AI-based biochemical detection result interpretation system and method

Abstract

The invention discloses a biochemical detection result interpretation system and method based on AI, and relates to the technical field of biochemical detection.The biochemical detection result interpretation system comprises a data acquisition module, a data preprocessing module, a data processing module and a data processing module, the data cleaning module cleans data, removes anomalies according to a 3 sigma criterion, normalizes the data according to items and sends the normalized data to the AI interpretation model module; an MLP and CNN architecture is built in the module, through massive sample training, an attention mechanism is introduced to focus key indexes, the accuracy is calculated by using cross entropy loss, and preliminary diagnosis suggestions are transmitted to the result verification and optimization module in seconds; and in combination with a knowledge base updated by experts, reasoning and semantic understanding verification optimization are carried out according to rules, and a visual report is made by a report generation module and is stored in a data storage and management module. Biochemical detection result interpretation and AI precise analysis are assisted, the diagnosis accuracy is improved, the process is accelerated, precise diagnosis can be achieved through a basic layer, and the doctor-patient experience is optimized.

Description

Technical Field

[0001] The present invention relates to the field of biochemical detection technologies, and particularly to an AI-based system and method for interpreting biochemical test results. Background Art

[0002] In the modern medical field, biochemical tests play a crucial role as a key means for disease diagnosis, condition monitoring, and health assessment. With the rapid development of detection technologies, various biochemical test instruments can quickly and accurately obtain biochemical index data of a large number of patients, covering many items such as blood routine, urine routine, liver function, kidney function, blood lipids, and blood glucose. However, what does not match is that the interpretation of these complex test results faces numerous difficulties.

[0003] On the one hand, traditional interpretation methods mainly rely on the professional knowledge and clinical experience of medical staff. However, the medical knowledge system is huge and complex, and the correlations between the symptoms of various diseases and biochemical indexes vary greatly. Even experienced doctors are difficult to comprehensively and immediately master all the latest information, and it is very easy to have omissions or misjudgments during the interpretation process. Especially in the face of some rare diseases, complex syndromes, or the co-occurrence of multiple diseases, the accuracy and timeliness of diagnosis are greatly reduced by relying solely on manual interpretation.

[0004] On the other hand, the professional levels of medical staff in medical institutions at different regions and different levels vary greatly. Due to limited resources in primary medical institutions, medical staff have few opportunities to contact difficult and complicated diseases, and the knowledge update is relatively lagging. They often feel powerless when interpreting biochemical test results, resulting in patients possibly needing to be referred to higher-level hospitals for re-examination and diagnosis, delaying the treatment time, and increasing the medical cost and the burden on patients.

[0005] Furthermore, with the improvement of people's health awareness and the popularization of physical examinations, the biochemical data of a large number of healthy people in physical examinations also urgently need professional interpretation. Currently, lacking efficient and intelligent tools for assistance, these data cannot fully play the role of warning health risks and guiding early intervention.

[0006] At the same time, most existing medical information systems focus on the collection and storage of test data and fail to deeply integrate intelligent interpretation functions. Even if some systems have simple reference range prompts, they far from meet the needs of clinical accurate diagnosis. The data in each link is isolated and cannot provide strong support for subsequent scientific research and disease prevention and control.

[0007] In addition, patients' requirements for the right to know and participation in their own test results are increasing day by day. They expect to quickly and intuitively understand their own health conditions, but the current mode is difficult to provide patients with convenient and easy-to-understand interpretation services, and the doctor-patient communication is blocked. In summary, there is an urgent need for an innovative AI-based system and method for interpreting biochemical test results to break the existing deadlock and improve the quality and efficiency of medical services. Summary of the Invention

[0008] The AI-based biochemical test result interpretation system and method proposed by the present invention are used to solve the problems mentioned in the above prior art.

[0009] To achieve the above object, the present invention adopts the following technical solutions: An AI-based biochemical test result interpretation system, comprising:

[0010] Data acquisition module: Equipped with a variety of adaptation interfaces, docked with biochemical test instruments, using a wireless transmission protocol, preferably Bluetooth Low Energy (BLE), to collect complete blood count, urine routine, and full biochemical index data in real time through a standardized data interface. The collection frequency is dynamically adjusted according to the detection throughput, and the collection frequency adjustment formula is set as f = f0 + k×ΔT, where f0 is the basic collection frequency, k is the throughput coefficient, and ΔT is the change value of the real-time detection throughput. At the same time, the patient's age, gender, and medical history information are recorded, and the CRC checksum algorithm is used to ensure the complete and accurate transmission of data to the data preprocessing module. In case of checksum failure, automatic retransmission is performed;

[0011] Data preprocessing module: Clean the collected data, use the 3σ criterion based on statistics to identify outliers. Let the data sequence of a certain index be x i , with the mean value μ and the standard deviation σ. When |x i - μ| > 3σ, it is determined as an outlier and corrected using the linear interpolation method; Normalize for different test items, including applying logarithmic transformation y new = log(y + 1) to the enzyme activity index y, so that the data is in a unified dimension. The processed data is pushed to the AI interpretation model module in real time;

[0012] AI interpretation model module: Built-in multi-layer perceptron (MLP) combined with convolutional neural network (CNN) architecture based on deep learning. The kernel size of the convolutional layer is set to 3×3, the stride is 1, and the padding is 1. It is trained using the clinical biochemical data of disease samples, and the attention mechanism is introduced during training. Let the attention weight formula s ij be the association score between the position i of the feature map and the category j, focusing on the key index features. The model outputs preliminary diagnosis suggestions, including disease risk warnings and disease trend judgments, and uses the cross-entropy loss function C is the number of categories, y c is the true label, p c is the predicted probability, and the result is transmitted to the result verification and optimization module in seconds;

[0013] Result verification and optimization module: Combined with the clinical knowledge base updated weekly by a medical expert team. The knowledge base is constructed using a knowledge graph, with nodes being diseases, symptoms, and indicators, and edges being association relationships; Verify the preliminary diagnosis suggestions output by the AI, using a rule-based reasoning system, and set the rule matching degree formula The number of currently successfully matched rules is matchedRules, and the total number of rules is totalRules. Optimize the diagnostic suggestions based on the verification results to improve the diagnostic accuracy, and push the optimized results to the report generation module;

[0014] Report generation module: Generate a visual report in a standard format based on the optimized diagnostic suggestions. The report includes charts and text descriptions. The charts are drawn using the Echarts library to intuitively display the changing trends of key indicators. At the same time, the report is stored in the data storage and management module;

[0015] Data storage and management module: Adopt the distributed storage architecture Ceph, set the data storage redundancy strategy to 3 replicas, and archive the patient's previous biochemical test data for a long time; Regularly use data mining techniques, including association rule mining, and set the support degree formula A and B are indicators or disease events. Mine the potential associations between indicators to provide data support for AI model optimization and clinical research, support fast data retrieval, and use the inverted index technology to accelerate the retrieval;

[0016] User interaction module: Design interaction interfaces for patients and doctors respectively. The patient interface is carried on the mobile APP, developed based on the ReactNative framework, and adapted to iOS and Android systems. It is used to query personal test reports and receive health popular science knowledge push. The popular science content is customized according to the patient's test results, and the collaborative filtering algorithm is used R u，i is the recommendation score of user u for knowledge, N u is the set of users similar to user u, w u，j is the user similarity weight, r j，i is the score of similar user j for knowledge i.

[0017] Furthermore, when the data acquisition module is docked with the detection instrument, it adopts a wireless transmission protocol, including Bluetooth Low Energy (BLE) or Wi-Fi6, and at the same time has an automatic reconnection mechanism. When the connection is interrupted, set the reconnection time interval formula T0 is the basic reconnection time, n is the number of consecutive reconnection failures, N is the maximum number of allowed reconnection failures, and reconnect within 10 seconds.

[0018] Furthermore, during the training process of the AI interpretation model module, in addition to the deep learning architecture, transfer learning technology is also introduced, and set the transfer learning weight formula W transfer =α×W source +(1-α)×W target ,W source is the weight of the source domain model, W targetLet the weight of the target domain model be \(\omega\), and \(\alpha\) be the transfer coefficient. By leveraging the transfer of knowledge in the field of similar diseases, the model training is accelerated, and the dependence on large-scale samples is reduced.

[0019] Furthermore, when the result verification and optimization module combines with the clinical knowledge base, it adopts semantic understanding technology to accurately extract key knowledge points, including identifying the key index thresholds and treatment plans in the clinical practice guidelines. Let the semantic understanding accuracy formula be where \(correctWords\) is the number of keywords correctly understood, and \(totalWords\) is the total number of keywords, which improves the efficiency and accuracy of verification and optimization.

[0020] Furthermore, the patient-side APP of the user interaction module has a social sharing function. Patients can share their test results and health experiences under anonymous protection to form a mutual assistance community. At the same time, the APP adopts a gamification design concept, sets health tasks and point rewards, and let the point acquisition formula be a i is the task type coefficient, b i is the number of tasks completed, which motivates patients to pay attention to their health and have regular tests.

[0021] Furthermore, during the data mining process of the data storage and management module, for chronic disease management, including diabetes and hypertension, it focuses on the long-term index changes and uses time series analysis. Let the autocorrelation function be x t is the index value at time \(t\), is the mean value, and \(k\) is the lag order. By mining the index change rules, it can predict the risk of disease deterioration in advance, providing a basis for personalized treatment of patients.

[0022] Furthermore, a method for applying an AI-based biochemical test result interpretation system includes the following steps:

[0023] Data collection: Use the data collection module to connect with the biochemical test instrument to collect the test results and the patient's basic information, and transmit them to the data preprocessing module;

[0024] Data preprocessing: Clean and normalize the collected data through the data preprocessing module, and then push the processed data to the AI interpretation model module;

[0025] AI interpretation: With the help of the AI interpretation model module, use the deep learning architecture to analyze the preprocessed data, output preliminary diagnosis suggestions, and transmit them to the result verification and optimization module;

[0026] Result verification and optimization: According to the result verification and optimization module, verify and optimize the preliminary diagnosis suggestions in combination with the clinical knowledge base, and push them to the report generation module;

[0027] Report generation: Use the report generation module to generate a visual report of the optimized diagnostic suggestions, store it in the data storage and management module, and make it available to patients and doctors for review;

[0028] Data storage and management: Use the data storage and management module to perform distributed storage and mining of data to provide support for system optimization and clinical research;

[0029] User interaction: Relying on the user interaction module, patients and doctors interact through the mobile APP and the web respectively to obtain information and assist in diagnosis.

[0030] Furthermore, in the data collection step, the collection efficiency is further optimized by adopting multi-threaded concurrent collection technology, dynamically allocating threads according to the number of detection instruments, and setting the thread allocation coefficient formula n instrument is the actual number of testing instruments, n max The maximum number of detection instruments supported by the system.

[0031] Furthermore, in the AI interpretation step, when encountering difficult cases, the expert system module in the model is activated, and the experience knowledge base of top medical experts is introduced. Through case reasoning, the diagnostic accuracy is improved to ensure accurate judgment of complex diseases.

[0032] Furthermore, in the user interaction step, when the doctor is analyzing the case on the web interface, he can invite a remote expert for consultation with one click, communicate in real time through the video conferencing system, share case information, and set a consultation response time formula. Basic consultation response time, m a is the number of experts participating in the consultation, n a is the network bandwidth coefficient.

[0033] Compared with the existing technology, the beneficial effects of the present invention are:

[0034] In terms of diagnostic accuracy, the AI interpretation model has been trained with massive data and combined with the attention mechanism to focus on key indicators. The interpretation accuracy rate exceeds 85%, far exceeding the limitations of manual interpretation. It can accurately identify disease risks and judge disease trends, especially in the diagnosis of difficult and rare diseases, laying the foundation for precision treatment.

[0035] Efficiency has been significantly improved. The data acquisition module uses multi-threaded concurrent data collection and connects to various instruments in real time to quickly obtain data. Diagnostic suggestions are output in seconds, which optimizes the diagnosis and treatment process, reduces patient waiting time, and primary medical institutions can also quickly obtain reliable diagnostic support.

[0036] With powerful clinical assistance, the result verification and optimization module combines with the cutting-edge knowledge base, and uses semantic understanding and rule reasoning to make the diagnostic suggestions conform to the clinical situation, providing reliable references for doctors, reducing the misdiagnosis rate. The functions of multi-case comparison and remote consultation facilitate communication and collaboration.

[0037] The patient experience is excellent. The mobile APP is convenient for querying reports and receiving popular science. Social sharing and gamification incentives encourage patients to pay attention to their health. The visual report is easy to understand, enhancing doctor-patient communication and comprehensively assisting the upgrade of medical services. Brief Description of the Drawings

[0038] Figure 1 It is a schematic block diagram of a biochemical test result interpretation system based on AI proposed by the present invention;

[0039] Figure 2 It is a schematic block diagram of a biochemical test result interpretation method based on AI proposed by the present invention. Detailed Embodiments

[0040] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0041] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", etc. indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention.

[0042] In addition, the terms "first" and "second" are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the said features. In the description of the present invention, the meaning of "a plurality" is two or more, unless otherwise specifically defined. In addition, the terms "mounted", "connected", and "coupled" should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium, and it may be the communication inside two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. The present invention will be further described in detail below with reference to the accompanying drawings.

[0043] Refer to Figure 1-2 : An AI-based biochemical test result interpretation system, comprising:

[0044] Data acquisition module: Equipped with a variety of adapter interfaces, such as RS232, USB, and network interfaces, seamlessly docked with mainstream brand biochemical test instruments such as Roche and Abbott, using a wireless transmission protocol, preferably Bluetooth Low Energy (BLE), with its transmission frequency band at 2.4 GHz and a transmission power not exceeding 10 mW, ensuring stable data transmission and low power consumption. Real-time collect various types of indicators covering blood routine, urine routine, and comprehensive biochemical tests, etc., through a standardized data interface. The collection frequency is dynamically adjusted according to the detection throughput, and a collection frequency adjustment formula f = f0 + k × ΔT is set, where f0 is the basic collection frequency, k is the throughput coefficient, and ΔT is the real-time change value of the detection throughput, up to 10 data per second. At the same time, basic information such as the patient's age, gender, and medical history is recorded, and the CRC checksum algorithm is used to ensure the complete and accurate transmission of data to the data preprocessing module. In case of checksum failure, automatic retransmission is performed.

[0045] Data preprocessing module: Clean the collected data, use the 3σ criterion based on statistics to identify outliers. Let the data sequence of a certain indicator be x i , with a mean of μ and a standard deviation of σ. When |x i - μ| > 3σ, it is determined as an outlier and corrected using linear interpolation; Normalize for different test items. For example, for the enzyme activity index y, a logarithmic transformation y new = log(y + 1) is used to make the data in a unified dimension. The processed data is pushed to the AI interpretation model module in real time.

[0046] AI Interpretation Model Module: It incorporates a multi-layer perceptron (MLP) based on deep learning combined with a convolutional neural network (CNN) architecture. The kernel size of the convolutional layer is set to 3×3, the stride is 1, and the padding is 1. It is trained using over 1 million pieces of clinical biochemical data covering normal samples and typical samples of various diseases. The attention mechanism is introduced during training, and the attention weight formula is set. s ij is the association score between the feature map position i and the category j, focusing on key indicator features. The model outputs preliminary diagnostic suggestions, covering disease risk warnings and the judgment of the disease trend. The interpretation accuracy rate is required to reach over 85%, and the cross-entropy loss function is adopted. C is the number of categories, y c is the true label, p c is the predicted probability, and the result is transmitted to the result verification and optimization module in seconds.

[0047] Result Verification and Optimization Module: It combines a clinical knowledge base updated weekly by a medical expert team. The knowledge base is constructed using a knowledge graph, with nodes being diseases, symptoms, indicators, etc., and edges being association relationships. It verifies the preliminary diagnostic suggestions output by the AI using a rule-based reasoning system, and sets the rule matching degree formula matchedRules is the number of currently successfully matched rules, and totalRules is the total number of rules. For example, if white blood cells and neutrophils are elevated and accompanied by fever symptoms, it is recommended to check for the source of infection, and the trigger rule matching degree needs to exceed 0.6. Optimize the diagnostic suggestions according to the verification results to improve the diagnostic accuracy, and push the optimized results to the report generation module.

[0048] Report Generation Module: It generates a visual report in a standard format based on the optimized diagnostic suggestions. The report includes charts and text descriptions. The charts are drawn using the Echarts library to intuitively display the change trends of key indicators. The text part elaborates in detail on the diagnostic conclusions and subsequent treatment suggestions, such as review items and referral departments. Paper or electronic reports can be generated as needed for the convenience of patients and doctors to view. At the same time, the report is stored in the data storage and management module.

[0049] Data Storage and Management Module: It adopts a distributed storage architecture Ceph, and the data storage redundancy strategy is set to 3 replicas to ensure high reliability and scalability. It archives the patient's previous biochemical test data for a long time. Regularly use data mining techniques, such as association rule mining, and set the support degree formula A and B are indicators or disease events, mining the potential associations between indicators, providing data support for AI model optimization and clinical research, supporting fast data retrieval, with the retrieval response time not exceeding 5 seconds, and using the inverted index technology to accelerate the retrieval.

[0050] User interaction module: Interaction interfaces are designed for patients and doctors respectively. The patient interface is installed on the mobile APP, developed based on the ReactNative framework, and is compatible with iOS and Android systems. It can query personal test reports and receive health science knowledge push notifications. The science content is personalized according to the patient's test results and adopts collaborative filtering algorithm. R u，i is the recommendation score of user u for knowledge, N u is a set of users similar to user u, w u，j is the user similarity weight, r j，i The doctor interface is a web-based application developed using the Vue.js framework. It features case analysis, AI diagnostic recommendations, and the ability to manually adjust diagnostic conclusions. It supports multi-case comparative analysis, facilitating rapid diagnosis by doctors, with an interface response time of less than 3 seconds.

[0051] In the present invention, when the data acquisition module is connected to the detection instrument, a wireless transmission protocol such as Bluetooth Low Energy (BLE) or Wi-Fi 6 is used to ensure the stability of data transmission. At the same time, it has an automatic reconnection mechanism. When the connection is interrupted, the reconnection time interval formula is set. T0 is the basic reconnection time, n is the number of consecutive reconnection failures, and N is the maximum allowed number of reconnection failures. The connection can be reestablished within 10 seconds to ensure the continuity of data collection.

[0052] In the present invention, in the training process of the AI interpretation model module, in addition to the deep learning architecture, the transfer learning technology is also introduced, and the transfer learning weight formula W is set transfer =α×W source +(1-α)×W target , W source is the source domain model weight, W target is the target domain model weight, α is the migration coefficient, and the knowledge migration of similar disease domains is used to accelerate model training and reduce dependence on large-scale samples. For rare disease detection, the diagnostic accuracy can be improved by more than 10% with limited samples.

[0053] In this invention, the result verification and optimization module uses semantic understanding technology when combined with the clinical knowledge base to accurately extract key knowledge points, such as identifying key indicator thresholds and treatment plans in the diagnosis and treatment guidelines, and setting a semantic understanding accuracy formula correctWords is the number of correctly understood keywords, and totalWords is the total number of keywords. This improves the efficiency and accuracy of verification and optimization, making diagnostic recommendations more in line with clinical practice.

[0054] In the system of the present invention, the patient-side APP of the user interaction module has a unique design. Among them, the social sharing function brings a brand-new experience to patients. When using the APP, patients can freely share their biochemical test results and health insights under a strict anonymous protection mechanism. This anonymous protection measure ensures the privacy and security of patients, so that they don't have to worry about personal information leakage. Through this sharing, patients can form a mutual assistance community. In this community, patients with similar health conditions can communicate about their conditions, share recovery experiences, encourage and support each other, creating a positive atmosphere.

[0055] At the same time, the APP also adopts a highly creative gamification design concept. It carefully sets various health-related tasks, such as reaching the daily step count standard, taking medications on time, and completing specific health exercises. For these tasks, the APP has established an integral reward mechanism. The integral acquisition formula is a i is the task type coefficient, b i is the number of tasks completed, to encourage patients to pay attention to their health and have regular check-ups; for example, tasks related to long-term chronic disease management may have a relatively high task type coefficient. It reflects the implementation situation of patients for specific tasks. If patients complete more tasks or complete more challenging tasks, they can obtain more points.

[0056] This gamification design and integral reward mechanism can effectively encourage patients to pay attention to their own health, making them more motivated to complete health tasks, and then forming good habits of regular check-ups and actively maintaining health, which helps to improve the overall health level of patients.

[0057] In the present invention, the data storage and management module undertakes an important data mining function. Especially in the aspect of chronic disease management, such as diseases like diabetes and hypertension, its data mining process is of great significance.

[0058] In this module, the focus is on the changes in the long-term biochemical test indicators of chronic disease patients. To achieve this goal, a time series analysis method is used. In time series analysis, an autocorrelation function is specifically set, and its expression is x t is the indicator value at time, is the mean value, and k is the lag order; by performing time series analysis and calculating the autocorrelation function on these long-term indicator data, the variation patterns of various indicators of chronic disease patients can be deeply explored. For example, it is possible to discover the fluctuation patterns of blood glucose levels in diabetic patients in different seasons and at different time periods, or the potential relationships between blood pressure values of hypertensive patients and factors such as living habits and medication use. Furthermore, the risk of patient's condition deterioration can be predicted in advance, such as predicting whether diabetic patients will develop diabetic complications, or whether hypertensive patients are at risk of sudden cardiovascular and cerebrovascular diseases.

[0059] Such prediction results obtained based on data mining can provide a strong basis for personalized treatment of patients. Doctors can adjust the medication dosage for patients and develop more reasonable life intervention plans according to the prediction situation, so as to better control the development of chronic diseases and improve the quality of life and health level of patients.

[0060] The present invention also discloses an AI-based method for interpreting biochemical test results, including the following steps:

[0061] Data collection: Use the data collection module to dock with the biochemical test instrument, collect the test results and the patient's basic information, and transmit them to the data preprocessing module.

[0062] Data preprocessing: Clean and normalize the collected data through the data preprocessing module, and push the processed data to the AI interpretation model module after processing.

[0063] AI interpretation: With the help of the AI interpretation model module, use the deep learning architecture to analyze the preprocessed data, output preliminary diagnosis suggestions, and transmit them to the result verification and optimization module.

[0064] Result verification and optimization: According to the result verification and optimization module, verify and optimize the preliminary diagnosis suggestions in combination with the clinical knowledge base, and push them to the report generation module.

[0065] Report generation: Use the report generation module to generate a visual report of the optimized diagnosis suggestions, store it in the data storage and management module, and at the same time make it available for patients and doctors to view.

[0066] Data storage and management: Use the data storage and management module to perform distributed storage and mining of data, providing support for system optimization and clinical research.

[0067] User interaction: Rely on the user interaction module, patients and doctors interact through the mobile APP and the Web side respectively to obtain information and assist in diagnosis.

[0068] In the present invention, the acquisition efficiency is further optimized, and the multi-threaded concurrent acquisition technology is adopted. Threads are dynamically allocated according to the number of test instruments, and the thread allocation coefficient formula is set ninstrument is the actual number of detection instruments, n max is the maximum number of detection instruments supported by the system, which can increase the acquisition speed by more than 30% to ensure timely data acquisition.

[0069] In the AI-based biochemical test result interpretation system constructed by the present invention, when encountering difficult cases during the diagnosis process, the system has a unique coping mechanism. At this time, the system will automatically activate the built-in expert system module in the model. This expert system module is not a simple program, but introduces an experience knowledge base of top medical experts. This knowledge base brings together the clinical experiences, diagnostic ideas, and knowledge achievements of dealing with complex diseases of many experts with profound attainments in the medical field for many years.

[0070] By deeply analyzing and reasoning a large number of cases in the knowledge base, the system can match, compare, and analyze the various indicators of the current case with similar cases in the knowledge base when facing difficult diseases. For example, for some rare diseases with complex symptom combinations, the system can make a comprehensive judgment based on the diagnostic processes and results of similar cases in the knowledge base and the specific biochemical test data of the current case.

[0071] This case-based reasoning method can make full use of the wisdom in the expert experience knowledge base and effectively make up for the limitations that may occur when simply relying on AI algorithms to deal with difficult diseases. Furthermore, it can significantly improve the accuracy of diagnosis, ensure accurate judgment when facing complex conditions, provide a reliable basis for formulating subsequent treatment plans, and strive for precious time for the treatment of patients.

[0072] In the system involved in the present invention, when a doctor conducts case analysis on the Web-based operation interface, the system is equipped with a convenient remote consultation function for him / her. The doctor only needs to click a button to invite remote experts to participate in the consultation. The system will achieve real-time communication between doctors and patients through an advanced video conferencing system. During this process, not only can voice and video communication be carried out, but also the sharing of case materials can be realized, such as the patient's biochemical test report, medical history record, imaging materials, etc., so that remote experts can comprehensively understand the case situation.

[0073] Regarding the consultation response time, there is a special formula. Let the consultation response time formula be the basic consultation response time, m a be the number of experts participating in the consultation, n a be the network bandwidth coefficient. To improve the diagnosis and treatment level, the consultation response time does not exceed 10 minutes;

[0074] This function can significantly improve the diagnosis and treatment level. Through optimization and design, it ensures that the consultation response time will not exceed 10 minutes, providing strong guarantee for the timely handling of emergency or complex cases and avoiding delaying the condition due to excessive waiting time for consultation.

[0075] The above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered within the protection scope of the present invention.

Claims

1. An AI-based biochemical test result interpretation system, characterized in that, include: Data acquisition module: Equipped with multiple adapter interfaces, it connects to biochemical testing instruments and adopts wireless transmission protocols, with Bluetooth low energy (BLE) being preferred. It uses a standardized data interface to collect blood routine, urine routine, and a full set of biochemical indicators in real time. The acquisition frequency is dynamically adjusted according to the detection flux. The acquisition frequency adjustment formula is f = f0 + k × ΔT, where f0 is the basic acquisition frequency, k is the flux coefficient, and ΔT is the real-time detection flux change value. The patient's age, gender, and medical history information are also recorded. The CRC check algorithm is used to ensure that the data is completely and accurately transmitted to the data preprocessing module. If the check fails, the data will be automatically retransmitted. Data preprocessing module: Clean the collected data, identify outliers using the 3σ criterion based on statistics. Let the data sequence of a certain index be x i , the mean is μ, and the standard deviation is σ. When |x i - μ| > 3σ, it is determined as an outlier and corrected using the linear interpolation method; Normalize for different detection items, including applying the logarithmic transformation y new = log(y + 1) to the enzyme activity index y, so that the data is in a unified dimension, and the processed data is pushed to the AI interpretation model module in real time; AI Interpretation Model Module: Built-in multi-layer perceptron MLP based on deep learning combined with convolutional neural network CNN architecture. The kernel size of the convolutional layer is set to 3×3, the stride is 1, and the padding is 1. It is trained using the clinical biochemical data of disease samples, and the attention mechanism is introduced during training. Set the attention weight formula s ij is the association score between the feature map position i and the category j, focusing on key indicator features. The model outputs preliminary diagnosis suggestions, including disease risk warnings and disease trend judgments, and uses the cross-entropy loss function C is the number of categories, y c is the true label, p c is the predicted probability, and the result is transmitted to the result verification and optimization module in seconds; Result Verification and Optimization Module: Combine with a clinical knowledge base updated weekly by a medical expert team. The knowledge base is constructed using a knowledge graph, with nodes being diseases, symptoms, and indicators, and edges being association relationships; verify the preliminary diagnosis suggestions output by AI using a rule-based reasoning system, and set a rule matching degree formula where matchedRules is the number of currently successfully matched rules and totalRules is the total number of rules. Optimize the diagnosis suggestions based on the verification results to improve the diagnostic accuracy, and push the optimized results to the report generation module; Report generation module: Generates a visual report based on the optimized diagnostic suggestions in a standardized format. The report contains charts and text descriptions. The charts are drawn using the Echarts library to intuitively display the trend of key indicators. The report is stored in the data storage and management module. Data storage and management module: Adopt the distributed storage architecture Ceph, set the data storage redundancy strategy to 3 replicas, and permanently archive the previous biochemical test data of patients; regularly apply data mining techniques, including association rule mining, and set the support formula A and B are indicators or disease events. Mine the potential associations between indicators to provide data support for AI model optimization and clinical research, support fast data retrieval, and use the inverted index technology to accelerate the retrieval; User Interaction Module: Interaction interfaces are designed for patients and doctors respectively. The patient interface is carried on the mobile APP, developed based on the ReactNative framework, and adapted to iOS and Android systems. It is used to query personal test reports and receive health science popularization knowledge push. The popularization content is customized according to the patient's test results, and the collaborative filtering algorithm is adopted. R u，i is the recommended score of knowledge for user u, N u is the set of users similar to user u, w u，j is the user similarity weight, r j，i is the score of similar user j for knowledge i.

2. The AI-based biochemical test result interpretation system according to claim 1, wherein When the data acquisition module is docked with the detection instrument, it uses a wireless transmission protocol, including Bluetooth Low Energy (BLE) or Wi-Fi 6, and also has an automatic reconnection mechanism. When the connection is interrupted, a reconnection time interval formula is set. T0 is the base reconnection time, n is the number of consecutive reconnection failures, N is the maximum allowable number of reconnection failures, and the connection is re-established within 10 seconds.

3. The AI-based biochemical test result interpretation system according to claim 1, wherein During the training process of the AI interpretation model module, in addition to the deep learning architecture, transfer learning technology is also introduced. Let the transfer learning weight formula be W transfer = α × W source + (1 - α) × W target , where W source is the weight of the source domain model, W target is the weight of the target domain model, and α is the transfer coefficient. By leveraging the knowledge transfer in the similar disease domain, the model training is accelerated and the dependence on large-scale samples is reduced.

4. The AI-based biochemical test result interpretation system according to claim 1, wherein When the result verification and optimization module combines with the clinical knowledge base, it uses semantic understanding technology to accurately extract key knowledge points, including identifying the key index thresholds and treatment plans in the diagnosis and treatment guidelines, and sets the semantic understanding accuracy formula correctWords is the number of keywords correctly understood, totalWords is the total number of keywords, which improves the efficiency and accuracy of verification and optimization.

5. The AI-based biochemical test result interpretation system according to claim 1, wherein The patient - side APP of the user interaction module has a social sharing function. Under anonymous protection, patients can share test results and health experiences to form a mutual - assistance community. At the same time, the APP adopts the concept of gamification design, sets health tasks and integral rewards, and sets an integral acquisition formula a i is the task - type coefficient, b i is the number of completed tasks, which encourages patients to pay attention to their health and have regular tests.

6. The AI-based biochemical test result interpretation system according to claim 1, characterized in that During the data mining process, the data storage and management module focuses on the long-term indicator changes for chronic disease management, including diabetes and hypertension. Using time series analysis, the autocorrelation function is set with x t being the indicator value at a certain moment, being the mean value, k being the lag order, to mine the law of indicator changes, predict the risk of disease deterioration in advance, and provide a basis for personalized treatment of patients.

7. A method of applying the AI-based biochemical test result interpretation system according to any one of claims 1-6, characterized in that, The following steps are involved: Data collection: Use the data collection module to connect with the biochemical testing instrument to collect test results and basic patient information and transmit them to the data preprocessing module; Data preprocessing: The collected data is cleaned and normalized through the data preprocessing module, and then pushed to the AI interpretation model module after processing; AI Interpretation: Leveraging the AI Interpretation Model module, a deep learning architecture is used to analyze pre-processed data, output preliminary diagnostic recommendations, and pass them to the result verification and optimization module. Result verification and optimization: Based on the result verification and optimization module, the preliminary diagnostic suggestions are verified and optimized in combination with the clinical knowledge base and pushed to the report generation module; Report generation: Use the report generation module to generate a visual report of the optimized diagnostic suggestions, store it in the data storage and management module, and make it available to patients and doctors for review; Data storage and management: Use the data storage and management module to perform distributed storage and mining of data to provide support for system optimization and clinical research; User interaction: Relying on the user interaction module, patients and doctors interact through the mobile APP and the web respectively to obtain information and assist in diagnosis.

8. The AI-based biochemical test result interpretation method according to claim 7, wherein In the data acquisition step, the acquisition efficiency is further optimized by adopting the multi-threaded concurrent acquisition technology. Threads are dynamically allocated according to the number of detection instruments. Let the thread allocation coefficient formula be n instrument be the actual number of detection instruments, and n max be the maximum number of detection instruments supported by the system.

9. The AI-based biochemical test result interpretation method according to claim 7, wherein In the AI interpretation step, when encountering difficult cases, the expert system module in the model is activated, and the experience knowledge base of top medical experts is introduced. Through case reasoning, the diagnostic accuracy is improved to ensure accurate judgment of complex diseases.

10. The AI-based biochemical test result interpretation method according to claim 7, wherein In the user interaction step, when a doctor analyzes a case on the Web interface, they can invite a remote expert for consultation with one click, communicate in real time through a video conferencing system, share case materials, and set a formula for the consultation response time is the basic consultation response time, m a is the number of experts participating in the consultation, n a is the network bandwidth coefficient.

Images