Traditional Chinese medicine acupuncture knowledge visual display system based on knowledge graph

Abstract

The invention belongs to the technical field of knowledge visualization, and discloses a traditional Chinese medicine acupuncture knowledge visual display system based on a knowledge graph. Comprising the steps that a traditional Chinese medicine ancient book data acquisition unit, a clinical case data acquisition unit and an ultrasonic image data acquisition unit acquire data and integrate the data into traditional Chinese medicine acupuncture data, and the traditional Chinese medicine acupuncture data is preprocessed to obtain perfect traditional Chinese medicine acupuncture data; constructing a knowledge graph based on the perfect traditional Chinese medicine acupuncture data; constructing an optimization network model to optimize the knowledge graph to obtain a high-quality knowledge graph; constructing an incremental learning framework, updating the high-quality knowledge graph based on the incremental learning framework, and outputting a real-time updated knowledge graph; constructing an interactive three-dimensional human body model based on the real-time updated knowledge graph; based on the real-time updated knowledge graph and the interactive three-dimensional human body model, outputting an executable medical strategy for the application case, and sending the executable medical strategy to the student terminal; and the teaching quality is improved while knowledge visualization is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of knowledge visualization. More specifically, the present invention relates to a knowledge visualization display system for traditional Chinese medicine acupuncture based on a knowledge graph. Background Art

[0002] The patent with the application publication number CN107357842A discloses a multi-screen knowledge visualization method based on knowledge classification, including the following steps: 1) Using a logistic regression classifier established by a classification method based on logistic regression in machine learning, classify knowledge into three categories: conceptual knowledge, process knowledge, and application knowledge; 2) Knowledge visualization: Use corresponding visualization tools to perform visualization operations on the knowledge and make different types of visualization images; 3) For the visualized knowledge, combined with corresponding text descriptions, use multi-screen technology to simultaneously display the visualization images and text information. Based on the knowledge visualization method, the view of classifying visualized knowledge is proposed, and it is displayed in combination with a multi-screen teaching system. At the same time, information is presented in both visual and text forms, allowing learners to process information through "dual coding", enhancing the understanding and memory of knowledge, and promoting the application of knowledge visualization technology in teacher teaching, knowledge transfer, thinking inspiration, etc.

[0003] Currently, in the field of education, it is often necessary to use knowledge visualization technology. However, the adaptability of current traditional methods is insufficient. For the traditional Chinese characters or variant Chinese characters in ancient traditional Chinese medicine acupuncture books, they cannot be recognized or converted into simplified Chinese characters, resulting in low semantic analysis accuracy; the data used is too single, unable to fuse multi-modal data to construct a unified knowledge representation, and it is not perfect enough; mostly static knowledge graphs are used, resulting in lagging knowledge updates, unable to ensure timeliness, and there are also contradictions in different knowledge, resulting in low stability of the knowledge graph; when constructing a three-dimensional model, due to the lack of consideration of the influence of patient body type differences or physical factors, the generated model has low accuracy and low interactivity; it is unable to give corresponding executable medical strategies for different cases, reducing the teaching effect.

[0004] In view of this, the present invention proposes a knowledge visualization display system for traditional Chinese medicine acupuncture based on a knowledge graph to solve the above problems. Summary of the Invention

[0005] In order to overcome the above defects of the prior art and to achieve the above object, the present invention provides the following technical solution: A knowledge visualization display system for traditional Chinese medicine acupuncture based on a knowledge graph, including:

[0006] The knowledge graph construction module includes a traditional Chinese medicine ancient book data collection unit, a clinical case data collection unit, and an ultrasonic image data collection unit; integrate the data of each collection unit to obtain traditional Chinese medicine acupuncture data and preprocess the traditional Chinese medicine acupuncture data to obtain improved traditional Chinese medicine acupuncture data; construct a knowledge graph based on the improved traditional Chinese medicine acupuncture data;

[0007] The knowledge graph optimization module constructs an optimization network model to optimize the knowledge graph to obtain a high-quality knowledge graph; constructs an incremental learning framework and updates the high-quality knowledge graph based on the incremental learning framework, and outputs a real-time updated knowledge graph;

[0008] The visualization module constructs an interactive three-dimensional human body model based on the real-time updated knowledge graph;

[0009] The medical strategy generation module outputs an executable medical strategy for the applied case based on the real-time updated knowledge graph and the interactive three-dimensional human body model, and sends the executable medical strategy to the student terminal.

[0010] Furthermore, the traditional Chinese medicine ancient book data collection unit extracts the description fields related to traditional Chinese medicine acupuncture knowledge, marks the traditional Chinese characters and variant Chinese characters in the description fields related to traditional Chinese medicine acupuncture knowledge, and constructs a mapping dictionary for the marked traditional Chinese characters and variant Chinese characters in the description fields related to traditional Chinese medicine acupuncture knowledge; based on the mapping dictionary, convert the marked traditional Chinese characters and variant Chinese characters into simplified Chinese to obtain traditional Chinese medicine ancient book data; the clinical case data collection unit collects acupuncture diagnosis and treatment records by docking with the hospital's case database, and performs word segmentation processing on each acupuncture diagnosis and treatment record to obtain clinical case data; the ultrasonic image data collection unit obtains ultrasonic image data by collecting ultrasonic images related to acupuncture treatment in the hospital's case database.

[0011] Furthermore, the method for preprocessing the traditional Chinese medicine acupuncture data includes:

[0012] Construct a human body coordinate system with the lower edge of the patella as the origin, map the traditional terms in the traditional Chinese medicine ancient book data to the human body coordinate system to obtain coordinate-converted ancient book data; construct a semantic model to perform semantic disambiguation on the coordinate-converted ancient book data to obtain unambiguous ancient book data; remove redundant data in the clinical case data to obtain streamlined case data; perform denoising processing on the ultrasonic image data to obtain high-quality ultrasonic image data; combine the unambiguous ancient book data, streamlined case data, and high-quality ultrasonic image data to obtain cleaned traditional Chinese medicine acupuncture data; extract features from the cleaned traditional Chinese medicine acupuncture data, and at the same time perform multi-modal feature fusion to obtain fusion features; set feature conflict resolution rules, and adjust the fusion features based on the feature conflict resolution rules to obtain fused multi-modal features; perform multi-tuple conversion on each fused multi-modal feature to generate structured fusion features, and integrate all structured fusion features to generate improved traditional Chinese medicine acupuncture data.

[0013] Furthermore, the method for the constructed semantic model to perform semantic disambiguation on the coordinate-transformed ancient book data includes:

[0014] Taking the BERT-BiLSTM network model as the basic structure of the semantic model and adding an attention mechanism; the semantic model includes a BERT layer, a BiLSTM layer, an attention channel, and a decoding layer; using a word segmentation tool to perform sentence splitting and word segmentation on the coordinate-transformed ancient book data successively to obtain sentence-split and word-segmented ancient book data; deploying a pre-trained language model in the field of traditional Chinese medicine knowledge in the BERT layer of the semantic model for global semantic modeling and converting the sentence-split and word-segmented ancient book data into multi-dimensional word vectors and sentence vectors based on the pre-trained language model in the field of traditional Chinese medicine knowledge, and then using the bidirectional LSTM model of the BiLSTM layer to obtain the forward and backward semantic dependency relationships of any multi-dimensional word vector in any sentence vector; dividing the attention channel into a word-level attention branch and a sentence-level attention branch, calculating the weight of each word in the multi-dimensional word vector in the sentence vector based on the word-level attention branch; calculating the weight of any sentence vector in the original text based on the sentence-level attention branch; constructing a joint loss function and optimizing the semantic model based on the joint loss function; constructing a sentence-word association matrix based on the weight of each word in the multi-dimensional word vector and the weight of each sentence vector in the original text, and decoding through the decoding layer to obtain an acupuncture knowledge vector; integrating all acupuncture knowledge vectors to obtain non-ambiguous ancient book data;

[0015] The method for removing redundant data in the clinical case data includes:

[0016] The clinical case data includes etiology, treatment acupoint plan, acupuncture depth, and curative effect evaluation; constructing a dictionary library, encoding the etiology based on the dictionary library to obtain an etiology code; representing the treatment acupoint plan using binary encoding and mapping the encoding to a vector space with a dimension equal to the number of acupoints to obtain an acupoint vector; discretizing the acupuncture depth to obtain a discrete acupuncture depth; quantifying the curative effect evaluation and representing it using a constant to obtain a quantified curative effect evaluation; combining the etiology code, acupoint vector, discrete acupuncture depth, and quantified curative effect evaluation into a clinical case matrix; initializing the number of cluster centers as K; regarding each clinical case matrix as a sample point and randomly selecting K sample points from all sample points as cluster centers; constructing a weighted distance formula to calculate the distance between any two sample points, and calculating the silhouette coefficient based on the distance between any two sample points; continuously adjusting the number of cluster centers during the clustering process, while calculating and recording the silhouette coefficient, and taking the number of cluster centers corresponding to the maximum silhouette coefficient as the optimal number of clusters; grouping all sample points based on the optimal number of clusters, setting a distance threshold, and regarding the sample points in each group with a distance less than the distance threshold from the cluster center as redundant points and removing them.

[0017] Furthermore, the method for denoising the ultrasonic image data includes:

[0018] Using wavelet multi - band decomposition, perform low - pass filtering and high - pass filtering on the rows of each ultrasound image in the ultrasound image data to generate horizontal low - frequency vectors and horizontal high - frequency vectors respectively; then perform low - pass filtering and high - pass filtering on the horizontal low - frequency vectors and horizontal high - frequency vectors in the column direction to generate ultrasound low - frequency sub - bands and ultrasound high - frequency sub - bands; introduce diffusion variance in the ultrasound high - frequency sub - band, and use Monte Carlo sampling to simulate the noise distribution and calculate the pixel gray - level diffusion variance of the ultrasound high - frequency sub - band at the same time; construct an adaptive threshold function, and based on the adaptive threshold function and the pixel gray - level diffusion variance, identify the noise regions in the ultrasound high - frequency sub - band and perform noise suppression to obtain a preliminary denoised high - frequency sub - band; determine the acupoint coordinates in the ultrasound low - frequency sub - band, and construct a local window centered on the acupoint coordinates; set a gray - level threshold, and based on the gray - level threshold, divide some sub - bands in the local window into background regions and main regions; perform histogram equalization on the main regions of some sub - bands in the local window to generate a contrast - enhanced low - frequency sub - band; add a dual - branch enhancement mechanism, including a spatial - domain branch and a frequency - domain branch, and the two branches are processed in parallel to perform dual - branch enhancement on the preliminary denoised high - frequency sub - band and the contrast - enhanced low - frequency sub - band; in the spatial - domain branch, adaptively adjust the shape of the convolution kernel using dynamic deformable convolution and perform convolution operations on the preliminary denoised high - frequency sub - band and the contrast - enhanced low - frequency sub - band using the adjusted convolution kernel; perform weighted fusion on the frequency - domain branch and the spatial - domain branch; in the frequency - domain branch, reconstruct the signal of the preliminary denoised high - frequency sub - band through wavelet inverse transform, and output the denoised ultrasound image after dual - branch parallel processing; use the local dynamic enhancement algorithm to enhance the denoised ultrasound image to obtain high - quality ultrasound image data.

[0019] Furthermore, the method for feature extraction of the cleaned traditional Chinese medicine acupuncture data includes:

[0020] Define the triple of ancient Chinese medical texts, which includes the object set, the attribute set, and the binary relation; extract the object set features, the attribute set features, and the binary relation features of each clause in the unambiguous ancient Chinese medical data, and form the triple of ancient Chinese medical texts for any clause; traverse each clause in the unambiguous ancient Chinese medical data, so that the triple of ancient Chinese medical texts for any clause can form a mapping with other clauses, thereby expanding the triple of ancient Chinese medical texts for any clause into a multi-dimensional association matrix; construct a semantic framework, and at the same time add the relationships unique to traditional Chinese medicine to the semantic framework; perform semantic expansion on the multi-dimensional association matrix based on the semantic framework to obtain the acupuncture semantic feature vector; traverse each piece of data in the streamlined case data, divide the streamlined case data into two categories based on the efficacy evaluation, including normal cases and boundary cases, calculate the efficacy score for each normal case and boundary case based on the efficacy evaluation of each case, and add a scoring item for each case based on the efficacy score to obtain the scored case feature vector; divide each high-quality ultrasound image in the high-quality ultrasound image data into adjacent pixel blocks of a preset size; compare the central pixel in any pixel block with the gray values of other pixels, and perform binary encoding on the pixel block based on the gray values of the central pixel and other pixels to obtain the binary sequence corresponding to each pixel block; perform decimal conversion on the binary sequence corresponding to each pixel block to obtain the encoded value of each pixel block; count the distribution of the encoded values of all pixel blocks in each high-quality ultrasound image to form the ultrasound image vector.

[0021] Further, the method for optimizing the knowledge graph by constructing an optimization network model includes:

[0022] Take the relationship-aware network model specialized for traditional Chinese medicine knowledge as the basic structure of the optimization network model, and at the same time add an attention mechanism improved based on traditional Chinese medicine acupuncture knowledge to the optimization network model; input the knowledge graph into the optimization network model in the form of multi-tuples; extract the relationships unique to traditional Chinese medicine in each multi-tuple, and construct a constraint matrix based on the relationships unique to traditional Chinese medicine; construct a meridian attribution vector and map the meridian attribution vector to the definition in the existing materials based on the existing materials; construct a time encoding based on the time features in the constraint matrix, and perform a three-dimensional Fourier transform on the time encoding to generate a three-dimensional time series matrix; calculate the attention weights based on the meridian attribution vector and the three-dimensional time series matrix, and add implicit relationships to the multi-tuples based on the attention weights; construct a dynamic topological structure, and take the Poincaré sphere model as the core structure of the dynamic topological structure; regard each multi-tuple after adding implicit relationships as a node, and calculate the hyperbolic distance between any two nodes based on the dynamic topological structure; perform hierarchical adjustment on the knowledge graph based on the hyperbolic distance to obtain a high-quality knowledge graph.

[0023] Further, the method for constructing an incremental learning framework and updating the high-quality knowledge graph based on the incremental learning framework includes:

[0024] In the incremental learning framework, a heterogeneous data channel is constructed to connect to the hospital's case database to collect new case data and new ultrasound images, and new ancient book data is obtained from a preset traditional Chinese medicine ancient book database. The new case data, new ultrasound images, and new ancient book data are combined into incremental data. An incremental data buffer is constructed, and the incremental data is stored in the incremental data buffer. At the same time, a buffer threshold is set. When the data volume of the incremental data is greater than the buffer threshold, incremental training is performed on the high-quality knowledge graph. A modular processing component library is constructed. In the modular processing component library, a BERT model specialized in the field of traditional Chinese medicine is used to perform entity recognition and relationship extraction on the ancient book text type data and case text type data in the incremental data. The UNet++ network model is used to accurately locate the acupoints corresponding to the image type data in the incremental data. The structured data in the incremental data is automatically converted into a multi-tuple format. At the same time, the same preprocessing process as the original knowledge graph is performed on the incremental data to obtain preprocessed incremental data. Calculate the confidence of the preprocessed incremental data, and perform incremental training using a hierarchical update strategy based on the confidence. When the confidence of the preprocessed incremental data is greater than the preset confidence threshold, hot update is performed; otherwise, cold update is performed. The hot update method includes: constructing a dynamic loss function and calculating the parameter adjustment amount based on the dynamic loss function. The cold update method includes: constructing a version optimization function and performing a full-scale iteration or rollback on the high-quality knowledge graph based on the version optimization function.

[0025] Further, the method for constructing an interactive three-dimensional human body model based on the real-time updated knowledge graph includes:

[0026] Construct an anatomical coordinate system; introduce traditional Chinese medicine topological parameters and construct a three-dimensional basic human body grid model based on the traditional Chinese medicine topological parameters and the anatomical coordinate system; construct a dynamic mapping function, and accurately associate the acupoints and meridians in the real-time updated knowledge graph with the coordinates in the three-dimensional basic human body grid model based on the dynamic mapping function; set a real-time incremental rendering mechanism, calculate the difference degree between different versions of the real-time updated knowledge graph, and construct a rendering priority function based on the difference degree; dynamically adjust the rendering level of the three-dimensional basic human body grid model based on the rendering priority function; construct a multi-physics field coupling equation to quantitatively process the response effect of each acupoint in the three-dimensional basic human body grid model, and add a display effect to the three-dimensional basic human body grid model; define a gesture feature vector, and construct a gesture space based on the gesture feature vector; associate the gesture space with the three-dimensional basic human body grid model, and define the operation instructions corresponding to any gesture feature vector in the gesture space; construct a labeling target optimization function, adjust the labeling layout in the three-dimensional basic human body grid model based on the labeling target optimization function, and add gesture-labeling linkage feedback to obtain an interactive three-dimensional human body model.

[0027] Further, the method for outputting an executable medical strategy for an application case based on a real-time updated knowledge graph and an interactive three-dimensional human body model includes:

[0028] The application case includes text-type case data, patient physical sign case data, and patient basic information; the BERT-BiLSTM network model is used to process the text-type case data to extract keyword vectors related to the disease; the patient physical sign case data includes tongue image and pulse waveform. The UNet++ network model is used to perform image segmentation on the tongue image to extract the HSV color features of the tongue texture, tongue coating, and crack density; the wavelet transform algorithm is used to perform wavelet packet decomposition on the pulse waveform to extract pulse features; the features of the tongue image and the pulse waveform are mapped to the acupuncture knowledge field, and disease information data is output; the keyword vectors and the disease information data are matched with the real-time updated knowledge graph. The vector-type data composed of the keyword vectors and the disease information data is regarded as a disease node, and each multi-tuple in the real-time updated knowledge graph is regarded as a diagnosis and treatment node. The disease similarity between the disease node and any diagnosis and treatment node is calculated. When the disease similarity is greater than a preset similarity threshold, the diagnosis and treatment plan represented by the relevant diagnosis and treatment node is output; the actual coordinates of the acupoints on the interactive three-dimensional human body model are dynamically adjusted based on the size of the interactive three-dimensional human body model and the patient basic information; a tissue deformation compensation function is constructed, and the diagnosis and treatment plan is adjusted based on the tissue deformation compensation function; the interactive three-dimensional human body model is rendered at multiple levels and dynamically parameter-labeled based on the adjusted diagnosis and treatment plan, and an executable medical strategy for the application case is obtained based on the adjusted diagnosis and treatment plan and the rendered interactive three-dimensional human body model.

[0029] The technical effects and advantages of a traditional Chinese medicine acupuncture knowledge visualization display system based on a knowledge graph according to the present invention:

[0030] By constructing a knowledge graph construction module, collecting the data required for constructing the knowledge graph based on three data collection units therein, and performing high-precision preprocessing at the same time, perfect and accurate traditional Chinese medicine acupuncture data are obtained to construct the knowledge graph. In view of the fact that there are still some problems in the knowledge graph even though it is constructed from preprocessed data, an optimization network model is selected to optimize the knowledge graph. At the same time, considering that the knowledge graph needs to be updated, an incremental learning framework is constructed to update the knowledge graph, and a real-time updated knowledge graph is obtained. To achieve knowledge visualization, a three-dimensional human body model is generated based on the real-time updated knowledge graph, and an interactive mechanism is added. Finally, a medical strategy generation module is added, which can generate executable medical strategies for the input application cases and display them on the interactive three-dimensional human body model, which is beneficial to promoting teaching. Compared with the existing field, data with a wider coverage range are collected and higher-precision preprocessing is performed. At the same time, the constructed knowledge graph is optimized and updated in real time. To facilitate students' learning of knowledge, an interactive mechanism is added to the constructed three-dimensional human body model. The system composed of the constructed knowledge graph and the interactive three-dimensional human body model can also output executable medical strategies for the input application cases and display them on the model, enhancing the practicability and facilitating the display of visual knowledge. Brief Description of the Drawings

[0031] Figure 1 Schematic diagram of a traditional Chinese medicine acupuncture knowledge visualization display system based on a knowledge graph according to the present invention;

[0032] Figure 2 Schematic diagram of a traditional Chinese medicine acupuncture knowledge visualization display method based on a knowledge graph according to the present invention. Detailed Embodiments

[0033] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0034] Embodiment 1

[0035] Please refer to Figure 1 As shown, a traditional Chinese medicine acupuncture knowledge visualization display system based on a knowledge graph in this embodiment includes:

[0036] A knowledge graph construction module, including a traditional Chinese medicine ancient book data collection unit, a clinical case data collection unit, and an ultrasonic image data collection unit; integrating the data of each collection unit to obtain traditional Chinese medicine acupuncture data and preprocessing the traditional Chinese medicine acupuncture data to obtain perfect traditional Chinese medicine acupuncture data; constructing a knowledge graph based on the perfect traditional Chinese medicine acupuncture data;

[0037] The knowledge graph optimization module constructs an optimization network model to optimize the knowledge graph and obtains a high-quality knowledge graph; constructs an incremental learning framework and updates the high-quality knowledge graph based on the incremental learning framework, and outputs a real-time updated knowledge graph;

[0038] The visualization module constructs an interactive three-dimensional human body model based on the real-time updated knowledge graph;

[0039] The medical strategy generation module outputs an executable medical strategy for the application case based on the real-time updated knowledge graph and the interactive three-dimensional human body model, and sends the executable medical strategy to the student terminal.

[0040] Collect data related to traditional Chinese medicine acupuncture knowledge from multiple data sources, covering three aspects: medical ancient books, clinical cases, and ultrasound image types, making the collected data cover a wider range; improving the efficiency of subsequent knowledge graph construction and making the knowledge graph more comprehensive.

[0041] The traditional Chinese medicine ancient books data collection unit extracts the description fields related to traditional Chinese medicine acupuncture knowledge from the digital versions of ancient books stored in the preset traditional Chinese medicine ancient books database (such as Huangdi Neijing, Zhenjiu Jiayijing, Zhenjiu Dacheng and other famous ancient medical books related to acupuncture knowledge), and uses a character recognition engine based on deep learning to mark the traditional Chinese characters and variant characters in the description fields related to traditional Chinese medicine acupuncture knowledge (the character recognition engine is such as Tencent Cloud Ancient Books OCR Engine; the writing format of ancient articles is complex and the typesetting methods are diverse, such as vertical typesetting and double-line plus annotation typesetting methods; Traditional Chinese characters are quite different from the simplified Chinese characters currently in use, and there are also many variants of variant characters, such as the traditional Chinese character "鍼" → simplified Chinese character "针", the traditional Chinese character "癥" → simplified Chinese character "症"; for example, the variants of "脉" include "脉" and "衇"; in order to make the constructed knowledge graph easy to view, it is necessary to use OCR to recognize the traditional Chinese characters and variant characters in ancient texts), and build a mapping dictionary for the traditional Chinese characters and variant characters marked in the description fields related to Chinese medicine and acupuncture knowledge (the preset ancient Chinese dictionary includes the ancient traditional Chinese characters or variant characters themselves and their corresponding modern simplified characters, so that the preset ancient Chinese characters can be easily viewed. The Chinese dictionary and the marked characters form a mapping to construct a mapping dictionary, which is convenient for the subsequent conversion of ancient Chinese into modern simplified Chinese characters and the subsequent construction and viewing of knowledge graphs); based on the mapping dictionary, the marked characters are converted into simplified Chinese to obtain ancient Chinese medical book data; the clinical case data collection unit collects acupuncture treatment records by connecting to the hospital's case database (screening cases of acupuncture treatment in the case database to ensure that the collected data is only related to acupuncture), and compares the hospital database with the acupuncture treatment records by querying the hospital database, and performs word segmentation on each acupuncture treatment record (because each acupuncture treatment record will have a large number of proper nouns, it is necessary to query the hospital database and compare the treatment records and perform word segmentation, extract the cause of disease, acupoint scheme, acupuncture depth and efficacy evaluation in the acupuncture treatment records, remove redundant information in the data, and indirectly improve the efficiency of subsequent operations), and obtain clinical case data; the ultrasound image data collection unit obtains ultrasound image data by collecting ultrasound images related to acupuncture treatment in the hospital's case database (the ultrasound image refers to the image of the anatomical structure of the acupoint, which is convenient for the subsequent construction of an interactive three-dimensional human body model).

[0042] Traditional Chinese medicine and acupuncture data include ancient Chinese medicine book data, clinical case data and ultrasound image data; ancient Chinese medicine book data and clinical case data are text type data, ultrasound image data are image type data, text type data may contain redundant fields or semantic ambiguity, and images in image type data may contain noise. In order to make the data more accurate, it is necessary to clean the traditional Chinese medicine and acupuncture data; feature extraction is performed on the cleaned data, and the pathological features or semantic features are extracted and fused to output the medical feature representation; the knowledge graph constructed based on the medical feature representation has the advantages of high precision and small size, and also has stronger readability.

[0043] The ways of preprocessing traditional Chinese medicine acupuncture data include:

[0044] Taking the lower edge of the human patella as the origin, a human body coordinate system is constructed (the human body coordinate system is based on the lower edge of the patella of a human body model with common specifications in the medical field and can be adjusted according to the body proportions of different cases). By querying existing traditional Chinese medicine materials, the traditional terms in the ancient Chinese medicine data are mapped to the human body coordinate system to obtain coordinate-transformed ancient medicine data (since there are often traditional terms representing distances or orientations in the ancient Chinese medicine data, and their measurement methods and standards are slightly different from modern medicine, it is necessary to construct a corresponding human body coordinate system based on modern medicine and map the traditional terms by querying the "Semantic Network Framework of the Traditional Chinese Medicine Language System"; for example, in the "Huangdi Neijing", there is a description of "Shousanli, two cun below Quchi". Determine the coordinates of the Quchi acupoint in the constructed human body coordinate system, calculate the actual length of two cun, and determine the three-dimensional coordinates of Shousanli in the human body coordinate system based on the above data to complete the mapping of traditional terms and solve the problem of spatial ambiguity of traditional terms); construct a semantic model to perform semantic disambiguation on the coordinate-transformed ancient medicine data to obtain non-ambiguous ancient medicine data (since there is often a problem of polysemy of the same term in ancient books, for example, "wind" can refer to exogenous wind pathogen or internal stirring of liver wind. At this time, it is necessary to perform semantic disambiguation on the term based on the semantic context to avoid ambiguity in the constructed knowledge graph); use the clustering algorithm to remove redundant data in the clinical case data to obtain refined case data (the clinical case data is obtained by combining the diagnosis and treatment records of each patient undergoing acupuncture treatment. Therefore, there may be patients with the same etiology adopting the same treatment strategy, resulting in an increase in the number of duplicate diagnosis and treatment records and data redundancy. Therefore, the clustering algorithm is selected for processing here); perform denoising processing on the ultrasonic image data based on wavelet decomposition and image enhancement algorithms to obtain high-quality ultrasonic image data (there may be noise in the ultrasonic image data, resulting in low data quality. Therefore, it is necessary to use wavelet decomposition and image enhancement algorithms to perform denoising processing on it and enhance the image details, which is beneficial for subsequent model construction); combine the non-ambiguous ancient medicine data, refined case data, and high-quality ultrasonic image data to obtain the cleaned traditional Chinese medicine acupuncture data; perform feature extraction on the cleaned traditional Chinese medicine acupuncture data and at the same time perform multi-modal feature fusion to obtain fusion features (perform feature extraction on the non-ambiguous ancient medicine data, refined case data, and high-quality ultrasonic image data respectively, and fuse the features of the three modalities to form a fusion feature form of ancient medicine data features - case data features - ultrasonic image data features); set feature conflict resolution rules and adjust the fusion features based on the feature conflict resolution rules to obtain fused multi-modal features (the feature conflict resolution rules are set for ancient medicine features and case features. When there is a conflict between the ancient medicine records and the modern medicine case data, the modern medicine case data is preferentially used to correct the ancient medicine records; for example, the ancient book records that "Fengchi acupoint is prohibited from moxibustion", while modern medicine allows acupuncture treatment. At this time, the modern medicine conclusion is used as the standard to correct the ancient medicine record to avoid conclusion conflicts when constructing the knowledge graph).Perform tuple transformation on each fused multi-modal feature to generate structured fused features, and integrate all structured fused features to generate complete traditional Chinese medicine acupuncture data (perform tuple transformation on each fused multi-modal feature, and at the same time establish an index for acupoint coordinates based on the Neo4j graph database, and label this index on high-quality ultrasound images before feature extraction to form an attribute graph, realizing the structured expression of traditional Chinese medicine acupuncture knowledge, such as (Zusanli, stomach deficiency-cold syndrome, warm acupuncture, traditional Chinese medicine knowledge relationship, efficacy score, attribute graph ID); strengthening the correlation between various types of data and facilitating the construction of a more professional knowledge graph).;

[0045] The ways to construct a semantic model for semantic disambiguation of coordinate-transformed ancient book data include:

[0046] Take the BERT-BiLSTM network model as the basic structure of the semantic model, and add an attention mechanism (the BERT-BiLSTM network model is a deep learning model that combines the BERT model and the bidirectional long short-term memory network model. The BERT model is used to understand the context of text-type data, and the BiLSTM model is used to process the context dependencies in the text for enhancement; build an attention mechanism based on the attention network to enhance the ability of the semantic model to capture key information and improve the accuracy of context understanding); the semantic model includes a BERT layer, a BiLSTM layer, an attention channel, and a decoding layer; use a tokenization tool to perform sentence splitting and tokenization on the coordinate-transformed ancient book data in sequence to obtain sentence-split and tokenized ancient book data (the tokenization tool is, for example, the Jieba tokenization tool. First, perform sentence splitting based on the punctuation marks in the text, and then perform tokenization on each sentence); deploy a pre-trained language model in the BERT layer of the semantic model for global semantic modeling and transform the sentence-split and tokenized ancient book data into multi-dimensional word vectors and sentence vectors based on the pre-trained language model in the field of traditional Chinese medicine knowledge (the pre-trained language model in the field of traditional Chinese medicine knowledge is, for example, the ZhongBERT model, which maps sentence splitting and tokenization to a vector space and generates a vector representation containing context associations), and then use the bidirectional LSTM model in the BiLSTM layer to obtain the forward and backward semantic dependencies of any multi-dimensional word vector in any sentence vector (use the bidirectional LSTM model to perform forward processing and reverse processing on the multi-dimensional word vector respectively, capture the forward semantic dependency and the reverse semantic dependency, and use them to strengthen the local modification relationship; for example, in "exogenous wind pathogen", forward processing associates "exogenous" with "wind", and reverse processing associates "pathogen" with "wind"; splice the forward semantic dependency and the reverse semantic dependency to obtain the forward and backward semantic dependency relationship); divide the attention channel into a word-level attention branch and a sentence-level attention branch, and calculate the weight of each word in the multi-dimensional word vector in the sentence vector based on the word-level attention branch (the calculation formula for this process is: where, a0 iDenote the weight of the $i$-th word in any sentence vector; Denote the transpose of the multi-dimensional word vector of the $i$-th word in any sentence vector; total denotes the number of multi-dimensional word vectors in any sentence vector; $U0$ denotes the randomly initialized word-level attention vector; word-level attention is used to screen key terms in each clause); Calculate the weight of any sentence vector in the original text based on the sentence-level attention branch (the calculation formula of the process is: where, $a1$ i0 Denote the weight of the $i0$-th sentence vector in the clause-segmented ancient book data; Denote the transpose of the $i0$-th sentence vector in the clause-segmented ancient book data; num denotes the number of sentence vectors in the clause-segmented ancient book data; $U2$ denotes the randomly initialized sentence-level attention vector; sentence-level attention is used to identify the clauses containing key information in the original text); Construct a joint loss function and optimize the semantic model based on the joint loss function (the calculation formula of the joint loss function is: LOSS = μ1×L BERT +μ2×L ZYL +μ3×L rule ; where, L BERT Denote the semantic dependency loss function, which is the cross-entropy loss function in this embodiment; L ZYL Denote the attention loss function, which is the binary cross-entropy loss function in this embodiment; L rule Denote the constraint rule loss function, and the calculation formula is: L rule =∑ r∈RULE Via(r)×PE(r)×U4; where, RULE denotes the preset traditional Chinese medicine theory rule database; r denotes any rule; via(r) denotes an indicator function, which takes the value of 1 when the rule r is violated and 0 otherwise; PE(r) denotes a penalty function, and the value is dynamically adjusted based on the rule importance; U4 denotes a word or sentence vector in any multi-dimensional word vector; when it is detected that a certain word or sentence vector violates the rule, this loss function is triggered; μ1, μ2, and μ3 respectively denote the weights of the semantic dependency loss function, the attention loss function, and the constraint rule loss function. In this embodiment, μ1 = 1, μ2 = 0.8, μ3 = 0.3; Improve the performance of the semantic model based on the constraints of the joint loss function and improve the accuracy of model prediction); Construct a word-sentence association matrix based on the weight of each word in the multi-dimensional word vector and the weight of each sentence vector in the original text, and decode through the decoding layer to obtain the acupuncture knowledge vector (in the word-sentence association matrix, the meaning of the polysemous word in each sentence in this context will be selected based on the weight. For example, in "He plays basketball", the disambiguation of "play" depends on "basketball" and forms a higher weight with the sports context in the sentence, so the activation probability of this word in the ambiguous path such as "fight" is suppressed); Integrate all acupuncture knowledge vectors to obtain the unambiguous ancient book data.

[0047] The methods for removing redundant data in clinical case data include:

[0048] The clinical case data includes etiology, treatment acupoint plan, acupuncture depth, and efficacy evaluation; construct a dictionary library, encode the etiology based on this dictionary library to obtain an etiology code (each etiology and its corresponding number in the relevant field materials are recorded in the dictionary library, and this number is used as the etiology code); represent the treatment acupoint plan using binary coding and map this coding to a vector space with a dimension equal to the number of acupoints to obtain an acupoint vector (for example, if a certain treatment acupoint plan requires acupuncture at Zusanli acupoint and Tianchi acupoint, then these two acupoints are represented as 1, and other acupoints are all represented as 0, forming an acupoint vector, and the dimension of this vector is related to the number of relevant acupoints); discretize the acupuncture depth to obtain a discrete acupuncture depth (the acupuncture depth is decomposed into shallow acupuncture, medium acupuncture, and deep acupuncture, which respectively represent that the acupuncture depth is less than or equal to 1 cm, the acupuncture depth is greater than 1 cm and less than 2 cm, and the acupuncture depth is greater than or equal to 2 cm; in this embodiment, the constant 1 is used to represent shallow acupuncture, the constant 2 is used to represent medium acupuncture, and the constant 3 is used to represent deep acupuncture); quantify the efficacy evaluation and represent it using a constant to obtain a quantified efficacy evaluation (the efficacy evaluation is divided into marked effect, effective, and ineffective, which are respectively represented by the constants 6, 4, and 2); combine the etiology code, acupoint vector, discrete acupuncture depth, and quantified efficacy evaluation into a clinical case matrix (one clinical case matrix corresponds to one piece of clinical case data); initialize the number of clustering centers as K; regard each clinical case matrix as a sample point, and randomly select K sample points from all sample points as clustering centers (use an improved K-means clustering algorithm to perform clustering processing on the sample points); construct a weighted distance formula to calculate the distance between any two sample points, and calculate the silhouette coefficient based on the distance between any two sample points (the weighted distance formula is: dis(x, y) = α1 × JA(XW x , XW y ) + α2 × OD(x, y) + α3 × HD(x, y) + α4 × DP(x, y); where dis(x, y) represents the weighted distance between sample points x and y; JA(XW x , XW y) represents the Jaccard coefficient of the acupoint vector corresponding to the sample point x and the acupoint vector corresponding to the sample point y, which is used to measure the difference between the two acupoint vectors; OD(x, y) represents the Euclidean distance between the sample points x and y in the efficacy evaluation dimension; HD(x, y) represents the Hamming distance between the sample points x and y in the etiology coding, and the Hamming distance is used to measure the difference between different etiology codings; DP(x, y) represents the distance between the sample points x and y in the discrete acupuncture depth dimension, that is, the sum of the discrete acupuncture depths of the two sample points is averaged); during the clustering process, the number of cluster centers is continuously adjusted, and at the same time, the silhouette coefficient is calculated and recorded, and the number of cluster centers corresponding to the maximum silhouette coefficient is used as the optimal number of clusters (the silhouette coefficient is calculated based on the compactness and separation of any two sample points, and the compactness represents the weighted distance of any sample point to other sample points; the optimal number of clusters is found by dynamically adjusting the number of cluster centers to make the clustering algorithm achieve the optimal effect); based on the optimal number of clusters, all sample points are grouped, a distance threshold is set, and the sample points in each group whose distance from the cluster center is less than the distance threshold are regarded as redundant points and removed (the sample points in each group are sample points with high similarity. If the distance between the sample point and the cluster center is less than the distance threshold, it means that the sample points in this range have extremely high similarity with the cluster center and belong to redundant data; removing redundant data is equivalent to compressing the original case data, and the required storage space and computing resources are both reduced synchronously, improving the data processing efficiency).

[0049] The methods for denoising the ultrasonic image data include:

[0050] (Select an appropriate wavelet basis function, such as the Daubechies wavelet basis function) Use wavelet multi-band decomposition to perform low-pass filtering and high-pass filtering on the row direction of each ultrasound image in the ultrasound image data, respectively generating a horizontal low-frequency vector and a horizontal high-frequency vector (by performing row decomposition and column decomposition on each ultrasound image successively, the frequency domain features are gradually refined, mapping the spatial features and frequency domain features of the ultrasound image into different sub-bands, facilitating the subsequent more obvious discovery of noise distribution in the sub-bands); then perform low-pass filtering and high-pass filtering on the horizontal low-frequency vector and the horizontal high-frequency vector in the column direction to generate an ultrasound low-frequency sub-band and an ultrasound high-frequency sub-band (where the ultrasound low-frequency sub-band includes a low-frequency approximation sub-band, and the ultrasound high-frequency sub-band includes a vertical high-frequency sub-band, a horizontal high-frequency sub-band, and a diagonal high-frequency sub-band; the low-frequency sub-band retains the overall structure of the image and the gradually slowly changing gray-scale information, such as the information of structures like organ contours and tissue regions; the high-frequency sub-band retains the edges, textures, and details of the image, such as structures like blood vessel walls or nerve endings, and also contains noise interference); introduce a diffusion variance in the ultrasound high-frequency sub-band, and use Monte Carlo sampling to simulate the noise distribution and calculate the pixel gray-scale diffusion variance of the ultrasound high-frequency sub-band (perform several random noise samplings on the same region in the ultrasound high-frequency sub-band, and calculate the pixel gray-scale diffusion variance of this region. In this embodiment, the number of random noise samplings is 50 times); construct an adaptive threshold function, and based on the adaptive threshold function and the pixel gray-scale diffusion variance, identify the noise region in the ultrasound high-frequency sub-band and perform noise suppression to obtain a preliminary denoised high-frequency sub-band (the calculation formula of the adaptive threshold function is: where, H Sbase represents a preset initial variance threshold, which takes a value of 0.05 in this embodiment; σ 2 represents the pixel gray-scale diffusion variance; β represents an adjustment coefficient used to control the noise sensitivity. When the pixel gray-scale diffusion variance is large, the adaptive threshold is adjusted to prevent under-suppression of high-noise regions. In this embodiment, the adjustment coefficient β takes a value of 0.1; when the pixel gray-scale diffusion variance is greater than the adaptive threshold function, the region is determined as a high-noise region and noise suppression is performed, and at the same time, the suppression intensity is increased; the noise suppression is achieved based on wavelet coefficient attenuation, and the calculation formula of wavelet coefficient attenuation is: coe after = coe before ×γ0; where, coe after represents the attenuated wavelet coefficient; coe before represents the wavelet coefficient before attenuation; γ0 represents the attenuation weight, and the calculation formula is: Among them, γ represents the attenuation intensity coefficient, which takes a value of 0.3 in this embodiment; when the variance of pixel gray-scale diffusion is greater than the adaptive threshold, the attenuation weight decreases with the decrease of the noise intensity, achieving an improvement in the noise suppression intensity); determine the acupoint coordinates in the low-frequency sub-band of ultrasound, and construct a local window centered on the acupoint coordinates (in this embodiment, the size of the original local window is 10×10, and the size of the local window can be adjusted according to the actual area size); set a gray-scale threshold, and divide some sub-bands in the local window into a background area and a main area based on the gray-scale threshold (for example, set the gray-scale threshold to 50, the area with a gray-scale value lower than 50 is the background area, and the area with a gray-scale value higher than 50 is the main area); perform histogram equalization on the main area of some sub-bands in the local window to generate a low-frequency sub-band with enhanced contrast (stretch the pixels in the main area to the 0-255 gray-scale value range, and only perform equalization on the main area to avoid over-enhancing background noise); add a dual-branch enhancement mechanism, including a spatial domain branch and a frequency domain branch, and the two branches are processed in parallel to perform dual-branch enhancement on the preliminarily denoised high-frequency sub-band and the low-frequency sub-band with enhanced contrast (jointly processing the sub-bands in the spatial domain and the frequency domain is more comprehensive, with a wider coverage range, which is conducive to obtaining better-quality sub-bands); in the spatial domain branch, use dynamic deformable convolution to adaptively adjust the shape of the convolution kernel and use the adjusted convolution kernel to perform convolution operations on the preliminarily denoised high-frequency sub-band and the low-frequency sub-band with enhanced contrast (the dynamic deformable convolution is the DCN algorithm, which adjusts the shape of the convolution kernel according to the local gradient direction; for example, at the bend of a blood vessel, the convolution kernel deforms into an ellipse to enhance the edge effect); perform weighted fusion on the frequency domain branch and the spatial domain branch (for example, in the blood vessel edge area, to enhance the image details, increase the weight of the frequency domain branch; in the muscle layer area, to make the image clear in levels, increase the weight of the spatial domain branch, achieving comprehensive enhancement of the high-frequency sub-band and the low-frequency sub-band); in the frequency domain branch, perform signal reconstruction on the preliminarily denoised high-frequency sub-band through inverse wavelet transform (only perform signal reconstruction on the preliminarily denoised high-frequency sub-band, retaining the high-frequency features), and output the denoised ultrasound image after dual-branch parallel processing; use the local dynamic enhancement algorithm to perform image enhancement on the denoised ultrasound image to obtain high-quality ultrasound image data (construct a dynamic window centered on the acupoint coordinates, calculate the gray-scale distribution entropy value in this dynamic window, set the gray-scale entropy value threshold, and divide some parts in the dynamic window into a low-entropy area and a high-entropy area based on the gray-scale entropy value threshold and the gray-scale distribution entropy value; the calculation formula for the gray-scale distribution entropy value is: Among them, H represents the gray-scale distribution entropy value of the dynamic window; Per i0It represents the occurrence probability of the gray level i0 in the dynamic window, that is, the ratio of the number of pixels with the gray level i0 to the total number of pixels in the dynamic window; in this embodiment, the threshold value of the gray entropy is 1.5. When the gray distribution entropy is less than 1.5, the area within the dynamic window is a low-entropy area, and vice versa, it is a high-entropy area; for the low-entropy area, power transformation is used to stretch the contrast and enhance the visibility of the structure; for the high-entropy area, logarithmic transformation is applied to compress the dynamic range and suppress over-enhancement).

[0051] The methods for feature extraction of the cleaned traditional Chinese medicine acupuncture data include:

[0052] Define the triples of ancient Chinese medical texts, which include an object set, an attribute set, and a binary relationship (the object set includes disease entities, such as descriptions like "cold in the stomach" and "headache"; the attribute set includes acupoints and treatment methods, such as descriptions like "Zusanli" and "warming acupuncture"; the binary relationship represents the mapping between the object set and the attribute set, such as "acupoint → disease"); extract the object set features, attribute set features, and binary relationship features of each clause in the unambiguous ancient Chinese medical data, and form the triples of ancient Chinese medical texts for any clause (for example, "Zusanli → cold in the stomach"); traverse each clause in the unambiguous ancient Chinese medical data to enable the triples of ancient Chinese medical texts for any clause to form a mapping with other clauses, so as to expand the triples of ancient Chinese medical texts for any clause into a multi-dimensional association matrix (in other clauses, there may be a connection with the content of the current clause. For example, expand "Zusanli → cold in the stomach" into a four-dimensional association matrix, introducing information on the disease state and treatment method, manifested as (cold in the stomach, cold pain in the epigastrium, Zusanli, warming acupuncture)); construct a semantic framework and add the relationships unique to traditional Chinese medicine to the semantic framework (such as seasonal association Taboo compatibility ("Fengchi acupoint is forbidden for moxibustion"), constitutional taboo ("Moxibustion is forbidden for yin-deficient constitution"), etc., which are specific knowledge in the field of traditional Chinese medicine acupuncture; based on the semantic framework, semantic expansion is performed on the multi-dimensional association matrix to obtain the acupuncture semantic feature vector (if the content of a certain multi-dimensional association matrix has relevant specific relationships in traditional Chinese medicine, relevant relationships such as seasonal association and taboo compatibility are added to the multi-dimensional association matrix to form the acupuncture semantic feature vector); traverse each piece of data in the refined case data, divide the refined case data into two categories based on the efficacy evaluation, including normal cases and boundary cases, calculate the efficacy score for each case based on the efficacy evaluation of each normal case and boundary case, and add a scoring item to each case based on the efficacy score to obtain the scored case feature vector (where a boundary case refers to a case with an unclear efficacy evaluation, such as a case with the words "partially effective"; set the efficacy score for marked improvement as P1, the efficacy score for effectiveness as P2, and the efficacy score for ineffectiveness as P3; then the normal case can directly obtain the score based on the efficacy score of the efficacy evaluation; construct a fuzzy compensation confidence function to calculate the efficacy score of the boundary case and use it as the score of the boundary case; the calculation formula of the fuzzy compensation confidence function is: Score = ρ1×P1 + ρ2×P2 + ρ3×P3; where Score represents the efficacy score of any boundary case; ρ1, ρ2, and ρ3 represent the weight for marked improvement, the weight for effectiveness, and the weight for ineffectiveness respectively; in this embodiment, ρ1 = 0.2, ρ2 = 0.6, ρ3 = 0.2; P1 = 6, P2 = 4, P3 = 2); divide each high-quality ultrasound image in the high-quality ultrasound image data into adjacent pixel blocks of a preset size (in this embodiment, the preset size is 3×3); compare the gray value of the central pixel in any pixel block with the gray values of other pixels, and perform binary encoding on the pixel block based on the gray values of the central pixel and other pixels to obtain the binary sequence corresponding to each pixel block (when the gray value of other pixels in any pixel block is greater than or equal to the gray value of the central pixel, the pixel is encoded as 1, otherwise it is encoded as 0; binary encoding is performed to quantify the microscopic texture features in the high-quality ultrasound image); convert the binary sequence corresponding to each pixel block into a decimal value to obtain the encoded value of each pixel block (the value range of the encoded value is (0, 255)); count the distribution of the encoded values of all pixel blocks in each high-quality ultrasound image to form the ultrasound image vector (this ultrasound image vector is a 256-dimensional vector, and this vector represents the microscopic texture distribution pattern of the fascia tissue).

[0053] Although the content in the constructed knowledge graph is formed through high-precision preprocessing, the hidden relationships between different pieces of knowledge cannot be discovered. Therefore, an optimized network model is constructed to capture the hidden relationships between different pieces of knowledge and add them to the original knowledge graph to form a high-quality knowledge graph.

[0054] The methods for optimizing the knowledge graph by constructing an optimization network model include:

[0055] Taking the relationship-aware network model specialized for traditional Chinese medicine knowledge as the basic structure of the optimization network model, and at the same time adding an attention mechanism improved based on traditional Chinese medicine acupuncture knowledge to the optimization network model (the relationship-aware network model specialized for traditional Chinese medicine knowledge is the RAGAT-TCM network model, which is an attention network embedded with knowledge related to the field of traditional Chinese medicine acupuncture; the attention mechanism improved based on traditional Chinese medicine acupuncture knowledge is a mechanism related to the unique relationships of traditional Chinese medicine in the knowledge graph, such as knowledge of seasonal associations and constitutional taboos, etc.); inputting the knowledge graph into the optimization network model in the form of multi-tuples (the knowledge graph is obtained by integrating and improving traditional Chinese medicine acupuncture data, and each piece of data therein is a multi-dimensional vector); extracting the unique relationships of traditional Chinese medicine in each multi-tuple, and constructing a constraint matrix based on the unique relationships of traditional Chinese medicine (the constraint matrix is constructed by the unique relationships of traditional Chinese medicine and is used to correct possible contradictory descriptions in the knowledge graph); constructing a meridian attribution vector and mapping the meridian attribution vector to the definitions in the existing materials based on the existing materials (the meridian attribution vector is obtained by performing one-hot encoding on the meridian positions; the existing materials are, for example, the extraordinary meridians defined in "The Spiritual Pivot - The Meridians" or the knowledge in the meridian field provided by other materials; mapping the meridian attribution vector to the existing materials means constructing a mapping between the relevant meridians and the acupoints related to them); constructing a time encoding based on the time features in the constraint matrix, and performing a three-dimensional Fourier transform on the time encoding to generate a three-dimensional time series matrix (the constraint matrix includes time features, such as variables like season, hour, and lunar calendar, etc.; the calculation formula for the three-dimensional Fourier transform is: where TI represents the time encoding; LunarPhase represents the time feature manifested as the lunar calendar).; calculating the attention weights based on the meridian attribution vector and the three-dimensional time series matrix, and adding implicit relationships to the multi-tuples based on the attention weights (the calculation formula for the attention weights is: where AZ represents the relationship-specific relationship, which is used to control the influence direction of different relationship types on the attention, such as "taboo", "synergy"; M 经络 represents the meridian attribution vector; HA i1 represents the i1-th multi-tuple; BR represents the bias term; N0 represents the number of multi-tuples).; constructing a dynamic topological structure, and taking the Poincaré sphere model as the core structure of the dynamic topological structure (the Poincaré sphere model is a model used to represent three-dimensional hyperbolic geometry and is used to calculate hyperbolic distances); regarding each multi-tuple after adding implicit relationships as a node, and calculating the hyperbolic distance between any two nodes based on the dynamic topological structure (the calculation formula for the hyperbolic distance is: Among them, DIS(u0, v0) represents the hyperbolic distance between node u0 and node v0; arcosh represents the inverse hyperbolic cosine function); the knowledge graph is hierarchically adjusted based on the hyperbolic distance to obtain a high-quality knowledge graph (since the hyperbolic distance grows exponentially in the hyperbolic space, this characteristic enables the core nodes to be closer to the center of the hyperbolic space, so the space is compressed based on the hyperbolic distance; the core nodes are, for example, the multi-tuples corresponding to the key acupoints related to certain meridians).

[0056] To ensure the timeliness of the high-quality knowledge graph, the high-quality knowledge graph is selected for synchronous update; in order to minimize the time and computing resources required for the update as much as possible, an incremental learning framework is constructed to perform incremental training on the high-quality knowledge graph. To ensure the stability of each update, a better version is selected and applied to the high-quality knowledge graph to obtain a real-time updated knowledge graph.

[0057] The method of constructing an incremental learning framework and updating the high-quality knowledge graph based on the incremental learning framework includes:

[0058] In the incremental learning framework, a heterogeneous data channel is constructed to dock with the hospital's case database to collect new case data and new ultrasound images, and new ancient book data is obtained from the preset traditional Chinese medicine ancient book database. The new case data, new ultrasound images, and new ancient book data are combined into incremental data (using the heterogeneous data channel to monitor in real time whether there is data update in the data source, and using the updated data from the data source as incremental data for incremental learning in the knowledge graph); an incremental data buffer is constructed, and the incremental data is stored in the incremental data buffer. At the same time, a buffer threshold is set. When the data volume of the incremental data is greater than the buffer threshold, incremental training is performed on the high-quality knowledge graph (in this embodiment, the buffer threshold is set to 1000 data volumes. Constructing the incremental data buffer is not only to store the newly collected incremental data, but also to ensure that the latency of the incremental data accessing the knowledge graph is reduced and the update efficiency of the knowledge graph is accelerated); a modular processing component library is constructed. In the modular processing component library, the BERT model specialized in the field of traditional Chinese medicine is used to perform entity recognition and relationship extraction on the ancient book text type data and case text type data in the incremental data (the BERT model specialized in the field of traditional Chinese medicine is connected to the term libraries of works in the field of traditional Chinese medicine such as "Huangdi Neijing" and "Compendium of Acupuncture and Moxibustion", and adopts an adversarial training strategy to improve the ability to handle semantic ambiguity and improve the recognition accuracy of the correct semantic description of professional terms); the UNet++ network model is used to accurately locate the acupoints corresponding to the image type data in the incremental data (the UNet++ network model is an improved version model based on the UNet network model. On the basis of the UNet network model structure, a spatial attention mechanism is added to perform dual-channel supervised learning on the acupoint positioning and anatomical markers of the image type data in the incremental data, so that the coordinate positioning error of the acupoints is controlled within the preset accuracy. In this embodiment, the preset accuracy is the 1mm accuracy required by the national standard); the structured data in the incremental data is automatically converted into a multi-tuple format (by extracting the key words or clauses in the incremental data, the structured data in the incremental data is converted into the data representation form in the knowledge graph, such as (Zusanli, stomach deficiency-cold syndrome, warm acupuncture,...). The dimension of this multi-tuple depends on the amount of key information contained); at the same time, the same preprocessing process as the original knowledge graph is performed on the incremental data to obtain preprocessed incremental data; calculate the confidence of the preprocessed incremental data, and perform incremental training using a hierarchical update strategy based on the confidence. When the confidence of the preprocessed incremental data is greater than the preset confidence threshold, hot update is performed, otherwise cold update is performed (judging the required hierarchical update strategy through the confidence, taking into account the real-time and stability of the knowledge graph update; at the same time, saving the computing resources of the system for updating the high-quality knowledge graph and ensuring that the update response time is at a low level; the formula for calculating the confidence is: Among them, NU represents the number of multi-tuples in the preprocessed incremental data; NU(hr, rel) represents the number of occurrences of the concept path hr and the relationship rel in the preprocessed incremental data. For example, Acupoint A0 → Treatment Method → Disease A1; NU(rel) represents the number of occurrences of the relationship rel in the preprocessed incremental data, that is, the number of all multi-tuples containing the relationship rel in the preprocessed incremental data; NU(hr, *) represents the number of occurrences of the concept path in all multi-tuples. For example, the concept path of the type Acupoint A2 → (...) → (...); in this embodiment, the confidence threshold is set to 0.9); The ways of hot update include: constructing a dynamic loss function and calculating the parameter adjustment amount based on the dynamic loss function (the calculation formula of the parameter adjustment amount is: Among them, Δθ represents the parameter adjustment amount of the model in the process of constructing a high-quality knowledge graph; η represents the dynamic learning rate of the model in the process of constructing a high-quality knowledge graph, and the value range is (0.01, 0.1), and the corresponding value is selected according to the specific situation; e -ε×time represents the time decay factor, ε represents the time decay coefficient, and the value in this embodiment is 0.05; time represents the acquisition timestamp of the incremental data; LS represents the dynamic loss function, and the calculation formula is: LS = δ × ||poi new -GAT(poi old )||2; where, δ represents the topological consistency weight, and the value in this embodiment is 0.7; poi new represents the multi-tuples in the preprocessed incremental data; GAT(poi old ) extracts the aggregated information of the multi-tuples in the high-quality knowledge graph by using the graph attention network); The ways of cold update include: constructing a version optimization function and performing a full-scale iteration or rollback on the high-quality knowledge graph based on the version optimization function (the calculation formula of the version optimization function is: FA(KG num , KG current ); where KG num represents the num-th version; KG curren represents the current version; this function compares the old version of the high-quality knowledge graph with the current version based on the graph matching algorithm; if the function value of the version optimization function of the old version is greater than the preset version threshold compared with the current version, the high-quality knowledge graph is rolled back, otherwise, a full-scale iterative update is performed).

[0059] Accurate acupoint mapping is achieved by constructing a three-dimensional human body model, which transforms abstract traditional Chinese medicine theoretical knowledge into an intuitive three-dimensional model. At the same time, an interactive mechanism is added to facilitate students' further understanding of traditional Chinese medicine theoretical knowledge, facilitate the learning of acupuncture methods and their action mechanisms, and reduce the learning cost of clinical operations.

[0060] The ways of constructing an interactive three-dimensional human body model based on a real-time updated knowledge graph include:

[0061] Construct an anatomical coordinate system and use the international standard human anatomical system as a template (the international standard is the ISB standard); introduce traditional Chinese medicine topological parameters and construct a three-dimensional basic human body grid model based on the traditional Chinese medicine topological parameters and the anatomical coordinate system (the calculation formulas of the traditional Chinese medicine topological parameters include: where JS represents the number of meridians; FI jl (dir) represents the meridian direction; represents the normal vector field; dir represents the coordinate position; xS represents the number of acupoints; loc xw0 (dir) represents the acupoint coordinate position; generate a three-dimensional basic human body grid model containing meridian directions and acupoints based on the traditional Chinese medicine topological parameters using a modeling tool); construct a dynamic mapping function, and based on the dynamic mapping function, accurately associate the acupoints and meridians in the real-time updated knowledge graph with the coordinates in the three-dimensional basic human body grid model (the calculation formula of the dynamic mapping function is: where EK represents the real-time updated knowledge graph entity; xw0(EK) represents the acupoint entity corresponding to the acupoint in the real-time updated knowledge graph entity; β1 represents the weight coefficient of the acupoint entity, and takes the value of 0.6 in this embodiment; represents the tissue offset energy field, which is used to characterize the influence of human tissue deformation on the acupoint position; β2 represents the weight coefficient of the tissue offset energy field, and takes the value of 0.4 in this embodiment; represents the tangent vector field; β3(ht) represents the meridian weight function, where ht represents the meridian active period); set a real-time incremental rendering mechanism, calculate the difference degree between different versions of the real-time updated knowledge graph, and construct a rendering priority function based on the difference degree (the calculation formula of the difference degree is: where ΔG represents the difference degree between different versions of the real-time updated knowledge graph; Vnew represents the set of nodes and edges of the new version or the current version of the real-time updated knowledge graph, where single information such as traditional Chinese medicine acupoints or disease entities is used as nodes, and path information such as acupoint-meridian attribution relationships or disease-acupoint treatment relationships is used as edges; Vold represents the set of nodes and edges of the old version of the real-time updated knowledge graph; the numerator uses the difference metric norm to calculate the exclusive OR result of the new version and the old version node and edge sets, reflecting the change amount of the nodes and edges in different versions; the denominator normalizes the new and old version node and edge sets to eliminate the influence of the graph scale on the difference degree); dynamically adjust the rendering level of the three-dimensional basic human body grid model based on the rendering priority function (the calculation formula of the rendering priority function is: Among them, Pri(jd) represents the rendering priority of any node jd in the three-dimensional basic human body mesh model; CS(jd) represents the node centrality index, which is the average value of the sum of degree centrality, closeness centrality, and betweenness centrality. Degree centrality represents the ratio of the number of associated edges of a node to the maximum number of edges; closeness centrality represents the distance between any two nodes in the three-dimensional basic human body mesh model; betweenness centrality represents the proportion of the number of times any node appears in any edge. represents the spatio-temporal relationship derivative, where space represents the spatial distribution of nodes within a certain model time window; tw represents any model time window; ω1 and ω2 respectively represent the weights of the node centrality index and the spatio-temporal relationship derivative. In this embodiment, ω1 = 0.5 and ω2 = 0.5); A multi-physical-field coupling equation is constructed to quantitatively process the response effects of each acupoint in the three-dimensional basic human body mesh model, and a display effect is added to the three-dimensional basic human body mesh model (the multi-physical-field coupling equation includes an electric potential field equation and a temperature field equation. The calculation formula of the electric potential field equation is: Among them, θ0 represents the electrical conductivity of human tissues; θ1 represents the electrical potential distribution of human tissues; Ia represents the acupuncture current intensity; ι(WZ) represents the acupuncture position; The electric potential field equation quantifies the electric potential changes generated in human tissues during acupuncture; The calculation formula of the temperature field equation is: Among them, represents the temperature change; te represents the temperature change period; γ2 represents the thermal diffusivity of human tissues, which is obtained by querying existing data based on the actual situation; Q0 represents the heat conduction coefficient, which is obtained by querying existing data based on the actual situation; The temperature field equation quantifies the temperature changes caused by acupuncture and reflects the temperature change law in the meridians); Define a gesture feature vector and construct a gesture space based on the gesture feature vector (the gesture feature vector is expressed as: GK = [Vh, jw, jg, jt]; where, Vh represents the sliding speed of the finger on the screen; jw represents the position where the finger clicks on the screen; jg represents the number of contact points on the screen; jt represents the pressing duration of the contact points on the screen; In this embodiment, the model is displayed on a visual display screen; The representation method of the gesture space is: Among them, GK i3 represents the i3-th element in any gesture feature vector; GK refDenote the preset standard operation gesture feature vector, such as standard operations like acupoint selection and scaling operations; τ represents the matching threshold, which is used to determine whether the gesture feature vector matches the preset standard operation gesture feature vector); associate the gesture space with the three-dimensional basic human body mesh model, and define the operation instruction corresponding to any gesture feature vector in the gesture space (by associating each gesture feature vector with an operation instruction, students can implement corresponding operation instructions through specific gestures when using the three-dimensional basic human body mesh model); construct a labeled target optimization function, and adjust the labeling layout in the three-dimensional basic human body mesh model based on the labeled target optimization function, and add gesture-labeling linkage feedback (the calculation formula of the labeled target optimization function is: where, Sim j1，j2 represents the semantic association degree between the j1-th entity and the j2-th entity in the real-time updated knowledge graph, that is, the shortest path length between the two entities in the real-time updated knowledge graph; DS(Y j1 , Y j2 ) represents the distance between the position vectors of the j1-th entity and the j2-th entity in the real-time updated knowledge graph in the three-dimensional basic human body model; ω3 represents the semantic loss weight, and ω4 represents the visual overlap weight, both of which are adjusted based on the student's viewing angle distance and the maximum viewing distance. In this embodiment, ω3 = 0.7 and ω4 = 0.3; when the student selects a specific acupoint, the system performs dynamic adjustment of the labeling, only displays the labels of the acupoints associated with this acupoint and gathers them towards the focus area, and does not display the labels of other acupoints, which is convenient for students to observe), and a three-dimensional interactive human body model is obtained.

[0062] In order to deepen students' understanding of traditional Chinese medicine acupuncture knowledge, teachers can input application cases through the teacher terminal during teaching, and then generate a diagnosis and treatment plan corresponding to the application case based on the real-time updated knowledge graph. At the same time, display all acupoints, meridians, and taboo areas on the three-dimensional interactive human body model, and generate labels on the model to form an executable medical strategy; students directly receive the generated executable medical strategy through the student terminal, which improves the teaching quality and is conducive to students' understanding of knowledge.

[0063] The ways to output an executable medical strategy for an application case based on the real-time updated knowledge graph and the three-dimensional interactive human body model include:

[0064] The application cases include text-type case data, patient physical sign case data, and patient basic information (the source of the application case data is the medical records of a certain patient. The text-type case data is the written description of the disease in the medical records; the patient physical sign data is the patient's physical examination data; the patient basic information is information such as the patient's constitution type and BMI index); use the BERT-BiLSTM network model to process the text-type case data and extract keyword vectors related to the disease (use the BERT-BiLSTM network model to extract text features from the text-type case data through the pre-trained traditional Chinese medicine semantic understanding layer in it to obtain keyword vectors related to the disease); the patient physical sign case data includes tongue image and pulse waveform. Use the UNet++ network model to perform image segmentation on the tongue image and extract the HSV color features of the tongue body, tongue coating, and crack density (when conducting traditional Chinese medicine interrogation, the condition of the disease is generally judged by examining the tongue image and pulse condition. Among them, the tongue image reflects the color of the tongue body and tongue coating as well as the crack density of the tongue coating); use the wavelet transform algorithm to perform wavelet packet decomposition on the pulse waveform and extract pulse features (extract the low-frequency energy ratio feature of the pulse waveform based on wavelet packet decomposition. Through the low-frequency energy ratio feature, the pulse type and intensity can be judged to identify the disease); map the features of the tongue image and the features of the pulse waveform to the acupuncture knowledge field by querying materials in the traditional Chinese medicine field and output disease information data related to the acupuncture knowledge field (compare the relevant disease descriptions in the traditional Chinese medicine field materials with the features of the tongue image and the features of the pulse waveform to judge the specific disease type and degree. Mapping to the acupuncture knowledge field is to ensure the accuracy of disease diagnosis. For example, some diseases may not be treatable by acupuncture or have different names in the acupuncture knowledge field, so it is necessary to align with the acupuncture knowledge field); match the keyword vectors and disease information data with the real-time updated knowledge graph. Regard the vector-type data composed of the keyword vectors and disease information data as disease nodes, regard each multi-tuple in the real-time updated knowledge graph as a diagnosis and treatment node, calculate the disease similarity between the disease node and any diagnosis and treatment node, and output the diagnosis and treatment plan represented by the relevant diagnosis and treatment node when the disease similarity is greater than the preset similarity threshold (the calculation formula for the disease similarity is: Among them, Sml(bz, zl) represents the disease similarity between the disease node bz and any diagnosis and treatment node zl. Here, the graph embedding algorithm is used to calculate the semantic similarity. In this embodiment, the similarity threshold is set to 0.7. When the similarity is greater than 0.7, the diagnosis and treatment node corresponding to the disease node is included in the solution candidate set, and finally, a primary executable strategy including the main and auxiliary acupoint combination plan, acupuncture parameters, and physical taboos is selected); Dynamically adjust the actual coordinates of the acupoints on the interactive three-dimensional human model based on the size of the interactive three-dimensional human model and the patient's basic information (Based on the ISB standard, the acupoint coordinates on the interactive three-dimensional human model are adjusted proportionally according to the BMI index in the patient's basic information); Construct a tissue deformation compensation function and adjust the diagnosis and treatment plan based on the tissue deformation compensation function (The calculation formula of the tissue deformation compensation function is: Among them, Δdeep represents the change in acupuncture depth; kd represents the elastic coefficient of the acupoint location, that is, the elastic coefficient of the skin near the acupoint area, which is determined based on specific circumstances; Fo represents the pressing force; EB represents the tissue density of the acupoint location; It is used to represent the deformation degree of the skin tissue near the acupoint location. When the skin tissue deforms, the acupuncture depth during acupuncture treatment needs to be adjusted to prevent injury or ineffective acupuncture treatment); Based on the adjusted diagnosis and treatment plan, perform multi-level rendering on the interactive three-dimensional human model and perform dynamic parameter annotation (Render the interactive three-dimensional human model with multiple colors. Render the main acupoints in the diagnosis and treatment plan with red; Render the auxiliary acupoints with blue; Render the meridians associated with the acupoints with a golden particle flow, and the golden particle flow flows along the meridian direction; Render the taboo area with a black grid area with superimposed flashing warnings; When the student clicks on an acupoint or meridian, generate an annotation next to the corresponding position of the acupoint or meridian on the model), and obtain an executable medical strategy for the application case based on the adjusted diagnosis and treatment plan and the rendered interactive three-dimensional human model.

[0065] In this embodiment, a knowledge graph construction module is built, and data required for constructing the knowledge graph is collected based on three data collection units therein. Meanwhile, high-precision preprocessing is performed to obtain complete and accurate traditional Chinese medicine acupuncture data for constructing the knowledge graph. Given that there are still some problems with the knowledge graph even though it is constructed from preprocessed data, an optimization network model is selected to optimize the knowledge graph. At the same time, considering that the knowledge graph needs to be updated, an incremental learning framework is built to update the knowledge graph, resulting in a real-time updated knowledge graph. To achieve knowledge visualization, a three-dimensional human body model is generated based on the real-time updated knowledge graph, and an interactive mechanism is added. Finally, a medical strategy generation module is added, which can generate executable medical strategies for the input application cases and display them on the interactive three-dimensional human body model, facilitating teaching. Compared with the existing field, data with a wider coverage range is collected and higher-precision preprocessing is performed. At the same time, the constructed knowledge graph is optimized and updated in real time. To facilitate students' learning of knowledge, an interactive mechanism is added to the constructed three-dimensional human body model. The system composed of the constructed knowledge graph and the interactive three-dimensional human body model can also output executable medical strategies for the input application cases and display them on the model, enhancing practicality and facilitating the display of visual knowledge.

[0066] Embodiment 2

[0067] Please refer to Figure 2 As shown, for the parts not described in detail in this embodiment, refer to the description content of Embodiment 1. A method for visual display of traditional Chinese medicine acupuncture knowledge based on a knowledge graph is provided, including:

[0068] S1. Collect traditional Chinese medicine acupuncture data and preprocess the traditional Chinese medicine acupuncture data to obtain complete traditional Chinese medicine acupuncture data; construct a knowledge graph based on the complete traditional Chinese medicine acupuncture data;

[0069] S2. Construct an optimization network model to optimize the knowledge graph to obtain a high-quality knowledge graph; construct an incremental learning framework and update the high-quality knowledge graph based on the incremental learning framework, and output a real-time updated knowledge graph;

[0070] S3. Construct an interactive three-dimensional human body model based on the real-time updated knowledge graph;

[0071] S4. Output executable medical strategies for application cases based on the real-time updated knowledge graph and the interactive three-dimensional human body model, and send the executable medical strategies to the student terminal; each module is connected by wired and / or wireless means.

[0072] Embodiment 3

[0073] This embodiment discloses an electronic device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the operation mode of the above-provided method for visual display of traditional Chinese medicine acupuncture knowledge based on a knowledge graph.

[0074] Since the electronic device introduced in this embodiment is the electronic device used to implement the method for visual display of traditional Chinese medicine acupuncture knowledge based on a knowledge graph in the embodiments of the present application, based on the method for visual display of traditional Chinese medicine acupuncture knowledge based on a knowledge graph introduced in the embodiments of the present application, those skilled in the art can understand the specific implementation manners and various variations of the electronic device in this embodiment. Therefore, the specific implementation of how this electronic device implements the method in the embodiments of the present application will not be described in detail here. As long as those skilled in the art implement the electronic device used in the method for visual display of traditional Chinese medicine acupuncture knowledge based on a knowledge graph in the embodiments of the present application, it falls within the scope of protection of the present application.

[0075] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain a formula that is closest to the actual situation. The preset parameters and threshold selection in the formulas are set by those skilled in the art according to the actual situation.

[0076] The above are only the preferred embodiments of the present invention. The protection scope of the present invention is not limited to the above embodiments. Any technical solutions falling within the concept of the present invention belong to the protection scope of the present invention. It should be noted that for ordinary technical users in the technical field, several improvements and refinements made without departing from the principle of the present invention should also be regarded as within the protection scope of the present invention.

Claims

1. A visualization display system for traditional Chinese medicine acupuncture knowledge based on a knowledge graph, characterized in that, Including: A knowledge graph construction module, including a traditional Chinese medicine ancient book data acquisition unit, a clinical case data acquisition unit, and an ultrasound image data acquisition unit; Integrate the data of each acquisition unit to obtain traditional Chinese medicine acupuncture data and preprocess the traditional Chinese medicine acupuncture data to obtain improved traditional Chinese medicine acupuncture data; construct a knowledge graph based on the improved traditional Chinese medicine acupuncture data; A knowledge graph optimization module, construct an optimization network model to optimize the knowledge graph to obtain a high-quality knowledge graph; construct an incremental learning framework and update the high-quality knowledge graph based on the incremental learning framework, and output a real-time updated knowledge graph; A visualization module, construct an interactive three-dimensional human body model based on the real-time updated knowledge graph; A medical strategy generation module, output an executable medical strategy for the application case based on the real-time updated knowledge graph and the interactive three-dimensional human body model, and send the executable medical strategy to the student terminal.

2. The visualization display system of traditional Chinese medicine acupuncture knowledge based on a knowledge graph according to claim 1, wherein, The traditional Chinese medicine ancient book data acquisition unit extracts the description fields related to traditional Chinese medicine acupuncture knowledge, marks the traditional Chinese characters and variant Chinese characters in the description fields related to traditional Chinese medicine acupuncture knowledge, and constructs a mapping dictionary for the marked traditional Chinese characters and variant Chinese characters in the description fields related to traditional Chinese medicine acupuncture knowledge; Convert the marked traditional Chinese characters and variant Chinese characters into simplified Chinese based on the mapping dictionary to obtain traditional Chinese medicine ancient book data; the clinical case data acquisition unit collects acupuncture diagnosis and treatment records by docking with the hospital's case database, and performs word segmentation on each acupuncture diagnosis and treatment record to obtain clinical case data; the ultrasound image data acquisition unit obtains ultrasound image data by collecting the ultrasound images related to acupuncture treatment in the hospital's case database.

3. The visualization display system of traditional Chinese medicine acupuncture knowledge based on a knowledge graph according to claim 2, characterized in that, The methods for preprocessing the traditional Chinese medicine acupuncture data include: Construct a human body coordinate system with the lower edge of the human patella as the origin, map the traditional terms in the traditional Chinese medicine ancient book data to the human body coordinate system to obtain coordinate-transformed ancient book data; construct a semantic model to perform semantic disambiguation on the coordinate-transformed ancient book data to obtain unambiguous ancient book data; remove the redundant data in the clinical case data to obtain refined case data; perform denoising processing on the ultrasound image data to obtain high-quality ultrasound image data; combine the unambiguous ancient book data, refined case data, and high-quality ultrasound image data to obtain pre-cleaned traditional Chinese medicine acupuncture data; extract features from the pre-cleaned traditional Chinese medicine acupuncture data, and at the same time perform multi-modal feature fusion to obtain fusion features; set feature conflict resolution rules, and adjust the fusion features based on the feature conflict resolution rules to obtain fused multi-modal features; perform multi-tuple conversion on each fused multi-modal feature to generate structured fusion features, and integrate all structured fusion features to generate improved traditional Chinese medicine acupuncture data.

4. A visualization display system for traditional Chinese medicine acupuncture knowledge based on a knowledge graph according to claim 3, characterized in that, The methods for constructing a semantic model to perform semantic disambiguation on the coordinate-transformed ancient book data include: Taking the BERT-BiLSTM network model as the basic structure of the semantic model and adding an attention mechanism; the semantic model includes a BERT layer, a BiLSTM layer, an attention channel, and a decoding layer; using a word segmentation tool to perform sentence splitting and word segmentation on the coordinate-transformed ancient medical books data to obtain the sentence-split and word-segmented ancient medical books data; deploying a pre-trained language model in the field of traditional Chinese medicine knowledge in the BERT layer of the semantic model for global semantic modeling and transforming the sentence-split and word-segmented ancient medical books data into multi-dimensional word vectors and sentence vectors based on the pre-trained language model in the field of traditional Chinese medicine knowledge, and then using the bidirectional LSTM model of the BiLSTM layer to obtain the forward and backward semantic dependency relationships of any multi-dimensional word vector in any sentence vector; dividing the attention channel into a word-level attention branch and a sentence-level attention branch, calculating the weight of each word in the multi-dimensional word vector in the sentence vector based on the word-level attention branch; calculating the weight of any sentence vector in the original text based on the sentence-level attention branch; constructing a joint loss function and optimizing the semantic model based on the joint loss function; constructing a sentence-word association matrix based on the weight of each word in the multi-dimensional word vector and the weight of each sentence vector in the original text, and decoding through the decoding layer to obtain the acupuncture knowledge vector; integrating all acupuncture knowledge vectors to obtain the unambiguous ancient medical books data; The methods for removing redundant data in the clinical case data include: The clinical case data includes etiology, treatment acupoint plan, acupuncture depth, and curative effect evaluation; constructing a dictionary library, encoding the etiology based on the dictionary library to obtain the etiology code; representing the treatment acupoint plan using binary encoding and mapping the encoding to a vector space with a dimension equal to the number of acupoints to obtain the acupoint vector; discretizing the acupuncture depth to obtain the discrete acupuncture depth; quantifying the curative effect evaluation and representing it using a constant to obtain the quantified curative effect evaluation; combining the etiology code, acupoint vector, discrete acupuncture depth, and quantified curative effect evaluation into a clinical case matrix; initializing the number of cluster centers as K; treating each clinical case matrix as a sample point and randomly selecting K sample points from all sample points as cluster centers; constructing a weighted distance formula to calculate the distance between any two sample points, and calculating the silhouette coefficient based on the distance between any two sample points; continuously adjusting the number of cluster centers during the clustering process, while calculating and recording the silhouette coefficient, and taking the number of cluster centers corresponding to the maximum silhouette coefficient as the optimal number of clusters; grouping all sample points based on the optimal number of clusters, setting a distance threshold, and treating the sample points in each group with a distance less than the distance threshold from the cluster center as redundant points and removing them.

5. The visualization display system of traditional Chinese medicine acupuncture knowledge based on a knowledge graph according to claim 4, characterized in that, The methods for denoising the ultrasonic image data include: Perform low-pass filtering and high-pass filtering on the row direction of each ultrasound image in the ultrasound image data using wavelet multi-band decomposition to generate a horizontal low-frequency vector and a horizontal high-frequency vector respectively; then perform low-pass filtering and high-pass filtering on the horizontal low-frequency vector and the horizontal high-frequency vector in the column direction to generate an ultrasound low-frequency sub-band and an ultrasound high-frequency sub-band; introduce diffusion variance in the ultrasound high-frequency sub-band, and use Monte Carlo sampling to simulate the noise distribution and calculate the pixel gray diffusion variance of the ultrasound high-frequency sub-band at the same time; construct an adaptive threshold function, identify the noise region in the ultrasound high-frequency sub-band based on the adaptive threshold function and the pixel gray diffusion variance and perform noise suppression to obtain a preliminary denoised high-frequency sub-band; determine the acupoint coordinates in the ultrasound low-frequency sub-band, and construct a local window centered on the acupoint coordinates; set a gray threshold, and divide some sub-bands in the local window into a background region and a main region based on the gray threshold; perform histogram equalization on the main region of some sub-bands in the local window to generate a contrast-enhanced low-frequency sub-band; add a dual-branch enhancement mechanism, including a spatial domain branch and a frequency domain branch, and the dual-branches are processed in parallel to perform dual-branch enhancement on the preliminary denoised high-frequency sub-band and the contrast-enhanced low-frequency sub-band; in the spatial domain branch, adaptively adjust the shape of the convolution kernel using dynamic deformable convolution and perform convolution operations on the preliminary denoised high-frequency sub-band and the contrast-enhanced low-frequency sub-band using the adjusted convolution kernel; perform weighted fusion on the frequency domain branch and the spatial domain branch; in the frequency domain branch, reconstruct the signal of the preliminary denoised high-frequency sub-band through inverse wavelet transform, and output the denoised ultrasound image after dual-branch parallel processing; use the local dynamic enhancement algorithm to enhance the denoised ultrasound image to obtain high-quality ultrasound image data.

6. The visualization display system of traditional Chinese medicine acupuncture knowledge based on a knowledge graph according to claim 5, wherein The method for feature extraction of the cleaned traditional Chinese medicine acupuncture data includes: Define the triple of ancient Chinese medical texts, which includes the object set, the attribute set, and the binary relation; extract the object set features, the attribute set features, and the binary relation features of each clause in the unambiguous ancient Chinese medical data, and form the triple of ancient Chinese medical texts for any clause; traverse each clause in the unambiguous ancient Chinese medical data, so that the triple of ancient Chinese medical texts for any clause can form a mapping with other clauses, thereby expanding the triple of ancient Chinese medical texts for any clause into a multi-dimensional association matrix; construct a semantic framework and add the relationships unique to traditional Chinese medicine to the semantic framework; perform semantic expansion on the multi-dimensional association matrix based on the semantic framework to obtain the semantic feature vector of acupuncture; traverse each piece of data in the refined case data, divide the refined case data into two categories based on the efficacy evaluation, including normal cases and boundary cases, calculate the efficacy score for each case based on the efficacy evaluation of each normal case and boundary case, and add a scoring item to each case based on the efficacy score to obtain the scored case feature vector; divide each high-quality ultrasound image in the high-quality ultrasound image data into adjacent pixel blocks of a preset size; compare the gray value of the central pixel in any pixel block with the gray values of other pixels, and perform binary encoding on the pixel block based on the gray values of the central pixel and other pixels to obtain the binary sequence corresponding to each pixel block; perform decimal conversion on the binary sequence corresponding to each pixel block to obtain the encoding value of each pixel block; count the distribution of the encoding values of all pixel blocks in each high-quality ultrasound image to form the ultrasound image vector.

7. The visualization display system of traditional Chinese medicine acupuncture knowledge based on a knowledge graph according to claim 6, characterized in that, The method for optimizing the knowledge graph by constructing an optimization network model includes: Use the relationship-aware network model specialized for traditional Chinese medicine knowledge as the basic structure of the optimization network model, and at the same time add an attention mechanism improved based on traditional Chinese medicine acupuncture knowledge to the optimization network model; input the knowledge graph into the optimization network model in the form of multi-tuples; extract the relationships unique to traditional Chinese medicine in each multi-tuple, and construct a constraint matrix based on the relationships unique to traditional Chinese medicine; construct a meridian attribution vector and map the meridian attribution vector to the definitions in the existing materials based on the existing materials; construct a time encoding based on the time features in the constraint matrix, and perform a three-dimensional Fourier transform on the time encoding to generate a three-dimensional time series matrix; calculate the attention weights based on the meridian attribution vector and the three-dimensional time series matrix, and add implicit relationships to the multi-tuples based on the attention weights; construct a dynamic topological structure, and use the Poincaré ball model as the core structure of the dynamic topological structure; regard each multi-tuple after adding implicit relationships as a node, and calculate the hyperbolic distance between any two nodes based on the dynamic topological structure; perform hierarchical adjustment on the knowledge graph based on the hyperbolic distance to obtain a high-quality knowledge graph.

8. A visualization display system for traditional Chinese medicine acupuncture knowledge based on a knowledge graph according to claim 7, characterized in that, The method for constructing an incremental learning framework and updating the high-quality knowledge graph based on the incremental learning framework includes: In the incremental learning framework, a heterogeneous data channel is constructed to connect to the hospital's case database to collect new case data and new ultrasound images, and new ancient book data is obtained from the preset traditional Chinese medicine ancient book database. The new case data, new ultrasound images, and new ancient book data are combined into incremental data. An incremental data buffer is constructed, and the incremental data is stored in the incremental data buffer. At the same time, a buffer threshold is set. When the data volume of the incremental data is greater than the buffer threshold, incremental training is performed on the high-quality knowledge graph. A modular processing component library is constructed. In the modular processing component library, the BERT model specialized in the field of traditional Chinese medicine is used to perform entity recognition and relationship extraction on the ancient book text type data and case text type data in the incremental data. The UNet++ network model is used to accurately locate the acupoints corresponding to the image type data in the incremental data. The structured data in the incremental data is automatically converted into a multi-tuple format. At the same time, the incremental data is subjected to the same preprocessing process as the original knowledge graph to obtain preprocessed incremental data. The confidence of the preprocessed incremental data is calculated, and incremental training is implemented based on the confidence using a hierarchical update strategy. When the confidence of the preprocessed incremental data is greater than the preset confidence threshold, hot update is performed; otherwise, cold update is performed. The hot update method includes: constructing a dynamic loss function and calculating the parameter adjustment amount based on the dynamic loss function. The cold update method includes: constructing a version optimization function and performing full-scale iteration or rollback on the high-quality knowledge graph based on the version optimization function.

9. The visualization display system of traditional Chinese medicine acupuncture knowledge based on a knowledge graph according to claim 8, characterized in that, The method for constructing an interactive three-dimensional human body model based on the real-time updated knowledge graph includes: Constructing an anatomical coordinate system; introducing traditional Chinese medicine topological parameters and constructing a three-dimensional basic human body grid model based on the traditional Chinese medicine topological parameters and the anatomical coordinate system; constructing a dynamic mapping function, and accurately associating the acupoints and meridians in the real-time updated knowledge graph with the coordinates in the three-dimensional basic human body grid model based on the dynamic mapping function; setting a real-time incremental rendering mechanism, calculating the difference degree between different versions of the real-time updated knowledge graph, and constructing a rendering priority function based on the difference degree; dynamically adjusting the rendering level of the three-dimensional basic human body grid model based on the rendering priority function; constructing a multi-physical field coupling equation to quantitatively process the response effects of each acupoint in the three-dimensional basic human body grid model, and adding a display effect on the three-dimensional basic human body grid model; defining a gesture feature vector and constructing a gesture space based on the gesture feature vector; associating the gesture space with the three-dimensional basic human body grid model, and defining the operation instructions corresponding to any gesture feature vector in the gesture space; constructing a labeling target optimization function, adjusting the labeling layout in the three-dimensional basic human body grid model based on the labeling target optimization function, and adding gesture-labeling linkage feedback to obtain an interactive three-dimensional human body model.

10. A visualization display system for traditional Chinese medicine acupuncture knowledge based on a knowledge graph according to claim 9, characterized in that, The method for outputting an executable medical strategy for an applied case based on the real-time updated knowledge graph and the interactive three-dimensional human body model includes: The application cases include text-type case data, patient physical sign case data, and patient basic information. The BERT-BiLSTM network model is used to process the text-type case data to extract keyword vectors related to the disease. The patient physical sign case data includes tongue image and pulse waveform. The UNet++ network model is used to perform image segmentation on the tongue image to extract the HSV color features of the tongue body, tongue coating, and crack density. The wavelet transform algorithm is used to perform wavelet packet decomposition on the pulse waveform to extract pulse features. The features of the tongue image and the pulse waveform are mapped to the acupuncture knowledge domain, and disease information data is output. The keyword vectors and disease information data are matched with the real-time updated knowledge graph. The vector-type data composed of the keyword vectors and disease information data is regarded as the disease node, and each multi-tuple in the real-time updated knowledge graph is regarded as the diagnosis and treatment node. The disease similarity between the disease node and any diagnosis and treatment node is calculated. When the disease similarity is greater than the preset similarity threshold, the diagnosis and treatment plan represented by the relevant diagnosis and treatment node is output. The actual coordinates of the acupoints on the interactive three-dimensional human model are dynamically adjusted based on the size of the interactive three-dimensional human model and the patient basic information. A tissue deformation compensation function is constructed, and the diagnosis and treatment plan is adjusted based on the tissue deformation compensation function. The interactive three-dimensional human model is rendered at multiple levels and dynamically parameter-labeled based on the adjusted diagnosis and treatment plan, and an executable medical strategy for the application case is obtained based on the adjusted diagnosis and treatment plan and the rendered interactive three-dimensional human model.

Images