An AI physical examination intelligent main examination diagnosis method and system

Abstract

The present invention discloses an AI physical examination intelligent main inspection diagnosis method and system, which relates to the field of data processing technology, including: obtaining a user's physical examination data stream, and performing multidimensional correction to obtain a physical examination correction stream; building a physical examination abnormality detection space, and performing iterative distillation optimization to establish Q individual physical examination abnormality detection channels; inputting the physical examination correction stream into the Q individual physical examination abnormality detection channels to obtain Q individual physical examination abnormality detection results, and performing accompanying interference tracing to obtain Q accompanying interference feature sequences; performing credible risk detection on the Q individual physical examination abnormality detection results according to the Q accompanying interference feature sequences to obtain Q abnormal credible risk sequences, performing closed-loop optimization on the Q individual physical examination abnormality detection results, and generating a physical examination analysis report. The present invention solves the technical problems of inaccurate physical examination data abnormality detection and insufficient consideration of interference factors in the prior art, and achieves the technical effect of improving the accuracy and credibility of the physical examination results.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to an AI physical examination intelligent main examination diagnosis method and system. Background Art

[0002] During physical examinations, data anomaly detection often faces the challenge of insufficient accuracy, especially when processing large amounts of data. This is particularly true when processing large amounts of data, which can be susceptible to various interfering factors, such as data entry errors, physiological differences, and equipment failures. Failure to fully account for these interfering factors can reduce the reliability of test results, thereby impacting the final outcome. Summary of the Invention

[0003] The present application provides an AI physical examination intelligent main inspection and diagnosis method and system, which is used to solve the technical problems in the existing technology of inaccurate abnormal detection of physical examination data and insufficient consideration of interference factors.

[0004] In view of the above problems, the present application provides an AI physical examination intelligent main examination diagnosis method and system.

[0005] The first aspect of the present application provides an AI physical examination intelligent primary examination and diagnosis method, the method comprising:

[0006] A user's physical examination data stream is obtained according to Q physical examination categories, and the physical examination data stream is multi-dimensionally corrected based on the data layer repair mechanism to obtain a physical examination correction stream, where Q is a positive integer greater than 1; a physical examination anomaly detection space is constructed according to the Q physical examination categories, and iterative distillation optimization is performed based on the physical examination anomaly detection space to establish Q individual physical examination anomaly detection channels; the physical examination correction stream is input into the Q individual physical examination anomaly detection channels to obtain Q individual physical examination anomaly detection results, and accompanying interference tracing is performed based on the Q individual physical examination anomaly detection results to obtain Q accompanying interference feature sequences; a trusted risk detection is performed on the Q individual physical examination anomaly detection results according to the Q accompanying interference feature sequences to obtain Q abnormal trusted risk sequences; a closed-loop optimization is performed on the Q individual physical examination anomaly detection results according to the Q abnormal trusted risk sequences to generate a physical examination analysis report.

[0007] In a possible implementation, the data layer repair mechanism includes: performing logical contradiction detection on the physical examination data stream to obtain a first data repair factor; performing physiological violation detection on the physical examination data stream to obtain a second data repair factor; performing conflict detection on the physical examination data stream to obtain a third data repair factor; fusing the first data repair factor, the second data repair factor and the third data repair factor to generate a fourth data repair factor; and correcting the physical examination data stream according to the fourth data repair factor to generate the physical examination correction stream.

[0008] In a possible implementation, a physical examination anomaly detection space is constructed based on the Q physical examination categories, including: performing anomaly detection history retrieval based on the Q physical examination categories to obtain Q physical examination anomaly detection record sets; performing supervised training on M learners based on each physical examination anomaly detection record set in the Q physical examination anomaly detection record sets to establish Q physical examination anomaly detection architectures, each physical examination anomaly detection architecture including M physical examination anomaly detection models corresponding to each physical examination category, where M is a positive integer greater than 1; performing optimization analysis on the Q physical examination anomaly detection architectures based on the anomaly detection loss threshold to establish Q anomaly detection optimization architectures; and establishing the physical examination anomaly detection space based on the Q anomaly detection optimization architectures.

[0009] In a possible implementation, iterative distillation optimization is performed based on the physical examination anomaly detection space to establish Q physical examination anomaly detection channels, including: activating a distillation decision factor, where the distillation decision factor includes the number of teacher models, student model attributes, and distillation hyperparameters; performing a distillation decision on the qth anomaly detection optimization architecture in the physical examination anomaly detection space based on the distillation decision factor to generate a qth distillation decision space, where q is a positive integer, 1≤q≤Q; performing a distillation loss analysis on the qth anomaly detection optimization architecture according to the qth distillation decision space to establish a qth distillation loss space; performing iterative distillation loss optimization on the qth distillation decision space according to the qth distillation loss space to determine the qth distillation decision optimization result, and generating a qth physical examination anomaly detection channel according to the qth distillation decision optimization result.

[0010] In a possible implementation, accompanying interference tracing is performed based on the Q individual physical examination abnormality detection results to obtain Q accompanying interference feature sequences, including: evaluating the abnormality triggering degree based on the Q individual physical examination abnormality detection results to obtain Q individual physical examination abnormality triggering degrees; activating the physical examination accompanying feature factors, the physical examination accompanying feature factors including the user accompanying status, the physical examination sample status and the physical examination instrument status; based on the Q individual physical examination abnormality triggering degrees, accompanying feature collection is performed on the Q individual physical examination abnormality detection results respectively according to the physical examination accompanying feature factors to generate Q individual physical examination accompanying feature sequences; accompanying benchmark mining is performed on the Q individual physical examination categories according to the physical examination accompanying feature factors to obtain Q benchmark accompanying feature sequences; interference feature identification is performed on the Q individual physical examination accompanying feature sequences according to the Q benchmark accompanying feature sequences to generate the Q accompanying interference feature sequences.

[0011] In a possible implementation, a credible risk test is performed on the Q physical examination abnormality detection results according to the Q accompanying interference feature sequences to obtain Q abnormal credible risk sequences, including: extracting the qth accompanying interference feature sequence according to the Q accompanying interference feature sequences, the qth accompanying interference feature sequence including the qth user accompanying interference feature, the qth sample interference feature and the qth instrument interference feature; performing a credible risk assessment on the qth physical examination abnormality detection result according to the qth user accompanying interference feature to obtain a first credible risk coefficient of physical examination abnormality; performing a credible risk assessment on the qth physical examination abnormality detection result according to the qth sample interference feature to obtain a second credible risk coefficient of physical examination abnormality; performing a credible risk assessment on the qth physical examination abnormality detection result according to the qth instrument interference feature to obtain a third credible risk coefficient of physical examination abnormality; and constructing the qth abnormal credible risk sequence according to the first credible risk coefficient of physical examination abnormality, the second credible risk coefficient of physical examination abnormality and the third credible risk coefficient of physical examination abnormality.

[0012] In a possible implementation, a trusted risk assessment is performed on the qth physical examination abnormality detection result based on the qth user accompanying interference feature to obtain a first coefficient of the physical examination abnormality trusted risk, including: performing a physical examination abnormality induction prediction based on the qth user accompanying interference feature to obtain the qth user interference abnormality induction feature; performing coupling detection on the qth physical examination abnormality detection result based on the qth user interference abnormality induction feature to obtain the qth user interference abnormality coupling feature; performing loss iterative training based on the user interference abnormality trusted risk history set to generate a user interference abnormality trusted risk detection model; and inputting the qth user interference abnormality coupling feature into the user interference abnormality trusted risk detection model to generate the first coefficient of the physical examination abnormality trusted risk.

[0013] In a possible implementation, obtaining a user's physical examination data stream according to Q physical examination categories includes: obtaining a physical examination data set of the user according to an AI physical examination system; cleaning and combing the physical examination data set according to the Q physical examination categories to generate the physical examination data stream.

[0014] Possible implementations include: interpreting the physical examination analysis report according to the RAG model, obtaining a physical examination interpretation report, and encrypting and sending the physical examination interpretation report to the user.

[0015] The second aspect of the present application provides an AI physical examination intelligent main inspection and diagnosis system, the system comprising:

[0016] A multidimensional correction module is used to obtain the user's physical examination data stream based on Q physical examination categories, and perform multidimensional correction on the physical examination data stream based on the data layer repair mechanism to obtain a physical examination correction stream, where Q is a positive integer greater than 1; a distillation optimization module is used to build a physical examination anomaly detection space based on the Q physical examination categories, and perform iterative distillation optimization based on the physical examination anomaly detection space to establish Q individual physical examination anomaly detection channels; an interference tracing module is used to input the physical examination correction stream into the Q individual physical examination anomaly detection channels to obtain Q individual physical examination anomaly detection results, and perform accompanying interference tracing based on the Q individual physical examination anomaly detection results to obtain Q accompanying interference feature sequences; a trusted risk detection module is used to perform trusted risk detection on the Q individual physical examination anomaly detection results based on the Q accompanying interference feature sequences to obtain Q abnormal trusted risk sequences; a closed-loop optimization module is used to perform closed-loop optimization on the Q individual physical examination anomaly detection results based on the Q abnormal trusted risk sequences to generate a physical examination analysis report.

[0017] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0018] The present application obtains the user's physical examination data stream according to Q physical examination categories, and performs multidimensional correction on the physical examination data stream based on the data layer repair mechanism to obtain a physical examination correction stream, where Q is a positive integer greater than 1; a physical examination anomaly detection space is built according to the Q physical examination categories, and iterative distillation optimization is performed based on the physical examination anomaly detection space to establish Q physical examination anomaly detection channels; the physical examination correction stream is input into the Q physical examination anomaly detection channel to obtain Q physical examination anomaly detection results, and accompanying interference tracing is performed based on the Q physical examination anomaly detection results to obtain Q accompanying interference feature sequences; credible risk detection is performed on the Q physical examination anomaly detection results according to the Q accompanying interference feature sequences to obtain Q abnormal credible risk sequences; closed-loop optimization is performed on the Q physical examination anomaly detection results according to the Q abnormal credible risk sequences to generate a physical examination analysis report. The present invention solves the technical problems of inaccurate physical examination data anomaly detection and insufficient consideration of interference factors in the prior art, and achieves the technical effect of improving the accuracy and credibility of physical examination results through multidimensional correction, abnormality detection channel optimization, accompanying interference tracing and credible risk detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0020] Figure 1A flowchart of an AI physical examination intelligent primary diagnosis method provided in an embodiment of the present application;

[0021] Figure 2 A schematic diagram of the structure of an AI physical examination intelligent main inspection and diagnosis system provided in an embodiment of the present application.

[0022] Explanation of the accompanying symbols: multi-dimensional correction module 11, distillation optimization module 12, interference tracing module 13, trusted risk detection module 14, closed-loop optimization module 15. DETAILED DESCRIPTION

[0023] This application provides an AI physical examination intelligent main inspection and diagnosis method and system to solve the technical problems in the existing technology of inaccurate abnormality detection of physical examination data and insufficient consideration of interference factors. Through multi-dimensional correction, abnormality detection channel optimization, accompanying interference tracing and credible risk detection, the technical effect of improving the accuracy and credibility of physical examination results is achieved.

[0024] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only some of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0025] It should be noted that any variations of the terms "include" and "have" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or are inherent to these processes, methods, products or devices.

[0026] Example 1, as Figure 1 As shown, the present application provides an AI physical examination intelligent main inspection and diagnosis method, the method comprising:

[0027] Step S100: obtaining a user's physical examination data stream according to Q physical examination categories, and performing multi-dimensional correction on the physical examination data stream based on a data layer repair mechanism to obtain a physical examination correction stream, where Q is a positive integer greater than 1.

[0028] In the embodiment of the present application, the user's physical examination data stream is first obtained through the AI ​​physical examination system, and the data is cleaned and sorted according to the physical examination category Q, thereby generating a physical examination data stream. The physical examination data stream contains the user's original data under different physical examination items.

[0029] Next, a multi-dimensional correction is performed on the health examination data stream based on the data layer repair mechanism. This data layer repair mechanism consists of three parts: first, logical contradiction detection is performed to obtain the first data repair factor. Next, physiological violation detection is performed to obtain the second data repair factor. Finally, conflict detection is performed to obtain the third data repair factor. These three repair factors are combined to generate the fourth data repair factor, which is used to correct the health examination data stream, ultimately obtaining the corrected health examination stream.

[0030] Furthermore, in the method provided in the embodiment of the application, obtaining the user's physical examination data stream according to the physical examination category Q also includes:

[0031] According to the AI ​​physical examination system, a physical examination data set of the user is obtained; the physical examination data set is cleaned and sorted according to the Q physical examination categories to generate the physical examination data stream.

[0032] In an embodiment of the present application, the user's physical examination data set is first obtained through the AI ​​physical examination system. The physical examination data set refers to all the raw data generated by the user during the physical examination process, including various physical examination indicators such as blood pressure, blood sugar, weight, vision, etc. The physical examination data set may contain multiple measurement values ​​of different physical examination items, and these data are stored in an unprocessed original state. Next, the data is cleaned and sorted according to the Q physical examination categories to generate a physical examination data stream. The Q physical examination categories refer to the classification information extracted from the physical examination items, such as blood pressure, blood sugar, liver function, etc., and each category represents a physical examination item.

[0033] During the cleaning and sorting process, the AI ​​system performs multiple processing steps on the dataset. First, it removes redundant data, such as duplicate records for the same physical examination. Second, it checks the format of each data item to ensure it complies with standards, such as ensuring that blood sugar is in mmol / L. Finally, this processed data is categorized by Q physical examination categories and organized into a physical examination data stream.

[0034] Furthermore, in the method provided in the embodiment of the application, the data layer repair mechanism includes:

[0035] Performing a logical contradiction check on the physical examination data stream to obtain a first data repair factor; performing a physiological violation check on the physical examination data stream to obtain a second data repair factor; performing a conflict check on the physical examination data stream to obtain a third data repair factor; fusing the first data repair factor, the second data repair factor, and the third data repair factor to generate a fourth data repair factor; and correcting the physical examination data stream according to the fourth data repair factor to generate the physical examination correction stream.

[0036] In this embodiment, the physical examination data stream is first checked for logical contradictions. The purpose of logical contradiction detection is to determine whether the physical examination data conforms to basic logic using a series of pre-defined rules. For example, if the weight data is negative, the blood sugar level is outside the reasonable range, or there are other violations of conventional logic, this data is marked as having a logical contradiction. For this abnormal data, a first data repair factor is generated.

[0037] Next, physiological violation detection is performed. This process compares the physical examination data with medical standards to check for any violations of physiological laws. For example, if the blood pressure value is outside the measurable range (e.g., below the device's lower limit or above the upper limit), this data is considered a physiological violation and a secondary data correction factor is generated to indicate the need for correction.

[0038] Conflict detection is then performed to check whether any physical examination items do not match the user's basic information. For example, if a male user's physical examination data includes a progesterone test, this conflicts with the user's gender information. A third data repair factor is generated to indicate that this physical examination item does not match the user's information.

[0039] Then the first data repair factor, the second data repair factor and the third data repair factor are fused. The purpose of the fusion process is to merge the physical examination data that needs to be corrected and remove duplicate repair information. Through this process, the fourth data repair factor is generated.

[0040] Finally, the medical examination data stream is corrected based on the fourth data repair factor. During this process, any abnormal data items that can be corrected are corrected by reacquiring the data or requiring the user to remeasure. Any abnormal data items that cannot be directly corrected are marked as pending and the user is notified that manual verification is required. After manual processing, the corrected information is obtained. This process ultimately generates the medical examination correction stream.

[0041] Step S200: constructing a physical examination abnormality detection space according to the Q physical examination categories, and performing iterative distillation optimization based on the physical examination abnormality detection space to establish Q physical examination abnormality detection channels.

[0042] In an embodiment of the present application, when building a physical examination anomaly detection space based on Q physical examination categories, the physical examination anomaly detection record set corresponding to each physical examination category is first obtained through an anomaly detection history search. Then, based on the data in the physical examination anomaly detection record set, M learners are supervised and trained separately to establish Q physical examination anomaly detection architectures. Then, the architecture is optimized based on the anomaly detection loss threshold, the model configuration is optimized, and Q anomaly detection optimization architectures are established. Based on these optimized architectures, a physical examination anomaly detection space is established.

[0043] An iterative distillation optimization is then performed based on the physical examination anomaly detection space. This process first activates the distillation decision factors, including the number of teacher models, student model attributes, and distillation hyperparameters. Guided by the distillation decision factors, a distillation decision is made on the qth physical examination anomaly detection architecture, generating the qth distillation decision space. Next, based on the qth distillation decision space, a distillation loss analysis is performed on the qth anomaly detection optimization architecture, analyzing the model loss and forming the qth distillation loss space. Based on the distillation loss space, iterative distillation loss optimization is continued. After multiple optimization steps, the qth distillation decision optimization result is determined. Based on the qth distillation decision optimization result, the qth physical examination anomaly detection channel is generated. By traversing all Q physical examination categories, Qth physical examination anomaly detection channels are ultimately established.

[0044] Furthermore, in the method provided in the embodiment of the application, building a physical examination abnormality detection space based on the Q physical examination categories also includes:

[0045] Anomaly detection history retrieval is performed according to the Q physical examination categories to obtain Q physical examination anomaly detection record sets; supervised training is performed on M learners according to each physical examination anomaly detection record set in the Q physical examination anomaly detection record sets to establish Q physical examination anomaly detection architectures, each physical examination anomaly detection architecture includes M physical examination anomaly detection models corresponding to each physical examination category, and M is a positive integer greater than 1; optimization analysis is performed on the Q physical examination anomaly detection architectures according to the anomaly detection loss threshold to establish Q anomaly detection optimization architectures; based on the Q anomaly detection optimization architectures, the physical examination anomaly detection space is established.

[0046] In this embodiment, a historical search for abnormality detection is first performed based on Q physical examination categories. This involves backtracking through the physical examination data to retrieve the corresponding abnormality detection record sets for each physical examination category (e.g., blood sugar, blood pressure, weight, etc.) from the physical examination database. These record sets contain data marked as abnormal during past examinations, such as blood sugar levels that have been diagnosed as abnormal in historical physical examinations. This process yields Q sets of abnormality detection record sets.

[0047] Next, supervised training is performed on each of the Q sets of anomaly detection records. This step utilizes supervised learning, which uses labeled data (including known anomalies) to train a model and learn how to identify abnormal patterns in the data. In this step, learners refer to a series of machine learning models (such as support vector machines, decision trees, random forests, and neural networks), each responsible for learning a specific anomaly detection task from the data. The M learners represent multiple models trained for each medical examination category, ensuring improved detection accuracy through the use of different algorithms and feature extraction methods. These learners are trained on the anomaly detection record sets to establish Q medical anomaly detection architectures. Each architecture corresponds to a medical examination category and contains multiple learners.

[0048] Then, based on the anomaly detection loss threshold, we perform an optimization analysis on the Q individual anomaly detection architectures. This step uses a loss function (such as the cross-entropy loss function) to evaluate the accuracy of each of the Q individual anomaly detection architectures. By calculating the loss value, we optimize the model parameters to ensure detection accuracy. Based on the set anomaly detection loss threshold, we filter out poorly trained models. By adjusting hyperparameters and optimizing the model structure, we establish Q optimal anomaly detection architectures.

[0049] Finally, based on the Q anomaly detection optimization architectures, a physical examination anomaly detection space is established. In this step, the optimized Q anomaly detection optimization architectures are combined to construct a multi-dimensional physical examination anomaly detection space. This space integrates detection rules, optimized models, and threshold settings for all physical examination categories, enabling comprehensive anomaly detection and analysis of physical examination data.

[0050] Furthermore, in the method provided in the embodiment of the application, iterative distillation optimization is performed based on the physical examination abnormality detection space to establish Q physical examination abnormality detection channels, and the method further includes:

[0051] Activate a distillation decision factor, wherein the distillation decision factor includes the number of teacher models, student model attributes, and distillation hyperparameters; perform a distillation decision on the qth anomaly detection optimization architecture in the physical examination anomaly detection space based on the distillation decision factor to generate a qth distillation decision space, where q is a positive integer, 1≤q≤Q; perform a distillation loss analysis on the qth anomaly detection optimization architecture according to the qth distillation decision space to establish a qth distillation loss space; perform an iterative distillation loss optimization on the qth distillation decision space according to the qth distillation loss space to determine a qth distillation decision optimization result; and generate a qth physical examination anomaly detection channel according to the qth distillation decision optimization result.

[0052] In an embodiment of the present application, the distillation decision factor is first activated. Among them, the distillation decision factor includes the number of teacher models, student model attributes and distillation hyperparameters. The number of teacher models refers to the number of teacher models used to provide guidance during the distillation process. The teacher model is a complex and efficient model that has been trained for a long time. The student model attributes refer to the structure and complexity of the student model. Usually, the student model is relatively simple and learns by imitating the teacher model. Distillation hyperparameters include parameters such as temperature and learning rate. These hyperparameters control the degree of softening of the teacher model output and the learning pace of the student model. By activating these distillation decision factors, the basic conditions are set for the subsequent distillation decision process.

[0053] Next, a distillation decision is made for the qth anomaly detection optimization architecture within the physical examination anomaly detection space based on the distillation decision factors. This process uses the number of teacher models, student model attributes, and distillation hyperparameters to make a distillation decision for the optimal architecture for physical examination anomaly detection for the qth physical examination category (such as blood glucose testing, blood pressure testing, etc.). This decision space trains and adjusts the student model based on the knowledge of the teacher model, ensuring that the student model effectively learns the anomaly detection patterns of the teacher model. For example, if the qth category is blood glucose, the teacher model provides probability distributions for different blood glucose levels (such as the probability of hyperglycemia and normoglycemia). The student model learns these probability distributions during the distillation process, improving its ability to detect blood glucose anomalies. This distillation decision generates the qth distilled decision space.

[0054] Then, based on the qth distilled decision space, a distillation loss analysis is performed on the qth anomaly detection optimization architecture. In this step, distillation loss analysis is used to assess the gap between the student model and the teacher model. Loss analysis analyzes the difference between the student model output and the teacher model output, quantifying this gap by calculating a loss function (such as cross-entropy loss). This process identifies areas where the student model needs further optimization and which features and parameters need to be adjusted to better mimic the teacher model's anomaly detection capabilities. At this point, loss analysis is used to establish the qth distillation loss space.

[0055] The qth distillation loss space is then used to iteratively optimize the qth distillation decision space. During this phase, multiple rounds of iterative optimization are used to adjust the student model's parameters, gradually reducing the loss value, bringing the student model closer to the teacher model's performance. After each iteration, the current loss value is evaluated and optimized based on the results of the loss analysis. This iterative optimization process continuously improves the performance of the qth individual inspection category model. After multiple iterations of optimization, the optimal anomaly detection model configuration for the qth individual inspection category is obtained, which is the qth distillation decision optimization result.

[0056] Finally, based on the optimization results of the qth distillation decision, the qth physical examination abnormality detection channel is generated. This channel is specifically responsible for processing the data stream under the qth physical examination category (such as blood sugar and blood pressure) and accurately identifying abnormalities. For example, in a blood sugar test, the optimized channel can accurately determine whether the blood sugar value is within the normal range and identify the risk of hyperglycemia or hypoglycemia. Each physical examination category is processed through a corresponding channel to ensure the accuracy and reliability of the test results.

[0057] Step S300: inputting the physical examination correction stream into the Q physical examination abnormality detection channels to obtain Q physical examination abnormality detection results, and performing accompanying interference tracing based on the Q physical examination abnormality detection results to obtain Q accompanying interference feature sequences.

[0058] In this embodiment, the physical examination correction stream is first input into the Q physical examination anomaly detection channel. Data anomaly detection is performed based on each physical examination category (such as blood sugar, blood pressure, cholesterol, etc.), generating Q physical examination anomaly detection results. These test results reflect the status of each physical examination category, such as whether blood sugar is abnormal or blood pressure is above the standard.

[0059] Next, based on the Q individual examination abnormality detection results, accompanying interference tracing is performed. In this process, the abnormal trigger degree of each result is first evaluated based on the Q individual examination abnormality detection results to obtain the Q individual examination abnormality trigger degree. Then, the physical examination accompanying feature factors are activated, including the user accompanying status, physical examination sample status, and physical examination instrument status, to provide additional information that may affect the test results. Based on the Q individual examination abnormality trigger degrees and accompanying feature factors, accompanying features are collected for the data of each physical examination category to generate Q individual examination accompanying feature sequences. Subsequently, accompanying benchmark mining is performed on the Q individual examination categories to obtain Q benchmark accompanying feature sequences. By comparing the benchmark feature sequence with the accompanying feature sequence, interference features are identified, and finally Q accompanying interference feature sequences are generated.

[0060] Furthermore, in the method provided in the embodiment of the application, based on the Q individual physical examination abnormality detection results, accompanying interference tracing is performed to obtain Q accompanying interference feature sequences, which also includes:

[0061] An abnormal trigger degree is evaluated based on the Q individual physical examination abnormality detection results to obtain the Q individual physical examination abnormality trigger degrees; the physical examination accompanying characteristic factors are activated, and the physical examination accompanying characteristic factors include the user accompanying status, the physical examination sample status and the physical examination instrument status; based on the Q individual physical examination abnormality trigger degrees, accompanying features are collected for the Q individual physical examination abnormality detection results according to the physical examination accompanying characteristic factors to generate Q individual physical examination accompanying feature sequences; accompanying benchmark mining is performed on the Q individual physical examination categories according to the physical examination accompanying characteristic factors to obtain Q benchmark accompanying feature sequences; interference features are identified on the Q individual physical examination accompanying feature sequences according to the Q benchmark accompanying feature sequences to generate the Q accompanying interference feature sequences.

[0062] In an embodiment of the present application, the abnormal trigger degree is first evaluated based on the abnormal test results of individual Q physical examinations. This step uses a preset rule table for comparative analysis, and determines the abnormal trigger degree of the physical examination by comparing the difference between the test results and the normal range. According to the normal range of each physical examination category, the degree of deviation between the test result and the normal value is calculated. For example, assuming that the normal range of blood sugar is 3.9 to 6.1mmol / L, if the test result is 8.0mmol / L, it deviates from the normal value by 2.0mmol / L. According to the preset rule table, the trigger degree of the value is determined. Different deviation ranges are set in the rule table to correspond to different trigger degree values. For example, if the deviation is within 1mmol / L, the trigger degree is low, the deviation exceeds 1mmol / L but is within 2mmol / L, the trigger degree is medium, and if it exceeds 2mmol / L, the trigger degree is high. Through these rules, the abnormal degree of each physical examination item is quantified, and finally the abnormal trigger degree of Q physical examination is obtained.

[0063] Next, the physical examination accompanying feature factors are activated. These factors include the user's accompanying status, the physical examination sample status, and the physical examination instrument status. The user's accompanying status refers to factors related to the user's lifestyle habits, such as diet, medication, exercise, staying up late, etc. These factors may directly affect the physical examination results. For example, staying up late can affect blood sugar levels, and medication use may change blood pressure. These accompanying features will be activated and recorded as a basis for subsequent analysis. The physical examination sample status focuses on possible problems in the sample collection and storage process, such as contamination, improper storage, etc., which may also affect the test results. The physical examination instrument status refers to the working status of the detection equipment, such as whether the equipment is operating normally and whether it has been calibrated regularly. These factors directly affect the accuracy of the test.

[0064] Based on the Q individual examination abnormality triggering degrees and the accompanying characteristic factors, accompanying features are collected for each physical examination category. In this step, interference features related to each abnormal detection result are collected based on the triggering degree and related accompanying characteristic factors. For example, if the blood sugar value is abnormal and the user has recently consumed a high-sugar diet or taken medications that affect blood sugar, this information is recorded and used as an accompanying feature to form a blood sugar accompanying feature sequence. Through this process, a Q individual examination accompanying feature sequence is generated.

[0065] Next, we conduct accompanying benchmark mining for Q physical examination categories based on the accompanying characteristic factors of physical examinations. During this process, standardized analysis methods are used to analyze the accompanying characteristic data of each physical examination category in a normal healthy state to generate a benchmark dataset. For example, for blood sugar testing, the user's diet and exercise habits in a healthy state are analyzed to generate a baseline accompanying characteristic sequence, such as the blood sugar value range under a normal diet and regular exercise. Using these baseline characteristics, the actual collected data is compared with the standards under a healthy state to identify possible interference factors. Finally, Q baseline accompanying characteristic sequences are generated, which represent the characteristics of each physical examination category in a normal healthy state.

[0066] Finally, interference features are identified based on the Q baseline accompanying feature sequences and the Q accompanying feature sequences of the physical examination. This step uses a comparative analysis method to compare the differences between the baseline accompanying feature sequences of each physical examination category and the actually collected accompanying feature sequences to identify interference factors that may affect the physical examination results. For example, if the blood sugar test value is abnormal and the user has recently consumed too much sugar or stayed up late for a long time, these changes in lifestyle habits are identified as interference features and recorded as accompanying interference features. Through interference feature identification, Q accompanying interference feature sequences are generated.

[0067] Step S400: performing credible risk detection on the Q individual physical examination abnormality detection results according to the Q accompanying interference feature sequences to obtain Q abnormal credible risk sequences.

[0068] In an embodiment of the present application, when performing credible risk detection on Q individual examination abnormality detection results based on Q accompanying interference feature sequences, the accompanying interference feature sequence for each physical examination category (the qth category) is first extracted based on the Q accompanying interference feature sequences, including user accompanying interference features, sample interference features, and instrument interference features. These features respectively describe the user's living habits, the physical examination sample status, and the equipment operating conditions. Then, based on these features, a risk assessment is performed on the qth individual examination abnormality detection result, obtaining a first coefficient of credible risk for the physical examination abnormality (based on the user's living habits), a second coefficient of credible risk for the physical examination abnormality (based on the sample status), and a third coefficient of credible risk for the physical examination abnormality (based on the instrument status). By combining these three coefficients, the qth abnormal credible risk sequence is constructed. Finally, the aforementioned steps are repeated to generate Q abnormal credible risk sequences.

[0069] Furthermore, in the method provided in the embodiment of the application, performing credible risk detection on the Q individual examination abnormality detection results according to the Q accompanying interference feature sequences to obtain Q abnormal credible risk sequences, further comprising:

[0070] A qth accompanying interference feature sequence is extracted based on the Q accompanying interference feature sequences, where the qth accompanying interference feature sequence includes the qth user accompanying interference feature, the qth sample interference feature, and the qth instrument interference feature; a credible risk assessment is performed on the qth physical examination abnormality detection result based on the qth user accompanying interference feature to obtain a first credible risk coefficient of the physical examination abnormality; a credible risk assessment is performed on the qth physical examination abnormality detection result based on the qth sample interference feature to obtain a second credible risk coefficient of the physical examination abnormality; a credible risk assessment is performed on the qth physical examination abnormality detection result based on the qth instrument interference feature to obtain a third credible risk coefficient of the physical examination abnormality; and a qth abnormality credible risk sequence is constructed based on the first credible risk coefficient of the physical examination abnormality, the second credible risk coefficient of the physical examination abnormality, and the third credible risk coefficient of the physical examination abnormality.

[0071] In an embodiment of the present application, the qth accompanying interference feature sequence is first extracted based on the Q accompanying interference feature sequences. This process extracts interference features related to the physical examination category, including the qth user accompanying interference feature, the qth sample interference feature, and the qth instrument interference feature. The user accompanying interference feature reflects the potential impact of the user's living habits (such as diet, medication, staying up late, etc.) on the physical examination results; the sample interference feature focuses on the problems that may arise during the collection and preservation of the physical examination samples (such as contamination, improper preservation, etc.); the instrument interference feature is related to the status of the physical examination equipment, such as whether the equipment has been calibrated or is working properly. By extracting these interference features, the qth accompanying interference feature sequence is generated for each physical examination category.

[0072] The qth physical examination abnormality detection result is evaluated for credibility based on the qth user's accompanying interference feature. This process first uses the physical examination abnormality induction prediction to analyze the impact of the user's lifestyle habits (such as diet, medication, exercise, and staying up late) on the physical examination abnormality based on the qth user's accompanying interference feature, generating the qth user interference abnormality induction feature. These interference features are then coupled with the physical examination abnormality detection results to generate the qth user interference abnormality coupling feature. Subsequently, based on the historical dataset of user interference abnormality credibility risks, the model is optimized through loss iterative training to generate a user interference abnormality credibility risk detection model. This allows for more accurate identification of the impact of interference factors on physical examination results. Finally, the qth user interference abnormality coupling feature is input into the trained user interference abnormality credibility risk detection model to generate the first coefficient of the physical examination abnormality credibility risk.

[0073] The steps for generating the second coefficient of credible risk for physical examination abnormalities are similar to those for generating the first coefficient. First, an assessment is performed based on the qth sample interference feature to analyze the status of the physical examination sample, such as whether there were any issues during sample collection and storage (such as contamination or improper storage). Then, using loss-based iterative training, the model is trained using sample interference factors from the historical dataset to generate a credible risk detection model for sample interference anomalies. Finally, the qth sample interference feature is input into the model to evaluate its impact on the physical examination abnormality detection results. This generates the second coefficient of credible risk for physical examination abnormalities, which represents the impact of the sample status on the confidence level of the results.

[0074] The steps for generating the third coefficient for the credible risk of a medical examination anomaly are similar to those for the first coefficient. First, an assessment is performed based on the qth instrument interference signature to analyze the status of the medical examination instrument, such as whether it is calibrated and whether there are any faults. Then, through equipment monitoring and fault diagnosis methods, the impact of the instrument status on the test results is evaluated. Similarly, loss-based iterative training is used to generate a credible risk detection model for instrument interference anomalies. The qth instrument interference signature is then input into this model to derive the third coefficient for the credible risk of a medical examination anomaly, representing the impact of the instrument status on the credibility of the test results.

[0075] Finally, the first coefficient of the credible risk of abnormal physical examination, the second coefficient of the credible risk of abnormal physical examination and the third coefficient of the credible risk of abnormal physical examination are integrated to generate the qth abnormal credible risk sequence.

[0076] Furthermore, in the method provided in the embodiment of the application, a credible risk assessment is performed on the qth physical examination abnormality detection result based on the interference characteristics of the qth user to obtain a first credible risk coefficient of the physical examination abnormality, and the method further includes:

[0077] A physical examination abnormality induction prediction is performed based on the qth user's accompanying interference feature to obtain the qth user interference abnormality induction feature; a coupling detection is performed on the qth physical examination abnormality detection result based on the qth user interference abnormality induction feature to obtain the qth user interference abnormality coupling feature; loss iterative training is performed based on the user interference abnormality credible risk history set to generate a user interference abnormality credible risk detection model; the qth user interference abnormality coupling feature is input into the user interference abnormality credible risk detection model to generate the first coefficient of the physical examination abnormality credible risk.

[0078] In this embodiment, a prediction of abnormalities in a physical examination is first performed based on the qth user's accompanying interference feature. Using a rule-based reasoning approach, the user's lifestyle habits, health status, and other factors (such as diet, medication use, exercise, and staying up late) are analyzed to predict the potential for these factors to cause abnormalities in the physical examination. For example, if a user consistently consumes a high-sugar diet, the user is predicted to have abnormal blood sugar levels. This generates the qth user interference anomaly-inducing feature, which represents the potential impact of lifestyle habits on physical examination abnormalities.

[0079] Next, the qth physical examination abnormality detection result is coupled with the qth feature induced by the user interference anomaly. In this step, the coupling analysis method is used to combine and analyze the qth feature induced by the qth user interference anomaly and the qth physical examination abnormality detection result. This method determines whether the interference factor has a significant impact on the test result by identifying the correlation between the user's living habits and the abnormal detection result. For example, if the blood sugar test result is abnormal and the user has a high-sugar diet, the impact of this lifestyle habit on the blood sugar abnormality is identified, and the qth user interference abnormality coupling feature is generated, which reflects the coupling relationship between the lifestyle habit and the abnormal detection result.

[0080] Then, loss iterative training is performed based on the user interference anomaly credible risk history set. In this step, the loss iterative training method is used to train a user interference anomaly credible risk detection model through the physical examination results, user interference features, and physical examination anomaly data that have been marked as credible or uncredible in the user interference anomaly credible risk history set. The user interference anomaly credible risk history set includes the user's living habit data, the corresponding physical examination anomaly detection results, and the credible risk coefficient annotated by technical experts. For example, if a user's blood sugar abnormality is related to a long-term high-sugar diet habit, and the data has been annotated in the history set, the model parameters will be iteratively optimized to improve the model's ability to identify the impact of user interference on the physical examination results. Ultimately, the model can evaluate the impact of the user's interference characteristics on the physical examination results and generate a user interference anomaly credible risk detection model.

[0081] Finally, the qth user interference abnormal coupling feature is input into the user interference abnormal credible risk detection model, and the model is used to analyze the qth user interference abnormal coupling feature and generate the first coefficient of the physical examination abnormal credible risk.

[0082] Step S500: performing closed-loop optimization on the Q physical examination abnormality detection results according to the Q abnormal credible risk sequences, and generating a physical examination analysis report.

[0083] In an embodiment of the present application, when closed-loop optimization is performed on Q abnormal physical examination detection results based on Q abnormal credible risk sequences, the three coefficients in the abnormal credible risk sequence of each physical examination category (the first coefficient of the abnormal credible risk of the physical examination, the second coefficient of the abnormal credible risk of the physical examination, and the third coefficient of the abnormal credible risk of the physical examination) are compared with the preset thresholds. If any coefficient exceeds the preset threshold, it is considered that the corresponding abnormal physical examination detection result may be interfered with or there is a potential problem, and the detection result is then marked. Through this closed-loop optimization, a physical examination analysis report is generated. The report not only contains the detection value of each abnormal physical examination detection result, but also marks the detection items that may be interfered with based on the threshold comparison results.

[0084] Furthermore, the method provided in the application embodiment also includes:

[0085] The physical examination analysis report is interpreted according to the RAG model to obtain a physical examination interpretation report, and the physical examination interpretation report is encrypted and sent to the user.

[0086] In an embodiment of the present application, when interpreting the physical examination analysis report according to the RAG model, the RAG model is first used to interpret the physical examination analysis report in combination with relevant information retrieved from the medical knowledge base. The RAG model automatically generates concise and professional interpretation content by analyzing the abnormal results in the physical examination analysis report. The model will combine the physical examination data and relevant knowledge in the medical literature to provide specific explanations and health recommendations for each abnormal result. For example, if the blood sugar test result is abnormal, the RAG model will search for relevant literature and provide potential causes, health risks and countermeasures for abnormal blood sugar. After this process, a physical examination interpretation report is generated, which contains a detailed explanation of the physical examination data, health recommendations and possible health risk prompts.

[0087] The medical examination interpretation report is then encrypted and sent to the user. To ensure user privacy, the medical examination interpretation report is encrypted using encryption technology and sent to the user via a secure transmission method (such as encrypted email or a dedicated health management application). After receiving the encrypted medical examination interpretation report, the user unlocks the report through authentication (such as a password or biometric recognition), thereby securely accessing the medical examination interpretation content.

[0088] In the embodiments of the present application, in summary, the embodiments of the present application have at least the following technical effects:

[0089] The present application obtains the user's physical examination data stream according to Q physical examination categories, and performs multidimensional correction on the physical examination data stream based on the data layer repair mechanism to obtain a physical examination correction stream, where Q is a positive integer greater than 1; a physical examination anomaly detection space is built according to the Q physical examination categories, and iterative distillation optimization is performed based on the physical examination anomaly detection space to establish Q physical examination anomaly detection channels; the physical examination correction stream is input into the Q physical examination anomaly detection channel to obtain Q physical examination anomaly detection results, and accompanying interference tracing is performed based on the Q physical examination anomaly detection results to obtain Q accompanying interference feature sequences; credible risk detection is performed on the Q physical examination anomaly detection results according to the Q accompanying interference feature sequences to obtain Q abnormal credible risk sequences; closed-loop optimization is performed on the Q physical examination anomaly detection results according to the Q abnormal credible risk sequences to generate a physical examination analysis report. The present invention solves the technical problems of inaccurate physical examination data anomaly detection and insufficient consideration of interference factors in the prior art, and achieves the technical effect of improving the accuracy and credibility of physical examination results through multidimensional correction, abnormality detection channel optimization, accompanying interference tracing and credible risk detection.

[0090] Example 2, based on the same inventive concept as the AI ​​physical examination intelligent main inspection and diagnosis method in the above embodiment, Figure 2 As shown, the present application provides an AI physical examination intelligent main inspection and diagnosis system. The system and method embodiments in the present application are based on the same inventive concept. The system includes:

[0091] A multidimensional correction module 11 is used to obtain the user's physical examination data stream based on Q physical examination categories, and perform multidimensional correction on the physical examination data stream based on the data layer repair mechanism to obtain a physical examination correction stream, where Q is a positive integer greater than 1; a distillation optimization module 12 is used to build a physical examination anomaly detection space based on the Q physical examination categories, and perform iterative distillation optimization based on the physical examination anomaly detection space to establish Q individual physical examination anomaly detection channels; an interference tracing module 13 is used to input the physical examination correction stream into the Q individual physical examination anomaly detection channels to obtain Q individual physical examination anomaly detection results, and perform accompanying interference tracing based on the Q individual physical examination anomaly detection results to obtain Q accompanying interference feature sequences; a trusted risk detection module 14 is used to perform trusted risk detection on the Q individual physical examination anomaly detection results based on the Q accompanying interference feature sequences to obtain Q abnormal trusted risk sequences; a closed-loop optimization module 15 is used to perform closed-loop optimization on the Q individual physical examination anomaly detection results based on the Q abnormal trusted risk sequences to generate a physical examination analysis report.

[0092] Furthermore, the system is also used to implement the following functions:

[0093] Performing a logical contradiction check on the physical examination data stream to obtain a first data repair factor; performing a physiological violation check on the physical examination data stream to obtain a second data repair factor; performing a conflict check on the physical examination data stream to obtain a third data repair factor; fusing the first data repair factor, the second data repair factor, and the third data repair factor to generate a fourth data repair factor; and correcting the physical examination data stream according to the fourth data repair factor to generate the physical examination correction stream.

[0094] Furthermore, the system is also used to implement the following functions:

[0095] Anomaly detection history retrieval is performed according to the Q physical examination categories to obtain Q physical examination anomaly detection record sets; supervised training is performed on M learners according to each physical examination anomaly detection record set in the Q physical examination anomaly detection record sets to establish Q physical examination anomaly detection architectures, each physical examination anomaly detection architecture includes M physical examination anomaly detection models corresponding to each physical examination category, and M is a positive integer greater than 1; optimization analysis is performed on the Q physical examination anomaly detection architectures according to the anomaly detection loss threshold to establish Q anomaly detection optimization architectures; based on the Q anomaly detection optimization architectures, the physical examination anomaly detection space is established.

[0096] Furthermore, the system is also used to implement the following functions:

[0097] Activate a distillation decision factor, wherein the distillation decision factor includes the number of teacher models, student model attributes, and distillation hyperparameters; perform a distillation decision on the qth anomaly detection optimization architecture in the physical examination anomaly detection space based on the distillation decision factor to generate a qth distillation decision space, where q is a positive integer, 1≤q≤Q; perform a distillation loss analysis on the qth anomaly detection optimization architecture according to the qth distillation decision space to establish a qth distillation loss space; perform an iterative distillation loss optimization on the qth distillation decision space according to the qth distillation loss space to determine a qth distillation decision optimization result; and generate a qth physical examination anomaly detection channel according to the qth distillation decision optimization result.

[0098] Furthermore, the system is also used to implement the following functions:

[0099] An abnormal trigger degree is evaluated based on the Q individual physical examination abnormality detection results to obtain the Q individual physical examination abnormality trigger degrees; the physical examination accompanying characteristic factors are activated, and the physical examination accompanying characteristic factors include the user accompanying status, the physical examination sample status and the physical examination instrument status; based on the Q individual physical examination abnormality trigger degrees, accompanying features are collected for the Q individual physical examination abnormality detection results according to the physical examination accompanying characteristic factors to generate Q individual physical examination accompanying feature sequences; accompanying benchmark mining is performed on the Q individual physical examination categories according to the physical examination accompanying characteristic factors to obtain Q benchmark accompanying feature sequences; interference features are identified on the Q individual physical examination accompanying feature sequences according to the Q benchmark accompanying feature sequences to generate the Q accompanying interference feature sequences.

[0100] Furthermore, the system is also used to implement the following functions:

[0101] A qth accompanying interference feature sequence is extracted based on the Q accompanying interference feature sequences, where the qth accompanying interference feature sequence includes the qth user accompanying interference feature, the qth sample interference feature, and the qth instrument interference feature; a credible risk assessment is performed on the qth physical examination abnormality detection result based on the qth user accompanying interference feature to obtain a first credible risk coefficient of the physical examination abnormality; a credible risk assessment is performed on the qth physical examination abnormality detection result based on the qth sample interference feature to obtain a second credible risk coefficient of the physical examination abnormality; a credible risk assessment is performed on the qth physical examination abnormality detection result based on the qth instrument interference feature to obtain a third credible risk coefficient of the physical examination abnormality; and a qth abnormality credible risk sequence is constructed based on the first credible risk coefficient of the physical examination abnormality, the second credible risk coefficient of the physical examination abnormality, and the third credible risk coefficient of the physical examination abnormality.

[0102] Furthermore, the system is also used to implement the following functions:

[0103] A physical examination abnormality induction prediction is performed based on the qth user's accompanying interference feature to obtain the qth user interference abnormality induction feature; a coupling detection is performed on the qth physical examination abnormality detection result based on the qth user interference abnormality induction feature to obtain the qth user interference abnormality coupling feature; loss iterative training is performed based on the user interference abnormality credible risk history set to generate a user interference abnormality credible risk detection model; the qth user interference abnormality coupling feature is input into the user interference abnormality credible risk detection model to generate the first coefficient of the physical examination abnormality credible risk.

[0104] Furthermore, the system is also used to implement the following functions:

[0105] According to the AI ​​physical examination system, a physical examination data set of the user is obtained; the physical examination data set is cleaned and sorted according to the Q physical examination categories to generate the physical examination data stream.

[0106] Furthermore, the system is also used to implement the following functions:

[0107] The physical examination analysis report is interpreted according to the RAG model to obtain a physical examination interpretation report, and the physical examination interpretation report is encrypted and sent to the user.

[0108] It should be noted that the order in which the embodiments of the present application are presented is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. Furthermore, the foregoing descriptions of specific embodiments of this specification are provided. The processes depicted in the accompanying drawings do not necessarily require the specific order or sequential sequence shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0109] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

[0110] This specification and drawings are merely illustrative of the present application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Obviously, those skilled in the art may make various modifications and variations to this application without departing from the scope of this application. Thus, this application is intended to include such modifications and variations as fall within the scope of this application and its equivalents.

Claims

1. An AI physical examination intelligent diagnosis method, characterized in that: The method comprises: Obtaining a user's physical examination data stream according to Q physical examination categories, and performing multi-dimensional correction on the physical examination data stream based on a data layer repair mechanism to obtain a physical examination correction stream, where Q is a positive integer greater than 1; Build a physical examination anomaly detection space based on the Q physical examination categories, and perform iterative distillation optimization based on the physical examination anomaly detection space to establish Q physical examination anomaly detection channels; Inputting the physical examination correction stream into the Q physical examination abnormality detection channels to obtain Q physical examination abnormality detection results, and performing accompanying interference tracing based on the Q physical examination abnormality detection results to obtain Q accompanying interference feature sequences; Performing credible risk detection on the Q individual physical examination abnormality detection results according to the Q accompanying interference feature sequences to obtain Q abnormal credible risk sequences; Performing closed-loop optimization on the Q physical examination abnormality detection results according to the Q abnormal credible risk sequences to generate a physical examination analysis report; The accompanying interference tracing is performed based on the Q individual physical examination abnormality detection results to obtain Q accompanying interference feature sequences, including: Perform abnormal triggering degree evaluation based on the abnormal detection results of the Q individual physical examinations to obtain the abnormal triggering degree of the Q individual physical examinations; Activate physical examination accompanying characteristic factors, wherein the physical examination accompanying characteristic factors include user accompanying status, physical examination sample status and physical examination instrument status; Based on the abnormal trigger degree of the Q individual physical examinations, accompanying feature collection is performed on the Q individual physical examination abnormality detection results according to the physical examination accompanying feature factor to generate a Q individual physical examination accompanying feature sequence; Perform accompanying benchmark mining on the Q physical examination categories according to the physical examination accompanying feature factors to obtain Q benchmark accompanying feature sequences; Interference feature identification is performed on the Q individual examination accompanying feature sequences according to the Q reference accompanying feature sequences to generate the Q accompanying interference feature sequences.

2. The AI ​​physical examination intelligent diagnosis method according to claim 1, characterized in that: The data layer repair mechanism includes: Performing a logical contradiction detection on the physical examination data stream to obtain a first data repair factor; Performing physiological violation detection on the physical examination data stream to obtain a second data repair factor; Performing conflict detection on the physical examination data stream to obtain a third data repair factor; Fusion of the first data repair factor, the second data repair factor, and the third data repair factor to generate a fourth data repair factor; The physical examination data stream is corrected according to the fourth repair factor of the data to generate the physical examination corrected stream.

3. The AI ​​physical examination intelligent diagnosis method according to claim 1, characterized in that: According to the Q physical examination categories, a physical examination abnormality detection space is built, including: Perform anomaly detection history retrieval based on the Q individual physical examination categories to obtain a Q individual physical examination anomaly detection record set; Perform supervised training on M learners according to each of the Q physical examination anomaly detection record sets, and establish Q physical examination anomaly detection architectures, each of which includes M physical examination anomaly detection models corresponding to each physical examination category, where M is a positive integer greater than 1; Performing optimization analysis on the Q individual anomaly detection architectures according to the anomaly detection loss threshold, and establishing Q anomaly detection optimization architectures; The physical examination anomaly detection space is established according to the Q anomaly detection optimization architectures.

4. The AI ​​physical examination intelligent diagnosis method according to claim 1, characterized in that: Based on the physical examination anomaly detection space, iterative distillation optimization is performed to establish Q physical examination anomaly detection channels, including: Activate distillation decision factors, which include the number of teacher models, student model attributes, and distillation hyperparameters; Performing a distillation decision on the qth abnormality detection optimization architecture in the physical examination abnormality detection space based on the distillation decision factor to generate a qth distillation decision space, where q is a positive integer, 1≤q≤Q; Performing distillation loss analysis on the qth anomaly detection optimization architecture according to the qth distillation decision space to establish a qth distillation loss space; Iteratively optimize the qth distillation loss on the qth distillation decision space according to the qth distillation loss space, determine the qth distillation decision optimization result, and generate the qth physical examination abnormality detection channel according to the qth distillation decision optimization result.

5. The AI ​​physical examination intelligent diagnosis method according to claim 1, characterized in that: Performing credible risk detection on the Q individual physical examination abnormality detection results according to the Q accompanying interference feature sequences to obtain Q abnormal credible risk sequences, including: Extracting a qth accompanying interference feature sequence according to the Q accompanying interference feature sequences, wherein the qth accompanying interference feature sequence includes a qth user accompanying interference feature, a qth sample interference feature, and a qth instrument interference feature; Performing a credibility risk assessment on the qth physical examination abnormality detection result according to the qth user's accompanying interference feature to obtain a first credibility risk coefficient of the physical examination abnormality; Performing a credibility risk assessment on the qth abnormal physical examination detection result according to the qth sample interference feature to obtain a second credibility risk coefficient of the abnormal physical examination; Performing a credibility risk assessment on the qth abnormal physical examination detection result according to the qth instrument interference feature to obtain a third credibility risk coefficient of the physical examination abnormality; A qth abnormal credible risk sequence is constructed according to the first credible risk coefficient of the physical examination abnormality, the second credible risk coefficient of the physical examination abnormality, and the third credible risk coefficient of the physical examination abnormality.

6. The AI ​​physical examination intelligent diagnosis method according to claim 5, characterized in that: Performing a credibility risk assessment on the qth physical examination abnormality detection result according to the qth user's accompanying interference feature to obtain a first credibility risk coefficient of the physical examination abnormality includes: Performing a physical examination abnormality induction prediction based on the qth user's accompanying interference feature to obtain the qth user interference abnormality induction feature; Perform coupling detection on the qth physical examination abnormality detection result according to the qth user interference abnormality-induced feature to obtain the qth user interference abnormality coupling feature; Perform loss iterative training based on the user interference anomaly credible risk history set to generate a user interference anomaly credible risk detection model; The qth user interference abnormal coupling feature is input into the user interference abnormal credible risk detection model to generate the first coefficient of the physical examination abnormal credible risk.

7. The AI ​​physical examination intelligent diagnosis method according to claim 1, characterized in that: Get the user's physical examination data stream based on the physical examination category of Q, including: Obtaining a physical examination data set of the user according to the AI ​​physical examination system; The physical examination data set is cleaned and sorted according to the Q physical examination categories to generate the physical examination data stream.

8. The AI ​​physical examination intelligent diagnosis method according to claim 1, characterized in that: The physical examination analysis report is interpreted according to the RAG model to obtain a physical examination interpretation report, and the physical examination interpretation report is encrypted and sent to the user.

9. An AI physical examination intelligent main inspection and diagnosis system, characterized by: The system is used to execute the AI ​​physical examination intelligent main inspection and diagnosis method according to any one of claims 1 to 8, and the system includes: A multidimensional correction module is used to obtain a user's physical examination data stream according to Q physical examination categories, and perform multidimensional correction on the physical examination data stream based on a data layer repair mechanism to obtain a physical examination correction stream, where Q is a positive integer greater than 1; A distillation optimization module is used to build a physical examination anomaly detection space based on the Q physical examination categories, and perform iterative distillation optimization based on the physical examination anomaly detection space to establish Q physical examination anomaly detection channels; an interference tracing module, configured to input the physical examination correction stream into the Q physical examination anomaly detection channels, obtain Q physical examination anomaly detection results, and perform accompanying interference tracing based on the Q physical examination anomaly detection results to obtain Q accompanying interference feature sequences; A credible risk detection module is configured to perform credible risk detection on the Q individual physical examination abnormality detection results according to the Q accompanying interference feature sequences to obtain Q abnormal credible risk sequences; A closed-loop optimization module is used to perform closed-loop optimization on the Q physical examination abnormality detection results according to the Q abnormal credible risk sequences to generate a physical examination analysis report.

Images