Acupuncture treatment scheme optimization method based on medical artificial intelligence

Abstract

The invention relates to the technical field of medical artificial intelligence and acupuncture treatment, and discloses an acupuncture treatment scheme optimization method based on medical artificial intelligence. According to the method, multi-dimensional physiological status indexes are analyzed and generated by collecting electronic medical records of patients and wearable equipment data, and a preliminary treatment prescription is generated by fusing acupuncture literature data. And analyzing an acupuncture point and parameter combination mode in the prescription by using a deep learning model, quantitatively evaluating contribution degrees of all elements, and optimizing and generating a final prescription in combination with historical treatment data of a patient. In clinical treatment, real-time feedback data are continuously collected, the corresponding relation between treatment parameters and symptom improvement is established, stimulation parameters are dynamically adjusted according to the corresponding relation, and closed-loop optimization is formed. According to the invention, quantitative intelligent optimization of the acupuncture treatment scheme and personalized dynamic fine adjustment in the treatment process are realized.

Description

Technical Field

[0001] This invention relates to the fields of medical artificial intelligence and acupuncture treatment technology, specifically to a method for optimizing acupuncture treatment plans based on medical artificial intelligence. Background Technology

[0002] Currently, acupuncture treatment plans primarily rely on physicians' personal experience and classical theories. With the development of medical informatization, some technologies attempt to use patients' electronic medical records for simple symptom classification and refer to fixed acupoint schemes in clinical literature databases to assist in generating preliminary treatment suggestions. However, this method is essentially a matching and retrieval of static knowledge, and the prescription generation logic remains at the level of empirical rules. Existing methods lack the ability to analyze the deep, non-linear correlations between the patient's complex physiological state and multi-dimensional indicators and the vast array of acupoint schemes in the literature, and also cannot quantitatively assess the expected contribution of each specific acupoint and treatment parameter within the prescription to a particular individual patient.

[0003] At the level of treatment plan optimization and execution, existing technologies mostly employ relatively fixed treatment parameters and assess efficacy through follow-up visits at the end of the entire treatment course. Although some studies mention collecting patient feedback, there is a lack of continuous, objective, real-time data collection and analysis mechanisms during treatment. Adjustments to treatment parameters are severely lagging, relying on physicians' subjective judgment and discrete follow-up points, failing to form a data-driven, dynamically responsive, and adaptive optimization loop. This prevents precise fine-tuning of treatment plans based on patients' immediate physiological responses during the treatment period, limiting personalization and precision. Therefore, existing technologies urgently need to address two core issues: how to achieve a leap from static experience matching to dynamic quantitative contribution analysis in prescription generation and optimization; and how to transform from fixed-parameter execution to a closed-loop treatment process based on real-time feedback and dynamic adjustment. Summary of the Invention

[0004] The purpose of this invention is to provide a method for optimizing acupuncture treatment plans based on medical artificial intelligence, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for optimizing acupuncture treatment plans based on medical artificial intelligence, the method comprising: Collect patients’ electronic medical record data and wearable device monitoring data, analyze patients’ clinical symptom characteristics and signs change trends, calculate the correlation strength between different symptoms, and obtain patients’ multi-dimensional physiological state indicators. By integrating acupuncture clinical literature data, extracting acupoint combination schemes and treatment parameters from the literature, analyzing the matching relationship between the multidimensional physiological state indicators of the patients and the acupoint combination schemes in the literature, evaluating the targeting of acupoint combinations to the patient's symptoms, and generating a preliminary acupuncture treatment prescription. The preliminary acupuncture treatment prescription is input into a deep learning model, which analyzes the combination pattern of acupoints and treatment parameters in the prescription, evaluates the contribution of each acupoint and parameter to the therapeutic effect, compares the effective patterns in the patient's historical treatment data, optimizes the acupoint combination and stimulation parameters in the prescription, and generates an optimized acupuncture treatment prescription. Based on the optimized acupuncture treatment prescription, real-time feedback data from patients is continuously collected during clinical treatment. The correlation between treatment parameters and the degree of improvement in patient symptoms is analyzed, and acupoint stimulation parameters are dynamically adjusted to form a closed-loop optimization process for the acupuncture treatment plan.

[0006] Preferably, the multidimensional physiological state indicators include symptom correlation, rate of change of signs, and range of fluctuation of physiological parameters; the preliminary acupuncture treatment prescription includes recommended acupoint combinations, initial treatment parameters, and a matching score with the patient's condition; the optimized acupuncture treatment prescription includes optimized acupoint combinations, optimized stimulation parameters, and an evaluation of the efficacy contribution of each parameter; and the real-time feedback data includes symptom improvement scores, changes in physiological parameters during treatment, and immediate feedback records after treatment.

[0007] Preferably, the collection of the patient's electronic medical record data and wearable device monitoring data, and the analysis of the patient's clinical symptom characteristics and signs and trends, specifically includes: The system collects patients' electronic medical record data from the hospital information system interface, analyzes the diagnostic records, medical history information, symptom descriptions and various examination and test results in the medical records, extracts the key features and duration of symptoms, calculates the severity level of symptom features, and obtains a structured symptom feature set. It synchronously receives physiological data continuously monitored by wearable devices, analyzes the change curves of physiological parameters over time, identifies the time periods and amplitudes of abnormal parameter fluctuations, and performs time correlation analysis between abnormal fluctuations and patients' symptoms and complaints to obtain the temporal change characteristics of vital signs. By integrating the structured symptom feature set with the temporal change features of the physical signs, the co-occurrence probability between different symptoms and specific abnormal physical signs is analyzed, the combined symptom and physical sign index is calculated, the overall condition of the patient is comprehensively assessed, and multi-dimensional physiological state indicators of the patient are generated.

[0008] Preferably, the integration of acupuncture clinical literature data with the patient's multi-dimensional physiological state indicators, through data structuring and artificial intelligence algorithm modeling, evaluates the targeting of acupoint combinations to the patient's symptoms and generates a preliminary acupuncture treatment prescription, specifically including: By using natural language processing technology to process acupuncture clinical literature in batches, we can identify and extract data on disease names, acupoints used, acupuncture techniques, treatment frequency and course of treatment recorded in the literature. We can also standardize and structure the data to build a knowledge base and obtain a set of standard acupoint schemes. The patient's multidimensional physiological state indicators are matched with the standard acupoint scheme set. Based on the disease diagnosis and symptom characteristics, the knowledge base is searched to select candidate acupoint schemes with a relevance higher than a preset threshold. The complete treatment parameters of each scheme are extracted to obtain a candidate treatment scheme set. Based on the candidate treatment plan set, the matching of the treatment parameters of each plan with the patient's specific physiological indicators is calculated by artificial intelligence algorithm modeling. Combined with the efficacy level information recorded in the literature, an applicability score for each plan is generated for the current patient. After sorting, a preliminary acupuncture treatment prescription is generated.

[0009] Preferably, the step of inputting the preliminary acupuncture treatment prescription into a deep learning model, and having the deep learning model analyze the combination pattern of acupoints and treatment parameters in the prescription, and evaluate the contribution of each acupoint and parameter to the therapeutic effect, specifically includes: The acupuncture clinical literature data is fused with the patient's multidimensional physiological state indicators, and the patient's multidimensional physiological state indicators are given high weight, while the acupuncture clinical literature data is given low weight. The preliminary acupuncture treatment prescription is converted into a vector representation that can be processed by a deep learning model. The vector includes acupoint encoding, stimulation intensity, needle retention time, and treatment frequency parameters. The trained deep learning model is invoked to process the vector. The deep learning model automatically analyzes the interaction between different acupoints in the prescription, resolves the potential treatment patterns under complex parameter combinations, and outputs the contribution weight of each acupoint and each treatment parameter to the expected therapeutic effect. Based on the aforementioned contribution weights, successful treatment cases of similar patient groups in the knowledge base are compared, parameter settings with low contribution are optimized, acupoints and parameter combinations with high contribution are strengthened, specific operational details in the prescription are adjusted, and an optimized acupuncture treatment prescription is generated.

[0010] Preferably, based on the optimized acupuncture treatment prescription, the patient's real-time feedback data is continuously collected during clinical treatment to analyze the correlation between treatment parameters and the degree of improvement in patient symptoms, specifically including: During each acupuncture treatment, the actual needle insertion angle, depth, stimulation technique and intensity are recorded using specialized equipment. At the same time, the patient's real-time subjective feeling score and local electromyographic changes are collected before and after treatment to obtain data for a single treatment process. After the treatment cycle is completed, the patient's self-reported symptom improvement scale score and the doctor's evaluation of the efficacy level are collected. Combined with all the single treatment process data recorded during the treatment, the mapping relationship between different combinations of treatment parameters and the final degree of symptom improvement is analyzed to obtain the parameter efficacy mapping table. Based on the parameter efficacy mapping table, the treatment parameter range with the highest positive correlation to symptom improvement and the parameter settings associated with poor efficacy are identified, providing a basis for the dynamic adjustment of subsequent treatment plans.

[0011] Preferably, the steps for constructing the deep learning model include: Collect historical acupuncture treatment case data, which includes patients' physiological indicators, the combination of acupoints used, treatment parameters, and corresponding efficacy evaluation results; The historical acupuncture treatment case data were preprocessed to convert unstructured efficacy evaluation results into structured labels, and the acupoint names and treatment parameters were standardized and coded. A neural network model is constructed, wherein the input layer dimension of the neural network model is matched with the dimension of the standardized encoded acupoint combination and treatment parameter features, and the output layer is set to multi-class probability prediction of the efficacy evaluation results. The neural network model is trained using preprocessed historical acupuncture treatment case data. The model weights are adjusted using the backpropagation algorithm until the model's classification accuracy for therapeutic effects reaches a preset threshold, resulting in a trained deep learning model.

[0012] Preferably, the synchronous reception of physiological data continuously monitored by the wearable device, analysis of the physiological parameter change curves over time, identification of the time periods and amplitudes of abnormal parameter fluctuations, and temporal correlation analysis of the abnormal fluctuations with the patient's symptoms and complaints are performed to obtain the temporal change characteristics of vital signs, including: Set a threshold for the normal fluctuation range of physiological parameters and perform real-time scanning of the physiological data stream continuously monitored by wearable devices; When a physiological data point is detected to exceed the normal fluctuation range threshold, the value of the physiological data point, the extent of the exceedance, and the time of occurrence are recorded and marked as an abnormal fluctuation event. Extract the patient's symptom complaints recorded within a preset time window before and after the occurrence of the abnormal fluctuation event; The correlation coefficient between the amplitude of abnormal fluctuation events and the severity described in the symptom complaints is calculated, and abnormal fluctuation events with high correlation coefficients are integrated with symptom complaints to form a temporal change feature of signs.

[0013] Preferably, the step of calculating the degree of fit between the treatment parameters of each treatment plan and the patient's specific physiological indicators based on the candidate treatment plan set includes: The treatment parameters of a candidate treatment plan are extracted from the set of candidate treatment plans. The treatment parameters include acupuncture technique, stimulation intensity, and treatment frequency. Specific physiological indicators corresponding to the treatment parameters are extracted from the patient's multidimensional physiological state indicators, including pain tolerance threshold, autonomic nerve response sensitivity, and fatigue recovery rate. The treatment parameters are quantitatively compared with the specific physiological indicators to calculate the matching degree between stimulation intensity and pain tolerance threshold, the compatibility between treatment frequency and fatigue recovery rate, and the fit between acupuncture technique and autonomic nerve response sensitivity. By combining the matching degree, fitness degree, and fit degree, the overall consistency between the treatment parameters of the candidate treatment plan and the patient's specific physiological indicators is calculated by weighted summation, and this process is repeated for all plans in the candidate treatment plan set.

[0014] Preferably, the patient's self-reported symptom improvement scale score and the physician's assessment of the efficacy level are collected, and combined with all the single treatment process data recorded during the treatment process, the mapping relationship between different combinations of treatment parameters and the final degree of symptom improvement is analyzed to obtain a parameter efficacy mapping table, including: After the treatment cycle is completed, the symptom improvement scale scores filled out by the patients after each treatment are summarized, and the final efficacy level given by the doctor based on clinical observation is obtained. The treatment parameters recorded in each individual treatment process, including needle insertion angle, depth, stimulation technique and intensity, are paired with the final degree of symptom improvement obtained after the entire treatment cycle. Using association rule mining algorithms, we analyzed all paired data to identify frequent patterns where specific combinations of treatment parameters occurred simultaneously with high symptom improvement scale scores or high efficacy levels. The frequently discovered patterns are organized to form a parameter efficacy mapping table with treatment parameter combinations as indexes and expected symptom improvement as values.

[0015] Compared with the prior art, the beneficial effects of the present invention are: A deep learning model is used to analyze the initially generated acupuncture treatment prescription, quantitatively evaluating the contribution of each specific acupoint and treatment parameter within the prescription to the expected therapeutic effect, and optimizing it by referring to effective patterns in the patient's individual historical treatment data. This transforms the vague relationship between acupoint combinations and parameter selection in traditional experience into a computable contribution weight relationship based on data patterns. Prescription optimization no longer relies on general empirical analogies, but is based on a quantitative analysis of the improvement of specific symptoms by specific parameters, making the final generated acupoint combinations and stimulation parameter settings have clear individualized orientation and interpretability.

[0016] During the treatment implementation phase, a dynamic correspondence model between specific treatment parameters and objective indicators of symptom improvement is established by continuously collecting real-time physiological feedback data from patients. Based on the output of this model, acupoint stimulation parameters are adjusted in real time during treatment. This achieves online matching and dynamic calibration between treatment intervention and patient physiological responses, compressing the traditional discrete assessment and adjustment cycle based on treatment courses into a near real-time, continuous optimization loop. The treatment plan can proactively adapt to the patient's response state during treatment, thereby improving the accuracy of intervention within a single treatment session and the adaptability and efficiency of the overall treatment course. Attached Figure Description

[0017] Figure 1 This is a schematic diagram illustrating the working principle of the acupuncture treatment plan optimization method based on medical artificial intelligence as described in this invention. Figure 2 A flowchart for collecting and analyzing data from electronic medical records and wearable devices; Figure 3 A flowchart for generating preliminary prescriptions by integrating literature data; Figure 4 Pearson correlation heatmap of acupuncture treatment parameters and structured efficacy labels; Figure 5 This is a graph showing the trend of symptom improvement and electromyography changes during acupuncture treatment for cervical and shoulder syndrome. Detailed Implementation

[0018] 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1This invention provides a method for optimizing acupuncture treatment plans based on medical artificial intelligence. The method includes: Upon initiation, the system first collects the patient's electronic medical record data and wearable device monitoring data. Through analysis of this data, the system extracts the patient's clinical symptom characteristics and tracks the trends in their vital signs, then calculates the correlation strength between different symptoms to obtain a multi-dimensional physiological state indicator that comprehensively reflects the patient's health status. Subsequently, the system integrates acupuncture clinical literature data with the patient's multi-dimensional physiological state indicator. Through data structuring and artificial intelligence algorithm modeling, it evaluates the targeting of different acupoint combinations to the patient's specific symptoms and generates a preliminary acupuncture treatment prescription. After generating the preliminary prescription, it is input into a pre-built deep learning model. This model is responsible for analyzing the combination patterns of acupoints and treatment parameters in the prescription, evaluating the contribution of each acupoint and parameter to the treatment effect, and optimizing the acupoint combinations and stimulation parameters in the prescription by comparing them with effective patterns in the patient's historical treatment data. Finally, an optimized acupuncture treatment prescription is output. During the clinical implementation of the optimized acupuncture treatment prescription, real-time feedback data from patients is continuously collected. The correlation between the treatment parameters used and the degree of improvement in patients' symptoms is analyzed, and the acupoint stimulation parameters in subsequent treatments are dynamically adjusted based on this relationship, thereby forming a complete closed-loop optimization process for acupuncture treatment plans that can continuously improve itself.

[0020] In one embodiment of the present invention, the multidimensional physiological state indicators include symptom correlation, rate of change of signs, and range of fluctuation of physiological parameters. Symptom correlation characterizes the statistical correlation strength between different clinical symptoms; rate of change of signs describes how fast a specific physiological indicator changes over time; range of fluctuation of physiological parameters defines the normal and abnormal boundaries of a certain physiological parameter within the observation period. The preliminary acupuncture treatment prescription includes a recommended acupoint combination, initial treatment parameters, and a matching score with the patient's condition. The recommended acupoint combination lists the acupoints to be acupunctured; the initial treatment parameters include stimulation intensity, needle retention time, etc.; the matching score quantifies the degree of fit between the prescription and the patient's current multidimensional physiological state. The optimized acupuncture treatment prescription includes optimized acupoint combination, optimized stimulation parameters, and efficacy contribution assessment of each parameter. The optimized acupoint combination is a combination of acupoints adjusted by the model; the optimized stimulation parameters are specific operational parameters adjusted by the model; the efficacy contribution assessment uses weights to indicate the influence of each component of the treatment plan on the expected efficacy. The real-time feedback data includes symptom improvement scores, changes in physiological parameters during treatment, and immediate feedback records after treatment. Symptom improvement scores are derived from patient subjective reports or standardized scales; physiological parameter changes during treatment are recorded as real-time physiological signals during acupuncture; and immediate feedback after treatment records the patient's immediate feelings after receiving treatment.

[0021] In practical implementation, taking a patient with chronic lower back pain and sleep disorders as an example, the system collects the patient's electronic medical record data and wearable device monitoring data. The electronic medical record data includes the patient's basic medical information, clinical symptom assessment scale data, physical and chemical test data, functional examination data, imaging examination data, and other related health data. The collected patient's basic medical information, clinical symptom assessment scale data, physical and chemical test data, functional examination data, imaging examination data, and other related health data are processed. During tagging, specific tags are assigned to each type of data according to the medical standard classification system. For example, demographic characteristic tags are assigned to age and gender in the basic medical information, symptom quantification tags are assigned to the clinical symptom assessment scale data, and indicator category tags are assigned to the physical and chemical test data. During data cleaning, duplicate records and invalid data lacking key information are removed, and data entry errors are corrected to ensure the accuracy and completeness of the data. Structured processing involves parsing unstructured text descriptions and test reports to extract core information and organize it according to preset data fields. Basic case information is transformed into standardized field records, clinical symptom assessment scale data is converted into quantitative values ​​in a unified format, and physical and chemical test results, functional examinations, and imaging examinations are refined into structured indicator names, test values, reference ranges, and other related information. This ultimately forms well-organized structured data, laying the foundation for subsequent extraction of key symptom features, calculation of severity levels, and generation of multi-dimensional physiological state indicators. Electronic medical record data recorded symptom descriptions of "persistent dull pain in the lower back" and "difficulty falling asleep," while wearable device monitoring data included nocturnal heart rate variability sequences and daytime activity level records. Through analysis, the system calculated a correlation strength of 0.78 between the two symptoms of "lower back pain" and "decreased sleep quality." The rate of change in vital signs was characterized by a nighttime autonomic balance index falling by more than 20% from baseline during periods of heightened pain, and the range of physiological parameter fluctuations defined the normal upper limit of daytime resting heart rate as 75 beats per minute.

[0022] In some embodiments, the system searches an acupuncture clinical literature knowledge base based on the aforementioned multi-dimensional physiological state indicators. The knowledge base contains multiple acupoint combination schemes for lumbar insomnia syndrome, and the system calculates the matching degree of each scheme with the current patient's condition. For example, a scheme containing the acupoints "Shenshu," "Weizhong," and "Shenmen" has treatment parameters including moderate stimulation intensity, needle retention for 25 minutes, and treatment every other day. After evaluating the matching degree of this scheme with the patient's pain characteristics and autonomic nervous system disorder features, the system generates a matching degree score. The preliminary acupuncture treatment prescription thus includes this recommended acupoint combination, the corresponding initial treatment parameters, and the matching degree score with the patient's condition. The matching degree score quantifies the degree of fit between the preliminary acupuncture treatment prescription and the current clinical situation.

[0023] Understandably, once the initial acupuncture treatment prescription is input into the deep learning model, the model parses the prescription content. The deep learning model analyzes the interaction patterns of the three acupoints "Shenshu," "Weizhong," and "Shenmen" in historical successful cases and evaluates the contribution weight of parameters such as needle retention time and stimulation intensity to the therapeutic effect. The optimization process adjusts the parameters based on the contribution weight, for example, optimizing the needle retention time from 25 minutes to 30 minutes and the stimulation intensity from medium to medium-strong. The optimized acupuncture treatment prescription therefore includes the optimized acupoint combination, the optimized stimulation parameters, and a report that clearly lists the therapeutic contribution assessment of each parameter. The therapeutic contribution assessment presents the influence of each decision element in numerical form.

[0024] In practice, real-time data collection is conducted simultaneously when clinical treatment is performed based on the optimized acupuncture prescription. In one treatment session, the actual depth of needle insertion at the "Shenshu" acupoint was recorded as 1.2 cun, and the intensity of the twisting technique used was quantified by the device as a torque of 0.6 N·m. These data constitute part of the physiological parameter changes during treatment. After treatment, the patient inputs a score on the degree of relief of lower back pain via a tablet computer; this score is recorded as a symptom improvement score. The doctor observes and records the patient's description of a "feeling of relaxation in the lower back," and this information is saved as an immediate post-treatment feedback record.

[0025] Optionally, the symptom correlation can be calculated based on the co-occurrence probability and independent occurrence probability of symptoms. One calculation method uses the following formula: Where: symbol Indicates the degree of association between symptom i and symptom j, with the symbol […]. This represents the probability that symptom i and symptom j co-occur in the patient population, with the symbol [symptom name missing]. The symbol represents the probability of symptom i occurring independently. This represents the probability of symptom j occurring independently.

[0026] In some embodiments, the generation of the matching score involves multi-level weighted calculations. The system considers not only a perfect match of the disease diagnosis name, but also the correspondence between specific symptom characteristics and the indications of acupoints, as well as the coverage of the range of fluctuations in the patient's physiological parameters and the applicable range of treatment parameters recorded in the literature. Optionally, the efficacy contribution assessment is performed within the deep learning model. The deep learning model first converts the initial acupuncture treatment prescription into a processable vector representation, which includes parameters such as acupoint encoding, stimulation intensity, needle retention time, and treatment frequency. The trained deep learning model then processes this vector, automatically analyzing the interactions between different acupoints in the prescription, resolving potential treatment patterns under complex parameter combinations, and outputting the contribution weight of each acupoint and each treatment parameter to the expected therapeutic effect. The deep learning model is built based on historical acupuncture treatment case data. By collecting patient physiological indicators, the acupoint combinations used, treatment parameters, and corresponding therapeutic evaluation results, the data is preprocessed, transforming unstructured therapeutic evaluation results into structured labels, and standardizing the encoding of acupoint names and treatment parameters. When constructing the neural network model, the input layer dimension matches the feature dimensions of the standardized encoded acupoint combinations and treatment parameters, and the output layer is set to predict the multi-class probability of the therapeutic evaluation results. The model is trained using the preprocessed historical data, and the model weights are adjusted through backpropagation until the model's classification accuracy for therapeutic effects reaches a preset threshold, thus obtaining a trained deep learning model capable of accurately parsing prescriptions and evaluating contribution weights. By analyzing a vast number of treatment cases, the model learned patterns such as the "Weizhong" acupoint having a higher weight in relieving radiating pain in the lower limbs but a lower weight in improving difficulty falling asleep. The efficacy contribution evaluation report attached to the optimized acupuncture treatment prescription will clearly indicate that the "Shenmen" acupoint has a weight of 0.85 in improving sleep, while the "Shenshu" acupoint has a weight of 0.90 in relieving core pain in the lower back, thus providing a transparent explanation for clinicians to understand the optimization logic.

[0027] In one embodiment of the present invention, see [reference] Figure 2The system collects patients' electronic medical record data from the hospital information system interface, analyzes the diagnostic records, medical history information, and symptom descriptions in the medical record text, extracts key symptom features and durations, and calculates the severity level of symptom features according to medical standards, thereby obtaining a structured symptom feature set. Simultaneously, it receives continuous physiological data streams from wearable devices, including heart rate, blood pressure, and skin temperature, analyzes the time-varying curves of these physiological parameters, and identifies the time periods and amplitudes of abnormal parameter fluctuations. A normal fluctuation range threshold for physiological parameters is set, and the continuous physiological data streams monitored by the wearable devices are scanned in real time. When a physiological data point is detected to exceed the normal fluctuation range threshold, the value of the physiological data point, the magnitude of the exceedance, and the time point of occurrence are recorded and marked as an abnormal fluctuation event. The system extracts the symptom complaints recorded by the patient within a preset time window before and after the occurrence of the abnormal fluctuation event; calculates the correlation coefficient between the amplitude of the abnormal fluctuation event and the severity described in the symptom complaint, and integrates abnormal fluctuation events with high correlation coefficients and symptom complaints into a temporal change feature of vital signs. By integrating the structured symptom feature set with the temporal change features of the physical signs, the co-occurrence probability between different symptoms and specific abnormal physical signs is analyzed, the joint symptom and physical sign index is calculated, the overall condition of the patient is comprehensively assessed, and finally, a multi-dimensional physiological state index of the patient is generated.

[0028] In practical implementation, the method for optimizing acupuncture treatment based on medical artificial intelligence involves the specific process of collecting patients' electronic medical record data and wearable device monitoring data, and analyzing the clinical symptom characteristics and signs. Taking a patient with paroxysmal palpitations and chest tightness as an example, the patient's electronic medical record data is collected from the hospital information system interface. The diagnostic records, medical history information and symptom descriptions in the medical record text are analyzed, and "palpitations" and "chest tightness" are extracted as key symptom features and their durations are recorded. According to clinical guidelines, the frequency of palpitations and the intensity of chest tightness are converted into numerical severity levels, thereby obtaining a structured symptom feature set containing symptom names, durations and quantitative severity levels.

[0029] In some embodiments, the system synchronously receives physiological data continuously monitored by a wearable device, including dynamic electrocardiogram recordings and data from a chest wall respiratory motion sensor. It analyzes the time-varying curves of heart rate and respiratory rate parameters in this physiological data, identifies the time periods and amplitudes of abnormal parameter fluctuations, sets a normal fluctuation range threshold of 60 to 100 beats per minute for the heart rate parameter, and performs real-time scanning of the physiological data stream continuously monitored by the wearable device. When a heart rate data point is detected to exceed the normal fluctuation range threshold of 60 to 100 beats per minute, the system records the specific value, the magnitude of the exceedance, and the time of occurrence of the heart rate data point, marking it as an abnormal fluctuation event. It extracts the symptom complaints recorded by the patient via a mobile application within a ten-minute time window before and after the occurrence of the abnormal fluctuation event, such as "sudden onset of palpitations" or "worsening of chest tightness." It calculates the correlation coefficient between the amplitude of the abnormal fluctuation event and the severity described in the symptom complaints, and integrates abnormal fluctuation events with high correlation coefficients and symptom complaints into a temporal change feature of vital signs.

[0030] It is understandable that the process of integrating structured symptom feature sets with temporal change features of signs involves data analysis, analyzing the co-occurrence probability between different symptoms and specific abnormal signs, such as calculating the probability of the symptom "palpitation" and the abnormal sign "heart rate exceeding 120 beats / minute" co-occurring in a time series, and calculating the symptom-sign joint index. The symptom-sign joint index is a comprehensive quantitative representation that comprehensively assesses the patient's overall condition and generates multi-dimensional physiological state indicators for the patient.

[0031] In practice, the correlation coefficient between the amplitude of abnormal fluctuation events and the severity described in the symptom complaints can be calculated using a quantitative formula. One method is as follows: Where: symbol The correlation coefficient represents the relationship between the magnitude of abnormal fluctuations and the severity of the reported symptoms. (Symbol: ) This represents the amplitude value of the k-th identified abnormal fluctuation event, with the sign... This represents the severity score of the symptom complaint within the time window corresponding to the k-th abnormal fluctuation event, with the symbol... This represents the arithmetic mean of the amplitude values ​​of all abnormal fluctuation events, with the sign... This represents the arithmetic mean of the severity scores for all symptom complaints, with the symbol [symbol missing]. This represents the total number of abnormal fluctuation events included in the calculation.

[0032] Optionally, when collecting electronic medical record data from the hospital information system interface, the parsing process involves natural language processing (NLP) technology. NLP technology identifies unstructured descriptions in the medical record text and converts them into structured fields. For example, "palpitations occur several times a day" is parsed into the key symptom features "palpitations," the duration "daily," and the frequency of occurrence "several times." The severity level is calculated based on a frequency-intensity comparison table, mapping "several times" to a level value. The structured symptom feature set is ultimately stored in the form of database records, containing symptom identifiers, time dimension information, and severity values. In some embodiments, identifying the time period and amplitude of abnormal fluctuations in physiological parameters can be combined with a sliding window statistical method. The sliding window statistical method calculates the mean and standard deviation of recent physiological data streams. When a real-time data point deviates from the mean by more than three standard deviations, an abnormal fluctuation event is triggered and recorded. The abnormal fluctuation event is marked and its duration is recorded. The temporal change characteristics of vital signs not only include point events but also pattern information of event sequences, such as a periodic cluster of nocturnal tachycardia events.

[0033] Optionally, the calculation of the symptom-sign joint indicator can integrate multiple statistical measures. The symptom-sign joint indicator includes the co-occurrence frequency of symptoms and signs, the time lag correlation coefficient, and the synchronicity intensity index. When comprehensively assessing the patient's overall condition, these sub-indicators are normalized and weighted and summed. The generated multi-dimensional physiological state indicator is a vector containing quantitative values ​​of multiple dimensions such as symptom correlation, rate of change of signs, and range of fluctuation of physiological parameters.

[0034] In one embodiment of the present invention, see [reference] Figure 3This study utilizes natural language processing (NLP) technology to batch process acupuncture clinical literature, identifying and extracting data on disease names, acupoints used, acupuncture techniques, treatment frequency, and treatment courses. The data is then standardized and structured to construct a knowledge base, resulting in a set of standard acupoint treatment plans. The patient's multi-dimensional physiological state indicators are matched against this set of standard acupoint treatment plans. Based on disease diagnosis and symptom characteristics, the knowledge base is searched to select candidate acupoint treatment plans with a relevance exceeding a preset threshold. Complete treatment parameters for each plan are extracted to obtain a candidate treatment plan set. Based on this candidate treatment plan set, artificial intelligence algorithms are used to model and calculate the degree of consistency between the treatment parameters of each plan and the patient's specific physiological indicators. A weighted summation algorithm combined with feature matching logic is used for quantitative calculation. First, the treatment parameters of a single plan are extracted from the candidate treatment plan set, while simultaneously extracting the corresponding specific physiological indicators from the patient's multi-dimensional physiological state indicators. The two types of indicators were then standardized and coded. Stimulus intensity was quantified into a score of 0-5, pain tolerance threshold was converted into a score of 0-5 according to the corresponding tolerance level, and acupuncture technique and autonomic nerve response sensitivity were mapped to a matching coefficient (between 0 and 1) according to a preset correspondence. Treatment frequency and fatigue recovery rate were quantified in weekly units and the fit ratio was calculated. Next, based on the clinical efficacy weighting rules, the matching degree of stimulation intensity and pain tolerance threshold was assigned a weight of 0.4, the matching degree of acupuncture technique and autonomic nerve response sensitivity was assigned a weight of 0.3, and the fit degree of treatment frequency and fatigue recovery rate was assigned a weight of 0.3. The overall fit was calculated using a weighted summation formula, the specific formula being: in: , , This indicates the preset weighting coefficient. Indicates matching degree, Indicates compatibility, The fit is then calculated. Finally, the above calculation process is repeated for all protocols in the candidate treatment set to obtain a quantified fit value for each protocol.

[0035] For example, for a patient with chronic low back pain, the candidate treatment parameters are "moderate stimulation intensity, once-daily treatment frequency, and twirling and reducing technique." The corresponding physiological indicators for the patient are "pain tolerance threshold of 3 points, fatigue recovery rate twice / week, and high autonomic nervous system sensitivity." After standardized coding, the matching degree between stimulation intensity and pain tolerance threshold is 0.8, the matching degree between treatment frequency and fatigue recovery rate is 0.6, and the matching degree between acupuncture technique and autonomic nervous system sensitivity is 0.9. The overall consistency is calculated by weighted summation. This provides a quantitative basis for subsequent applicability scoring.

[0036] Treatment parameters for one candidate treatment plan are extracted from the set of candidate treatment plans. These parameters include acupuncture technique, stimulation intensity, and treatment frequency. Specific physiological indicators corresponding to the treatment parameters are extracted from the patient's multi-dimensional physiological state indicators, including pain tolerance threshold, autonomic nervous system response sensitivity, and fatigue recovery rate. The treatment parameters and specific physiological indicators are quantitatively compared to calculate the matching degree between stimulation intensity and pain tolerance threshold, the fit between treatment frequency and fatigue recovery rate, and the compatibility between acupuncture technique and autonomic nervous system response sensitivity. The overall consistency between the treatment parameters of the candidate treatment plan and the patient's specific physiological indicators is calculated by weighted summation of the matching degree, fit, and compatibility. This process is repeated for all plans in the set of candidate treatment plans. Combining the efficacy level information recorded in the literature, an applicability score is generated for each plan for the current patient. After sorting, a preliminary acupuncture treatment prescription is generated.

[0037] In practice, taking a patient diagnosed with migraine due to liver yang hyperactivity accompanied by dizziness and irritability as an example, the acupuncture clinical literature is processed in batches using natural language processing technology. Natural language processing technology identifies and extracts data on disease name, acupoints used, acupuncture techniques, treatment frequency and course of treatment recorded in the literature. For example, from a literature, the disease name "liver yang hyperactivity headache", the acupoints used "Taichong", "Fengchi", "Baihui", the acupuncture technique "reducing method", the treatment frequency "once a day" and the course of treatment "10 times as a course of treatment" are extracted to construct a standardized acupuncture treatment knowledge base, resulting in a set of standard acupoint schemes containing a large number of such standardized entries.

[0038] In some embodiments, the patient's multidimensional physiological state indicators are matched with a set of standard acupoint schemes. The patient's multidimensional physiological state indicators include the disease diagnosis "liver yang hyperactivity type migraine", symptom characteristics "headache location on the temporal side", "dizziness during attack", and physiological parameter "fluctuating increase in blood pressure". Based on the disease diagnosis and symptom characteristics, the acupuncture treatment knowledge base is searched, and the system selects candidate acupoint schemes that completely match the disease name and are highly related to the symptom description. For example, multiple schemes that meet the conditions of "liver yang hyperactivity" and "headache" are selected. Candidate acupoint schemes with a relevance higher than a preset threshold are selected and the complete treatment parameters of each scheme are extracted to obtain a candidate treatment scheme set containing several specific treatment schemes.

[0039] Understandably, based on the candidate treatment plan set, it is necessary to calculate the degree of consistency between the treatment parameters of each plan and the patient's specific physiological indicators. Treatment parameters for a candidate treatment plan are extracted from the set, including the acupuncture technique "reducing method," the stimulation intensity "moderate," and the treatment frequency "once daily." Specific physiological indicators corresponding to the treatment parameters are extracted from the patient's multi-dimensional physiological state indicators, including a "low" pain tolerance threshold, a "high" autonomic nerve response sensitivity, and a "moderate" fatigue recovery rate. The treatment parameters are then quantitatively compared with these specific physiological indicators. The matching degree between "moderate" stimulation intensity and "low" pain tolerance threshold, the fit between "once daily" treatment frequency and "moderate" fatigue recovery rate, and the fit between "reducing method" acupuncture technique and "high" autonomic nerve response sensitivity are calculated. The overall consistency between the treatment parameters of this candidate treatment plan and the patient's specific physiological indicators is calculated by combining the matching degree, fit, and fit through a weighted summation. This calculation process is repeated for all plans in the candidate treatment plan set.

[0040] In practice, the process of calculating the overall fit by combining matching degree, adaptability, and compatibility can be expressed using a mathematical formula. One calculation method uses the following formula: Where: symbol Indicates the overall degree of agreement between treatment parameters and specific physiological indicators of the patient, with the symbol […]. The symbol represents the result of the matching degree between stimulus intensity and pain tolerance threshold. The result of the fit calculation between treatment frequency and fatigue recovery rate is represented by the symbol. The symbol represents the calculated result of the compatibility between acupuncture techniques and autonomic nervous system response sensitivity. Indicates matching degree Preset weighting coefficients, symbols Indicates fit Preset weighting coefficients, symbols Indicates compatibility The preset weighting coefficients.

[0041] Optional, matching degree Adaptability and compatibility The calculation of each has its own rules, and the matching degree The calculation is based on a mapping table between the gap between stimulus intensity grades and pain tolerance threshold grades, and the fit is... The calculation is based on the degree of consistency between the treatment frequency and the recommended recovery cycle based on the fatigue recovery rate, and the degree of fit. The calculations are based on a database of correspondences between different acupuncture techniques and autonomic nerve response sensitivity categories, and the calculation results of these sub-items are all normalized values. In some embodiments, the process of generating and ranking applicability scores combines overall consistency with literature efficacy levels. The system retrieves efficacy level information recorded in the original literature for each candidate treatment plan, such as descriptions of "significantly effective" or "effective," and converts them into grade scores. The calculated overall consistency and literature efficacy level scores are combined according to new weights to generate a final applicability score for each plan for the current patient. All candidate treatment plans are sorted in descending order according to the final applicability score, and the highest-ranked plan and its complete parameters are output as a preliminary acupuncture treatment prescription.

[0042] It is understandable that building a standardized acupuncture treatment knowledge base is an ongoing process. Natural language processing technology needs to handle inconsistent terminology in different documents. For example, it needs to map "flat needling" and "transverse needling" to the standard technique "skin-to-skin needling," and unify "once a day" and "qd" to the standard frequency "once a day." The establishment of a set of standard acupoint schemes ensures the consistency of subsequent matching and calculation. Optionally, when screening candidate acupoint schemes with relevance higher than a preset threshold, various search strategies can be adopted. These strategies include exact matching based on disease diagnosis names, fuzzy matching based on symptom keywords, and semantic matching based on TCM syndrome terminology. The preset threshold can be dynamically adjusted according to clinical needs to control the size and quality of the candidate treatment scheme set. The candidate treatment scheme set serves as an intermediary bridge connecting massive amounts of literature knowledge with individual patient conditions.

[0043] In one embodiment of the present invention, acupuncture clinical literature data and patients' multi-dimensional physiological state indicators are fused. The patients' multi-dimensional physiological state indicators are given high weight, while the acupuncture clinical literature data is given low weight. The weight allocation ratio is determined through historical treatment effect verification to ensure that individual patient data dominates the fusion result. The preliminary acupuncture treatment prescription is converted into a vector representation that can be processed by a deep learning model. The vector includes acupoint codes, stimulation intensity, needle retention time, and treatment frequency parameters. The acupoint codes adopt the internationally accepted acupoint standard coding system, stimulation intensity is quantified and graded from 0 to 5 points, needle retention time is directly quantified in minutes, and treatment frequency is quantified by the number of treatments per week. The trained deep learning model is invoked to process the vector. The deep learning model automatically analyzes the interaction relationships between different acupoints in the prescription, analyzes the potential treatment patterns under complex parameter combinations, and outputs the contribution weight of each acupoint and each treatment parameter to the expected therapeutic effect. The contribution weight is represented by a value between 0 and 1, with higher values ​​indicating greater influence on the therapeutic effect. Based on contribution weights, the system compares successful treatment cases of similar patient groups in the knowledge base, optimizes parameter settings with low contribution, strengthens acupoints and parameter combinations with high contribution, adjusts specific operational details in the prescription, and generates an optimized acupuncture treatment prescription. The optimization process strictly follows the contribution weight ranking results and does not introduce additional empirical adjustment factors.

[0044] In practice, taking a patient with primary dysmenorrhea as an example, the patient's multidimensional physiological state indicators include symptom correlation, rate of change of signs, and range of fluctuation of physiological parameters. Symptom correlation is the quantitative value of the correlation between menstrual abdominal pain and lumbosacral distension; rate of change of signs is the rate of change of prostaglandin levels before and after menstruation; and range of fluctuation of physiological parameters is the normal fluctuation range of heart rate and blood pressure during menstruation. Acupuncture clinical literature data covers acupuncture treatment literature related to primary dysmenorrhea published in core journals in recent years, including different acupoint combinations, treatment parameters, and corresponding efficacy descriptions. During the fusion process, the patient's multidimensional physiological state indicators were assigned a weight of 0.7, and the acupuncture clinical literature data were assigned a weight of 0.3. The data fusion was completed through weighted summation. The initial acupuncture treatment prescription includes the recommended acupoint combination of "Guanyuan," "Sanyinjiao," and "Taichong." The initial treatment parameters are a stimulation intensity of 3 points, a needle retention time of 20 minutes, and a treatment frequency of 3 times per week. When converting the initial acupuncture treatment prescription into a vector representation, "Guanyuan" is encoded as CV4, "Sanyinjiao" as SP6, and "Taichong" as LR3. The stimulation intensity of 3 points is quantified as a value of 3, the needle retention time of 20 minutes as a value of 20, and the treatment frequency of 3 times per week as a value of 3, forming the vector [CV4,SP6,LR3,3,20,3]. A trained deep learning model is then used to process this vector. The deep learning model outputs the contribution weights of the "Guanyuan" acupoint as 0.22, "Sanyinjiao" as 0.28, "Taichong" as 0.15, the stimulation intensity parameter as 0.18, the needle retention time parameter as 0.12, and the treatment frequency parameter as 0.05. Comparing successful treatment cases of similar patient groups in the knowledge base, the successful cases showed that for this type of primary dysmenorrhea, adding the "Zusanli" acupoint (coded ST36) and extending the needle retention time to 25 minutes resulted in better efficacy feedback. Therefore, the optimization process retained the "Sanyinjiao" and "Guanyuan" acupoints with high contribution weights, added the "Zusanli" acupoint, adjusted the needle retention time from 20 minutes to 25 minutes, maintained the stimulation intensity of 3 points and the treatment frequency of 3 times per week, and generated an optimized acupuncture treatment prescription. The optimized acupuncture treatment prescription includes the optimized acupoint combination of "Guanyuan", "Sanyinjiao" and "Zusanli", and the optimized stimulation parameters are stimulation intensity of 3 points, needle retention time of 25 minutes and treatment frequency of 3 times per week. The efficacy contribution evaluation of each parameter is "Sanyinjiao" 0.28, "Guanyuan" 0.22, "Zusanli" 0.20, stimulation intensity 0.18, needle retention time 0.10, and treatment frequency 0.02.

[0045] The construction of a deep learning model requires the collection of historical acupuncture treatment case data. This data includes patient physiological indicators, the acupoint combinations used, treatment parameters, and corresponding efficacy evaluation results. Patient physiological indicators cover various physiological parameter values ​​related to acupuncture treatment; the acupoint combinations used are clinically applied pairings; treatment parameters include specific operational data such as stimulation intensity, needle retention time, and treatment frequency; and efficacy evaluation results describe the symptom improvement after treatment. The historical acupuncture treatment case data is preprocessed to transform unstructured efficacy evaluation results into structured labels. These labels are divided into four levels based on the degree of symptom improvement and assigned a label value of 1-4. Acupoint names and treatment parameters are standardized and encoded to eliminate differences in recording methods among different medical institutions. A neural network model is then constructed using a three-layer fully connected structure. The input layer dimension matches the dimension of the standardized encoded acupoint combinations and treatment parameters. The number of hidden layer nodes is set to twice the input layer dimension. The output layer is configured for multi-class probability prediction of the efficacy evaluation results, with four nodes corresponding to the four efficacy levels. The neural network model is trained using preprocessed historical acupuncture treatment case data. The model weights are adjusted through backpropagation until the model's classification accuracy for therapeutic effects reaches a preset threshold, resulting in a trained deep learning model. Cross-validation is used during training to ensure the model's generalization ability.

[0046] In some embodiments, historical acupuncture treatment case data are derived from clinical records of the acupuncture departments of multiple tertiary hospitals over the past 5 years, totaling 8,000 valid cases. These cases cover common acupuncture treatment indications, and each case includes complete records of patient physiological indicators, acupoint combinations, treatment parameters, and efficacy evaluation results. During the preprocessing, unstructured efficacy evaluation results such as "symptoms completely disappeared" correspond to a label value of 4, "symptoms significantly improved" correspond to a label value of 3, "symptoms somewhat improved" correspond to a label value of 2, and "symptoms not improved" correspond to a label value of 1. The standardized coding of acupoint names strictly follows the national standard "Names and Locations of Acupoints." When standardizing treatment parameters, stimulation intensity is uniformly quantified into a score of 0-5, needle retention time is uniformly converted to minutes, and treatment frequency is uniformly converted to weekly treatments. The constructed neural network model has an input layer dimension of 12, corresponding to 6 acupoint encoding dimensions and 6 treatment parameter dimensions, and 24 hidden layer nodes. The activation function is the ReLU function, and the output layer uses the Softmax function to achieve multi-class probability prediction. During training, the learning rate is set to 0.001, the number of iterations is set to 1000, and the preset threshold is set to 85%. When the model's classification accuracy for therapeutic effects is stable above 85% for 10 consecutive iterations, training is stopped and the model parameters are saved.

[0047] It is understandable that the training effect of a deep learning model directly affects the accuracy of parameter contribution weight evaluation. The quantity and quality of historical acupuncture treatment case data are the foundation of model training. A sufficient number of cases ensures that the model learns treatment patterns in different scenarios, and complete case information can prevent bias in model training. Standardized encoding in the preprocessing process is key to eliminating data heterogeneity. Unified encoding rules enable the model to correctly identify the same acupoints and parameters in different cases. The transformation of structured labels can convert vague descriptions of efficacy into quantitative data that the model can process, providing a clear target output for model training. The structural design of the neural network model needs to be combined with the dimensions of data features. A reasonable number of hidden layer nodes can balance the model's fitting ability and generalization ability. The selection of activation functions can enhance the model's ability to learn complex nonlinear relationships. The backpropagation algorithm continuously adjusts the weights to make the model's prediction results gradually approach the real efficacy. The setting of preset thresholds can ensure that the model achieves practical prediction accuracy.

[0048] Optionally, during the preprocessing of historical acupuncture treatment case data, for cases lacking a small amount of non-critical data, imputation is performed using the mean of similar cases. Cases lacking critical data or with contradictory data are directly discarded to ensure the quality of the preprocessed data. During the neural network model training process, stochastic gradient descent is used to optimize the loss function, with cross-entropy loss function selected. The model training error is measured by calculating the difference between the model's predicted probability and the true label. After each iteration, the classification accuracy of the training set and validation set is calculated. When the validation set accuracy shows a continuous decline, an early stopping strategy is adopted to prevent model overfitting (see Table 1).

[0049] Table 1: Comparison Table of Standardized Processing of Historical Acupuncture Treatment Case Data Understandably, the standardized processing checklist provides a unified basis for the preprocessing of historical acupuncture treatment case data, ensuring that data from different sources and recording methods can be converted into a unified format, facilitating model reading and learning. Standardized coding of acupoint names using national standards ensures industry-wide compatibility; quantitative grading of stimulation intensity transforms qualitative descriptions into quantitative data; standardized processing of needle retention time and treatment frequency eliminates differences in units and expressions; and labeling of efficacy evaluation provides clear supervisory signals for model training.

[0050] Optionally, during vector transformation, for combinations containing multiple acupoints, they are sorted by importance in the prescription and then encoded sequentially to ensure consistency in the encoding order of acupoints in the vector, avoiding model misjudgment due to different orders. The contribution weights output by the deep learning model can be displayed through a visual heatmap, intuitively presenting the contribution degree of each acupoint and parameter, facilitating medical staff's understanding of the model optimization logic. The visualization is only for auxiliary understanding and does not participate in the prescription optimization calculation process.

[0051] During model training, weight adjustments follow the formula below: in, This indicates the adjusted model weights. This indicates the model weights before adjustment. Indicates the learning rate. Represents the loss function. This represents the partial derivative of the loss function with respect to the unadjusted model weights. Learning rate. This is used to control the magnitude of each weight adjustment, avoiding excessively large adjustments that cause model training oscillations or excessively small adjustments that lead to low training efficiency. The loss function... Partial derivatives are used to measure the difference between model predictions and actual efficacy labels. Used to indicate the direction of weight adjustment, causing the model weights to be adjusted in a direction that reduces the value of the loss function.

[0052] In some embodiments, the learning rate A dynamic adjustment strategy is adopted. Initially, the learning rate is set to 0.001. As the number of training iterations increases, when the rate of decrease of the loss function falls below the preset value, the learning rate is adjusted to 0.0005 to ensure stable convergence of the model in the later stages of training. Loss function The cross-entropy loss function is chosen, which is calculated based on the multi-class probability output by the model and the one-hot encoding of the true efficacy label. It can effectively measure the prediction error in classification tasks, and its partial derivatives are also used. By using a chain rule to calculate layer by layer, the algorithm propagates backward from the output layer to the input layer, ensuring that each weight can be adjusted in a targeted manner.

[0053] See Figure 4In the efficacy correlation analysis of acupuncture treatment plans, the heatmap visually presents the linear correlation between core treatment parameters such as stimulation intensity, needle retention time, treatment frequency, and total treatment course, and structured efficacy labels. Specifically, within the matrix formed by each parameter and efficacy label, the numerical values ​​represent Pearson correlation coefficients, and the color gradient corresponds to the correlation strength (dark red indicates a strong positive correlation, and dark blue indicates a strong negative correlation). The graph shows that the absolute values ​​of the correlation coefficients between each treatment parameter are generally below 0.15, indicating a weak linear correlation between parameters. The correlation coefficient between the total treatment course and the structured efficacy labels is 0.18, making it the most significant indicator of efficacy correlation among all parameters, but overall, it remains at a weak correlation level. This result suggests that the linear correlation between a single treatment parameter and efficacy is limited, and subsequent optimization needs to consider the synergistic effect of parameter combinations (such as the compatibility patterns analyzed by deep learning models).

[0054] In one embodiment of the present invention, during each acupuncture treatment, the actual needle insertion angle, depth, stimulation technique, and intensity are recorded using specialized equipment. Simultaneously, the patient's immediate subjective feeling scores and local electromyographic changes are collected before and after treatment to obtain single-treatment process data. After the treatment cycle, the patient's self-reported symptom improvement scale score and the doctor's evaluation of the efficacy level are collected. Combined with all the single-treatment process data recorded during the treatment, the mapping relationship between different combinations of treatment parameters and the final degree of symptom improvement is analyzed. The symptom improvement scale scores filled out by the patient after each treatment are summarized, and the final efficacy level given by the doctor based on clinical observation is obtained. The treatment parameter combinations recorded in each single treatment process, including needle insertion angle, depth, stimulation technique, and intensity, are paired with the final degree of symptom improvement obtained after the entire treatment cycle. An association rule mining algorithm is used to analyze all paired data to identify frequent patterns where treatment parameter combinations and high symptom improvement scale scores or high efficacy levels occur simultaneously. The mined frequent patterns are organized to form a parameter efficacy mapping table with treatment parameter combinations as the index and the expected degree of symptom improvement as the value. Based on the parameter efficacy mapping table, the treatment parameter range with the highest positive correlation to symptom improvement and the parameter settings associated with poor efficacy are identified, providing a basis for the dynamic adjustment of subsequent treatment plans.

[0055] In practice, taking a patient diagnosed with cervicobrachial syndrome and receiving optimized acupuncture treatment as an example, the optimized acupuncture treatment prescription includes the acupoints "Fengchi", "Jianjing", and "Tianzong" as well as corresponding optimized stimulation parameters. During each acupuncture treatment, the actual needle insertion angle, depth, stimulation technique, and intensity are recorded using needles equipped with angle and depth sensors and a dedicated electromyography (EMG) recorder. For example, the angle of insertion into the "Jianjing" acupoint is recorded as straight insertion, the depth as 0.8 cun, the stimulation technique as twisting and reinforcing, and the intensity as a torque of 0.3 N·m. At the same time, the patient's immediate subjective feeling score and local EMG changes are collected before and after treatment. The patient's soreness and distension score immediately after receiving acupuncture at the "Tianzong" acupoint and the change in the amplitude of trapezius muscle EMG before and after acupuncture together constitute the data for a single treatment process.

[0056] In some embodiments, after the treatment cycle, the system collects the patient's self-reported symptom improvement scale score and the physician's assessment of the efficacy level. The patient's self-reported symptom improvement scale score comes from a standardized questionnaire that includes dimensions such as pain and range of motion. After ten treatment cycles, the total score of the patient's self-reported symptom improvement scale is 45 points. The physician's efficacy level, assessed based on clinical examination, is "significantly effective." Combining all the data from each individual treatment process recorded during the treatment, the mapping relationship between different combinations of treatment parameters and the final degree of symptom improvement is analyzed. The symptom improvement scale scores filled in by the patient after each treatment are summarized, and the final efficacy level given by the physician based on clinical observation is obtained. The combination of treatment parameters recorded in each individual treatment process is paired with the final degree of symptom improvement obtained after the entire treatment cycle. For example, the parameter combination recorded in the first treatment is paired with the final efficacy level of "significantly effective." This pairing is performed on the data from all ten treatments.

[0057] Understandably, association rule mining algorithms are used to analyze all paired data. These algorithms identify frequent patterns where treatment parameter combinations coexist with high symptom improvement scale scores or high efficacy levels. The mined frequent patterns are then organized into a parameter efficacy mapping table, indexed by treatment parameter combinations and set to the expected degree of symptom improvement. Based on this table, the treatment parameter ranges with the highest positive correlation to symptom improvement, as well as parameter settings associated with poor efficacy, are identified. For example, the parameter efficacy mapping table shows that the parameter combination "insertion at Fengchi acupoint 0.5-0.8 cun combined with twisting manipulation, intensity 0.2-0.4 N·m" frequently appears alongside "significantly effective" results, while the parameter setting "deep insertion at Jianjing acupoint greater than 1.0 cun" rarely appears in high efficacy records. These identification results serve as a basis for dynamic adjustment of subsequent treatment plans.

[0058] In practice, association rule mining algorithms calculate the support for the co-occurrence of specific combinations of treatment parameters and specific therapeutic effects. One method for calculating support uses the following formula: Where: symbol The symbol represents the support level for a specific combination of treatment parameters co-occurring with a specific therapeutic outcome. Represents a combination of treatment parameters, symbol Symbols indicating a therapeutic outcome This indicates the combination of treatment parameters in historical treatment records. With therapeutic effect The number of times they appear together, and the symbol This indicates the total number of treatments involved in the analysis.

[0059] Optionally, the real-time subjective feeling scores collected during each acupuncture treatment can use a visual analog scale or a numerical rating scale. Local electromyography (EMG) change data can be collected at fixed locations around the acupuncture points using surface EMG sensors. The percentage change in the root mean square value of EMG before and after treatment is recorded as part of the local EMG change data. Data from a single treatment session is stored in a structured format and associated with patient identification and treatment timestamps. In some embodiments, the definition of treatment parameter combinations can have different granularities. A treatment parameter combination can be a complete set of parameters for a single acupuncture point or a common parameter feature across multiple acupuncture points. Optionally, the efficacy level assessed by the physician can be divided according to industry-recognized efficacy evaluation standards, such as dividing the efficacy level into four levels: "clinically cured," "significantly effective," "effective," and "ineffective." The final degree of symptom improvement can be expressed in the form of grades or continuous values ​​of scale scores. The method of determining the expected degree of symptom improvement in the parameter efficacy mapping table needs to be consistent with the selected analysis target.

[0060] See Figure 5 In Phase 1 of optimized acupuncture treatment for patients with cervical and shoulder syndrome, the dynamic correlation between symptom improvement scale scores (left ordinate) and the percentage of electromyographic changes (right ordinate) during 10 treatment sessions was simultaneously presented. Specifically, the symptom improvement scale score generally increased with the number of treatments (from 34 to 46 points), reflecting the continuous relief of patients' symptoms; while the percentage of electromyographic changes showed fluctuating characteristics, peaking at treatment sessions 2, 7, and 9 (reaching a maximum of 30%), corresponding to significant changes in local electromyographic activity after a single treatment, and the peak points were temporally correlated with the phased improvement of the symptom improvement scale score. The dual-axis design of this figure enabled parallel visualization of subjective symptom scores and objective electromyographic physiological indicators, providing a direct temporal dimension for analyzing the mapping relationship between treatment parameter combinations and efficacy.

[0061] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0062] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for optimizing acupuncture treatment plans based on medical artificial intelligence, characterized in that, The method includes: Collect patients’ electronic medical record data and wearable device monitoring data, analyze patients’ clinical symptom characteristics and signs change trends, calculate the correlation strength between different symptoms, and obtain patients’ multi-dimensional physiological state indicators. By integrating acupuncture clinical literature data with the patients' multi-dimensional physiological state indicators, and through data structuring and artificial intelligence algorithm modeling, the pertinence of acupoint combinations on the patients' symptoms is evaluated, and a preliminary acupuncture treatment prescription is generated. The preliminary acupuncture treatment prescription is input into a deep learning model. The deep learning model analyzes the combination pattern of acupoints and treatment parameters in the prescription, evaluates the contribution of each acupoint and parameter to the therapeutic effect, compares the effective patterns in the patient's historical treatment data, optimizes the acupoint combination and stimulation parameters in the prescription, and generates an optimized acupuncture treatment prescription. Based on the optimized acupuncture treatment prescription, real-time feedback data from patients is continuously collected during clinical treatment. The correlation between treatment parameters and the degree of improvement in patient symptoms is analyzed, and acupoint stimulation parameters are dynamically adjusted to form a closed-loop optimization process for the acupuncture treatment plan.

2. The method for optimizing acupuncture treatment plans based on medical artificial intelligence according to claim 1, characterized in that, The multidimensional physiological state indicators include symptom correlation, rate of change of signs, and range of fluctuation of physiological parameters. The preliminary acupuncture treatment prescription includes recommended acupoint combinations, initial treatment parameters, and matching degree score with the patient's condition. The optimized acupuncture treatment prescription includes optimized acupoint combinations, optimized stimulation parameters, and efficacy contribution assessment of each parameter. The real-time feedback data includes symptom improvement score, changes in physiological parameters during treatment, and immediate feedback records after treatment.

3. The method for optimizing acupuncture treatment plans based on medical artificial intelligence according to claim 1, characterized in that, The collection of patients' electronic medical record data and wearable device monitoring data, and the analysis of patients' clinical symptom characteristics and trend of changes in signs, specifically include: The system collects patients' electronic medical record data from the hospital information system interface, analyzes the diagnostic records, medical history information, symptom descriptions and various examination and test results in the medical records, extracts the key features and duration of symptoms, calculates the severity level of symptom features, and obtains a structured symptom feature set. It synchronously receives physiological data continuously monitored by wearable devices, analyzes the change curves of physiological parameters over time, identifies the time periods and amplitudes of abnormal parameter fluctuations, and performs time correlation analysis between abnormal fluctuations and patients' symptoms and complaints to obtain the temporal change characteristics of vital signs. By integrating the structured symptom feature set with the temporal change features of the physical signs, the co-occurrence probability between different symptoms and specific abnormal physical signs is analyzed, the combined symptom and physical sign index is calculated, the overall condition of the patient is comprehensively assessed, and multi-dimensional physiological state indicators of the patient are generated.

4. The method for optimizing acupuncture treatment plans based on medical artificial intelligence according to claim 1, characterized in that, The integrated acupuncture clinical literature data and the patient's multi-dimensional physiological state indicators are used to model the data through data structuring and artificial intelligence algorithms. This assesses the relevance of acupoint combinations to the patient's symptoms and generates a preliminary acupuncture treatment prescription, specifically including: By using natural language processing technology to process acupuncture clinical literature in batches, we can identify and extract data on disease names, acupoints used, acupuncture techniques, treatment frequency and course of treatment recorded in the literature. We can also standardize and structure the data to build a knowledge base and obtain a set of standard acupoint schemes. The patient's multidimensional physiological state indicators are matched with the standard acupoint scheme set. Based on the disease diagnosis and symptom characteristics, the knowledge base is searched to select candidate acupoint schemes with a relevance higher than a preset threshold. The complete treatment parameters of each scheme are extracted to obtain a candidate treatment scheme set. Based on the candidate treatment plan set, the matching of the treatment parameters of each plan with the patient's specific physiological indicators is calculated by artificial intelligence algorithm modeling. Combined with the efficacy level information recorded in the literature, an applicability score for each plan is generated for the current patient. After sorting, a preliminary acupuncture treatment prescription is generated.

5. The method for optimizing acupuncture treatment plans based on medical artificial intelligence according to claim 1, characterized in that, The process of inputting the preliminary acupuncture treatment prescription into a deep learning model, which then analyzes the combination patterns of acupoints and treatment parameters in the prescription and evaluates the contribution of each acupoint and parameter to the therapeutic effect, specifically includes: The acupuncture clinical literature data is fused with the patient's multidimensional physiological state indicators, and the patient's multidimensional physiological state indicators are given high weight, while the acupuncture clinical literature data is given low weight. The preliminary acupuncture treatment prescription is converted into a vector representation that can be processed by a deep learning model. The vector includes acupoint encoding, stimulation intensity, needle retention time, and treatment frequency parameters. The trained deep learning model is invoked to process the vector. The deep learning model automatically analyzes the interaction between different acupoints in the prescription, resolves the potential treatment patterns under complex parameter combinations, and outputs the contribution weight of each acupoint and each treatment parameter to the expected therapeutic effect. Based on the aforementioned contribution weights, successful treatment cases of similar patient groups in the knowledge base are compared, parameter settings with low contribution are optimized, acupoints and parameter combinations with high contribution are strengthened, specific operational details in the prescription are adjusted, and an optimized acupuncture treatment prescription is generated.

6. The method for optimizing acupuncture treatment plans based on medical artificial intelligence according to claim 1, characterized in that, Based on the optimized acupuncture treatment prescription, real-time feedback data from patients is continuously collected during clinical treatment to analyze the correlation between treatment parameters and the degree of improvement in patient symptoms. Specifically, this includes: During each acupuncture treatment, the actual needle insertion angle, depth, stimulation technique and intensity are recorded using specialized equipment. At the same time, the patient's real-time subjective feeling score and local electromyographic changes are collected before and after treatment to obtain data for a single treatment process. After the treatment cycle is completed, the patient's self-reported symptom improvement scale score and the doctor's evaluation of the efficacy level are collected. Combined with all the single treatment process data recorded during the treatment, the mapping relationship between different combinations of treatment parameters and the final degree of symptom improvement is analyzed to obtain the parameter efficacy mapping table. Based on the parameter efficacy mapping table, the treatment parameter range with the highest positive correlation to symptom improvement and the parameter settings associated with poor efficacy are identified, providing a basis for the dynamic adjustment of subsequent treatment plans.

7. The method for optimizing acupuncture treatment plans based on medical artificial intelligence according to claim 1, characterized in that, The steps for constructing the deep learning model include: Collect historical acupuncture treatment case data, which includes patients' physiological indicators, the combination of acupoints used, treatment parameters, and corresponding efficacy evaluation results; The historical acupuncture treatment case data were preprocessed to convert unstructured efficacy evaluation results into structured labels, and the acupoint names and treatment parameters were standardized and coded. A neural network model is constructed, wherein the input layer dimension of the neural network model is matched with the dimension of the standardized encoded acupoint combination and treatment parameter features, and the output layer is set to multi-class probability prediction of the efficacy evaluation results. The neural network model is trained using preprocessed historical acupuncture treatment case data. The model weights are adjusted using the backpropagation algorithm until the model's classification accuracy for therapeutic effects reaches a preset threshold, resulting in a trained deep learning model.

8. The method for optimizing acupuncture treatment plans based on medical artificial intelligence according to claim 3, characterized in that, The system synchronously receives physiological data continuously monitored by wearable devices, analyzes the changes in physiological parameters over time, identifies the time periods and amplitudes of abnormal parameter fluctuations, and performs a time correlation analysis between abnormal fluctuations and the patient's symptoms and complaints to obtain the temporal characteristics of vital signs, including: Set a threshold for the normal fluctuation range of physiological parameters and perform real-time scanning of the physiological data stream continuously monitored by wearable devices; When a physiological data point is detected to exceed the normal fluctuation range threshold, the value of the physiological data point, the extent of the exceedance, and the time of occurrence are recorded and marked as an abnormal fluctuation event. Extract the patient's symptom complaints recorded within a preset time window before and after the occurrence of the abnormal fluctuation event; The correlation coefficient between the amplitude of abnormal fluctuation events and the severity described in the symptom complaints is calculated, and abnormal fluctuation events with high correlation coefficients are integrated with symptom complaints to form a temporal change feature of signs.

9. The method for optimizing acupuncture treatment plans based on medical artificial intelligence according to claim 4, characterized in that, The step of calculating the degree of agreement between the treatment parameters of each treatment plan and the patient's specific physiological indicators based on the candidate treatment plan set includes: The treatment parameters of a candidate treatment plan are extracted from the set of candidate treatment plans. The treatment parameters include acupuncture technique, stimulation intensity, and treatment frequency. Specific physiological indicators corresponding to the treatment parameters are extracted from the patient's multidimensional physiological state indicators, including pain tolerance threshold, autonomic nerve response sensitivity, and fatigue recovery rate. The treatment parameters are quantitatively compared with the specific physiological indicators to calculate the matching degree between stimulation intensity and pain tolerance threshold, the compatibility between treatment frequency and fatigue recovery rate, and the fit between acupuncture technique and autonomic nerve response sensitivity. By combining the matching degree, fitness degree, and fit degree, the overall consistency between the treatment parameters of the candidate treatment plan and the patient's specific physiological indicators is calculated by weighted summation, and this process is repeated for all plans in the candidate treatment plan set.

10. The method for optimizing acupuncture treatment plans based on medical artificial intelligence according to claim 6, characterized in that, The patient's self-reported symptom improvement scale scores and physician-assessed efficacy levels were collected, and combined with all single treatment process data recorded during the treatment, the mapping relationship between different combinations of treatment parameters and the final degree of symptom improvement was analyzed to obtain a parameter efficacy mapping table, including: After the treatment cycle is completed, the symptom improvement scale scores filled out by the patients after each treatment are summarized, and the final efficacy level given by the doctor based on clinical observation is obtained. The treatment parameters recorded in each individual treatment process, including needle insertion angle, depth, stimulation technique and intensity, are paired with the final degree of symptom improvement obtained after the entire treatment cycle. Using association rule mining algorithms, we analyzed all paired data to identify frequent patterns where specific combinations of treatment parameters occurred simultaneously with high symptom improvement scale scores or high efficacy levels. The frequently discovered patterns are organized to form a parameter efficacy mapping table with treatment parameter combinations as indexes and expected symptom improvement as values.

Images