System and method for evaluating traditional Chinese medicine diabetes dry eye treatment difference based on artificial intelligence

Abstract

The invention relates to the technical field of artificial intelligence, and discloses a traditional Chinese medicine diabetes dry eye treatment difference evaluation system and method based on artificial intelligence, and the system comprises a four-diagnosis integrated module, a syndrome modeling module, an intelligent decision module, a curative effect feedback module, a knowledge base management module and a self-adaptive learning module. Multi-mode traditional Chinese medicine four-diagnosis data of tongue condition, pulse condition and eye diagnosis and modern physiological parameters of metabolic indexes are fused, a deep learning method is adopted for standardization processing, the problem that traditional Chinese medicine syndrome differentiation depends on subjective experience is solved, traditional Chinese medicine syndrome feature extraction of diabetes dry eye has objectivity and quantifiability, and the traditional Chinese medicine syndrome feature extraction efficiency is improved. The comprehensiveness and the accuracy of the diagnosis basis are improved; and the curative effect feedback module compares a prediction result with revival curative effect data in real time, and the knowledge base management module is linked to carry out case feature matching and strategy optimization, so that continuous optimization and long-term reliability of the treatment decision model are ensured.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to an evaluation system and method for assessing the differences in traditional Chinese medicine treatment of diabetic dry eye based on artificial intelligence. Background Technology

[0002] Diabetes mellitus is a chronic disease characterized by hyperglycemia, caused by absolute or relative insulin deficiency and impaired insulin utilization. Dry eye syndrome is a common and complex ophthalmic disease caused by insufficient tear secretion, excessive evaporation, or abnormal tear composition, which can cause eye discomfort and visual impairment. Diabetic dry eye is a common complication of diabetes, manifested as dryness of the ocular surface, inflammation, and visual impairment.

[0003] Currently, the diagnosis and treatment of diabetic dry eye in traditional Chinese medicine mainly relies on the subjective interpretation of information from the four diagnostic methods by physicians based on their experience. There is a lack of a standardized fusion mechanism for multi-source heterogeneous data such as tongue appearance, pulse appearance, and eye diagnosis, which leads to deviations in the quantitative analysis of core syndrome elements such as yin deficiency and dryness-heat, qi deficiency and blood stasis.

[0004] Therefore, this paper proposes an AI-based evaluation system and method for assessing the differences in TCM treatment of diabetic dry eye to address the aforementioned issues. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an AI-based evaluation system and method for assessing the differences in TCM treatment of diabetic dry eye. This solves the problems mentioned in the background technology, such as the lack of a standardized fusion mechanism for multi-source heterogeneous data including tongue appearance, pulse appearance, and eye diagnosis, and the bias in the quantitative analysis of core syndrome elements like Yin deficiency and dryness-heat, and Qi deficiency and blood stasis.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an evaluation system and method for assessing the differences in TCM treatment of diabetic dry eye based on artificial intelligence, wherein the method includes the following steps: S1. Collect diagnostic data, physiological indicators, and treatment history data of diabetic dry eye patients. The diagnostic data includes tongue appearance, pulse appearance, eye diagnosis characteristics, and chief complaints. S2. Perform multimodal fusion preprocessing on the collected data, including tongue image region segmentation, pulse waveform denoising, and symptom semantic vectorization conversion, to generate a standardized TCM feature dataset; S3. Based on the standardized dataset, the syndrome element decomposition model is used to extract the quantitative features of three core syndromes: Yin deficiency and dryness-heat, Qi deficiency and blood stasis, and Liver and kidney deficiency. S4. Construct a treatment response correlation model to analyze the regulatory patterns of different Chinese medicine prescriptions on syndrome elements; S5. Establish an individual difference assessment model to quantify the influence coefficient of patients' metabolic indicators and microcirculation status on drug absorption. S6. Input the quantified characteristics of syndromes, the output of the treatment response correlation model, and the individual difference assessment results into the deep learning decision model; S7. Use a genetic algorithm to optimize the weight allocation strategy of the decision-making model; S8. Output individualized treatment suggestions, including optimized Chinese medicine prescriptions, acupuncture point combinations, and treatment cycle adjustment strategies; S9. Based on the follow-up visit data, compare the actual therapeutic effect with the predicted results and dynamically update the decision model parameters.

[0007] Preferably, the data collected in S1 from diabetic dry eye patients include: diagnostic data, physiological index data, and treatment history data. S11. Data on tongue color, coating, and crack distribution are obtained using a high-resolution tongue image acquisition instrument, and the tongue tip, tongue side, and tongue middle are divided into three diagnostic areas. S12. Use a multi-channel pulse imager to collect pulse waveform data at the cun, guan, and chi positions, and extract dynamic characteristics of pulse rate, pulse strength, and pulse momentum. S13. Meibomian gland morphology, tear film breakup time, and corneal staining score were obtained using a slit-lamp microscopy system. S14. Integrate metabolic indicator data, including glycated hemoglobin levels, insulin resistance index, and concentrations of oxidative stress markers; S15. Store all data in a structured manner and establish a spatiotemporal correlation index.

[0008] Preferably, the multimodal fusion preprocessing in S2 includes: S21. Use U-Net network to segment the tongue image region and remove the interference region of lips and teeth; S22. Apply wavelet thresholding to denoise the pulse waveform while preserving the main wave characteristics of the pulse position. S23. Convert the chief symptom complaint into a syndrome semantic vector using the BERT model; S24. Perform multi-scale normalization on the visual features to unify the dimensions of corneal staining and meibomian gland scoring. Use Min-Max scaling to map scores from different sources to the [0,1] interval. The formula is as follows: ; in These are the normalized eigenvalues. The original score. , These are the minimum and maximum possible values ​​defined for this rating scale, respectively.

[0009] Preferably, the syndrome element decomposition model in S3 includes: S31. Establish a tongue appearance-syndrome mapping model and quantify the correlation coefficient between the purplish-dark tongue characteristics and blood stasis syndrome. S32. Construct a pulse-organ correlation model to identify the correspondence between the characteristics of the wiry pulse and liver yin deficiency; S33, Integrating visual diagnosis features with the semantic vector of the chief complaint; S34. Output a feature matrix that includes syndrome weights and severity levels.

[0010] Preferably, the construction of the treatment response correlation model in S4 specifically includes: S41. Establish a TCM-syndrome regulation matrix and quantify the regulatory efficacy coefficients of Rehmannia glutinosa and Ophiopogon japonicus on Yin deficiency syndrome; S42. Analyze the time-effect curves of acupuncture point combinations on tear secretion and optimize the combination of stimulation parameters; S43. Use the random forest algorithm to evaluate the interaction between traditional Chinese medicine and Western medicine, and avoid the risk of drug antagonism. S44. Generate decision reference indicators that include treatment sensitivity and contraindication warnings.

[0011] Preferably, the establishment of the individual difference assessment model in S5 includes: S51. Establish a capillary perfusion assessment model based on microcirculation imaging data to predict drug penetration efficiency; S52. Quantify the effect of liver CYP450 enzyme activity on drug metabolism through metabolomics analysis; S53. Construct an age-body mass index correction model and optimize the treatment dose conversion formula; The model is a linear correction function: Age BMI ; in To adjust the dosage, For standard dose, , , The correction coefficient is obtained through regression analysis of historical data. Age represents the patient's age, and BMI represents the body mass index.

[0012] S54. Output individualized dosing adjustment coefficients and treatment cycle correction suggestions.

[0013] Preferably, step S6, which involves inputting the syndrome quantification features, treatment response correlation model output, and individual difference assessment results into the deep learning decision model, includes the following steps: S61. Perform time-sensitive weighting on the quantitative features of syndromes, and dynamically adjust the feature weights according to the interval between the collection time of the syndrome features and the current time, assigning higher weight coefficients to features collected recently. S62. Establish a treatment response priority queue and arrange the output items of the treatment response association model in descending order according to the adjustment efficacy value of the traditional Chinese medicine prescription. S63. Adjust the confidence level of individual difference assessment results. When the confidence level of microcirculation data is lower than the threshold, automatically reduce its decision weight. S64. The weighted syndrome characteristics, the ranked treatment response items, and the corrected individual difference data are synchronously input into the input layer of the decision model.

[0014] Preferably, the weight allocation strategy in S7 includes: S71. Initializing the population includes three sets of hyperparameters: network depth, convolutional kernel size, and number of attention heads. S72. Set the fitness function as a weighted sum of the prediction bias of treatment effectiveness and the incidence of side effects; S73. Employ an elite retention strategy to iteratively optimize the model topology; S74. When a new treatment method is detected, the topology reconstruction mechanism is triggered.

[0015] Preferably, the individualized treatment recommendations output in S8 include: S81. Output the Chinese medicine prescription optimization command to adjust the ratio of principal drug and the synergistic scheme of drug pairs; S82. Generate acupuncture plan dynamic adjustment strategy, and change the main acupoint and acupoint combination according to the transformation of syndrome; S83. Provides a flexible adjustment range for the treatment cycle and marks key efficacy evaluation time nodes; S84. Visualize the relationship between the evolution trajectory of symptoms and treatment interventions.

[0016] Preferably, the system includes: The four diagnostic methods integrated module acquires tongue diagnosis features through the tongue image acquisition unit, extracts dynamic pulse diagnosis data through the pulse perception unit, quantifies microscopic eye lesions through the eye diagnosis analysis unit, and integrates the symptom input unit to generate the original dataset. The syndrome modeling module receives the raw dataset, decomposes the core syndrome features through the syndrome element analysis unit, constructs the Chinese medicine-syndrome interaction network using the treatment response modeling unit, and outputs a feature vector set through the individual difference assessment unit. The intelligent decision-making module receives a set of feature vectors, builds a decision model architecture through a multimodal graph network construction unit, dynamically adjusts network parameters through a genetic optimization unit, and outputs individualized suggestions using a treatment plan generation unit. The efficacy feedback module receives follow-up efficacy data, calculates prediction errors through the deviation analysis unit, generates optimization instructions through the model iteration unit, and displays error attribution reports through the decision tracing unit. The knowledge base management module stores a database of classic prescriptions and medical case data of famous doctors. It calls up similar cases through the case feature matching unit and feeds the results back to the intelligent decision-making module through the strategy optimization unit. The adaptive learning module integrates the output data from the efficacy feedback module, identifies optimization directions through the model defect diagnosis unit, and optimizes the acquisition frequency of the four diagnostic parameters by the data acquisition strategy control unit.

[0017] Beneficial effects Compared with existing technologies, this invention provides an artificial intelligence-based evaluation system and method for assessing the differences in TCM treatment of diabetic dry eye, which has the following beneficial effects: 1. In this invention, by integrating multimodal TCM diagnostic data such as tongue appearance, pulse appearance, and eye diagnosis with modern physiological parameters of metabolic indicators, and by using deep learning methods for standardization processing, the problem of traditional TCM syndrome differentiation relying on subjective experience is solved, making the extraction of TCM syndrome features of diabetic dry eye objective and quantifiable, and improving the comprehensiveness and accuracy of diagnostic basis.

[0018] 2. In this invention, by constructing a syndrome element decomposition model and a treatment response correlation model, the regulatory patterns of traditional Chinese medicine prescriptions and acupuncture programs on core syndromes such as yin deficiency and dryness-heat and qi deficiency and blood stasis are quantitatively analyzed. Combined with individual metabolic differences, the drug absorption efficiency is evaluated, so that the treatment plan design can accurately match the patient's syndrome characteristics and individual physiological state, thereby improving the pertinence and effectiveness of the treatment.

[0019] 3. In this invention, the efficacy feedback module compares the predicted results with the follow-up efficacy data in real time, links the knowledge base management module to perform case feature matching and strategy optimization, and uses the adaptive learning module to continuously correct the model parameters, so that the system can iteratively upgrade the diagnosis and treatment rules based on the actual clinical effect, ensuring the continuous optimization and long-term reliability of the treatment decision model. Attached Figure Description

[0020] Figure 1 This is a flowchart of the evaluation method for the differences in TCM treatment of diabetic dry eye based on artificial intelligence, as described in this invention. Figure 2 This is a diagram illustrating the architecture of the AI-based TCM-based assessment system for evaluating differences in dry eye treatment for diabetes. Detailed Implementation

[0021] 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.

[0022] Specific Implementation: An AI-based evaluation system and method for assessing differences in TCM treatment of diabetic dry eye, the method comprising the following steps: S1. Collect diagnostic data, physiological indicators, and treatment history data of diabetic dry eye patients. The diagnostic data includes tongue appearance, pulse appearance, eye diagnosis characteristics, and chief complaints. S2. Perform multimodal fusion preprocessing on the collected data, including tongue image region segmentation, pulse waveform denoising, and symptom semantic vectorization conversion, to generate a standardized TCM feature dataset; S3. Based on the standardized dataset, the syndrome element decomposition model is used to extract the quantitative features of three core syndromes: Yin deficiency and dryness-heat, Qi deficiency and blood stasis, and Liver and kidney deficiency. S4. Construct a treatment response correlation model to analyze the regulatory patterns of different Chinese medicine prescriptions on syndrome elements; S5. Establish an individual difference assessment model to quantify the influence coefficient of patients' metabolic indicators and microcirculation status on drug absorption. S6. Input the quantified characteristics of syndromes, the output of the treatment response correlation model, and the individual difference assessment results into the deep learning decision model; S7. Use a genetic algorithm to optimize the weight allocation strategy of the decision-making model; S8. Output individualized treatment suggestions, including optimized Chinese medicine prescriptions, acupuncture point combinations, and treatment cycle adjustment strategies; S9. Based on the follow-up visit data, compare the actual therapeutic effect with the predicted results and dynamically update the decision model parameters.

[0023] S1 collects diagnostic data, physiological indicators, and treatment history data from diabetic dry eye patients, including: S11. Data on tongue color, coating, and crack distribution are obtained using a high-resolution tongue image acquisition instrument, and the tongue tip, tongue side, and tongue middle are divided into three diagnostic areas. S12. Use a multi-channel pulse imager to collect pulse waveform data at the cun, guan, and chi positions, and extract dynamic characteristics of pulse rate, pulse strength, and pulse momentum. S13. Meibomian gland morphology, tear film breakup time, and corneal staining score were obtained using a slit-lamp microscopy system. S14. Integrate metabolic indicator data, including glycated hemoglobin levels, insulin resistance index, and concentrations of oxidative stress markers; S15. Store all data in a structured manner and establish a spatiotemporal correlation index.

[0024] The multimodal fusion preprocessing in S2 includes: S21. Use U-Net network to segment the tongue image region and remove the interference region of lips and teeth; The training objective of the U-Net network is to minimize the composite loss function between the predicted segmentation map and the expert-annotated gold standard, and its formula is: ; in This represents the total loss value during model training. The Dice loss function, The binary cross-entropy loss function is... To balance the weight hyperparameters; S22. Apply wavelet thresholding to denoise the pulse waveform while preserving the main wave characteristics of the pulse position. The specific steps are as follows: The original pulse signal was decomposed into 5 levels using the "sym8" wavelet basis to obtain the detail coefficients of each level. and approximation coefficients ; Secondly, an adaptive threshold function is applied to shrink the detail coefficients: sign ; Where the threshold , For the first Noise standard deviation estimation of layer coefficients, For signal length, The threshold for the j-th layer, Let j be the detail coefficients of the j-th layer. This represents the value of the detail coefficient at layer j after thresholding. Finally, wavelet reconstruction is performed using the processed coefficients to obtain the denoised pulse signal. S23. Convert the chief symptom complaint into a syndrome semantic vector using the BERT model; The specific implementation steps are as follows: Using a pre-trained Chinese BERT model as the encoder, the patient's chief complaint text is input into the model. The hidden state corresponding to the [CLS] label of the last Transformer layer is taken as the semantic vector representation of the whole sentence. Its dimension is 768. This vector captures the deep semantic information of the symptoms and is suitable for subsequent syndrome classification. S24. Perform multi-scale normalization on the visual features to unify the dimensions of corneal staining and meibomian gland scoring. Use Min-Max scaling to map scores from different sources to the [0,1] interval. The formula is as follows:

[0025] in These are the normalized eigenvalues. The original score. , These are the minimum and maximum possible values ​​defined for this rating scale, respectively.

[0026] The S3 syndrome element decomposition model includes: S31. Establish a tongue appearance-syndrome mapping model and quantify the correlation coefficient between the purplish-dark tongue characteristics and blood stasis syndrome. This model is a logistic regression classifier, and its output is the probability of the syndrome: Blood stasis syndrome ; Blood stasis syndrome Given a tongue image feature vector The probability of later being diagnosed as blood stasis syndrome was predicted. This is the feature vector extracted from the segmented tongue image. This is the intercept term of the logistic regression model. It is the dot product of vectors; S32. Construct a pulse-organ correlation model to identify the correspondence between the characteristics of the wiry pulse and liver yin deficiency; This model uses support vector machines for classification, and its decision function is: ; Among them, the input pulse feature vector Prediction categories, The input feature vector, For the Lagrange multipliers corresponding to the support vectors, For the true label of training sample i, For the bias term of the decision function, It is a radial basis kernel function; S33. Integrating visual diagnosis features with the semantic vector of the chief complaint enhances the accuracy of identifying heat syndrome; S34. Output a feature matrix that includes syndrome weights and severity levels.

[0027] The treatment response correlation model is constructed in S4, specifically including: S41. Establish a TCM-syndrome regulation matrix and quantify the regulatory efficacy coefficients of Rehmannia glutinosa and Ophiopogon japonicus on Yin deficiency syndrome; The TCM-syndrome regulation matrix is ​​as follows: ; matrix elements Indicates the first Chinese herbal medicine pair The moderating strength of each syndrome element; the rows of this matrix correspond to Chinese herbal medicines, and the columns correspond to syndrome elements. S42. Analyze the time-effect curves of acupuncture point combinations on tear secretion and optimize the combination of stimulation parameters; S43. Use the random forest algorithm to evaluate the interaction between traditional Chinese medicine and Western medicine, and avoid the risk of drug antagonism. The random forest model contains 100 decision trees, and each tree is considered during its growth. One characteristic, The total number of features includes the components of traditional Chinese medicine, the dosage of Western medicine, and the patient's metabolic type; the final prediction result is the mode and mean of the outputs of all trees; S44. Generate decision reference indicators that include treatment sensitivity and contraindication warnings.

[0028] The individual difference assessment model established in S5 includes: S51. Establish a capillary perfusion assessment model based on microcirculation imaging data to predict drug penetration efficiency; S52. Quantify the effect of liver CYP450 enzyme activity on drug metabolism through metabolomics analysis; S53. Construct an age-body mass index correction model and optimize the treatment dose conversion formula; The model is a linear correction function: Age BMI ; in To adjust the dosage, For standard dose, , , The correction coefficient is obtained through regression analysis of historical data. Age represents the patient's age, and BMI represents the body mass index.

[0029] S54. Output individualized dosing adjustment coefficients and treatment cycle correction suggestions.

[0030] In S6, the quantitative characteristics of syndromes, the output of the treatment response correlation model, and the results of individual difference assessment are input into the deep learning decision model, including the following steps: S61. Perform time-sensitive weighting on the quantitative features of syndromes, and dynamically adjust the feature weights according to the interval between the collection time of the syndrome features and the current time, assigning higher weight coefficients to features collected recently. S62. Establish a treatment response priority queue and arrange the output items of the treatment response association model in descending order according to the adjustment efficacy value of the traditional Chinese medicine prescription. S63. Adjust the confidence level of individual difference assessment results. When the confidence level of microcirculation data is lower than the threshold, automatically reduce its decision weight. S64. The weighted syndrome characteristics, the ranked treatment response items, and the corrected individual difference data are synchronously input into the input layer of the decision model.

[0031] The weight allocation strategies in S7 include: S71. Initializing the population includes three sets of hyperparameters: network depth, convolutional kernel size, and number of attention heads. S72. Set the fitness function as a weighted sum of the prediction bias of treatment effectiveness and the incidence of side effects: ; in For comprehensive fitness score, To account for the bias in predicting therapeutic efficacy, As the dominant factor in therapeutic efficacy, As the primary factor for safety, For the incidence of side effects, As an economic factor, Treatment cost index; S73. An elite retention strategy is adopted to iteratively optimize the model topology. In each generation of evolution, the top 10% of individuals with the highest fitness are directly selected and unconditionally retained in the next generation of the population to avoid the loss of excellent genes. S74. When a new treatment method is detected, the topology reconstruction mechanism is triggered.

[0032] S8 outputs individualized treatment recommendations, including: S81. Output the Chinese medicine prescription optimization command to adjust the ratio of principal drug and the synergistic scheme of drug pairs; S82. Generate acupuncture plan dynamic adjustment strategy, and change the main acupoint and acupoint combination according to the transformation of syndrome; S83. Provides a flexible adjustment range for the treatment cycle and marks key efficacy evaluation time nodes; S84. Visualize the relationship between the evolution trajectory of symptoms and treatment interventions.

[0033] The system includes: The four diagnostic methods integrated module acquires tongue diagnosis features through the tongue image acquisition unit, extracts dynamic pulse diagnosis data through the pulse perception unit, quantifies microscopic eye lesions through the eye diagnosis analysis unit, and integrates the symptom input unit to generate the original dataset. The syndrome modeling module receives the raw dataset, decomposes the core syndrome features through the syndrome element analysis unit, constructs the Chinese medicine-syndrome interaction network using the treatment response modeling unit, and outputs a feature vector set through the individual difference assessment unit. The intelligent decision-making module receives a set of feature vectors, builds a decision model architecture through a multimodal graph network construction unit, dynamically adjusts network parameters through a genetic optimization unit, and outputs individualized suggestions using a treatment plan generation unit. The efficacy feedback module receives follow-up efficacy data, calculates prediction errors through the deviation analysis unit, generates optimization instructions through the model iteration unit, and displays error attribution reports through the decision tracing unit. The knowledge base management module stores a database of classic prescriptions and medical case data of famous doctors. It calls up similar cases through the case feature matching unit and feeds the results back to the intelligent decision-making module through the strategy optimization unit. The adaptive learning module integrates the output data from the efficacy feedback module, identifies optimization directions through the model defect diagnosis unit, and optimizes the acquisition frequency of the four diagnostic parameters by the data acquisition strategy control unit.

[0034] The operation steps of this system and method are as follows: Step 1: Multimodal Data Acquisition and Standardization Processing The system first collects dynamic waveforms of the patient's pulse and microscopic indicators of the eyes, while integrating metabolic data and structured symptom complaints. It then uses a U-Net network to segment the tongue region, performs wavelet thresholding for denoising the pulse wave, and employs a BERT model to transform symptom text into syndrome semantic vectors, ultimately achieving dimensional unification and spatiotemporal correlation indexing of multi-source heterogeneous data.

[0035] Step 2: Quantification of core syndromes and modeling of treatment response A three-tiered analysis was conducted on the standardized dataset: a syndrome element decomposition model was used to identify the severity weights of three core syndromes: Yin deficiency with dryness and heat, Qi deficiency with blood stasis, and Liver and Kidney deficiency, and to establish the mapping relationship between the purplish-dark tongue appearance and blood stasis syndrome, and between the wiry pulse and Liver Yin deficiency; a treatment response association model was used to analyze the regulatory efficacy of classic Chinese medicine prescriptions on syndrome elements, to evaluate the time-effect curve of acupuncture point combinations on tear secretion, and to avoid the risk of interaction antagonism between Chinese and Western medicines using a random forest algorithm; and a body difference assessment model was used to combine microcirculation perfusion data and metabolomics analysis to predict drug penetration efficiency and liver metabolic capacity, and to optimize dose conversion.

[0036] Step 3: Dynamic Decision Making and Personalized Treatment Generation The deep learning decision model receives weighted syndrome features, priority-ranked treatment response items, and confidence-adjusted individual difference data. The output layer generates three-dimensional individualized suggestions, including adjustments to the ratio of principal Chinese medicine herbs, dynamic changes in acupuncture points, and flexible treatment cycle settings, and visualizes the correlation between syndrome evolution and treatment intervention.

[0037] Step 4: Closed-loop optimization and knowledge iteration The efficacy feedback module compares follow-up visit data with predicted results and uses deviation analysis to pinpoint the source of model errors. The knowledge base management module accesses databases of classic prescriptions and renowned physicians' medical records, matching similar case characteristics to optimize decision-making strategies. The adaptive learning module dynamically adjusts the collection frequency of the four diagnostic methods (inspection, auscultation and olfaction, inquiry, and palpation) based on clinical feedback data and triggers model topology reconstruction, forming a continuous evolutionary closed loop.

[0038] 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 a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0039] 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. An evaluation method for the differences in TCM treatment of diabetic dry eye based on artificial intelligence, characterized by: The method includes the following steps: S1. Collect diagnostic data, physiological indicators, and treatment history data of diabetic dry eye patients. The diagnostic data includes tongue appearance, pulse appearance, eye diagnosis characteristics, and chief complaints. S2. Perform multimodal fusion preprocessing on the collected data, including tongue image region segmentation, pulse waveform denoising, and symptom semantic vectorization conversion, to generate a standardized TCM feature dataset; S3. Based on the standardized dataset, the syndrome element decomposition model is used to extract the quantitative features of three core syndromes: Yin deficiency and dryness-heat, Qi deficiency and blood stasis, and Liver and kidney deficiency. S4. Construct a treatment response correlation model to analyze the regulatory patterns of different Chinese medicine prescriptions on syndrome elements; S5. Establish an individual difference assessment model to quantify the influence coefficient of patients' metabolic indicators and microcirculation status on drug absorption. S6. Input the quantified characteristics of syndromes, the output of the treatment response correlation model, and the individual difference assessment results into the deep learning decision model; S7. Use a genetic algorithm to optimize the weight allocation strategy of the decision-making model; S8. Output individualized treatment suggestions, including optimized Chinese medicine prescriptions, acupuncture point combinations, and treatment cycle adjustment strategies; S9. Based on the follow-up visit data, compare the actual therapeutic effect with the predicted results and dynamically update the decision model parameters.

2. The method for evaluating the differences in TCM treatment of diabetic dry eye based on artificial intelligence as described in claim 1, characterized in that: The S1 process collects diagnostic data, physiological index data, and treatment history data from diabetic dry eye patients, including: S11. Data on tongue color, coating, and crack distribution are obtained using a high-resolution tongue image acquisition instrument, and the tongue tip, tongue side, and tongue middle are divided into three diagnostic areas. S12. Use a multi-channel pulse imager to collect pulse waveform data at the cun, guan, and chi positions, and extract dynamic characteristics of pulse rate, pulse strength, and pulse momentum. S13. Meibomian gland morphology, tear film breakup time, and corneal staining score were obtained using a slit-lamp microscopy system. S14. Integrate metabolic indicator data, including glycated hemoglobin levels, insulin resistance index, and concentrations of oxidative stress markers; S15. Store all data in a structured manner and establish a spatiotemporal correlation index.

3. The method for evaluating the differences in TCM treatment of diabetic dry eye based on artificial intelligence according to claim 1, characterized in that: The multimodal fusion preprocessing in S2 includes: S21. Use U-Net network to segment the tongue image region and remove the interference region of lips and teeth; S22. Apply wavelet thresholding to denoise the pulse waveform while preserving the main wave characteristics of the pulse position. S23. Convert the chief symptom complaint into a syndrome semantic vector using the BERT model; S24. Perform multi-scale normalization on the visual features to unify the dimensions of corneal staining and meibomian gland scoring. Use Min-Max scaling to map scores from different sources to the [0,1] interval. The formula is as follows: ； in These are the normalized eigenvalues. The original score. , These are the minimum and maximum possible values ​​defined for this rating scale, respectively.

4. The method for evaluating the differences in TCM treatment of diabetic dry eye based on artificial intelligence as described in claim 1, characterized in that: The syndrome element decomposition model in S3 includes: S31. Establish a tongue appearance-syndrome mapping model and quantify the correlation coefficient between the purplish-dark tongue characteristics and blood stasis syndrome. S32. Construct a pulse-organ correlation model to identify the correspondence between the characteristics of the wiry pulse and liver yin deficiency; S33, Integrating visual diagnosis features with the semantic vector of the chief complaint; S34. Output a feature matrix that includes syndrome weights and severity levels.

5. The method for evaluating the differences in TCM treatment of diabetic dry eye based on artificial intelligence as described in claim 1, characterized in that: The treatment response correlation model constructed in S4 specifically includes: S41. Establish a TCM-syndrome regulation matrix and quantify the regulatory efficacy coefficients of Rehmannia glutinosa and Ophiopogon japonicus on Yin deficiency syndrome; S42. Analyze the time-effect curves of acupuncture point combinations on tear secretion and optimize the combination of stimulation parameters; S43. Use the random forest algorithm to evaluate the interaction between traditional Chinese medicine and Western medicine, and avoid the risk of drug antagonism. S44. Generate decision reference indicators that include treatment sensitivity and contraindication warnings.

6. The method for evaluating the differences in TCM treatment of diabetic dry eye based on artificial intelligence as described in claim 1, characterized in that: The individual difference assessment model established in S5 includes: S51. Establish a capillary perfusion assessment model based on microcirculation imaging data to predict drug penetration efficiency; S52. Quantify the effect of liver CYP450 enzyme activity on drug metabolism through metabolomics analysis; S53. Construct an age-body mass index correction model and optimize the treatment dose conversion formula; The model is a linear correction function: Age BMI ; in To adjust the dosage, For standard dose, , , The correction coefficients are obtained through regression analysis of historical data, where Age is the patient's age and BMI is the body mass index. S54. Output individualized dosing adjustment coefficients and treatment cycle correction suggestions.

7. The method for evaluating the differences in TCM treatment of diabetic dry eye based on artificial intelligence according to claim 1, characterized in that: The S6 step involves inputting the syndrome quantification characteristics, treatment response correlation model output, and individual difference assessment results into the deep learning decision model, including the following steps: S61. Perform time-sensitive weighting on the quantitative features of syndromes, and dynamically adjust the feature weights according to the interval between the collection time of the syndrome features and the current time, assigning higher weight coefficients to features collected recently. S62. Establish a treatment response priority queue and arrange the output items of the treatment response association model in descending order according to the adjustment efficacy value of the traditional Chinese medicine prescription. S63. Adjust the confidence level of individual difference assessment results. When the confidence level of microcirculation data is lower than the threshold, automatically reduce its decision weight. S64. The weighted syndrome characteristics, the ranked treatment response items, and the corrected individual difference data are synchronously input into the input layer of the decision model.

8. The method for evaluating the differences in TCM treatment of diabetic dry eye based on artificial intelligence according to claim 1, characterized in that: The weight allocation strategy in S7 includes: S71. Initializing the population includes three sets of hyperparameters: network depth, convolutional kernel size, and number of attention heads. S72. Set the fitness function as a weighted sum of the prediction bias of treatment effectiveness and the incidence of side effects; S73. Employ an elite retention strategy to iteratively optimize the model topology; S74. When a new treatment method is detected, the topology reconstruction mechanism is triggered.

9. The method for evaluating the differences in TCM treatment of diabetic dry eye based on artificial intelligence according to claim 1, characterized in that: The S8 outputs individualized treatment recommendations, including: S81. Output the command to optimize the Chinese medicine prescription, and adjust the ratio of principal drug and the synergistic scheme of drug pairs. S82. Generate acupuncture plan dynamic adjustment strategy, and change the main acupoint and acupoint combination according to the transformation of syndrome; S83. Provides a flexible adjustment range for the treatment cycle and marks key efficacy evaluation time nodes; S84. Visualize the relationship between the evolution trajectory of symptoms and treatment interventions.

10. An AI-based evaluation system for assessing the differences in TCM treatment of diabetic dry eye, used to implement any one of the AI-based evaluation methods for assessing the differences in TCM treatment of diabetic dry eye as described in claims 1-9, characterized in that: The system includes: The four diagnostic methods integrated module acquires tongue diagnosis features through the tongue image acquisition unit, extracts dynamic pulse diagnosis data through the pulse perception unit, quantifies microscopic eye lesions through the eye diagnosis analysis unit, and integrates the symptom input unit to generate the original dataset. The syndrome modeling module receives the raw dataset, decomposes the core syndrome features through the syndrome element analysis unit, constructs the Chinese medicine-syndrome interaction network using the treatment response modeling unit, and outputs a feature vector set through the individual difference assessment unit. The intelligent decision-making module receives a set of feature vectors, builds a decision model architecture through a multimodal graph network construction unit, dynamically adjusts network parameters through a genetic optimization unit, and outputs individualized suggestions using a treatment plan generation unit. The efficacy feedback module receives follow-up efficacy data, calculates prediction errors through the deviation analysis unit, generates optimization instructions through the model iteration unit, and displays error attribution reports through the decision tracing unit. The knowledge base management module stores a database of classic prescriptions and medical case data of famous doctors. It calls up similar cases through the case feature matching unit and feeds the results back to the intelligent decision-making module through the strategy optimization unit. The adaptive learning module integrates the output data from the efficacy feedback module, identifies optimization directions through the model defect diagnosis unit, and optimizes the acquisition frequency of the four diagnostic parameters by the data acquisition strategy control unit.

Images