's possible future health problems, different health management plans and personalized health recommendations are generated that take into account physical characteristics and the current solar terms.

7. The TCM gynecology auxiliary diagnosis and treatment system based on an attention mechanism neural network according to claim 1 is characterized by: In step S5, an iterative optimization module is introduced, which includes a data construction step, an incremental learning step, and a quality control step; The data collection specifications in the data construction step are: structured four-diagnosis information and quantified efficacy feedback; In the incremental learning step, the symptom-syndrome attention mapping is updated in stages, the attention association between new cases and existing knowledge is calculated and regional adaptation is enhanced, and the experience of different schools of thought is integrated; In the quality control step, a dual-channel verification mechanism is set up, the model recommends a plan, experts conduct random sampling and rule engine verification, determine whether to enter the learning queue, design a feedback closed-loop design, and derive an efficacy score based on clinical results and implementation plans. If the score exceeds the set threshold, a difference analysis is performed on the predicted features and actual features, and the database adds a revised case to trigger model retraining.

8. The TCM gynecology auxiliary diagnosis and treatment system based on attention mechanism neural network according to claim 1 is characterized in that: include: A TCM gynecology multimodal clinical manifestation fusion module is used to obtain patient information, extract and align multimodal features, and dynamically assign weights; A dynamic identification module for TCM gynecological syndromes, which is used to dynamically identify syndromes based on the symptom-syndrome relationship using the patient's multimodal characteristics; A TCM gynecological prescription recommendation and addition / subtraction module is used to recommend prescriptions and dynamically add / subtract prescriptions based on the syndrome identification results; A TCM gynecology disease prevention prediction module, used to predict future symptom changes in patients and generate health management plans; The iterative optimization module of TCM gynecological diagnosis and treatment experience is used to digitize the case diagnosis and treatment process and iteratively optimize the auxiliary diagnosis and treatment system.

9. The TCM gynecology auxiliary diagnosis and treatment system based on an attention mechanism neural network according to claim 2, characterized in that: The multimodal feature fusion extracts the patient's clinical manifestation characteristics as input features through different encoders, namely text, image, and time series signal; the dimension of the attention head is set to 8, namely menstrual disorders, leucorrhea, pregnancy disorders, postpartum diseases, gynecological diseases, overall syndrome differentiation and individualized analysis, complex syndrome type-treatment method reasoning, and symptom contribution evaluation; 8 linear transformations are performed on the three groups of input features to generate 8 independent input feature matrices, namely: Where, is the text feature linear projection matrix, is the image feature linear projection matrix, is the linear projection matrix of the time series signal characteristics; Cross-modal multi-head attention fusion, parallel calculation of the scaled dot product attention of each head: Where, s is the scaling factor to prevent the softmax gradient from disappearing due to excessive dot product results; multimodal features are spliced, and through multi-head attention interaction, the outputs of the eight attention heads are spliced ​​and linearly transformed to obtain the patient's corresponding output data: Where, For the output projection matrix, ensure the dimensions align with the input.

10. The TCM gynecology auxiliary diagnosis and treatment system based on an attention mechanism neural network according to claim 6, characterized in that: The three-level prevention prediction model has a parallel triggering mechanism, focusing on the prediction of regularity of the menstrual cycle or the assessment of pregnancy risks, taking into account gynecological diseases and key elements in gynecological diagnosis; the first level is prevention before illness occurs, warning of health risks, using solar term sensitive position coding in the attention mechanism, dynamically adjusting according to the twenty-four solar terms, and predicting the probability of future gynecological diseases and identifying sub-health status based on temporal dependence and cross-modal interaction; the second level is prevention before illness occurs, intervening in the early stages of the disease, when the patient's menstrual cycle begins to become disordered but does not meet the diagnostic criteria, imaging examinations show changes in gynecological appendages but no clinical manifestations, monitoring the patient's various clinical manifestations, and using dynamic thresholds to warn of disease occurrence; the third level is prevention before illness occurs, managing the patient's overall course of illness, monitoring sudden abnormal indicators based on the patient's existing symptoms, and monitoring physiological indicators that are likely to lead to worsening of the disease.

Images