An intelligent diagnosis-based acupuncture diagnosis and treatment system

Abstract

The application discloses an acupuncture system based on intelligent diagnosis, which comprises an intelligent diagnosis module, an acupoint auxiliary positioning module and an acupuncture treatment module. The intelligent diagnosis module collects multi-source diagnosis and treatment information and obtains an acupuncture diagnosis and treatment scheme of a patient. The multi-source diagnosis and treatment information comprises physiological data, image data and text data. The acupoint auxiliary positioning module collects acupuncture positioning information according to the acupuncture diagnosis and treatment scheme of the patient and outputs the acupuncture acupoint position of the patient. The acupuncture positioning information comprises an acupoint image and acupoint data. The acupoint image is the corresponding acupuncture acupoint position in the acupuncture diagnosis and treatment scheme of the patient. The acupoint data is the resistance value of the corresponding acupuncture acupoint position in the acupoint image. The acupuncture treatment module is used for positioning the acupuncture needle at the acupuncture acupoint position of the patient and performing acupuncture treatment on the patient. The system utilizes advanced technologies such as machine learning and natural language processing to perform deep analysis and feature fusion on multi-modal data and realizes accurate judgment of disease modes.

Description

Technical Field

[0001] This invention relates to the field of intelligent acupuncture technology, and in particular to an acupuncture system based on intelligent diagnosis. Background Technology

[0002] With the continuous advancement of medical technology, the demand for Traditional Chinese Medicine (TCM) acupuncture is increasing. TCM acupuncture has advantages such as wide applicability, rapid effectiveness, and few side effects. However, it also has disadvantages such as strong stimulation, the risk of cross-infection from skin puncture, high technical difficulty in operation, and the need for the practitioner's experience to determine efficacy, making it difficult to replicate. Statistics show that acupuncture is currently used in 196 countries and regions worldwide, becoming an important part of global medicine. Against this backdrop, the integration of artificial intelligence (AI) with TCM acupuncture diagnosis and treatment has opened new avenues for personalized acupuncture medicine, allowing every patient to receive more precise and effective treatment.

[0003] A patented invention, CN202122567713.4, describes an acupuncture robot with an intelligent acupuncture system. This system comprises an acupuncture treatment host, treatment electrodes, and a palm-fiber fixation frame. The palm-fiber fixation frame stabilizes the patient's hand, and the treatment electrodes deliver acupuncture and massage therapy at holographic acupuncture points on the second metacarpal bone. The treatment host receives operating instructions from a server and guides the patient to correctly use the acupuncture robot to achieve standard therapeutic effects. However, this approach relies on pre-set programs for treatment, neglecting individual differences in physiological and pathological conditions and failing to adequately consider the patient's immediate physiological responses during treatment.

[0004] For example, the invention patents for acupuncture robots and intelligent acupuncture systems (patent numbers JP2004357813A and KR200472501Y1) disclose an acupuncture robot and its supporting intelligent acupuncture system, including an acupuncture treatment host, treatment electrodes, and a hand fixation frame. The hand fixation frame is used to stabilize the hand, while the treatment electrodes are used to perform acupuncture and massage therapy. The robot receives operating instructions through a server to ensure that the patient receives a suitable treatment plan during use. Furthermore, the intelligent acupuncture system and method can provide personalized acupuncture and massage therapy for different patients, and has remote assisted treatment capabilities. It can also upload treatment effect data, facilitating doctors to better serve patients.

[0005] The aforementioned existing technologies offer a range of solutions for acupuncture robots, including actuators for stabilizing needles, assisting in needle insertion, and providing treatment. However, these technologies primarily focus on improving the convenience and accuracy of acupuncture procedures, lacking corresponding intelligent diagnostic components and self-service capabilities in emergency medical situations.

[0006] To improve the level of intelligent diagnosis and treatment and provide patients with targeted and personalized treatment plans, a multi-dimensional Tibetan medicine AI intelligent auxiliary decision-making device and system (patent number: 2022110891257) has been invented. The device discloses a multi-dimensional Tibetan medicine AI intelligent auxiliary decision-making device including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it performs the following steps: collecting tongue image information, pulse image information, electronic questionnaire information, and urine examination information; inputting the above information into a weighted model and outputting diagnostic elements and their contribution; inputting the diagnostic elements and their contribution into an auxiliary decision-making model and outputting a prediction result for the disease. While the system of this invention can be integrated into a single unit for portability, its intelligent diagnostic scheme is too one-sided and cannot provide accurate treatment plans based on the patient's actual condition.

[0007] In summary, existing technologies offer a range of acupuncture robot solutions, primarily focusing on improving the convenience and accuracy of acupuncture procedures. However, they lack corresponding intelligent diagnostic components and self-service capabilities in emergency medical situations. Furthermore, existing intelligent diagnostic algorithms rely on relatively limited data sources, resulting in low algorithm accuracy and an inability to provide precise treatment plans based on the patient's actual condition. Summary of the Invention

[0008] The purpose of this invention is to address the shortcomings of existing technologies by providing an acupuncture system based on intelligent diagnosis. This method integrates multimodal data analysis technology, combines the principles of syndrome differentiation and treatment in traditional Chinese medicine, and utilizes advanced data analysis methods to conduct in-depth analysis of multi-source data from patients, thereby providing patients with precise and personalized acupuncture treatment plans.

[0009] This application provides an acupuncture system based on intelligent diagnosis, including: an intelligent diagnosis module, an acupoint auxiliary positioning module, and an acupuncture treatment module;

[0010] The intelligent diagnostic module collects multi-source diagnostic and treatment information and inputs it into the intelligent diagnostic model. The intelligent diagnostic model analyzes the multi-source diagnostic and treatment information to obtain the patient's acupuncture treatment plan.

[0011] The multi-source diagnostic information includes: physiological data, image data, and text data; the physiological data refers to the patient's physiological parameters acquired by sensors; the image data includes photographs of the patient's lesions or facial images; and the text data is a description of the patient's symptoms.

[0012] The acupoint auxiliary positioning module collects acupuncture positioning information according to the patient's acupuncture treatment plan, inputs the acupuncture positioning information into the acupoint auxiliary positioning model, and outputs the location of the acupuncture points for the patient.

[0013] The acupuncture positioning information includes acupoint images and acupoint data; the acupoint images are the acupuncture acupoint locations corresponding to the patient's acupuncture treatment plan captured by a camera; the acupoint data are the resistance values ​​of the acupuncture acupoint locations corresponding to the acupuncture acupoint locations in the acupoint images obtained by an electric acupoint detector.

[0014] The acupuncture treatment module is used to position the acupuncture needles at the acupuncture points on the patient's body to perform acupuncture treatment.

[0015] Furthermore, the specific steps for the intelligent diagnostic module to obtain the patient's acupuncture treatment plan include:

[0016] S11: Data collection, gathering multi-source medical information about patients;

[0017] S12: Data preprocessing, preprocessing the patient's multi-source diagnosis and treatment information to obtain the preprocessed multi-source diagnosis and treatment information of the patient;

[0018] S13: Potential disease pattern recognition, which inputs the pre-processed multi-source diagnosis and treatment information of the patient into the intelligent diagnostic model and outputs the patient's potential disease pattern.

[0019] S14: Acquisition of acupuncture treatment plans. Combining the principles of TCM syndrome differentiation and treatment, the patient's potential disease pattern is mapped to TCM syndrome, and the patient's acupuncture treatment plan is obtained based on the patient's TCM syndrome.

[0020] Furthermore, the preprocessing of the multi-source diagnostic information includes preprocessing of physiological data, preprocessing of image data, and preprocessing of text data;

[0021] The steps for preprocessing physiological data include at least one of normalization, noise removal, and missing value processing.

[0022] The steps of image data preprocessing include at least one of resampling, normalization, data augmentation, and conversion to grayscale.

[0023] The steps for preprocessing text data include vectorization, and also include at least one of word segmentation, stop word removal, stemming, and word form restoration.

[0024] Furthermore, the intelligent diagnostic model includes a feature extraction module I, a feature fusion module I, and an output module I;

[0025] The feature extraction module I includes three different neural network architectures, which are used to extract features from physiological data, image data, and text data respectively, to obtain corresponding physiological features, image features, and text features.

[0026] The feature fusion module I includes a fusion layer that fuses physiological features, image features, and text features to obtain diagnostic fusion features;

[0027] The output module I includes one or more fully connected layers. Diagnostic fusion features are input into one or more fully connected layers for further information synthesis and output of potential disease patterns.

[0028] Furthermore, recurrent neural networks, long short-term memory networks, or gated recurrent units are used to process the physiological data, convolutional neural networks are used to process the image data, and recurrent neural networks are used to process the text data.

[0029] Furthermore, the feature fusion module I acquires diagnostic fusion features based on a multimodal attention mechanism, and the specific steps include:

[0030] First, we define three attention networks: physiological feature attention network, image feature attention network, and text feature attention network.

[0031] Secondly, based on the physiological feature attention network, image feature attention network, and text feature attention network, physiological feature modal weights, image feature modal weights, and text feature modal weights are obtained respectively.

[0032] α phy =σ(W phy f phy +b phy )

[0033] α img =σ(w ing f img +b img )

[0034] α txt =σ(w txt f txt +b txt )

[0035] Where: σ is the Sigmoid function, α phy f represents the modal weights for physiological features. phy As a physiological characteristic, w phy b phy The learning parameters of the attention network are physiological characteristics; α img f represents the image feature modal weights. img For image features, w img b img These are the learning parameters for the image feature attention network; a txt f represents the modal weights of the text features. txt For text features, w txt b txt These are the learning parameters for the text feature attention network;

[0036] Finally, diagnostic fusion features are obtained based on physiological features, image features, text features, and the corresponding physiological feature modality weights, image feature modality weights, and text feature modality weights.

[0037] f fuse =α phy f phy +α img f img +α txt f txt

[0038] Where: f fuse To diagnose fusion features.

[0039] Furthermore, the training steps of the intelligent diagnostic model include:

[0040] Step 1: Collect the training dataset for the intelligent diagnostic model. The training dataset for the intelligent diagnostic model includes multi-source diagnostic and treatment information of patients of different genders, ages, and diseases.

[0041] The intelligent diagnostic model training dataset is divided into an intelligent diagnostic model training set and an intelligent diagnostic model validation set.

[0042] Step 2: Preprocess the multi-source diagnostic and treatment information in the training set of the intelligent diagnostic model;

[0043] Step 3: Define the loss function of the intelligent diagnostic model and train the network of the intelligent diagnostic model using the training set of the intelligent diagnostic model;

[0044] Step 4: Evaluate the network of the intelligent diagnostic model using the intelligent diagnostic model validation set, and adjust the network architecture or training strategy of the intelligent diagnostic model based on the evaluation results.

[0045] Furthermore, the specific steps for the acupoint-assisted positioning module to obtain the location of the acupoints for acupuncture treatment include:

[0046] S21: Data collection, collecting the patient's acupuncture location information;

[0047] S22: Input the patient's acupuncture location information into the acupoint-assisted location model, and output the patient's acupuncture point location through feature extraction, feature fusion, and acupoint prediction.

[0048] Furthermore, the acupoint-assisted positioning model includes a feature extraction module II, a feature fusion module II, and an output module II;

[0049] The feature extraction module II obtains acupoint image features and acupoint data features from the acupoint image and acupoint data, respectively;

[0050] The feature fusion module II fuses the acupoint image features and acupoint data features to obtain acupoint fusion features;

[0051] The output module II outputs the location of the acupuncture points for the patient based on the acupuncture point fusion characteristics.

[0052] Furthermore, the acupoint image feature is the image feature vector corresponding to the acupoint image;

[0053] The acupoint data feature is the resistance feature vector corresponding to the acupoint data;

[0054] Furthermore, the feature fusion module II concatenates the image feature vector and the resistance feature vector to obtain the acupoint fusion feature through feature-level connection;

[0055] f fxw =[f i ,f r ]

[0056] Where: f fxw Acupoint fusion characteristics; f i f is the image feature vector; r For resistance eigenvectors; [f i ,f r The symbol ] represents the concatenation of image feature vectors and resistance feature vectors.

[0057] Furthermore, the feature fusion module II learns and acquires acupoint fusion features through a self-attention mechanism or a gating unit;

[0058] f fxw =ReLU(w xw [f i f r ]+b xw )

[0059] Where: f fxw Acupoint fusion characteristics; f i f is the image feature vector; r w is the resistance eigenvector; xw and b xw These are the weights and biases learned through self-attention mechanisms or gating units, respectively; ReLU is a non-linear activation function.

[0060] Furthermore, it also includes at least one of the following: a feedback and optimization module, a teaching video and technique explanation generation module, and an acupuncture first aid module;

[0061] The feedback and optimization module is used to adjust and optimize the acupuncture treatment plan based on patient feedback and treatment results after the acupuncture treatment module has completed the acupuncture treatment.

[0062] The teaching video and technique explanation generation module is used to automatically generate corresponding teaching videos or technique explanation videos based on the patient's acupuncture treatment plan.

[0063] The acupuncture emergency module monitors the patient's physiological state in real time based on multi-source diagnostic and treatment information, and activates emergency mode when it is determined that the patient needs emergency assistance.

[0064] The beneficial effects of this invention are as follows:

[0065] (1) Utilizing advanced technologies such as machine learning and natural language processing, this solution performs in-depth analysis and feature fusion of multimodal data to achieve accurate judgment of disease patterns. Compared with existing technologies, this fusion analysis can process and understand more complex health data, making diagnostic and treatment recommendations more refined and accurate.

[0066] (2) This approach not only employs modern medical monitoring and analysis techniques but also incorporates the traditional Chinese medicine concept of syndrome differentiation and treatment. This ensures that the treatment recommendations meet the objectivity requirements of modern medical data analysis while also respecting the individualized and holistic treatment characteristics of traditional Chinese medicine. This combination provides a more comprehensive treatment perspective and enhances the applicability of the treatment plan.

[0067] (3) A high-precision acupoint-assisted positioning system, which combines resistance value and image analysis to achieve accurate acupoint positioning.

[0068] (4) The acupuncture emergency module takes into account emergency response measures, which can significantly improve the system's ability to handle emergency medical situations, thereby providing timely and effective emergency assistance to patients in urgent need. This will greatly enhance the system's practical value and social benefits, filling the gap in rapid response and intelligent processing of emergency medical equipment. Attached Figure Description

[0069] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0070] Figure 1 This is a block diagram of an acupuncture system based on intelligent diagnosis.

[0071] Figure 2 A flowchart for the intelligent diagnostic module to obtain the patient's acupuncture treatment plan.

[0072] Figure 3 This is a flowchart of the training process for the intelligent diagnostic module.

[0073] Figure 4 This is a flowchart illustrating how the position-assisted positioning module obtains the location of acupuncture points for a patient. Detailed Implementation

[0074] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0075] like Figure 1 As shown, this embodiment of the invention provides an acupuncture system based on intelligent diagnosis, including: an intelligent diagnosis module, an acupoint auxiliary positioning module, and an acupuncture treatment module.

[0076] The intelligent diagnostic module collects multi-source diagnostic and treatment information and inputs it into the intelligent diagnostic model. The intelligent diagnostic model analyzes the multi-source diagnostic and treatment information to obtain the patient's acupuncture treatment plan.

[0077] Multi-source diagnostic information includes: physiological data, image data, and text data; physiological data refers to the patient's physiological parameters acquired by sensors; image data includes photos of the patient's lesions or facial images; and text data consists of descriptions of the patient's symptoms.

[0078] This embodiment integrates multiple sensors, such as heart rate monitors, blood pressure monitors, pulse oximeters, and skin resistance meters, which can comprehensively monitor the patient's physiological condition, while also combining visual images, medical images, and self-reported symptoms.

[0079] The acupoint-assisted positioning module collects acupuncture positioning information based on the patient's acupuncture treatment plan, inputs the acupuncture positioning information into the acupoint-assisted positioning model, and outputs the location of the acupuncture points for the patient.

[0080] Acupuncture location information includes acupoint images and acupoint data; the acupoint images are the acupoint locations corresponding to the acupuncture treatment plan of the patient, which are captured by a camera; the acupoint data are the resistance values ​​of the acupoint locations corresponding to the acupuncture points in the acupoint images obtained by an electroacupuncture detector.

[0081] Specifically, the acupoint-assisted positioning module based on image and resistance measurement aims to help users accurately locate precise acupuncture points on a patient's body. The system includes a device equipped with a high-resolution camera and a skin-contact resistance measuring instrument. Users use this system to capture images of the patient's body corresponding to the acupuncture points in their acupuncture treatment plan and perform resistance measurements on the same area. Multiple measurements can be taken at several different points to construct an resistance map.

[0082] The acupuncture treatment module is used to position acupuncture needles at the acupuncture points on the patient's body to perform acupuncture treatment.

[0083] This embodiment provides an acupuncture system based on intelligent diagnosis. This system can process and interpret multi-source data from patients, such as physiological measurements (e.g., heart rate, blood pressure, and skin electrical activity), visual images, medical images, and self-reported symptoms. Through the analysis of multi-source diagnostic information, it utilizes advanced data analysis methods to conduct in-depth research on the patient's specific health condition and provide precise acupuncture treatment plans. Simultaneously, by collecting images of the patient's acupoints and their resistance values, it provides precise acupuncture needle placement locations, achieving precise acupuncture treatment.

[0084] like Figure 2 As shown, the specific steps for the intelligent diagnosis module to obtain the patient's acupuncture treatment plan in this embodiment include:

[0085] S11: Data collection, gathering multi-source medical information about patients;

[0086] S12: Data preprocessing, preprocessing the patient's multi-source diagnosis and treatment information to obtain the preprocessed multi-source diagnosis and treatment information of the patient;

[0087] S13: Potential disease pattern recognition, which inputs the pre-processed multi-source diagnosis and treatment information of the patient into the intelligent diagnostic model and outputs the patient's potential disease pattern.

[0088] S14: Acquisition of acupuncture treatment plans. Combining the principles of TCM syndrome differentiation and treatment, the patient's potential disease pattern is mapped to TCM syndrome, and the patient's acupuncture treatment plan is obtained based on the patient's TCM syndrome.

[0089] The intelligent diagnostic module in this embodiment can analyze physiological parameters such as heart rate, blood pressure, blood oxygen and skin resistance based on the patient's multi-source medical information, and can comprehensively monitor the patient's physiological condition. At the same time, it can build a personalized health profile for the patient by combining images such as lesion photos, facial features and symptom descriptions.

[0090] Furthermore, the module's AI processing unit not only provides diagnostic suggestions based on preliminary data analysis, but also provides real-time feedback and adjusts recommended treatment plans to ensure the timeliness and accuracy of medical decisions. By comprehensively analyzing patient feedback and treatment outcomes, it learns and optimizes its diagnostic model, continuously improving the personalization and precision of treatment plans.

[0091] In order for the intelligent diagnostic module to accurately output personalized acupuncture treatment plans, the multi-source diagnostic and treatment information of the patients collected in S12 needs to be preprocessed. This involves screening, cleaning, and standardizing the collected multi-source data to facilitate subsequent analysis.

[0092] In this embodiment, the preprocessing of multi-source diagnostic and treatment information includes preprocessing of physiological data, image data, and text data;

[0093] The steps for preprocessing physiological data include at least one of normalization, noise removal, and missing value handling;

[0094] Image data preprocessing steps include at least one of resampling, normalization, data augmentation, and conversion to grayscale.

[0095] The steps for preprocessing text data include vectorization, as well as at least one of word segmentation, stop word removal, stemming, and lemmatization.

[0096] In this embodiment, physiological data typically includes physiological indicators such as heart rate, blood pressure, and blood glucose. The preprocessing steps for this physiological data include:

[0097] Normalization: Adjusting the data size to a standard range. Common methods include max-min normalization and Z-score normalization. This prevents certain features from having a disproportionate impact on model training due to their large numerical range.

[0098] Noise removal: Using filters or other techniques to remove noise from data. For example, moving average or median filters can be used to smooth time series data;

[0099] Missing value handling: Fill in or remove missing data. Filling methods can include using the mean, median, or more complex interpolation methods.

[0100] Image data is frequently used in medical diagnosis to process images such as X-rays, CT scans, or MRI scans. Image data preprocessing steps typically include:

[0101] Resampling: Adjusting all images to the same resolution and size to eliminate differences between devices;

[0102] Normalization: Pixel values ​​are typically normalized to between 0 and 1, and sometimes standardized to have a mean of 0 and a unit variance;

[0103] Data augmentation: Increasing data diversity by rotating, scaling, flipping, or cropping images helps improve the generalization ability of models.

[0104] Convert to grayscale: For analyses that do not require color information, converting the image to grayscale can reduce computational load.

[0105] In intelligent diagnosis, text data can originate from medical record summaries, doctors' diagnostic reports, etc. Text data preprocessing steps typically include:

[0106] Word segmentation: Dividing continuous text data into lexical units, especially for languages such as Chinese or Japanese that do not use spaces to separate words;

[0107] Stop word removal: Deleting common but low-semantic-value words, such as "de", "shi", etc.;

[0108] Stemming or lemmatization: Reducing words to their base forms to reduce the total number of words that the model needs to process;

[0109] Vectorization: Converting text into a numerical form for processing by machine learning models, and common methods include bag-of-words model, TF-IDF, Word2Vec, or pre-trained models such as BERT.

[0110] The preprocessed multi-source diagnosis and treatment information of the patient can be further input into the intelligent diagnosis model for multi-dimensional analysis. The intelligent diagnosis model is a core component of this system, which relies on deep learning, machine learning or other advanced algorithms to analyze the multi-source diagnosis and treatment information and output potential disease patterns. This process involves extracting patterns or trends from a large amount of data and converting such information into usable medical decision support.

[0111] In this embodiment, the intelligent diagnosis model includes a feature extraction module I, a feature fusion module I, and an output module I;

[0112] The feature extraction module I includes 3 different neural network architectures, which are respectively used to perform feature extraction on physiological data, image data and text data to obtain corresponding physiological features, image features and text features;

[0113] The feature fusion module I includes a fusion layer, which fuses the physiological features, the image features and the text features to obtain diagnostic fusion features;

[0114] The output module I includes one or more fully connected layers, the diagnostic fusion features are input into the one or more fully connected layers for further information integration, and potential disease patterns are output.

[0115] Assume that the multi-source diagnosis and treatment information of a patient includes the following data:

[0116] Physiological data: heart rate 80bpm, blood pressure 130 / 85mmHg, body temperature 36.5°C;

[0117] Image data: the facial image shows slight swelling;

[0118] Text data: the patient describes persistent stomach pain and indigestion.

[0119] The intelligent diagnostic model analyzes this data to obtain the patient's potential disease patterns. Combining a medical knowledge base with the principles of syndrome differentiation and treatment in traditional Chinese medicine, it customizes treatment plans according to each patient's specific situation, improving treatment efficiency and patient satisfaction. The patient's acupuncture treatment plan includes acupuncture procedures and lifestyle suggestions, providing comprehensive health management for the patient.

[0120] Feature extraction module I extracts features from multiple sources. In this embodiment, the features are derived from the analysis of physiological data, image data, and text data. Different neural network architectures need to be selected for data of different modalities. Each modality is extracted separately through a deep learning model to capture the unique characteristics of the data of this modality.

[0121] In this embodiment, a recurrent neural network, a long short-term memory network, or a gated recursive unit is used to process physiological data, a convolutional neural network is used to process image data, and a recurrent neural network is used to process text data.

[0122] Physiological data (such as continuous physiological signals like heart rate, blood pressure, and body temperature) typically exhibit time-dependent and periodic characteristics. The ideal choice for processing this type of data is to use recurrent neural networks (RNNs) or their more advanced variants, such as long short-term memory networks (LSTMs) or gated recurrent units (GRUs). These network structures are particularly well-suited for processing time-series data because they can capture the temporal dynamics and long-term dependencies within the data.

[0123] In this embodiment, for the extraction of physiological features, a Long Short-Term Memory (LSTM) network is selected. Multiple LSTM layers are configured to extract features from physiological time-series data layer by layer, with each layer capturing different levels of temporal dependencies. LSTM is well-suited for processing and predicting important events with long intervals and delays in time series. Physiological data often contains key but infrequent physiological changes, and LSTM can effectively identify these patterns.

[0124] In another embodiment, a gated recurrent unit (GRU) is chosen. GRU is a variant of LSTM that simplifies the model structure, is generally more computationally efficient, and performs similarly to LSTM in some cases. GRU is used for some less complex time-series physiological data.

[0125] Key time-series features can be extracted from physiological data using LSTM or GRU networks. These features can reflect changes or trends in pathological states, such as arrhythmia and periodic fluctuations in blood pressure.

[0126] In the multimodal learning of intelligent diagnostic models, the feature fusion module I is a core module. This module analyzes data from different sources to extract physiological, image, and textual features, effectively integrating these multimodal features into a unified representation for deeper analysis and prediction. For complex fusion techniques, such as multimodal attention mechanisms, it can more intelligently determine the importance of each modality in the final output.

[0127] In this embodiment, feature fusion module I obtains diagnostic fusion features based on a multimodal attention mechanism, and the specific steps include:

[0128] First, we define three attention networks: physiological feature attention network, image feature attention network, and text feature attention network.

[0129] Secondly, based on the physiological feature attention network, image feature attention network, and text feature attention network, the physiological feature modal weights, image feature modal weights, and text feature modal weights are obtained respectively.

[0130] α phy =σ(w phy f phy +b phy )

[0131] α img =σ(w img f img +b img )

[0132] α txt =σ(w txt f txt +b txt )

[0133] Where: σ is the Sigmoid function, α phy f represents the modal weights for physiological features. phy As a physiological characteristic, w phy b phy The learning parameters of the attention network are physiological characteristics; α img f represents the image feature modal weights. img For image features, w img b img These are the learning parameters for the image feature attention network; a txt f represents the modal weights of the text features. txt For text features, w txt b txt These are the learning parameters for the text feature attention network;

[0134] Finally, diagnostic fusion features are obtained based on physiological features, image features, text features, and the corresponding physiological feature modality weights, image feature modality weights, and text feature modality weights.

[0135] f fuse =α phy f phy +α img f img +α txt f txt

[0136] Where: f fuse To diagnose fusion features.

[0137] In this embodiment, feature fusion module I does not simply merge features, but dynamically adjusts the contribution of each data modality based on task relevance. For example, for some diseases, image data may be more informative than text descriptions, and vice versa. The fusion mechanism in this embodiment can effectively improve the system's information utilization and the accuracy of the final decision, especially in the face of complex and unevenly distributed medical data scenarios.

[0138] Output module I is responsible for further processing the fused features and outputting the final diagnostic result. In this embodiment, the output task is to identify and classify the patient's potential disease patterns. Output module I includes one or more fully connected layers and ends with a softmax output layer. The diagnostic fusion features are input into one or more fully connected layers for further information synthesis. Through softmax, the predicted probabilities of different potential disease patterns are output.

[0139] Specifically, assuming we have N possible potential disease patterns, the structure of output module I can be described as follows:

[0140] (1) Fully connected layer:

[0141] It can contain one or more fully connected layers, each of which is usually followed by an activation function, such as ReLU, to introduce non-linearity and enhance the expressive power of the network.

[0142] Each layer can have a different number of neurons, which is usually determined based on the specific application and the complexity of the data.

[0143] (2) Dropout layer (optional):

[0144] Dropout layers may be added between fully connected layers to reduce model overfitting.

[0145] (3) Output layer:

[0146] The final layer is a fully connected layer with N neurons, matching the number of potential disease patterns.

[0147] Using the softmax activation function, the output can be transformed into a probability distribution, where the output of each neuron represents the probability of belonging to the corresponding potential disease pattern.

[0148] Let the output of the fully connected layer before the output layer be x. q If the weights and biases of the output layer are h and g, respectively, then the calculation of the output layer can be expressed as:

[0149] z i =hx q +g

[0150] Where: z i The output of the output layer for the i-th potential disease pattern, i = 1, 2, 3, 4...N;

[0151] Then, the probabilities p of the N potential disease patterns are obtained using the softmax function. i :

[0152]

[0153] Where: p i It is the probability that the model predicts that the input sample belongs to one of the i potential disease patterns.

[0154] If the model is designed to diagnose diseases including heart disease, diabetes, and hypertension, then N=3, and the final output layer of the model will have 3 output neurons, each corresponding to the predicted probability of a potential disease pattern.

[0155] With this design, the model can provide intuitive probability assessments, helping doctors better understand the diagnostic basis of the model and improving the reliability and transparency of diagnosis.

[0156] To improve the recognition accuracy of the intelligent diagnostic model, model training is crucial. In this embodiment, as shown... Figure 3 As shown, the training steps for the intelligent diagnostic model include:

[0157] Step 1: Collect training dataset for intelligent diagnostic model. The training dataset for intelligent diagnostic model includes multi-source diagnostic and treatment information of patients of different genders, ages, and diseases.

[0158] The training dataset for the intelligent diagnostic model is divided into the intelligent diagnostic model training set and the intelligent diagnostic model validation set.

[0159] Step 2: Preprocess the multi-source diagnostic and treatment information in the training set of the intelligent diagnostic model;

[0160] Step 3: Define the loss function of the intelligent diagnostic model and train the network of the intelligent diagnostic model using the training set of the intelligent diagnostic model;

[0161] Step 4: Evaluate the network of the intelligent diagnostic model using the intelligent diagnostic model validation set, and adjust the network architecture or training strategy of the intelligent diagnostic model based on the evaluation results.

[0162] The training dataset for the intelligent diagnostic model in the first step typically includes multi-source diagnostic information, covering patients of different genders, ages, and conditions. This ensures that the information learned by the model comprehensively covers various types of disease conditions and patient characteristics. Specifically, the training dataset for the intelligent diagnostic model may include, but is not limited to:

[0163] Data from patients of different genders: Ensure that the model can identify and process disease characteristics that may manifest differently in male and female patients.

[0164] Patient data across different age groups: The manifestations of the disease may vary with age; for example, the disease characteristics may differ significantly between children and the elderly.

[0165] Patient data for different conditions: including heart disease, diabetes, hypertension and other diseases, as well as other possible rare diseases, so that the model can handle a variety of clinical situations.

[0166] Multi-source diagnostic information: such as physiological parameters (heart rate, blood pressure, etc.), imaging data (X-ray, MRI, etc.), laboratory test results, and medical records.

[0167] The training dataset for the intelligent diagnostic model is divided into a training set and a validation set. The training set is used to train the network of the intelligent diagnostic model, while the validation set is used for performance evaluation during the model training process, helping to detect whether the model is overfitting and its performance on unseen data. The validation set should also include multi-source diagnostic information, and its sample composition should be similar to that of the training set but strictly separated to ensure that the model evaluation is fair and effective.

[0168] In another embodiment, the entire intelligent diagnostic model training dataset is divided into an intelligent diagnostic model training set, an intelligent diagnostic model validation set, and an intelligent diagnostic model test set. The distribution ratio of the intelligent diagnostic model training dataset is: approximately 70%-80% for the intelligent diagnostic model training set; approximately 20%-30% for the intelligent diagnostic model validation set; and approximately 10%-20% for the intelligent diagnostic model test set. It is important to ensure that the data is correctly divided and that the distribution among the groups is balanced. This is usually achieved through random sampling or stratified sampling to ensure that each subset can properly represent the characteristics of the overall data.

[0169] The second step, data preprocessing, is crucial in the training of the intelligent diagnostic model network because it directly affects the model's training effectiveness and final performance. Different types of data—such as physiological data, image data, and text data—each have their unique preprocessing methods. In this embodiment, the data processing step can use the method in S12 to preprocess the multi-source diagnostic information of the intelligent diagnostic model training dataset.

[0170] Thirdly, in classification tasks, the cross-entropy loss function is one of the most commonly used choices. It measures the difference between the probability distribution predicted by the model and the probability distribution of the actual labels. For multi-class classification problems, specifically for a classification task with N potential disease patterns in this embodiment, the formula can be expressed as:

[0171]

[0172] Where: y is a one-hot encoded vector (representing the true label), and It is the probability output predicted by the intelligent diagnostic model. This is the output probability of the i-th potential disease pattern predicted by the intelligent diagnostic model.

[0173] In this embodiment, the training method includes:

[0174] (1) Gradient descent and its variants

[0175] Gradient descent is the most basic optimization algorithm used to update the parameters in a model to minimize the loss function.

[0176] Adam optimizer: An optimization algorithm with an adaptive learning rate that combines the advantages of momentum from gradient descent and RMSprop. Adam performs well in many situations, especially when dealing with very large datasets or neural networks with a large number of parameters.

[0177] (2) Techniques to avoid overfitting

[0178] Regularization (such as L1 and L2 regularization) can penalize the magnitude of model weights, making the weight values ​​smoother and smaller, and preventing the model from becoming too complex.

[0179] Dropout: During training, the activation values ​​of some hidden nodes in the network are randomly deactivated with a certain probability. This can be regarded as a model averaging during training, which helps to reduce the model's overfitting to the training data.

[0180] (3) Strategies for dealing with small sample data

[0181] Transfer learning: When training data is insufficient, a pre-trained model, such as one pre-trained on ImageNet, can be used and then fine-tuned on a smaller dataset. Through transfer learning, the model can leverage the rich features learned on a large dataset to accelerate and optimize the learning process on new tasks.

[0182] The fourth step is to evaluate and optimize the network of the trained intelligent diagnostic model. The network of the intelligent diagnostic model is evaluated using the validation set of the intelligent diagnostic model. Evaluation metrics include accuracy, recall, F1 score, etc. The network architecture or training strategy is adjusted according to the model performance.

[0183] After obtaining the patient's acupuncture treatment plan through an intelligent diagnostic model, the acupuncture point locations can be accurately output using an acupuncture point-assisted positioning model, such as... Figure 4 As shown, the specific steps for the acupoint-assisted positioning module to obtain the location of the acupuncture points for a patient include:

[0184] S21: Data collection, collecting the patient's acupuncture location information;

[0185] S22: Input the patient's acupuncture location information into the acupoint-assisted location model, and output the patient's acupuncture point location through feature extraction, feature fusion, and acupoint prediction.

[0186] Specifically, in this embodiment, the acupoint-assisted positioning model includes a feature extraction module II, a feature fusion module II, and an output module II;

[0187] Feature extraction module II focuses on identifying key visual features from images, such as skin texture, surface markers, and joint positions, which are crucial information for understanding and locating acupoints. Feature fusion module II integrates features extracted from different image or sensor sources to enhance the model's accurate identification of acupoint locations. For example, it combines surface electromyography signals and high-resolution images to improve localization accuracy. Output module II typically outputs the specific location of the acupoint, which may be its coordinate position on the image or a description of its position relative to a surface marker.

[0188] Specifically, feature extraction module II obtains acupoint image features and acupoint data features from acupoint images and acupoint data, respectively;

[0189] The acupoint image feature is the image feature vector corresponding to the acupoint image;

[0190] The acupoint data feature is the resistance feature vector corresponding to the acupoint data;

[0191] Feature fusion module II fuses acupoint image features and acupoint data features to obtain acupoint fusion features;

[0192] Output module II outputs the location of the acupuncture points for the patient based on the acupuncture point fusion characteristics.

[0193] The final output of an acupoint-assisted localization model is typically the specific location of the acupoint, which can be the coordinates of an image or the actual body surface. The output format depends on the technology used and the purpose. In this embodiment, the model output is set to the two-dimensional coordinates of the acupoint relative to the pixel position of the image. For example, the model can output the coordinates of the acupuncture point at (350, 420) in the image. This location information can then be used to assist the user in accurately placing the acupuncture needle.

[0194] In practical applications, location information can also be directly mapped onto the patient using augmented reality technology, providing more intuitive guidance and assistance. For example, the device can mark acupoint locations directly in the operator's field of vision via a screen or special glasses. This can greatly improve the accuracy and efficiency of medical procedures.

[0195] The acupoint-assisted localization model in this embodiment combines image features with electrical resistance features, employing a multimodal learning method aimed at extracting the most useful information from different types of data for more accurate prediction or classification. Image features contain visual information about the acupoint's surroundings, and key features are extracted using image processing techniques (such as edge detection, texture analysis, and color distribution) to assist in identifying acupoint regions. Electrical resistance features provide physiological information related to the body surface conductivity or resistance, which is particularly useful in acupoint identification because some acupoint regions are distinguished from their surrounding areas due to different electrical resistance characteristics. By analyzing data from the electrical acupoint detector, the location and characteristic values ​​of low-resistance points (potential acupoints) are determined.

[0196] Specifically, the feature extraction module II acquires acupoint image features and acupoint data features. The acupoint image features are processed using advanced computer vision techniques such as convolutional neural networks (CNN) to extract spatial features at multiple levels.

[0197] Use architectures such as ResNet, VGG, or custom CNNs to convert the input acupoint images into high-dimensional feature vectors.

[0198] For acupoint data features, this implementation uses the resistance characteristics of acupoints, measuring the resistance value of the skin surface. The resistance is usually different in acupoint areas and non-acupoint areas. The resistance value is obtained through a simple sensor and can be converted into a feature vector useful for acupoint identification.

[0199] For Feature Fusion Module II, the fusion of acupoint image features and acupoint data features can be achieved in various ways. The simplest fusion method is Feature Concatenation, which directly concatenates the image feature vector and the resistance feature vector into a long vector.

[0200] More advanced methods include feature transformation or synthetic feature generation, such as using self-attention mechanisms or gating units to learn to extract the most informative parts from each type of feature.

[0201] For a simple fusion method, specifically, the feature fusion module II concatenates the image feature vector and the resistance feature vector through feature-level connections to obtain the acupoint fusion feature;

[0202] f fxw =[f i ,f r ]

[0203] Where: f fxw Acupoint fusion characteristics; f i f is the image feature vector; r For resistance eigenvectors; [f i ,f r The symbol ] represents the concatenation of image feature vectors and resistance feature vectors.

[0204] For advanced fusion methods, Feature Fusion Module II learns and acquires acupoint fusion features through self-attention mechanisms or gating units;

[0205] f fxw =ReLU(w xw [f i ,f r ]+b xw )

[0206] Where: f fxw Acupoint fusion characteristics; f i f is the image feature vector; r w is the resistance eigenvector; xw and b xw These are the weights and biases learned through self-attention mechanisms or gating units, respectively; ReLU is a non-linear activation function.

[0207] In another embodiment, the system further includes at least one of the following: a feedback and optimization module, a teaching video and technique explanation generation module, and an acupuncture emergency module;

[0208] The feedback and optimization module is used to adjust and optimize the acupuncture treatment plan based on patient feedback and treatment results after the acupuncture treatment module has completed the acupuncture treatment.

[0209] The teaching video and technique explanation generation module is used to automatically generate corresponding teaching videos or technique explanation videos based on the patient's acupuncture treatment plan.

[0210] The acupuncture emergency module monitors the patient's physiological state in real time based on multi-source diagnostic and treatment information. When it is determined that the patient needs emergency assistance, the emergency mode is activated.

[0211] The system in this embodiment also includes a feedback and optimization module, which can provide real-time feedback and adjust recommended treatment plans to ensure the timeliness and accuracy of medical decisions. It continuously improves the personalization and precision of treatment plans by comprehensively analyzing patient feedback and treatment outcomes and learning to optimize the diagnostic model.

[0212] Specifically, in the feedback and optimization module, patient feedback includes:

[0213] (1) Subjective feelings and changes in symptoms: the degree of pain relief, the recovery of function (such as mobility, sleep quality, etc.), and the improvement of symptoms (such as muscle tension, joint stiffness, etc.).

[0214] (2) Changes in quality of life: the degree of improvement in quality of life and the impact on daily activities;

[0215] (3) Side effects or discomfort: any discomfort or side effects caused by treatment; specific description of discomfort experienced during treatment (such as local redness and swelling, increased pain, etc.).

[0216] The evaluation of treatment outcomes is usually more objective and quantitative. In this embodiment, the treatment outcomes specifically include:

[0217] (1) Changes in physiological indicators: Changes in physiological data (blood pressure, heart rate, temperature, etc.), changes in blood indicators and other related biochemical test results;

[0218] (2) Imaging or physical measurement results: changes in tissues or organs monitored using imaging techniques such as ultrasound, X-ray or MRI, and changes in bodily functions monitored by means of skin resistance or electromyography.

[0219] (3) Symptom relief level: assessed using quantitative tools such as pain rating scales and functional status assessments.

[0220] The methods for obtaining treatment results include the following:

[0221] (1) Clinical assessment tools: pain rating scales (such as VAS score), functional status scales (such as SF36 quality of life questionnaire), patient logs or diaries to record changes in symptoms and quality of life;

[0222] (2) Physical and physiological monitoring equipment: using sensors to monitor physiological parameters such as heart rate, blood pressure, and respiratory rate, and using imaging equipment (such as MRI and X-ray) for periodic assessment of structure and function.

[0223] (3) Biochemical testing: analysis of chemical and biomarkers in blood and other body fluids;

[0224] (4) Software and benign algorithms: Use data analysis and machine learning tools to analyze the collected large-scale data to identify changes and trends before and after treatment.

[0225] After comprehensively analyzing patient feedback and treatment results, the feedback and optimization module will further adjust and optimize the acupuncture treatment plan. Specifically, the feedback and optimization module includes:

[0226] Real-time data monitoring and comprehensive analysis module: This module collects patients' physiological data (heart rate, blood pressure, etc.) in real time using various physiological monitoring devices (such as sensors). It combines this data with multi-source data including patients' physiological data, historical medical records, and treatment feedback to extract patient feedback and treatment outcomes. Furthermore, it uses data fusion technology to improve diagnostic accuracy based on this data.

[0227] The personalized treatment plan optimization model building module uses machine learning algorithms (such as decision trees, support vector machines, and neural networks) to analyze and process collected patient feedback and treatment results, extract features that are decisive for disease status and treatment response, and build a personalized treatment plan optimization model based on these features.

[0228] Dynamic feedback module: Establish a closed-loop feedback system, that is, collect patient feedback (such as pain score, functional status) and treatment results in a timely manner during the treatment process.

[0229] The model inputs patient feedback and treatment results into a personalized treatment plan optimization model, and outputs an optimized personalized treatment plan, such as adjusting the frequency and location of acupuncture.

[0230] Self-learning optimization module: Over time and with the accumulation of data, the feedback and optimization module can learn and optimize itself, continuously adjusting the personalized treatment plan optimization model through algorithms to improve the accuracy of prediction and the effectiveness of treatment.

[0231] In another embodiment of this application, the teaching video and technique explanation generation module automatically generates corresponding teaching videos and technique explanations based on the patient's acupuncture treatment plan. This provides additional value, not only improving the accuracy of treatment but also helping medical staff and patients better understand and implement the treatment plan.

[0232] In emergency medical situations, to enhance the system's rapid response capabilities, provide users with timely medical intervention, alleviate symptoms, and gain valuable treatment time, this embodiment also includes an acupuncture emergency module. This module comprises an emergency detection module, an emergency mode module, an automatic adjustment and optimization of treatment plans module, and a user interaction and guidance module.

[0233] Emergency Detection Module: Utilizes physiological monitoring sensors (such as heart rate, blood pressure, and body temperature sensors) to monitor the patient's condition in real time. Set thresholds, and once the sensor data exceeds the preset safety range, the system automatically identifies it as an emergency medical situation.

[0234] It can also integrate artificial intelligence models to identify potential signs of emergencies and improve the system's early warning capabilities.

[0235] Emergency Mode Module: When the system detects an emergency medical situation, it will automatically switch to emergency mode and suspend the current treatment plan.

[0236] In emergency mode, the system assesses the severity of the situation and recommends appropriate first aid measures based on the built-in emergency medical database, such as activating the automatic ambulance call service and notifying emergency contacts.

[0237] Different emergency situations require dedicated emergency response algorithms to ensure accurate and timely rescue operations.

[0238] Automatic adjustment and optimization of treatment plan module: The system dynamically adjusts and optimizes the acupuncture treatment plan based on the patient's response in emergency medical situations.

[0239] By leveraging machine learning and reinforcement learning algorithms, we can learn from historical emergency response data to optimize future emergency response plans.

[0240] User interaction and guidance module: In emergency medical situations, the system provides audio and visual guidance to help patients or non-professionals present perform initial first aid measures.

[0241] Design an easy-to-understand and easy-to-use interactive interface to ensure that users can respond quickly in urgent situations.

[0242] The acupuncture emergency module thoroughly tests the system's response capabilities and handling effectiveness in real-world and simulated emergency scenarios to ensure adequate preparation and system reliability. Simultaneously, laboratory testing and field drills are used to verify the system's emergency response efficiency and safety.

[0243] Regularly conduct system security reviews and risk assessments to ensure that emergency response measures comply with the latest medical standards and regulations.

[0244] The acupuncture emergency module in this embodiment can significantly improve the system's ability to handle emergency medical situations, thereby providing timely and effective emergency assistance to patients in urgent need. This will greatly enhance the system's practical value and social benefits, filling the gap in rapid response and intelligent processing capabilities for emergency medical equipment.

[0245] It should be noted that the modules (or units) in this embodiment are logical in nature. In actual implementation, multiple modules (or units) can be merged into one module (or unit), and one module (or unit) can also be split into multiple modules (or units).

[0246] The acupuncture system based on intelligent diagnosis in this implementation achieves automated identification, is low in cost, simple and convenient, and has obvious effects. Its application enables accurate and rapid estimation of key points of the human body in densely populated application scenarios, which has the advantages of reducing the workload of staff and improving work efficiency.

[0247] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0248] The above descriptions are merely preferred examples of the invention and are not intended to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. All modifications and equivalent substitutions made within the spirit and principles of the invention should be included within the scope of protection of the invention.

Claims

1. An acupuncture system based on intelligent diagnosis, characterized in that, include: Intelligent diagnosis module, acupoint auxiliary positioning module, acupuncture treatment module; The intelligent diagnostic module collects multi-source diagnostic and treatment information and inputs it into the intelligent diagnostic model. The intelligent diagnostic model analyzes the multi-source diagnostic and treatment information to obtain the patient's acupuncture treatment plan. The multi-source diagnostic information includes: physiological data, image data, and text data; the physiological data refers to the patient's physiological parameters acquired by sensors; the image data includes photographs of the patient's lesions or facial images; and the text data is a description of the patient's symptoms. The acupoint auxiliary positioning module collects acupuncture positioning information according to the patient's acupuncture treatment plan, inputs the acupuncture positioning information into the acupoint auxiliary positioning model, and outputs the location of the acupuncture points for the patient. The acupoint-assisted localization model includes a feature extraction module II, a feature fusion module II, and an output module II; The feature extraction module II obtains acupoint image features and acupoint data features from the acupoint image and acupoint data, respectively; The acupoint image feature is the image feature vector corresponding to the acupoint image; The acupoint data feature is the resistance feature vector corresponding to the acupoint data; The feature fusion module II fuses the acupoint image features and acupoint data features to obtain acupoint fusion features; The feature fusion module II concatenates the image feature vector and the resistance feature vector to obtain the acupoint fusion feature through feature-level connection; in: This is a characteristic of acupoint fusion; Image feature vector; This is the resistance eigenvector; This represents the concatenation of image feature vectors and resistance feature vectors; The output module II outputs the location of the acupuncture points for the patient based on the acupuncture point fusion characteristics. The acupuncture positioning information includes acupoint images and acupoint data; the acupoint images are the acupuncture acupoint locations corresponding to the patient's acupuncture treatment plan captured by a camera; the acupoint data are the resistance values ​​of the acupuncture acupoint locations corresponding to the acupuncture acupoint locations in the acupoint images obtained by an electric acupoint detector. The acupuncture treatment module is used to position the acupuncture needles at the acupuncture points on the patient's acupuncture site to perform acupuncture treatment on the patient; It also includes at least one of the following: a feedback and optimization module, a teaching video and technique explanation generation module, and an acupuncture first aid module; The feedback and optimization module is used to adjust and optimize the acupuncture treatment plan based on patient feedback and treatment results after the acupuncture treatment module has completed the acupuncture treatment. The instructional video and technique explanation generation module is used to automatically generate corresponding instructional videos or technique explanation videos based on the patient's acupuncture treatment plan. The acupuncture emergency module monitors the patient's physiological state in real time based on multi-source diagnostic and treatment information, and activates emergency mode when it is determined that the patient needs emergency assistance.

2. The acupuncture system based on intelligent diagnosis according to claim 1, characterized in that, The specific steps by which the intelligent diagnostic module obtains the patient's acupuncture treatment plan include: S11: Data collection, gathering multi-source medical information about patients; S12: Data preprocessing, preprocessing the patient's multi-source diagnosis and treatment information to obtain the preprocessed multi-source diagnosis and treatment information of the patient; S13: Potential disease pattern recognition, which inputs the pre-processed multi-source diagnosis and treatment information of the patient into the intelligent diagnostic model and outputs the patient's potential disease pattern. S14: Acquisition of acupuncture treatment plans. Combining the principles of TCM syndrome differentiation and treatment, the patient's potential disease pattern is mapped to TCM syndrome, and the patient's acupuncture treatment plan is obtained based on the patient's TCM syndrome.

3. The acupuncture system based on intelligent diagnosis according to claim 2, characterized in that, The preprocessing of the multi-source diagnostic and treatment information includes preprocessing of physiological data, image data, and text data. The steps for preprocessing physiological data include at least one of normalization, noise removal, and missing value processing. The steps of image data preprocessing include at least one of resampling, normalization, data augmentation, and conversion to grayscale. The steps for preprocessing text data include vectorization, and also include at least one of word segmentation, stop word removal, stemming, and word form restoration.

4. The acupuncture system based on intelligent diagnosis according to claim 2, characterized in that, The intelligent diagnostic model includes a feature extraction module I, a feature fusion module I, and an output module I; The feature extraction module I includes three different neural network architectures, which are used to extract features from physiological data, image data, and text data respectively, to obtain corresponding physiological features, image features, and text features. The feature fusion module I includes a fusion layer that fuses physiological features, image features, and text features to obtain diagnostic fusion features; The output module I includes one or more fully connected layers, which input diagnostic fusion features to the one or more fully connected layers for further information synthesis and output potential disease patterns.

5. The acupuncture system based on intelligent diagnosis according to claim 4, characterized in that, The physiological data is processed using recurrent neural networks, long short-term memory networks, or gated recurrent units; the image data is processed using convolutional neural networks; and the text data is processed using recurrent neural networks.

6. The acupuncture system based on intelligent diagnosis according to claim 4, characterized in that, The feature fusion module I is based on a multimodal attention mechanism to obtain diagnostic fusion features; The specific steps for the feature fusion module I to obtain diagnostic fusion features include: First, we define three attention networks: physiological feature attention network, image feature attention network, and text feature attention network. Secondly, based on the physiological feature attention network, image feature attention network, and text feature attention network, physiological feature modal weights, image feature modal weights, and text feature modal weights are obtained respectively. in: For the Sigmoid function, For physiological feature modal weights, , These are the learning parameters for the physiological characteristics attention network; For image feature modal weights, , These are the learning parameters for the image feature attention network; For text feature modal weights, , These are the learning parameters for the text feature attention network; Finally, diagnostic fusion features are obtained based on physiological features, image features, text features, and the corresponding physiological feature modality weights, image feature modality weights, and text feature modality weights. in: To diagnose fusion features.

7. An acupuncture system based on intelligent diagnosis according to claim 2, characterized in that, The training steps of the intelligent diagnostic model include: Step 1: Collect the training dataset for the intelligent diagnostic model. The training dataset for the intelligent diagnostic model includes multi-source diagnostic and treatment information of patients of different genders, ages, and diseases. The intelligent diagnostic model training dataset is divided into an intelligent diagnostic model training set and an intelligent diagnostic model validation set. Step 2: Preprocess the multi-source diagnostic and treatment information in the training set of the intelligent diagnostic model; Step 3: Define the loss function of the intelligent diagnostic model and train the network of the intelligent diagnostic model using the training set of the intelligent diagnostic model; Step 4: Evaluate the network of the intelligent diagnostic model using the intelligent diagnostic model validation set, and adjust the network architecture or training strategy of the intelligent diagnostic model based on the evaluation results.

Images