Multi-symptom cooperative disease diagnosis system based on fuzzy reasoning

Abstract

The invention provides a multi-symptom cooperative disease diagnosis system based on fuzzy reasoning, and relates to the technical field of disease diagnosis systems. The multi-symptom collaborative disease diagnosis system based on fuzzy reasoning comprises a symptom input module, a three-level fuzzy rule base module, a reasoning calculation module, a diagnosis result output module and a patient information database module. Through a three-level semantic mapping mechanism, fuzzy symptoms such as intermittent abdominal pain and the like are converted into quantized values in a 0-1 interval, and compared with a traditional method, the precision is remarkably improved. A T-norm operator is adopted to dynamically process symptom association, a product operator is used for strong association combination, and a minimum operator is used for weak association, so that the accuracy of association calculation is improved, and the missed diagnosis rate is reduced.

Description

Technical Field

[0001] This invention relates to the field of disease diagnosis system technology, specifically to a multi-symptom collaborative disease diagnosis system based on fuzzy reasoning. Background Technology

[0002] In the field of modern medical diagnosis, computer-aided diagnostic systems have become an important auxiliary tool for clinical diagnosis and treatment. However, existing technologies still have significant limitations when dealing with complex clinical scenarios, mainly in the following four dimensions: I. Insufficient precision in quantifying vague symptoms Existing systems often handle ambiguous symptoms such as intermittent abdominal pain and moderate fatigue using simple binary classifications (e.g., presence / absence) or single-dimensional scoring (e.g., a 1-10 scale), lacking a systematic semantic parsing mechanism. For example, a mainstream diagnostic system uniformly maps intermittent symptoms to a fixed value of 0.5, failing to differentiate between 3 episodes per day and 5 episodes per day; the determination of moderate fatigue relies solely on subjective scoring by doctors, with errors exceeding 40% between different operators. This approach leads to significant discrepancies between the quantitative results of ambiguous symptoms and clinical reality, directly impacting the accuracy of subsequent reasoning.

[0003] II. Rigid Multisymptom Association Analysis Mechanism Traditional systems typically employ linear weighted algorithms when analyzing relationships between multiple symptoms, failing to reflect the nonlinear synergistic effects between symptoms. For instance, in diagnosing diabetes-related abdominal pain, the combined effect of abnormal blood glucose and abdominal pain is not a simple additive result of their individual effects. Clinical data shows that when fasting blood glucose exceeds 7.0 mmol / L and abdominal pain occurs more than four times daily, the association strength is 3.2 times that of a single symptom, while existing systems can only calculate an association strength of 1.5 times. This deficiency leads to insufficient diagnostic sensitivity in complex comorbidity scenarios, resulting in a false negative rate as high as 28%.

[0004] III. Lack of Personalized Correction Mechanism Current diagnostic thresholds are mostly fixed values, failing to consider individual differences such as patient age and underlying diseases. For example, the clinical significance of a 65-year-old diabetic patient presenting with the same blood glucose level of 8.0 mmol / L and abdominal pain symptoms as a 30-year-old healthy person is significantly different; the risk factor for the elderly patient is 2.1 times that of the younger patient, yet current systems will give the same diagnostic result. Furthermore, the impact of underlying diseases is only qualitatively assessed, lacking quantitative correction models, leading to a 35% decrease in diagnostic accuracy for patients with chronic diseases.

[0005] IV. Poor interpretability of results Most systems only output disease names and matching probabilities, without showing the reasoning behind them. Clinicians cannot understand why the diagnosis was reached or which symptoms contributed most, leading to low confidence in the diagnostic results. A survey showed that due to the lack of visual association maps, the doctors' acceptance rate of system diagnostic results was only 41%. Furthermore, the logic behind generating probability values ​​is opaque; for example, the calculation process for a 72% matching probability cannot be traced, making it difficult to meet the interpretability requirements of medical decision-making.

[0006] The aforementioned shortcomings collectively limit the application of existing diagnostic systems in complex clinical scenarios, especially in primary healthcare institutions where the actual usage rate is less than 30%. Therefore, developing a diagnostic system that can accurately quantify fuzzy symptoms, dynamically analyze the correlation of multiple symptoms, enable personalized reasoning, and provide interpretable results has become an urgent technical problem to be solved. Summary of the Invention

[0007] Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a multi-symptom collaborative disease diagnosis system based on fuzzy reasoning, which solves the defects and deficiencies in existing technologies.

[0008] Technical solution To achieve the above objectives, the present invention provides the following technical solution: a multi-symptom collaborative disease diagnosis system based on fuzzy reasoning, the system comprising a symptom input module, a three-level fuzzy rule base module, a reasoning calculation module, a diagnosis result output module, and a patient information database module; The symptom input module is used to receive the patient's symptom information, which includes vague symptoms and specific symptoms, and transmits the symptom information to the inference calculation module in a structured data format; The patient information database module is used to store the patient's age, past medical history, and family medical history information. It adopts a distributed storage architecture and provides a real-time dynamic correction data call interface for the inference calculation module. The three-level fuzzy rule base module includes a first-level rule base, a second-level rule base, and a third-level rule base. The first-level rule base has a built-in membership function model for more than 120 common symptoms, which is used to convert the input symptoms into membership values ​​in the 0-1 range. The second-level rule base contains more than 200 sets of multi-symptom association rules, which fuse multi-symptom association degrees through the T-norm operator and calculate the coupling weight of different symptom combinations. The third-level rule base has a preset dynamic correction factor matrix of 8 age segments and 32 common medical histories, which is used to adjust the inference threshold according to the patient's age and medical history information. The inference calculation module receives symptom information transmitted by the symptom input module, calls the three-level fuzzy rule base module through the API interface, first converts symptoms into membership degrees through the first-level rule base, then fuses the association degree of multiple symptoms using the T-norm operator based on the second-level rule base to obtain the association value of multiple symptom combinations, and finally combines the third-level rule base with the patient age and medical history information provided by the patient information database module, uses a dynamic correction factor to adjust the inference threshold, compares the association value of multiple symptom combinations with the adjusted inference threshold to obtain the diagnosis result, and transmits the diagnosis result to the diagnosis result output module in JSON format. The diagnostic result output module is used to receive the diagnostic results transmitted by the inference calculation module and present them on a high-definition display screen in a combination of text and graphics, including text diagnostic conclusions, probability values, and symptom correlation heatmaps.

[0009] Preferably, the specific implementation of the distributed storage architecture includes the following: 1) Data partitioning design Hash partitioning: Distributes data evenly across different nodes using a hash function, avoiding hotspot issues; Range partitioning: Divides data based on the range of field values, suitable for range queries; 2) Data replication and redundancy Triple-replica redundancy: Each data block is stored in 3 replicas, with 2 replicas in the same rack and 1 replica in a different rack, ensuring high availability; Raft protocol: guarantees data consistency and enables automatic failover; 3) Real-time interface updates RESTful API: Provides dynamic API calls and supports real-time data updates; 4) Privacy Protection Field-level encryption: Sensitive fields are stored using AES-256 encryption; Desensitization: Displaying some fields through desensitized views avoids plaintext leakage.

[0010] Preferably, the membership function model construction is a core step in fuzzy logic applications, and includes the following implementation steps: 1) Define the domain: Clearly define the range of the input variables; 2) Method selection: Choose either the statistical method or the assignment method based on the nature of the data; 3) Parameterization: Adjusting function parameters to match the actual distribution; 4) Verification: Test the rationality of the membership function using actual data.

[0011] Preferably, the symptom information is transmitted in a structured data format through the following methods: 1) Converting unstructured natural language into structured data is a key step in achieving intelligent diagnosis. This process mainly relies on natural language processing technology and a predefined medical ontology library. 2) Symptom Entity Recognition: The system automatically identifies and extracts key symptom entities from the text; 3) Attribute and Relationship Extraction: For the identified symptoms, the symptom input module will further extract their attributes to construct a multi-dimensional symptom profile; 4) Intensity / Degree: The membership function model is used to quantify subjective descriptions into numerical values; 5) Property Modification: Classification is mapped to standard terms in a predefined medical lexicon; 6) Forming structured data objects: After processing, each symptom is converted into a standardized JSON object to ensure the machine readability and computability of the data.

[0012] Preferably, the symptom input module employs a three-level universal mapping mechanism for semantic parsing of ambiguous symptoms, covering more than 90% of common ambiguous description scenarios: First-level mapping: Classify fuzzy descriptive words according to semantic type and map them to general feature dimensions. Frequency class corresponds to the "attack frequency" dimension, intensity class corresponds to the "symptom intensity" dimension, duration class corresponds to the "duration" dimension, and degree class corresponds to the "degree of impact" dimension. Secondary mapping: Establish 3-5 level quantitative standards for each type of fuzzy descriptive term. In the frequency category, "occasional" is defined as the number of attacks per unit time being less than 30% of the clinical mean of the symptom, "intermittent" is the number of attacks being between 30% and 70% of the mean, and "frequent" is the number of attacks being more than 70% of the mean. In the intensity category, "somewhat" is the subjective feeling intensity of 2-3 points, "moderate" is 4-6 points, and "severe" is 7-10 points. The persistence and degree categories are graded according to the same logic. Three-level mapping: Establish a general association model between the grade value and the feature interval of the first-level rule base. The frequency-type grade value is directly mapped to the basic score of the "attack frequency" feature, and the intensity-type is mapped to the basic score of the "symptom intensity" feature. It also supports mapping adaptation across symptom types. The fuzzy semantic recognition of text input is achieved through a pre-trained medical semantic model. This model is a deep learning model based on the Transformer architecture, containing an input layer, a 6-layer encoder module, and an output layer. The training corpus covers more than 50,000 fuzzy symptom descriptions from 12 departments related to internal medicine, surgery, and endocrinology, including labeled data for more than 1,200 common fuzzy descriptive words. The model is trained through the following steps: 1) Clean the original corpus, remove duplicate and invalid data, and label the semantic type and corresponding feature dimensions of fuzzy words; 2) The Word2Vec tool was used to convert the text into word vectors, which were then input into the model for pre-training. The learning rate was set to 5e-5, and the training rounds were 50. 3) Design a multi-label classification task for cross-semantic descriptions, and output the semantic parsing results through the Softmax function; The model performance metrics are: an accuracy rate of no less than 95% for identifying frequency- and intensity-based core words, an accuracy rate of no less than 90% for parsing cross-semantic descriptions, and a single text parsing response time of no more than 0.3 seconds.

[0013] Preferably, the patient information database module stores medical history information including disease name, year since diagnosis, and current control status, which is used to support the fuzzy correction logic of the three-level rule base. When a patient has a single underlying disease, the basic correction factor corresponding to that disease is retrieved from the level 3 rule base, and the calculation formula is as follows: Single disease correction factor formula: ; Where F represents a single disease correction factor; This represents the baseline correction factor corresponding to the disease; K represents the correction coefficient for the years since diagnosis; when a patient has multiple underlying diseases, a fuzzy overlay algorithm is used to calculate the comprehensive correction factor, and the formula is: Comprehensive correction factor formula: ; Where m is the number of basic diseases; Indicates the overall correction factor; , ... These represent the single disease correction factors for diseases 1 to m, respectively. , ... These represent the weights of diseases 1 to m, determined based on the correlation between the disease and the disease to be diagnosed. A high correlation results in a weight of 0.6-0.8, while a medium correlation results in a weight of 0.4-0.6. The overall correction factor is controlled within the range of 0.1-0.5, and 0.5 is used when it exceeds the upper limit.

[0014] Preferably, the primary rule base pre-defines corresponding membership degree calculation standards for different symptoms, and the membership degree of all symptoms is obtained through the following unified calculation process: 1) Determine the core feature dimensions of the symptoms. Each symptom contains 2-3 core features, and each feature is assigned a weight based on its clinical relevance. 2) Set quantitative grading standards for each core feature, divide the feature value into 3-5 intervals, and each interval corresponds to a basic score in the range of 0-1; 3) Fine-tune the eigenvalues ​​using linear interpolation. If the eigenvalue is at the critical value of two graded intervals, use the following formula: Corrected score formula: ; Where S represents the modified score of the feature; V represents the base score for the lower interval; V represents the actual value of the feature. Indicates the upper limit of the lower interval; This represents the lower limit of the high interval; This represents the base score for the higher interval; 4) Calculate the symptom membership degree using the following formula: Membership formula: ; Where n is the number of features, taking the value 2 or 3; M represents the membership degree of the symptom; , ... These represent the corrected scores for the 1st to nth features, respectively. , ... These represent the weights of the 1st to nth features, respectively. The results are rounded to two decimal places and are controlled within the range of 0-1.

[0015] Preferably, the coupling weight calculation of multiple symptom combinations in the secondary rule base adopts fuzzy association reasoning logic, and the fuzzy association strength is calculated through the T-norm operator: Operator selection: T-norm operators are automatically matched based on the symptom combination type. For strongly correlated symptom combinations, the product operator is used; for weakly correlated symptom combinations, the minimum operator is used. The product operator is calculated as follows: ; Where R represents the fuzzy association strength; Indicates the degree of membership of symptom A; Let C represent the membership degree of symptom B; let C represent the weight coefficient; the minimum operator is calculated as follows: ; Where R represents the fuzzy association strength; Indicates the degree of membership of symptom A; represents the membership degree of symptom B; C represents the weight coefficient; min() represents the minimum value function; Weighting coefficients were determined based on the symptom co-occurrence frequency distribution of more than 100,000 cases. The weighting coefficient C for combinations with a co-occurrence frequency of not less than 35% was 1.2-1.3, C for combinations with a co-occurrence frequency of 15%-35% was 1.0-1.2, and C for combinations with a co-occurrence frequency of less than 15% was 0.8-1.0.

[0016] Preferably, the dynamic correction factor matrix of the three-level rule base is constructed based on fuzzy logic, and uses a trapezoidal membership function to divide multi-dimensional fuzzy sets, which is suitable for common disease diagnosis scenarios in various departments. Age dimension: Divided into segments of 10 years each, into children, adolescents, youth, middle-aged and young adults, middle-aged, middle-aged and elderly, pre-elderly, and elderly, with corresponding basic adjustment factors of 1.1, 1.05, 1.0, 1.0, 1.1, 1.15, 1.2, and 1.3, respectively; Medical history dimension: 32 common medical history categories were divided into four fuzzy sets according to disease risk level: low risk, medium risk, medium-high risk, and high risk, corresponding to membership degrees of 0.5, 0.65, 0.8, and 0.95, respectively, with basic correction factors of 0.05-0.1, 0.1-0.15, 0.15-0.2, and 0.2-0.25, respectively. Correction rules: A weighted summation algorithm is used to fuse multi-dimensional correction factors. The age correction factor has a weight of 0.3, the comprehensive medical history correction factor has a weight of 0.7, and the final dynamic correction factor is... for: In the formula, Age correction factor As a comprehensive correction factor for medical history, The results were controlled within the range of 1.0-1.5.

[0017] Preferably, the probability value presented by the diagnostic result output module is the confidence level of fuzzy inference, which is obtained by fuzzy comparison between multi-symptom association values ​​and dynamic thresholds, and the fuzzy association strength calculated by the T-norm operator corresponding to different color depths in the dynamic symptom association map.

[0018] Beneficial effects This invention provides a multi-symptom collaborative disease diagnosis system based on fuzzy reasoning. It has the following beneficial effects: This invention utilizes a three-level semantic mapping mechanism to transform ambiguous symptoms such as intermittent abdominal pain into quantified values ​​within the 0-1 range, significantly improving accuracy compared to traditional methods. It employs a T-norm operator to dynamically process symptom associations, using a product operator for strong associations and a minimum operator for weak associations, thereby enhancing the accuracy of association calculations and reducing the false negative rate.

[0019] This invention combines age and medical history to generate dynamic correction factors, making the diagnostic results for the same symptoms more closely match individual risks for different patients, thus improving the accuracy of diagnosis for patients with chronic diseases. Presented in multiple formats such as text, pie charts, and heatmaps, it clearly demonstrates the contribution of symptom associations, increasing clinical adoption rates. It is compatible with existing hospital information systems, ensuring rapid diagnostic response, reducing deployment costs in primary healthcare institutions, and improving usage rates. Attached Figure Description

[0020] Figure 1 The core architecture diagram of the system of this invention; Figure 2 The core reasoning flowchart of this invention; Figure 3 A schematic diagram of the diagnostic result output of this invention. 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] Example 1: like Figures 1-3 As shown, this embodiment of the invention provides a multi-symptom collaborative disease diagnosis system based on fuzzy reasoning. The system includes a symptom input module, a three-level fuzzy rule base module, a reasoning calculation module, a diagnosis result output module, and a patient information database module. The symptom input module is used to receive the patient's symptom information, including vague symptoms (such as intermittent abdominal pain and moderate fatigue) and specific symptoms (such as body temperature of 38.5℃ and blood glucose level of 6.8mmol / L), and transmits the symptom information to the inference calculation module in a structured data format; Symptom information is transmitted in a structured data format through the following methods: 1) Converting unstructured natural language (especially vague symptoms) into structured data is a key step in achieving intelligent diagnosis. This process mainly relies on natural language processing technology and predefined medical ontology libraries. 2) Symptom Entity Recognition: The system automatically identifies and extracts key symptom entities from the text, such as "headache", "fatigue", and "cough"; 3) Attribute and Relationship Extraction: For the identified symptoms, the symptom input module will further extract their attributes to construct a multi-dimensional symptom profile; 4) Intensity / Degree: Use membership function models (such as trigonometric or Gaussian functions) to quantify subjective descriptions into numerical values; for example, the patient's description of "mild headache" and "severe headache" can be mapped to 0.3 and 0.9 in the interval [0,1], respectively; 5) Nature Modification: Descriptions such as "distending pain", "burning sensation", and "needle-like sensation" are categorized and mapped to standard terms in a predefined medical lexicon; 6) Forming structured data objects: After processing, each symptom is converted into a standardized JSON object to ensure the machine readability and computability of the data; The patient information database module is used to store patients’ age, past medical history (including disease name, diagnosis time, and treatment plan), and family medical history information. It adopts a distributed storage architecture and provides a real-time dynamic data correction call interface for the inference and calculation module. The specific implementation methods of distributed storage architecture include the following: 1) Data partitioning design Hash partitioning: Data is evenly distributed across different nodes using a hash function to avoid hotspot issues. For example, the patient information database module uses hash partitioning to ensure that data is evenly distributed across multiple nodes. Range partitioning: Divide data based on the range of field values ​​(such as age), suitable for range queries; for example, the user table is partitioned by age, and nodes store data for different age groups; 2) Data replication and redundancy Triple redundancy: Each data block is stored in 3 copies, with 2 copies in the same rack and 1 copy in a different rack, to ensure high availability. For example, the patient information module uses a 3-replica mechanism to prevent single point of failure. Raft protocol: ensures data consistency and enables automatic failover. For example, data changes are synchronized through the Raft protocol to ensure data consistency. 3) Real-time interface updates RESTful API: Provides a dynamic calling interface that supports real-time data updates. For example, the inference calculation module calls real-time data from the patient information database through the RESTful API. 4) Privacy Protection Field-level encryption: Sensitive fields (such as ID card numbers and mobile phone numbers) are stored using AES-256 encryption. For example, the patient information module encrypts ID card numbers and mobile phone numbers; De-identification: Displaying only the last four digits of a phone number through a de-identified view avoids plaintext leakage. For example, a de-identified view can only display the last four digits of a phone number. The three-level fuzzy rule base module includes a first-level rule base, a second-level rule base, and a third-level rule base. The first-level rule base has a built-in membership function model for more than 120 common symptoms, which is used to convert the input symptoms into membership values ​​in the 0-1 interval. The second-level rule base contains more than 200 sets of multi-symptom association rules. It uses T-norm operators (including product operators and minimum operators) to fuse multi-symptom association degrees and calculate the coupling weights of different symptom combinations. The third-level rule base has a preset dynamic correction factor matrix of 8 age segments (each segment is 10 years) and 32 common medical histories, which is used to adjust the inference threshold according to the patient's age and medical history information. The construction of the membership function model is a core step in the application of fuzzy logic, and includes the following implementation steps: 1) Define the domain: Clearly define the range of the input variables (e.g., age 0-100 years). 2) Method selection: Choose either the statistical method or the assignment method based on the nature of the data; 3) Parameterization: Adjust function parameters (such as vertex and standard deviation) to match the actual distribution; 4) Verification: Test the rationality of the membership function using actual data; The inference calculation module receives symptom information transmitted from the symptom input module. It calls the three-level fuzzy rule base module via an API interface. First, it converts symptoms into membership degrees using the first-level rule base (with a calculation error not exceeding 0.05). Then, based on the second-level rule base, it fuses multiple symptom association degrees using the T-norm operator to obtain the association value of multiple symptom combinations (precision retained to three decimal places). Finally, combining the third-level rule base, and based on the patient's age and medical history information provided by the patient information database module, it adjusts the inference threshold using a dynamic correction factor (correction range 1.0-1.5). The association value of the multiple symptom combinations is compared with the adjusted inference threshold to obtain a diagnostic result (including disease name, matching probability, and symptom association graph). The diagnostic result is then transmitted to the diagnostic result output module in JSON format. The diagnostic results output module receives the diagnostic results transmitted by the inference calculation module and presents them on a high-definition display screen in a combination of text and graphics, including text diagnostic conclusions (font size not less than 12pt), probability values ​​(displayed in a pie chart), and symptom association heatmaps.

[0023] The symptom input module employs a three-level universal mapping mechanism for semantic parsing of ambiguous symptoms, covering over 90% of common ambiguous description scenarios: First-level mapping: Classify fuzzy descriptive words according to semantic type and map them to general feature dimensions. Frequency category (such as "intermittent", "occasional", "frequent") corresponds to the "attack frequency" dimension, intensity category (such as "moderate", "somewhat", "severe") corresponds to the "symptom intensity" dimension, duration category (such as "short", "continuous", "long-term") corresponds to the "duration" dimension, and degree category (such as "mild", "obvious", "serious") corresponds to the "degree of impact" dimension. Secondary mapping: Establish 3-5 level quantitative standards for each category of fuzzy descriptive terms. In the frequency category, "occasional" is defined as the number of attacks per unit time being less than 30% of the clinical mean of the symptom (corresponding value 0.2), "intermittent" is the number of attacks being between 30% and 70% of the mean (1-3 times / day corresponding to 0.3, 4-6 times / day corresponding to 0.6), and "frequent" is the number of attacks being more than 70% of the mean (more than 7 times / day corresponding to 0.9). In the intensity category, "somewhat" is the subjective feeling intensity of 2-3 points (VAS score) (corresponding value 0.3-0.4), "moderate" is 4-6 points (corresponding value 0.5-0.7), and "severe" is 7-10 points (corresponding value 0.8-1.0). The persistence and severity categories are graded according to the same logical classification. Three-level mapping: Establish a general association model between the grade value and the feature interval of the first-level rule base. The frequency-type grade value is directly mapped to the base score of the "attack frequency" feature, and the intensity-type is mapped to the base score of the "symptom intensity" feature. It also supports mapping adaptation across symptom types (e.g., "occasional headache" and "occasional diarrhea" both follow the frequency-type mapping rules, and only the unit time base is adjusted according to the symptom characteristics). The fuzzy semantic recognition of text input is achieved through a pre-trained medical semantic model. This model is a deep learning model based on the Transformer architecture, which includes an input layer, a 6-layer encoder module (each layer contains a multi-head self-attention mechanism and a feedforward neural network), and an output layer. The training corpus covers more than 50,000 fuzzy symptom descriptions from 12 departments related to internal medicine, surgery, and endocrinology, and includes labeled data for more than 1,200 common fuzzy descriptive words. The model is trained through the following steps: 1) Clean the original corpus, remove duplicate and invalid data, and label the semantic type (frequency, intensity, etc.) and corresponding feature dimensions of fuzzy words; 2) The Word2Vec tool was used to convert the text into word vectors, which were then input into the model for pre-training. The learning rate was set to 5e-5, and the training rounds were 50. 3) Design a multi-label classification task for cross-semantic descriptions (such as "persistent mild abdominal pain"), and output the semantic parsing results through the Softmax function; The model performance metrics are: an accuracy rate of no less than 95% for identifying frequency- and intensity-based core words, an accuracy rate of no less than 90% for parsing cross-semantic descriptions, and a single text parsing response time of no more than 0.3 seconds.

[0024] The patient information database module stores medical history information including disease name, year since diagnosis, and current control status (e.g., "good," "fair," "poor"), which supports the fuzzy correction logic of the three-level rule base. When a patient has a single underlying disease, the basic correction factor corresponding to that disease is retrieved from the level 3 rule base, and the calculation formula is as follows: Formula for single disease correction factor: Where F represents a single disease correction factor; This represents the baseline correction factor for the disease (e.g., 0.2 for diabetes, 0.15 for hypertension); K represents the correction coefficient for the number of years since diagnosis (1.0 for 1-5 years after diagnosis, 1.1 for 6-10 years, and 1.2 for more than 10 years); when a patient has multiple underlying diseases, a fuzzy overlay algorithm is used to calculate the comprehensive correction factor, as shown in the formula: Comprehensive correction factor formula: Where m is the number of basic diseases; Indicates the overall correction factor; , ... These represent the single disease correction factors for diseases 1 to m, respectively. , ... These represent the weights of diseases 1 through m, determined based on the correlation between the disease and the disease to be diagnosed. A high correlation results in a weight of 0.6-0.8, while a moderate correlation results in a weight of 0.4-0.6. The overall correction factor is controlled within the range of 0.1-0.5, and is set to 0.5 if it exceeds the upper limit. The primary rule base pre-defines corresponding membership degree calculation standards for different symptoms. The membership degree of all symptoms is obtained through the following unified calculation process (reflecting fuzzy quantitative reasoning): 1) Determine the core feature dimensions of the symptoms. Each symptom contains 2-3 core features (such as "body temperature value" and "fever duration" for fever; "frequency of onset" and "duration" for abdominal pain). Each feature is assigned a weight based on its clinical relevance (the total weight is 1.0). 2) Set quantitative grading standards for each core feature, divide the feature value into 3-5 intervals, and each interval corresponds to a basic score in the range of 0-1 (e.g., "daily number of abdominal pain attacks" is divided into 1-3 times corresponding to 0.3, 4-6 times corresponding to 0.6, and more than 7 times corresponding to 0.9). 3) Fine-tune the eigenvalues ​​using linear interpolation. If the eigenvalue is at the critical value of two graded intervals, use the following formula: Corrected score formula: Where S represents the modified score of the feature; V represents the base score for the lower interval; V represents the actual value of the feature. Indicates the upper limit of the lower interval; This represents the lower limit of the high interval; This represents the base score for the higher interval; 4) Calculate the symptom membership degree using the following formula: Membership formula: Where n is the number of features, taking the value 2 or 3; M represents the membership degree of the symptom; , ... These represent the corrected scores for the 1st to nth features (calculated through step three); , ... These represent the weights of the 1st to nth features (the sum of the weights of each feature is 1.0). The result is rounded to two decimal places and is within the range of 0-1 (0 is taken if the result is below 0 and 1 is taken if the result is above 1).

[0025] The coupling weight calculation for multiple symptom combinations in the secondary rule base employs fuzzy relational reasoning logic, using the T-norm operator to calculate the fuzzy relational strength. Operator selection: T-norm operators are automatically matched based on the symptom combination type. For strongly correlated symptom combinations, the product operator is used; for weakly correlated symptom combinations, the minimum operator is used. The product operator is calculated as follows: Where R represents the fuzzy association strength; Indicates the degree of membership of symptom A; Let C represent the membership degree of symptom B; let C represent the weight coefficient; the minimum operator is calculated as follows: Where R represents the fuzzy association strength; Indicates the degree of membership of symptom A; represents the membership degree of symptom B; C represents the weight coefficient; min() represents the minimum value function; Weighting coefficients were determined as follows: Based on the symptom co-occurrence frequency distribution of 100,000+ cases, the weighting coefficient C for combinations with a co-occurrence frequency of not less than 35% was 1.2-1.3, the C for combinations with a co-occurrence frequency of 15%-35% was 1.0-1.2, and the C for combinations with a co-occurrence frequency of not more than 15% was 0.8-1.0. The dynamic correction factor matrix of the three-level rule base is constructed based on fuzzy logic and uses a trapezoidal membership function to divide multi-dimensional fuzzy sets, making it suitable for common disease diagnosis scenarios in various departments. Age dimension: Divided into segments of 10 years each, the age groups are: "Children" (0-10 years old, membership degree 0.9), "Adolescents" (11-20 years old, membership degree 0.8), "Young Adults" (21-30 years old, membership degree 0.7), "Middle-aged and Young Adults" (31-40 years old, membership degree 0.6), "Middle-aged" (41-50 years old, membership degree 0.7), "Middle-aged and Elderly" (51-60 years old, membership degree 0.8), "Early Old Age" (61-70 years old, membership degree 0.9), and "Elderly" (71 years old and above, membership degree 1.0), with corresponding basic adjustment factors of 1.1, 1.05, 1.0, 1.0, 1.1, 1.15, 1.2, and 1.3, respectively. Medical history dimension: 32 common medical history categories were divided into four fuzzy sets according to disease risk level: "low risk," "medium risk," "medium-high risk," and "high risk," corresponding to membership degrees of 0.5, 0.65, 0.8, and 0.95, respectively, with basic correction factors of 0.05-0.1, 0.1-0.15, 0.15-0.2, and 0.2-0.25, respectively. Among them, "low risk" includes chronic gastritis, "medium risk" includes thyroid dysfunction, "medium-high risk" includes hypertension, and "high risk" includes diabetes. Correction rules: A weighted summation algorithm is used to fuse multi-dimensional correction factors. The age correction factor has a weight of 0.3, the comprehensive medical history correction factor has a weight of 0.7, and the final dynamic correction factor is... for: In the formula, Age correction factor As a comprehensive correction factor for medical history, The results were controlled within the range of 1.0-1.5.

[0026] The probability values ​​presented in the diagnostic results output module represent the confidence level of fuzzy inference, which is obtained through a fuzzy comparison between the multi-symptom association values ​​and the dynamic threshold (e.g., the comparison result of the association value of 0.72 and the threshold of 0.7 is expressed as "high probability match"). In the dynamic symptom association map, different color depths correspond to the fuzzy association strength calculated by the T-norm operator (the 0-1 interval is mapped to a gradient from blue to red, 0.0-0.3 is dark blue, 0.3-0.6 is light blue, 0.6-0.8 is orange, and 0.8-1.0 is red).

[0027] Example 2: Case Information The patient is a 58-year-old female with an 8-year history of hypertension (currently under fair control). Her chief symptoms include: vague symptoms, frequent headaches, 6 times a day; moderate dizziness, VAS score of 5; and definite symptoms, blood pressure of 150 / 95 mmHg and heart rate of 85 beats / minute.

[0028] System processing 1. Symptom Input Module Processing A three-level semantic mapping was used for "frequent headaches": the frequency category corresponds to the "attack frequency" dimension, with 6 times per day corresponding to a base score of 0.6; A three-level semantic mapping was used for "moderate dizziness": the intensity class corresponds to the "symptom intensity" dimension, and a VAS score of 5 points corresponds to a base score of 0.6. Standardize the symptoms: blood pressure of 150 / 95 mmHg is converted to a characteristic value of 0.7 according to the hypertension classification standard, and heart rate of 85 beats / minute is converted to a characteristic value of 0.5.

[0029] 2. Calculation of membership degree of the first-level rule base The core characteristics of headache symptoms: frequency of onset weight 0.7, pain intensity weight 0.3: membership degree = 0.6 × 0.7 + 0.5 × 0.3 = 0.57; The core characteristic of dizziness symptoms: VAS score weight 1.0: membership degree = 0.6; The core characteristics of abnormal blood pressure: blood pressure value weight 0.8, medical history association weight 0.2: membership degree = 0.7×0.8+0.6×0.2=0.68.

[0030] 3. Calculation of association strength in the second-level rule base Headache and abnormal blood pressure co-occurred frequently, indicating a strong association combination. The product operator was used with a weighting coefficient of 1.2: Association strength = 0.57 × 0.68 × 1.2 = 0.465; Dizziness and abnormal blood pressure co-occurred at a moderate frequency, which is a weak association combination. The minimum operator was used with a weight coefficient of 1.0: association strength = min(0.6, 0.68) × 1.0 = 0.6.

[0031] 4. Dynamic correction of the three-level rule base Age 58 corresponds to the "middle-aged and elderly" segment, with an age correction factor of 1.15; The patient has a history of hypertension for 8 years, and the single disease correction factor is 0.15 × 1.1 = 0.165. Overall correction factor = 0.165 × 0.7 = 0.1155; The final dynamic correction factor = 1.15 × 0.3 + 0.1155 = 0.4605.

[0032] 5. Reasoning, Calculation, and Result Output The association value for multiple symptoms was taken as the strong association combination of 0.465; Dynamic threshold = base threshold 0.5 × (1 + 0.4605) = 0.730; Comparison result: 0.465 is less than 0.730, output "Hypertension-related headache and dizziness"; Diagnostic results output: The textual conclusion explains the diagnostic basis, the pie chart shows a matching probability of 46.5%, the heat map shows a light blue line connecting headache and abnormal blood pressure (correlation strength 0.465), and the line connecting dizziness and abnormal blood pressure (correlation strength 0.6).

[0033] As can be seen from this embodiment, the system can accurately quantify ambiguous symptoms, perform personalized reasoning based on the patient's age and medical history, and finally present the diagnostic results in an intuitive way, which reflects the core advantages of the present invention.

[0034] 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 a reference structure" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0035] 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 multi-symptom collaborative disease diagnosis system based on fuzzy reasoning, characterized in that: The system includes a symptom input module, a three-level fuzzy rule base module, an inference calculation module, a diagnosis result output module, and a patient information database module; The symptom input module is used to receive the patient's symptom information, which includes vague symptoms and specific symptoms, and transmits the symptom information to the inference calculation module in a structured data format; The patient information database module is used to store the patient's age, past medical history, and family medical history information. It adopts a distributed storage architecture and provides a real-time dynamic correction data call interface for the inference calculation module. The three-level fuzzy rule base module includes a first-level rule base, a second-level rule base, and a third-level rule base. The first-level rule base has a built-in membership function model for more than 120 common symptoms, which is used to convert the input symptoms into membership values ​​in the 0-1 range. The second-level rule base contains more than 200 sets of multi-symptom association rules, which fuse multi-symptom association degrees through the T-norm operator and calculate the coupling weight of different symptom combinations. The third-level rule base has a preset dynamic correction factor matrix of 8 age segments and 32 common medical history categories, which is used to adjust the inference threshold according to the patient's age and medical history information. The inference calculation module receives symptom information transmitted by the symptom input module, calls the three-level fuzzy rule base module through the API interface, first converts symptoms into membership degrees through the first-level rule base, then fuses the association degree of multiple symptoms using the T-norm operator based on the second-level rule base to obtain the association value of multiple symptom combinations, and finally combines the third-level rule base with the patient age and medical history information provided by the patient information database module, uses a dynamic correction factor to adjust the inference threshold, compares the association value of multiple symptom combinations with the adjusted inference threshold to obtain the diagnosis result, and transmits the diagnosis result to the diagnosis result output module in JSON format. The diagnostic result output module is used to receive the diagnostic results transmitted by the inference calculation module and present them on a high-definition display screen in a combination of text and graphics, including text diagnostic conclusions, probability values, and symptom correlation heatmaps.

2. The multi-symptom collaborative disease diagnosis system based on fuzzy reasoning according to claim 1, characterized in that: The specific implementation methods of the distributed storage architecture include the following: 1) Data partitioning design Hash partitioning: Distributes data evenly across different nodes using a hash function, avoiding hotspot issues; Range partitioning: Divides data based on the range of field values, suitable for range queries; 2) Data replication and redundancy Triple-replica redundancy: Each data block is stored in 3 replicas, with 2 replicas in the same rack and 1 replica in a different rack, ensuring high availability; Raft protocol: guarantees data consistency and enables automatic failover; 3) Real-time interface updates RESTful API: Provides dynamic API calls and supports real-time data updates; 4) Privacy Protection Field-level encryption: Sensitive fields are stored using AES-256 encryption; Desensitization: Displaying some fields through a desensitized view avoids plaintext leakage.

3. The multi-symptom collaborative disease diagnosis system based on fuzzy reasoning according to claim 1, characterized in that: The construction of the membership function model is a core step in the application of fuzzy logic, and includes the following implementation steps: 1) Define the domain: Clearly define the range of the input variables; 2) Method selection: Choose either the statistical method or the assignment method based on the nature of the data; 3) Parameterization: Adjusting function parameters to match the actual distribution; 4) Verification: Test the rationality of the membership function using actual data.

4. The multi-symptom collaborative disease diagnosis system based on fuzzy reasoning according to claim 1, characterized in that: The symptom information is transmitted in a structured data format through the following methods: 1) Converting unstructured natural language into structured data is a key step in achieving intelligent diagnosis. This process mainly relies on natural language processing technology and a predefined medical ontology library. 2) Symptom Entity Recognition: The system automatically identifies and extracts key symptom entities from the text; 3) Attribute and Relationship Extraction: For the identified symptoms, the symptom input module will further extract their attributes to construct a multi-dimensional symptom profile; 4) Intensity / Degree: The membership function model is used to quantify subjective descriptions into numerical values; 5) Property Modification: Classification is mapped to standard terms in a predefined medical lexicon; 6) Forming structured data objects: After processing, each symptom is converted into a standardized JSON object to ensure the machine readability and computability of the data.

5. The multi-symptom collaborative disease diagnosis system based on fuzzy reasoning according to claim 1, characterized in that: The symptom input module employs a three-level universal mapping mechanism for semantic parsing of ambiguous symptoms, covering over 90% of common ambiguous description scenarios: First-level mapping: Classify fuzzy descriptive words according to semantic type and map them to general feature dimensions. Frequency class corresponds to the "attack frequency" dimension, intensity class corresponds to the "symptom intensity" dimension, duration class corresponds to the "duration" dimension, and degree class corresponds to the "degree of impact" dimension. Secondary mapping: Establish 3-5 level quantitative standards for each type of fuzzy descriptive term. In the frequency category, "occasional" is defined as the number of attacks per unit time being less than 30% of the clinical mean of the symptom, "intermittent" is the number of attacks being between 30% and 70% of the mean, and "frequent" is the number of attacks being more than 70% of the mean. In the intensity category, "somewhat" is the subjective feeling intensity of 2-3 points, "moderate" is 4-6 points, and "severe" is 7-10 points. The persistence and degree categories are graded according to the same logic. Three-level mapping: Establish a general association model between the grade value and the feature interval of the first-level rule base. The frequency-type grade value is directly mapped to the basic score of the "attack frequency" feature, and the intensity-type is mapped to the basic score of the "symptom intensity" feature. It also supports mapping adaptation across symptom types. The fuzzy semantic recognition of text input is achieved through a pre-trained medical semantic model. This model is a deep learning model based on the Transformer architecture, containing an input layer, a 6-layer encoder module, and an output layer. The training corpus covers more than 50,000 fuzzy symptom descriptions from 12 departments related to internal medicine, surgery, and endocrinology, including labeled data for more than 1,200 common fuzzy descriptive words. The model is trained through the following steps: 1) Clean the original corpus, remove duplicate and invalid data, and label the semantic type and corresponding feature dimensions of fuzzy words; 2) The Word2Vec tool was used to convert the text into word vectors, which were then input into the model for pre-training. The learning rate was set to 5e-5, and the training rounds were 50. 3) Design a multi-label classification task for cross-semantic descriptions, and output the semantic parsing results through the Softmax function; The model performance metrics are: an accuracy rate of no less than 95% for identifying frequency- and intensity-based core words, an accuracy rate of no less than 90% for parsing cross-semantic descriptions, and a single text parsing response time of no more than 0.3 seconds.

6. The multi-symptom collaborative disease diagnosis system based on fuzzy reasoning according to claim 1, characterized in that: The patient information database module stores medical history information including disease name, year since diagnosis, and current control status, which is used to support the fuzzy correction logic of the three-level rule base. When a patient has a single underlying disease, the basic correction factor corresponding to that disease is retrieved from the level 3 rule base, and the calculation formula is as follows: Single disease correction factor formula: ； Where F represents a single disease correction factor; This represents the baseline correction factor corresponding to the disease; K represents the correction coefficient for the years since diagnosis; when a patient has multiple underlying diseases, a fuzzy overlay algorithm is used to calculate the comprehensive correction factor, and the formula is: Comprehensive correction factor formula: ； Where m is the number of basic diseases; Indicates the overall correction factor; , ... These represent the single disease correction factors for diseases 1 to m, respectively. , ... These represent the weights of diseases 1 to m, determined based on the correlation between the disease and the disease to be diagnosed. A high correlation results in a weight of 0.6-0.8, while a medium correlation results in a weight of 0.4-0.

6. The overall correction factor is controlled within the range of 0.1-0.5, and 0.5 is used when it exceeds the upper limit.

7. The multi-symptom collaborative disease diagnosis system based on fuzzy reasoning according to claim 1, characterized in that: The primary rule base pre-defines corresponding membership degree calculation standards for different symptoms, and the membership degree of all symptoms is obtained through the following unified calculation process: 1) Determine the core feature dimensions of the symptoms. Each symptom contains 2-3 core features, and each feature is assigned a weight based on its clinical relevance. 2) Set quantitative grading standards for each core feature, divide the feature value into 3-5 intervals, and each interval corresponds to a basic score in the range of 0-1; 3) Fine-tune the eigenvalues ​​using linear interpolation. If the eigenvalue is at the critical value of two graded intervals, use the following formula: Corrected score formula: ； Where S represents the modified score of the feature; V represents the base score for the lower interval; V represents the actual value of the feature. Indicates the upper limit of the lower interval; This represents the lower limit of the high interval; This represents the base score for the higher interval; 4) Calculate the symptom membership degree using the following formula: Membership formula: ； Where n is the number of features, taking the value 2 or 3; M represents the membership degree of the symptom; , ... These represent the corrected scores for the 1st to nth features, respectively. , ... These represent the weights of the 1st to nth features, respectively. The results are rounded to two decimal places and are controlled within the range of 0-1.

8. The multi-symptom collaborative disease diagnosis system based on fuzzy reasoning according to claim 1, characterized in that: The coupling weight calculation for multiple symptom combinations in the secondary rule base employs fuzzy relational reasoning logic, and the fuzzy relational strength is calculated using the T-norm operator: Operator selection: T-norm operators are automatically matched based on the symptom combination type. For strongly correlated symptom combinations, the product operator is used; for weakly correlated symptom combinations, the minimum operator is used. The product operator is calculated as follows: ； Where R represents the fuzzy association strength; Indicates the degree of membership of symptom A; Let C represent the membership degree of symptom B; let C represent the weight coefficient; the minimum operator is calculated as follows: ； Where R represents the fuzzy association strength; Indicates the degree of membership of symptom A; represents the membership degree of symptom B; C represents the weight coefficient; min() represents the minimum value function; Weighting coefficients were determined based on the symptom co-occurrence frequency distribution of more than 100,000 cases. The weighting coefficient C for combinations with a co-occurrence frequency of not less than 35% was 1.2-1.3, C for combinations with a co-occurrence frequency of 15%-35% was 1.0-1.2, and C for combinations with a co-occurrence frequency of less than 15% was 0.8-1.

0.

9. A multi-symptom collaborative disease diagnosis system based on fuzzy reasoning according to claim 1, characterized in that: The dynamic correction factor matrix of the three-level rule base is constructed based on fuzzy logic and uses a trapezoidal membership function to divide multi-dimensional fuzzy sets, making it suitable for common disease diagnosis scenarios in various departments. Age dimension: Divided into segments of 10 years each, into children, adolescents, youth, middle-aged and young adults, middle-aged, middle-aged and elderly, pre-elderly, and elderly, with corresponding basic adjustment factors of 1.1, 1.05, 1.0, 1.0, 1.1, 1.15, 1.2, and 1.3, respectively; Medical history dimension: 32 common medical history categories were divided into four fuzzy sets according to disease risk level: low risk, medium risk, medium-high risk, and high risk, corresponding to membership degrees of 0.5, 0.65, 0.8, and 0.95, respectively, with basic correction factors of 0.05-0.1, 0.1-0.15, 0.15-0.2, and 0.2-0.25, respectively. Correction rules: A weighted summation algorithm is used to fuse multi-dimensional correction factors. The age correction factor has a weight of 0.3, the comprehensive medical history correction factor has a weight of 0.7, and the final dynamic correction factor is... for: In the formula, Age correction factor As a comprehensive correction factor for medical history, The results were controlled within the range of 1.0-1.

5.

10. A multi-symptom collaborative disease diagnosis system based on fuzzy reasoning according to claim 1, characterized in that: The probability values ​​presented by the diagnostic result output module are the confidence levels of fuzzy inference, which are obtained by fuzzy comparison between multi-symptom association values ​​and dynamic thresholds. The fuzzy association strengths calculated by the T-norm operator correspond to different color depths in the dynamic symptom association map.

Images