Pet medical data ai desensitization processing and intelligent analysis management method

Abstract

This invention provides an AI-based de-identification and intelligent analysis management method for pet medical data, belonging to the field of medical data processing technology. It includes employing a multi-source data acquisition strategy to comprehensively acquire basic pet information, diagnostic and treatment data, and disease prevention and control data; preprocessing the collected data; classifying and labeling the preprocessed data based on privacy leakage risks and data structuring levels; dynamically matching the optimal de-identification strategy using AI algorithms to construct a hybrid de-identification model to complete data de-identification; constructing a secure management system encompassing hierarchical storage, secure transmission, and data backup to ensure the security and availability of data storage and transmission; and using intelligent analysis algorithms to predict disease trends, recommend treatment plans, assess medical quality, and identify abnormal behaviors from the de-identified data. This invention addresses the current challenges in data management by employing the aforementioned AI-based de-identification and intelligent analysis management method for pet medical data.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, and in particular to an AI-based desensitization processing and intelligent analysis and management method for pet medical data. Background Technology

[0002] Driven by the digital wave, the pet healthcare industry is undergoing profound changes and actively embarking on a new journey of digital transformation. In this process, pet hospitals, health management platforms, and other related institutions have accumulated a massive amount of pet medical data. This data is broad in scope, including not only basic pet information such as breed, age, and unique identifiers; but also detailed information about pet owners, such as name, contact information, and address; medical records are also a crucial component, including medical history, various examination reports, and medication plans; furthermore, disease diagnosis data also plays a significant role.

[0003] These pet medical data possess a dual nature. On one hand, they contain a wealth of sensitive and private information, such as pet owners' contact information and precise pet identification details. Leakage of this data could cause unnecessary distress to pet owners and even lead to serious consequences like financial loss. On the other hand, this data holds immense value, playing a crucial role in medical research, disease prevention and control, and the optimization of treatment methods. In-depth analysis of this data can provide strong data support and decision-making basis for the development of the pet medical industry.

[0004] However, current pet medical data management faces two major pain points that urgently need to be addressed. First, privacy protection is significantly inadequate. Traditional data anonymization methods, such as simply masking sensitive information or using fixed replacement methods, struggle to find a proper balance between data security and usability. If the anonymization level is too high, the data loses its original key information, making it unsuitable for subsequent analysis and rendering it worthless; conversely, insufficient anonymization easily leads to privacy leaks, posing potential security threats to pet owners and their pets. Second, data value mining is insufficient. Currently, a comprehensive "anonymization-storage-analysis-application" end-to-end management system has not been established. A large amount of valuable medical data remains idle, unable to be effectively transformed into practical support for optimizing diagnosis and treatment and improving disease prevention and control capabilities, thus failing to fully realize its potential value. Summary of the Invention

[0005] The purpose of this invention is to provide an AI-based de-identification processing and intelligent analysis management method for pet medical data, solving the current challenges in data management and promoting high-quality development in the pet medical industry.

[0006] To achieve the above objectives, this invention provides an AI-based de-identification processing and intelligent analysis management method for pet medical data, comprising the following steps: S1. Adopt a multi-source data collection strategy and comprehensively utilize multiple data collection channels to obtain various key information in the pet medical process, including basic pet information, diagnosis and treatment data, and disease prevention and control data. S2. Preprocess the collected multi-source pet medical data, including deleting duplicate records, processing data with abnormal formats, and unifying the data format; S3. Classify and label the pre-processed pet medical data. Based on the risk of data privacy leakage, the data is divided into core sensitive data, highly sensitive data, general sensitive data and non-sensitive data. According to the degree of data structure, the data is further classified and labeled into three categories: structured data, semi-structured data and unstructured data. S4. Based on the hierarchical classification results, use AI algorithms to dynamically match the optimal desensitization strategy, construct a hybrid desensitization model, and desensitize pet medical data. S5. Construct a data security management system, covering hierarchical storage mechanisms, secure transmission mechanisms, and data backup strategies, to ensure the security, integrity, and availability of de-identified pet medical data during storage and transmission. S6. Intelligent analysis algorithms are used to conduct in-depth analysis of desensitized pet medical data, including disease epidemic trend prediction, treatment plan recommendation, medical quality assessment, and abnormal behavior identification.

[0007] Preferred multi-source data acquisition strategies include: It connects with the pet hospital management system to obtain basic information about pets, medical treatment data, and disease prevention and control data. The veterinarian's mobile diagnostic terminal can quickly record the pet's symptoms and preliminary examination results on-site and upload them to the data center in a timely manner. It interfaces with laboratory testing equipment to automatically collect test data, including blood biochemical indicators and microbial culture results; A self-reporting portal is set up, allowing owners to fill in details about their pets' daily eating habits and living environment via a mobile app or webpage.

[0008] Preferably, preprocessing of the collected multi-source pet medical data includes: Set a unique identifier field to compare and filter data, and delete duplicate records; For medical records missing key fields, supplement or mark them as invalid data and delete them as appropriate; for erroneous detection values, make a comprehensive judgment based on the pet's clinical symptoms and other test results, and correct or delete them if they are confirmed to be erroneous data. The date format is standardized as “YYYY-MM-DD”, and the units for testing indicators are standardized.

[0009] Preferably, the preprocessed pet medical data is graded, classified, and labeled, including: Based on the risk of data privacy breaches, data is categorized into core sensitive data, highly sensitive data, generally sensitive data, and non-sensitive data. Based on the degree of data structuring, the data is divided into three categories: structured data, semi-structured data, and unstructured data, and then labeled accordingly.

[0010] Preferably, the desensitization treatment includes: For character fields in generally sensitive data, a substitution algorithm is used for desensitization. This algorithm is suitable for certain fields in highly sensitive data and employs a masking algorithm. For core sensitive data, the AES symmetric encryption algorithm is used for encryption. For structured and semi-structured data that require analysis support, a generalization algorithm based on AI models to dynamically adjust the generalization coefficient is used for desensitization processing.

[0011] Preferably, the hierarchical storage mechanism in S5 includes: For core sensitive data that has been encrypted with AES, a distributed encrypted database is used for storage, and a triple key protection system is set up, namely system key, user key and device key. For highly sensitive and generally sensitive data that have undergone generalization, masking, or replacement processing, a partitioned storage strategy is adopted. The data is divided into different storage partitions according to data type and time dimension, and strict access control is configured for each partition. For non-sensitive data, object storage services are used. Secure transmission mechanisms include: In the process of data transmission across terminals and organizations, a secure transmission channel is established using the SSL / TLS encryption protocol; During transmission, a data checksum is added to each data item, and the receiving end verifies the integrity of the data based on the checksum. Data backup strategies include: For core and sensitive data, daily incremental backups are used; other data is backed up fully every week. This is achieved through a combination of local backup and off-site disaster recovery.

[0012] Preferably, the epidemic trend prediction uses an LSTM neural network to perform time series analysis on desensitized disease diagnosis data, extracts generalized key information to train the model, predicts the disease risk of specific regions and varieties in the next 1-3 months, and combines epidemiological characteristics to analyze the transmission path and influencing factors. The treatment plan recommendation constructs a medical knowledge graph, processes desensitized medical record data based on collaborative filtering algorithm, extracts symptom and diagnosis information to recommend the optimal treatment plan, and forms standardized templates for common diseases; The medical quality assessment and abnormal behavior identification uses the K-means clustering algorithm to evaluate the treatment efficiency of pet hospitals and veterinarians, generate medical quality reports, establish abnormal behavior models to monitor treatment data, and identify abnormal behaviors.

[0013] Therefore, the present invention employs the above-mentioned AI-based desensitization processing and intelligent analysis management method for pet medical data, and the technical effects are as follows: 1. Balancing privacy protection and data availability: By classifying data privacy leakage risks and combining AI algorithms to dynamically match the optimal de-identification strategy, the contradiction between "over-de-identification leading to data invalidation" or "insufficient de-identification causing privacy leakage" in traditional de-identification methods is resolved.

[0014] 2. Construct a full-process security management system: Core sensitive data is protected by a distributed encrypted database and triple keys; highly sensitive data is stored in partitions and subject to strict access control; non-sensitive data uses object storage services; SSL / TLS encrypted channels and data verification codes are used during transmission; the backup strategy combines daily incremental backups and weekly full backups, and supports off-site disaster recovery; ensuring the security, integrity and availability of data during storage and transmission.

[0015] 3. In-depth data value mining and intelligent application: LSTM neural networks are used to predict the disease risk of specific regions and breeds in the next 1-3 months. Collaborative filtering algorithms are combined to recommend the optimal treatment plan, and K-means clustering algorithm is used to evaluate the treatment efficiency of pet hospitals and veterinarians, generating medical quality reports. Idle medical data is transformed into practical support for disease prevention and control and treatment optimization, significantly improving the application value of the data. Attached Figure Description

[0016] Figure 1 This is a flowchart of an AI-based desensitization processing and intelligent analysis management method for pet medical data according to the present invention; Figure 2 This is a flowchart of the AI ​​adaptive desensitization processing method of the present invention; Figure 3 This is a flowchart of the method for secure storage and transmission of de-identified data according to the present invention; Figure 4 This is a flowchart of the intelligent analysis method for de-identified data according to the present invention. Detailed Implementation

[0017] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0018] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0019] Example 1 like Figures 1-4 As shown, this invention provides an AI-based de-identification processing and intelligent analysis management method for pet medical data, comprising the following steps: A multi-source data acquisition strategy is adopted, which comprehensively utilizes multiple data acquisition channels to obtain various key information in the pet medical process. The information includes basic pet information, diagnosis and treatment data, and disease prevention and control data. Preprocessing of collected multi-source pet medical data includes deleting duplicate records, processing data with abnormal formats, and standardizing data formats; The preprocessed pet medical data is classified and labeled according to its level. Based on the risk of data privacy leakage, the data is divided into core sensitive data, highly sensitive data, general sensitive data and non-sensitive data. According to the degree of data structure, the data is further classified and labeled into three categories: structured data, semi-structured data and unstructured data. Based on the hierarchical classification results, AI algorithms are used to dynamically match the optimal desensitization strategy and construct a hybrid desensitization model to desensitize pet medical data. Establish a data security management system that includes a tiered storage mechanism, a secure transmission mechanism, and a data backup strategy to ensure the security, integrity, and availability of de-identified pet medical data during storage and transmission. Intelligent analysis algorithms are used to conduct in-depth analysis of desensitized pet medical data, including disease epidemic trend prediction, treatment plan recommendation, medical quality assessment, and abnormal behavior identification.

[0020] To obtain comprehensive and accurate information related to pet medical care, a multi-source data collection strategy is adopted, which comprehensively utilizes various data collection channels to cover all kinds of key information in the pet medical care process.

[0021] As the primary storage platform for pet medical information, this system records the complete process of a pet's care from admission to discharge. Through integration with the pet hospital management system, it can obtain basic pet information, including pet ID, breed, age, gender, microchip number, owner's name, contact information, and address. It can also access treatment data, such as medical record number, appointment time, symptom description, examination items, test results, diagnosis, medication regimen, and surgical records. Furthermore, disease prevention and control data can also be obtained from this system, including disease diagnosis results, vaccination records, antibody test data, and epidemiological information.

[0022] Veterinarians can use mobile diagnostic terminals to quickly record pet symptoms, preliminary examination results, and other information on-site, and upload this data to a data center in a timely manner. This data includes basic information, diagnostic data, and disease prevention information, providing a strong supplement to the completeness of pet medical data. In outdoor emergency rescue or home visits, mobile diagnostic terminals can record key pet information in real time, ensuring the timeliness and accuracy of the data.

[0023] Laboratory testing is a crucial part of veterinary disease diagnosis, and various advanced laboratory testing equipment can provide accurate test results. By interfaceing with laboratory testing equipment, data such as blood biochemistry indicators and microbial culture results can be automatically collected. This data is stored in a structured format, ensuring high accuracy and reliability, and providing a scientific basis for veterinary disease diagnosis and treatment. For example, blood biochemistry tests can reveal indicators such as liver and kidney function, blood glucose, and blood lipids, helping to assess the pet's health and the severity of any illness.

[0024] To obtain detailed information that is difficult for pet hospital management systems to record, such as pets' daily eating habits and living environment details, a self-reporting portal for pet owners has been set up. Owners can fill in their pets' relevant information themselves via a mobile app or website. While this information has a degree of subjectivity, it provides more comprehensive background data for pet medical care, helping veterinarians to gain a deeper understanding of their pets' health. For example, the pet's daily eating habits reported by the owner may be related to certain digestive system diseases, providing important clues for disease diagnosis and prevention.

[0025] The collected multi-source pet medical data, due to its wide range of sources and diverse formats, may contain issues such as duplication, anomalies, and invalidity, affecting data quality and the accuracy of analysis results. Therefore, preprocessing of the collected data is essential before subsequent data analysis and application.

[0026] By setting a unique identifier field, data is compared and filtered to remove duplicate records, ensuring the uniqueness of each data entry. Secondly, data with abnormal formats is handled. For medical records lacking key fields, such as those missing diagnostic conclusions or medication plans, these are supplemented or marked as invalid and deleted, depending on the actual situation. For erroneous test values, such as blood test indicators exceeding the normal range, a comprehensive judgment is made based on the pet's clinical symptoms and other test results; if confirmed as erroneous data, it is corrected or deleted.

[0027] To facilitate data storage, management, and analysis, data formats from different sources are standardized. For example, the date format is standardized to "YYYY-MM-DD" to ensure that all date data has a consistent representation; the units of detection indicators are standardized, such as unifying the unit of blood glucose concentration to "mmol / L" to avoid data analysis errors caused by inconsistent units.

[0028] Data preprocessing effectively improved data quality and consistency, laying a solid foundation for subsequent data classification, labeling, and in-depth analysis. After preprocessing, data integrity and accuracy were significantly enhanced, the proportion of duplicate and outlier data was greatly reduced, and the uniformity of data format facilitated smoother data interaction and sharing between different systems.

[0029] To better protect pet owners' privacy while meeting the needs of different data analysis and application scenarios, preprocessed pet medical data is graded and categorized. Based on the risk of data privacy leakage, the data is divided into four levels: Core Sensitive Data: This type of data directly involves the personal privacy of pet owners and the precise identity information of their pets. Its leakage could lead to serious consequences. This includes the owner's contact information, address, pet microchip number, and precise identification details.

[0030] Highly sensitive data: This includes detailed medical information about the pet and some key information about the owner. A leak of this data could significantly impact the pet owner's life and the pet's health. Examples include complete medical records, the owner's name, and a combination of precise appointment times and hospital location data.

[0031] Generally sensitive data: This type of data has a relatively small impact on the privacy of pet owners, but it still involves pet health information to some extent. This includes pet breed, age, symptom descriptions, and types of medications used.

[0032] Non-sensitive data: After anonymization, this data does not contain any information that could identify pet owners or individual pets. It is mainly used for macro-level statistical analysis of diseases and research on medical visit trends. Examples include anonymized disease statistics and non-location-based medical visit trend data.

[0033] By quantifying the risk of data leakage, the sensitivity level can be automatically labeled. The specific expression is as follows: ; in, Data sensitivity is scored as follows: ≥8 points indicates core sensitivity, 6-7 points indicates high sensitivity, 3-5 points indicates moderate sensitivity, and ≤2 points indicates non-sensitivity. Privacy breaches can affect weighting; Restore risk weights to the data association; Weighting for compliance requirements; Risk scores for each dimension are output based on the BERT text classification model.

[0034] Based on their degree of structure, data is categorized into three types: structured data, semi-structured data, and unstructured data, and then labeled accordingly. Structured data has a clear data structure and format, facilitating computer storage and processing. Structured data is typically stored in tables in databases, offering high query and analysis efficiency. Semi-structured data has some structure, but it is not as strictly standardized as structured data. Semi-structured data is usually stored in text files or specific document formats, requiring parsing and processing to extract useful information. Unstructured data lacks a clear structure and format, such as imaging examination images and audio medical records. Processing this type of data is relatively complex, requiring specialized image processing and audio analysis techniques to extract its feature information.

[0035] By categorizing and labeling data hierarchically, the sensitivity levels and data types of different data are clearly defined, providing an important basis for the secure storage, access control, and effective utilization of data. In practical applications, different access permissions can be set according to the sensitivity level of the data to ensure that core sensitive data and highly sensitive data are strictly protected; at the same time, appropriate analysis and processing methods can be adopted for different types of data to improve the efficiency and accuracy of data analysis and application.

[0036] After completing the classification and grading of pet medical data, in order to ensure the security and usability of the data in different application scenarios, this study uses AI algorithms to dynamically match the optimal desensitization strategy based on the classification and grading results, and constructs a hybrid desensitization model of "basic algorithm + AI optimization" to achieve efficient and accurate desensitization processing of pet medical data.

[0037] For character fields in generally sensitive data, such as pet nicknames and non-core identifiers, a substitution algorithm is used for desensitization. This algorithm effectively hides the original data information while maintaining the integrity of the field format by randomly replacing characters at specified positions. For example, the pet nickname "Qiuqiu" might become "Qiuyuan" after randomly replacing two characters, thus preserving the basic characteristics of the nickname while preventing the leakage of the original information.

[0038] The expression for the permutation algorithm is: ; in, The number of characters that can be replaced for sensitive fields; This is the original character position; This represents the position of the character after the replacement.

[0039] This algorithm is suitable for certain fields in highly sensitive data, such as the owner's name and contact information. It protects critical information by blocking core character segments while retaining some non-sensitive characters. For example, in a mobile phone number, the middle four digits are replaced with quotation marks (""), such as "138". "5678" reduces the risk of information leakage without affecting the basic data identification function; for the owner's name, the last character can be masked, such as "Zhang San" becoming "Zhang".

[0040] The masking algorithm expression is: .

[0041] For critical data such as pet microchip numbers and owners' complete addresses, the AES symmetric encryption algorithm is used for encryption. This algorithm uses a 128-bit key for block encryption, offering high security and encryption efficiency. Only authorized users with the correct key can decrypt and retrieve the original data, effectively preventing the unauthorized acquisition and misuse of critical data.

[0042] For structured and semi-structured data that need to support analysis, such as age, consultation time, test values, and symptom descriptions, a generalization algorithm based on AI models to dynamically adjust the generalization coefficient is used for desensitization processing.

[0043] For numerical data such as age, weight, and test results, the AI ​​algorithm first analyzes the data's distribution characteristics and calculates its mean. and standard deviation Then, the generalization coefficient is adaptively set. (Values ​​range from 1 to 3), generate interval data according to the formula.

[0044] ; For example, if a pet is 3 years old, and the calculated mean of this age data set is 3.2 years and the standard deviation is 0.5 years, then if the generalization coefficient is set to 2, the generalization interval is [2.2, 4.2], ultimately generalizing the age "3 years old" to "2-4 years old". Similarly, for a detection value of "120 mg / dL", if the mean of this data set is 125 mg / dL and the standard deviation is 10 mg / dL, then with a generalization coefficient of 2, the generalization interval is [105, 145], generalizing the detection value to "100-140 mg / dL". This generalization method preserves the approximate range of the data while concealing the precise numerical value, thus meeting the needs of data analysis.

[0045] For text-based data such as symptom descriptions and diagnostic conclusions, AI Natural Language Processing (NLP) technology is used for semantic generalization. By removing precise descriptions and retaining key semantic information, the text data is desensitized. For example, "visited XX pet hospital on March 15, 2025" is generalized to "visited a pet hospital in a certain region in Q1 2025," which retains the time range and location information of the visit while hiding the specific date and hospital name; "positive for canine parvovirus infection (CPV)" is generalized to "positive for canine enterovirus infection," reducing the sensitivity of the information without affecting the disease type judgment.

[0046] For spatiotemporal data such as consultation time and hospital location, time data is generalized to a "year-quarter" or "year-month" granularity, and location data is generalized to a "city-region" granularity. For example, "Beijing Chaoyang District XX Road Pet Hospital" is generalized to "Beijing Chaoyang District Pet Hospital," hiding the specific street information and retaining only the city and region information. This satisfies the needs of data analysis and statistics while protecting the hospital's detailed location information.

[0047] To achieve dynamic adaptation of desensitization strategies to data attributes and application scenarios, this invention constructs a desensitization algorithm matching model based on machine learning. This model uses data sensitivity level, data type, and application scenario as input features and is trained using a large amount of sample data. During training, the model continuously learns the mapping relationship between different feature combinations and the optimal desensitization algorithm and parameters. After training, the model can automatically output the optimal desensitization algorithm and corresponding parameters based on the input data features. For example, for core sensitive data, the model will match the AES encryption algorithm; in a diagnostic analysis scenario, for general sensitive data, the model will match a generalization algorithm. In this way, dynamic and accurate adaptation of "data attributes - application scenario - desensitization strategy" is achieved, improving the targeting and effectiveness of desensitization processing.

[0048] To ensure that the anonymized data meets the requirements of security, usability, and accuracy, AI algorithms are used to automatically verify the anonymization results. The verification process includes three aspects: accuracy verification, security verification, and usability verification.

[0049] For structured data, the integrity of the anonymized data format is checked, requiring an accuracy of 99.5% or higher. For text data, the semantic preservation accuracy is evaluated using a semantic similarity algorithm, requiring an accuracy of 98% or higher. Only data meeting these two conditions is considered to meet the accuracy requirements. A simulated restoration attack is used to test the success rate of restoring the anonymized data to its original form. The success rate of restoration attacks is required to be no more than 0.5% to ensure high security and difficulty in unauthorized restoration of the anonymized data. In specific application scenarios, the usability of the anonymized data is evaluated. In analytical scenarios, the data usability is required to reach 92% or higher, meaning the anonymized data can meet the needs of most data analysis tasks and will not significantly affect the analytical value of the data due to anonymization.

[0050] For data that fails verification, the system will automatically adjust the de-identification parameters, such as increasing the generalization coefficient or switching the de-identification algorithm, and then re-process the data for de-identification. This process is repeated multiple times until the de-identification result meets all verification criteria.

[0051] To ensure the security, integrity, and availability of desensitized pet medical data during storage and transmission, a comprehensive and sophisticated data security management system has been established, covering three key aspects: hierarchical storage mechanism, secure transmission mechanism, and data backup strategy.

[0052] For core sensitive data, such as pet microchip numbers and owners' complete addresses, which have undergone AES encryption, a distributed encrypted database is used for storage. To further enhance data security, a triple-key protection system is established: a system key, a user key, and a device key. The system key is centrally managed by the data center and serves as the basic encryption layer; the user key is held by authorized users and used for secondary encryption; and the device key is bound to the storage device, serving as the final line of defense. Only when all three keys are simultaneously verified can an authorized administrator trigger the decryption operation. This multi-layered key management mechanism greatly improves the security of core sensitive data and effectively prevents the risk of data leakage. For example, in practical applications, even if a key at one stage is illegally obtained, attackers cannot break through the protection of the other two keys, thus ensuring the security of core data.

[0053] For highly sensitive and generally sensitive data, such as owner names, contact information, and pet nicknames, after generalization, masking, or replacement processing, a partitioned storage strategy is adopted. Data is divided into different storage partitions according to data type (medical data, disease data) and time dimension (year of treatment, quarter), and each partition is configured with strict access control. Different levels of users can only access data partitions within their authorized scope, thus achieving granular data management. For example, researchers can access the disease data partition for disease statistical analysis, while internal management personnel mainly access the medical data partition for routine management operations. This partitioned storage and access control method not only improves data security but also facilitates data management and use, improving work efficiency.

[0054] For non-sensitive data, such as basic pet information (breed, gender, etc.), object storage services, such as a Hadoop-based distributed file system, are used. This storage method offers advantages in high-concurrency access and batch analysis, meeting the needs of large-scale data processing. For example, when performing statistical analysis of pet breeds, large amounts of non-sensitive data can be quickly retrieved from object storage for analysis, greatly improving the efficiency and accuracy of data analysis. Furthermore, object storage services also possess good scalability and fault tolerance, adapting to the continuous growth of data volume and the stable operation of the system.

[0055] To ensure the security of anonymized data during cross-device and cross-institutional transmission and prevent data theft, tampering, or loss, SSL / TLS encryption protocols are used to establish secure transmission channels. By encrypting transmitted data, SSL / TLS ensures that the data exists in ciphertext form during transmission, preventing attackers from obtaining the original data content even if intercepted. For example, when anonymized medical data is transmitted between a veterinary hospital and a research institution, encryption via SSL / TLS effectively guarantees the security of the data transmitted over the public internet, avoiding the risk of data leakage.

[0056] During transmission, a data checksum is added to each data item. Upon receiving the data, the receiving end verifies its integrity based on the checksum. If data is detected as tampered with or lost during transmission, the receiving end will promptly issue an alarm and request retransmission. This data verification mechanism ensures the accuracy and integrity of transmitted data, improving data transmission reliability. For example, in a large-scale data transmission task, some data was lost due to network instability. Through the detection of the data checksum, the system promptly detected and retransmitted the lost data, ensuring complete data reception.

[0057] For core sensitive data, daily incremental backups are used, meaning only data that changes on that day is backed up each day to reduce backup time and storage space usage. Other data is backed up weekly in full, ensuring data integrity and consistency. Backup data is also anonymized and has the same security level as the original data to prevent risks from data leakage.

[0058] By combining local backups and off-site disaster recovery, even if the local data center experiences a catastrophic failure such as a fire or earthquake, data can still be quickly restored from the off-site disaster recovery center, ensuring business continuity and data integrity. For example, in a simulated disaster recovery exercise, after a failure in the local data center, the system successfully restored all anonymized data from the off-site disaster recovery center within a short period of time, ensuring the normal operation of pet medical services. Meanwhile, regular backup operations facilitate data traceability and recovery, providing strong support for data management and use.

[0059] Accurate prediction of disease trends is crucial for the prevention and control of pet diseases. This invention employs an LSTM (Long Short-Term Memory) neural network from time series models to conduct in-depth analysis of desensitized disease diagnostic data, thereby achieving precise prediction of disease trends in specific regions and for specific pet breeds.

[0060] The anonymized disease diagnosis data is preprocessed to extract key information such as time, region, pet breed, and disease type, which is then used as input to the LSTM neural network. This generalization process effectively protects pet owners' privacy while preserving key data features, providing a reliable data foundation for model training. For example, specific dates are generalized to months, and detailed addresses to regions, ensuring both the analysis of disease trends and data security.

[0061] An LSTM neural network was trained using a large amount of historical disease diagnosis data, and the model's parameters were adjusted to accurately capture the time-series features and potential patterns in the disease data. After thorough training, the model can predict the disease risk of specific regions and breeds of pets over the next 1-3 months based on the input current disease data. For example, by analyzing the incidence data of canine distemper in a certain region over the past few years, the model can predict the incidence risk level of canine distemper in that region over the next three months, providing a scientific basis for relevant departments to formulate prevention and control measures in advance.

[0062] Based on epidemiological characteristics, this study further analyzes the transmission routes and influencing factors of diseases. It considers the impact of seasonal changes on disease transmission, such as the fact that some diseases are more easily transmitted in summer due to high temperature and humidity; it analyzes the role of pet activity range in disease spread, as pets with wider ranges have more opportunities to come into contact with other pets, thus increasing their risk of infection and transmission; and it studies the relationship between vaccination rates and disease incidence, showing that high vaccination rates can effectively reduce disease morbidity. Through a comprehensive analysis of these influencing factors, this study provides comprehensive decision support for disease prevention and control, helping to formulate more precise and effective prevention and control strategies.

[0063] To improve the accuracy and efficiency of pet diagnosis and treatment, a medical knowledge graph was constructed. Based on anonymized medical record data, a collaborative filtering algorithm was used to recommend the optimal treatment plan for pets. At the same time, key treatment parameters were optimized for common diseases to form standardized treatment templates.

[0064] By collecting a wealth of interconnected information, including pet breeds, symptoms, diseases, examination items, medication regimens, and surgical methods, a comprehensive medical knowledge graph is constructed. This knowledge graph graphically displays the relationships between various entities, providing a rich knowledge foundation for recommending treatment plans. For example, the knowledge graph clearly shows the association between a certain symptom and multiple diseases, as well as the different examination items and medication regimens for each disease.

[0065] The desensitized medical record data is processed to extract generalized information such as symptoms, test results, and diagnostic conclusions. Based on this information, a collaborative filtering algorithm is used to analyze treatment plans for similar medical records and recommend the optimal treatment plan for the current pet. The collaborative filtering algorithm calculates the similarity between pets, finds historical medical records with similar symptoms to the current pet, and refers to the treatment plans in these records for recommendations. For example, when a pet exhibits symptoms such as vomiting and diarrhea, the system uses a collaborative filtering algorithm to find medical records of other pets with similar symptoms, analyzes their diagnostic results and medication plans, and recommends the most suitable treatment drugs and dosages for the current pet.

[0066] For common diseases such as canine parvovirus infection and feline panleukopenia, we analyze treatment data after desensitization, including key parameters such as drug dosage and treatment duration. By statistically analyzing the treatment effects of different cases, we identify the optimal combination of treatment parameters and develop standardized treatment templates. For example, for canine parvovirus infection, the analysis of a large amount of treatment data determined the optimal antiviral drug dosage and treatment duration, providing clear guidance for veterinarians in clinical treatment and improving treatment effectiveness and cure rates.

[0067] To strengthen the supervision of the pet medical industry and improve the service quality of medical institutions, this invention uses a clustering algorithm based on anonymized consultation data and treatment effect data to evaluate the treatment efficiency of different pet hospitals and veterinarians and identify abnormal behaviors during the treatment process.

[0068] Collect anonymized patient data, including consultation duration, examination items, medication use, and treatment effectiveness data such as cure rate and recurrence rate. Preprocess this data to remove noise and outliers, ensuring accuracy and reliability. For example, cases with significantly excessively long or short consultation durations are screened and corrected to avoid interfering with the evaluation results.

[0069] Clustering algorithms, such as K-means clustering, were used to perform cluster analysis on the treatment efficiency data of different pet hospitals and veterinarians. Hospitals and veterinarians were categorized based on indicators such as treatment time, cure rate, and recurrence rate, generating a medical quality assessment report. Cluster analysis clearly identifies which hospitals and veterinarians have high treatment efficiency and which have room for improvement. For example, hospitals and veterinarians with short treatment times, high cure rates, and low recurrence rates were grouped together as industry benchmarks; hospitals and veterinarians with long treatment times, low cure rates, and high recurrence rates were grouped into another category, requiring focused attention and improvement.

[0070] Establishing an abnormal behavior identification model, combined with medical knowledge and industry standards, allows for the monitoring and analysis of data during the diagnosis and treatment process. This model identifies abnormal behaviors such as irrational medication use and excessive testing, providing a basis for industry regulation and internal institutional optimization. For example, analyzing medication records may reveal instances of veterinarians overdosing medication or frequently changing drugs, which could be indicative of irrational medication use. Comparing examination items and diagnostic results may reveal unnecessary examinations in some cases, constituting excessive testing. Timely detection of these abnormal behaviors and taking appropriate measures helps regulate the pet medical market and protect the health rights of pets.

[0071] Therefore, the present invention adopts the above-mentioned AI desensitization processing and intelligent analysis management method for pet medical data. Through multi-source data collection, preprocessing, hierarchical classification and labeling, dynamic desensitization, secure storage and transmission, and intelligent analysis, it achieves efficient, secure and intelligent management of pet medical data, providing comprehensive technical support for the pet medical industry.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for AI-based de-identification processing and intelligent analysis management of pet medical data, characterized in that, Includes the following steps: S1. Adopt a multi-source data collection strategy and comprehensively utilize multiple data collection channels to obtain various key information in the pet medical process, including basic pet information, diagnosis and treatment data, and disease prevention and control data. S2. Preprocess the collected multi-source pet medical data, including deleting duplicate records, processing data with abnormal formats, and unifying the data format; S3. Classify and label the pre-processed pet medical data. Based on the risk of data privacy leakage, the data is divided into core sensitive data, highly sensitive data, general sensitive data and non-sensitive data. According to the degree of data structure, the data is further classified and labeled into three categories: structured data, semi-structured data and unstructured data. S4. Based on the hierarchical classification results, use AI algorithms to dynamically match the optimal desensitization strategy, construct a hybrid desensitization model, and desensitize pet medical data. S5. Construct a data security management system, covering hierarchical storage mechanisms, secure transmission mechanisms, and data backup strategies, to ensure the security, integrity, and availability of de-identified pet medical data during storage and transmission. S6. Intelligent analysis algorithms are used to conduct in-depth analysis of desensitized pet medical data, including disease epidemic trend prediction, treatment plan recommendation, medical quality assessment, and abnormal behavior identification.

2. The method for AI-based desensitization processing and intelligent analysis management of pet medical data according to claim 1, characterized in that, Multi-source data acquisition strategies include: It connects with the pet hospital management system to obtain basic information about pets, medical treatment data, and disease prevention and control data. The veterinarian's mobile diagnostic terminal can quickly record the pet's symptoms and preliminary examination results on-site and upload them to the data center in a timely manner. It interfaces with laboratory testing equipment to automatically collect test data, including blood biochemical indicators and microbial culture results; A self-reporting portal is set up, allowing owners to fill in details about their pets' daily eating habits and living environment via a mobile app or webpage.

3. The method for AI-based desensitization processing and intelligent analysis management of pet medical data according to claim 1, characterized in that, Preprocessing of the collected multi-source pet medical data includes: Set a unique identifier field to compare and filter data, and delete duplicate records; For medical records missing key fields, supplement or mark them as invalid data and delete them as appropriate; for erroneous detection values, make a comprehensive judgment based on the pet's clinical symptoms and other test results, and correct or delete them if they are confirmed to be erroneous data. The date format is "YYYY-MM-DD", and the units for testing indicators are standardized.

4. The method for AI-based desensitization processing and intelligent analysis management of pet medical data according to claim 1, characterized in that, The preprocessed pet medical data is graded, classified, and labeled, including: Based on the risk of data privacy breaches, data is categorized into core sensitive data, highly sensitive data, generally sensitive data, and non-sensitive data. Based on the degree of data structuring, the data is divided into three categories: structured data, semi-structured data, and unstructured data, and then labeled accordingly.

5. The method for AI-based desensitization processing and intelligent analysis management of pet medical data according to claim 1, characterized in that, Desensitization treatment includes: For character fields in generally sensitive data, a substitution algorithm is used for desensitization. This algorithm is suitable for certain fields in highly sensitive data and employs a masking algorithm. For core sensitive data, the AES symmetric encryption algorithm is used for encryption. For structured and semi-structured data that require analysis support, a generalization algorithm based on AI models to dynamically adjust the generalization coefficient is used for desensitization processing.

6. The method for AI-based desensitization processing and intelligent analysis management of pet medical data according to claim 1, characterized in that, The tiered storage mechanism in S5 includes: For core sensitive data that has been encrypted with AES, a distributed encrypted database is used for storage, and a triple key protection system is set up, namely system key, user key and device key. For highly sensitive and generally sensitive data that have undergone generalization, masking, or replacement processing, a partitioned storage strategy is adopted. The data is divided into different storage partitions according to data type and time dimension, and strict access control is configured for each partition. For non-sensitive data, object storage services are used. Secure transmission mechanisms include: In the process of data transmission across terminals and organizations, a secure transmission channel is established using the SSL / TLS encryption protocol; During transmission, a data checksum is added to each data item, and the receiving end verifies the integrity of the data based on the checksum. Data backup strategies include: For core and sensitive data, daily incremental backups are used; other data is backed up fully every week. This is achieved through a combination of local backup and off-site disaster recovery.

7. The method for AI-based desensitization processing and intelligent analysis management of pet medical data according to claim 1, characterized in that, The epidemic trend prediction uses LSTM neural network to perform time series analysis on desensitized disease diagnosis data, extracts generalized key information to train the model, predicts the disease risk of specific regions and varieties in the next 1-3 months, and combines epidemiological characteristics to analyze the transmission path and influencing factors. The treatment plan recommendation constructs a medical knowledge graph, processes desensitized medical record data based on collaborative filtering algorithm, extracts symptom and diagnosis information to recommend the optimal treatment plan, and forms standardized templates for common diseases; The medical quality assessment and abnormal behavior identification uses the K-means clustering algorithm to evaluate the treatment efficiency of pet hospitals and veterinarians, generate medical quality reports, establish abnormal behavior models to monitor treatment data, and identify abnormal behaviors.

Images