Clinical test CRC management method and system based on artificial intelligence

Abstract

The invention discloses a clinical test CRC management method and system based on artificial intelligence, and belongs to the technical field of clinical test management, and the method specifically comprises the steps: obtaining a structured parameter and an unstructured feedback, carrying out the time alignment matching, and calculating a time deviation value between the structured parameter and the unstructured feedback; extracting description vocabularies from the unstructured feedback, performing matching detection on the description vocabularies and structured parameters in corresponding time windows, and identifying logic consistency; when it is detected that the structural parameter missing rate exceeds a preset threshold value, a supplementary question request is sent to the subject terminal; when detecting that an event related to any structured parameter is missed in the unstructured feedback, adjusting a sampling frequency and detection item combination of the detection equipment; according to the time deviation value and the logic consistency analysis result, the structured parameters and the unstructured feedback are corrected, and a final clinical test report containing a correction mark is generated; the accuracy of clinical test result judgment is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of clinical trial management, and in particular to an artificial intelligence-based clinical trial CRC management method and system. Background Art

[0002] In the biomedical field, clinical trials are a critical step in determining the safety and effectiveness of drugs, medical devices, and treatments. With the rapid development of science and technology, the scale and complexity of clinical trials are constantly increasing, and the amount of data involved is also exploding. This data is not only the basis for evaluating trial results, but also an important cornerstone for promoting medical progress and protecting public health. Among them, structured data (such as objective parameters such as physiological indicators and test values) can provide precise quantitative information, while unstructured data (such as subjective feelings such as text descriptions and voice feedback from subjects) can reflect more details and context. Both play an indispensable role in fully and accurately understanding the effects of trials.

[0003] Currently, structured and unstructured data are typically handled differently in clinical trial data management. Structured data is primarily collected using advanced medical equipment and stored and analyzed according to unified formats and standards. This ensures high accuracy and standardization, but has limitations in reflecting subjects' subjective experiences and complex scenarios. Unstructured data, on the other hand, is primarily collected through manual recording and interviews and contains a wealth of subjective information.

[0004] Because structured and unstructured data are processed relatively independently and lack effective mutual verification mechanisms, insufficient objective parameter collection or unclear subjective feedback can easily occur. When objective parameter collection is incomplete, it may not accurately reflect the subject's true condition; conversely, when subjective feedback is unclear, it is difficult to extract valuable information. This can lead to bias in the final clinical trial report, affecting the accurate interpretation of trial results and hindering the progress of biomedical research. Summary of the Invention

[0005] The purpose of the present invention is to provide an artificial intelligence-based clinical trial CRC management method and system to solve the following technical problems:

[0006] Structured data and unstructured data are relatively independent and lack an effective mutual verification mechanism, which makes it easy for objective parameters to be insufficiently collected or subjective feedback to be unclear.

[0007] The purpose of the present invention can be achieved through the following technical solutions:

[0008] An artificial intelligence-based clinical trial CRC management method includes the following steps:

[0009] Obtain structured parameters and unstructured feedback from clinical trials, align the event timestamps of the structured parameters with the semantic time stamps in the unstructured feedback, and calculate the time offset between the two;

[0010] Based on natural language processing technology, emotional tendency values ​​and symptom description words are extracted from unstructured feedback, and matched with structured parameters within the corresponding time window to identify the logical consistency between the description content and the parameter values;

[0011] When it is detected that the structured parameter missing rate exceeds a preset threshold, a supplementary question request is sent to the subject terminal, wherein the question request includes symptom inquiry content associated with the missing parameter;

[0012] When it is detected that an event related to any structured parameter is omitted in the unstructured feedback, the sampling frequency of the detection equipment and the combination of detection items are adjusted to increase the collection density of the structured parameters;

[0013] According to the time deviation value and logical consistency analysis results, the structured parameters and unstructured feedback are revised and the final clinical trial report containing the revised marks is generated.

[0014] As a further solution of the present invention: the structured parameters include physiological indicator time series data, drug metabolism kinetic parameters and equipment detection values, and the unstructured feedback includes researcher voice records, subject electronic diary text and image evaluation annotations.

[0015] As a further solution of the present invention: the alignment and matching process is:

[0016] Identify all sentences containing time descriptions from the text content of subjective feedback, extract the time expressions involved and convert them into standardized time intervals; arrange the precise timestamps recorded by medical equipment in the order of acquisition to generate a structured parameter timeline; establish a mapping relationship between each standardized time interval and the nearest neighbor timestamp on the structured parameter timeline, and calculate the overlapping time length between the two; when the overlapping time length is less than the preset ratio, insert a virtual time node on the structured parameter timeline, and recalculate the average value of each parameter in the time interval as supplementary data; record all events in which the time deviation value exceeds the allowable range, and generate a time calibration log file containing the original timestamp, the corrected timestamp and the deviation value.

[0017] As a further solution of the present invention: the logical consistency identification process includes:

[0018] A medical vocabulary database containing symptom names, physical signs, and drug reactions was constructed, with each word associated with a corresponding physiological indicator type and expected numerical range. The subjective feedback text was segmented, and keywords existing in the medical vocabulary database were screened out and their frequency of occurrence was recorded. The numerical records of the physiological indicators corresponding to the keywords within the same time interval were searched in the structured parameter database, and a time series change curve of the indicator was drawn. When the value of the physiological indicator corresponding to the keyword did not fall within the expected range or the change trend did not match the description, the time interval was marked as a contradictory interval to be verified. A list of contradictory intervals was generated, and each entry in the list contained the contradiction type, the associated parameter name, the original subjective feedback fragment, and the suggested verification method.

[0019] As a further solution of the present invention: the generation process of the supplementary question is:

[0020] According to the type of missing structured parameters, related symptom descriptions and common question templates are retrieved from the preset medical knowledge base; question templates are sorted according to the probability of symptom occurrence, and high-priority questions that are strongly associated with missing parameters are given priority; professional medical terms are converted into everyday expressions that are easy for subjects to understand, and examples are added to improve question comprehension; the operation records and usage habits of the subject's terminal are analyzed, and its active time period is selected as the question sending window; the response status of the subject's terminal is continuously monitored after the question is sent, and a reminder notification is automatically triggered when no reply is received after the timeout.

[0021] As a further solution of the present invention, adjusting the sampling frequency of the test equipment specifically includes:

[0022] Identify the types of structured parameters that need to be additionally collected and their corresponding medical device models; query the current working status, remaining storage capacity, and battery life of the target device; analyze the historical usage data of the device to predict the subject's degree of cooperation and activity patterns in the future time period; when the subject is in a stationary state and the device is idle, gradually increase the sampling frequency until the maximum carrying capacity of the device is reached; align the timestamps of the additionally collected data with the original data, and update them to the structured parameter database.

[0023] As a further solution of the present invention, the modification of the structured parameters and the unstructured feedback specifically includes:

[0024] Create a temporary data copy for events where the time deviation value exceeds the threshold, retain the original record and generate a revised version; convert the descriptive content in the subjective feedback into a numerical score, and the scoring criteria include the degree of descriptive detail, word accuracy and logical coherence; establish a linear regression model between the structured parameter values ​​and the subjective scores, and calculate the correlation coefficient between the two; when the correlation coefficient is lower than the preset standard, use a sliding window algorithm to smooth the parameter values, and the adjustment range is proportional to the subjective score; highlight all revised data points in the final report, and display a comparison chart of the values ​​before and after the adjustment.

[0025] As a further solution of the present invention: the generation of the final clinical trial report includes:

[0026] Construct a hierarchical display structure, with the first layer showing the original collected uncorrected data and the second layer overlaying the corrected data points; add a detailed information panel next to each corrected data point; display the original value, corrected value, adjustment basis, and associated subjective feedback fragments of the data point in the detailed information panel; generate a complete verification process record file, including all time calibration logs, conflict interval lists, and correction operation records; set a unique version identifier for each report generation, recording the generation time, operator, and number of corrections.

[0027] As a further solution of the present invention: create an independent event record for each data correction operation, including the triggering conditions, operation time and impact scope; establish a data modification association network, and mark other related parameter nodes affected by the corrected data point; when a correction error is found, trace back to the original data acquisition node along the association network and re-trigger the verification process; generate a correction audit report, recording in detail the error discovery time, impact assessment results and the correct value after re-verification; and store the audit report as an independent attachment bound to the final test report.

[0028] The present invention also includes an artificial intelligence-based clinical trial CRC management system for implementing the above-mentioned artificial intelligence-based clinical trial CRC management method, comprising:

[0029] The time matching module is used to obtain structured parameters and unstructured feedback in clinical trials, align the event timestamps of structured parameters with the semantic time stamps in unstructured feedback, and calculate the time deviation between the two;

[0030] The symptom verification module is used to extract sentiment values ​​and symptom description words from unstructured feedback based on natural language processing technology, and match them with structured parameters within the corresponding time window to identify the logical consistency between the description content and the parameter values;

[0031] The test adjustment module is configured to, when it is detected that the missing rate of structured parameters exceeds a preset threshold, send a supplementary question request to the subject terminal, wherein the question request includes symptom inquiry content associated with the missing parameters; and when it is detected that an event related to any structured parameter is omitted in the unstructured feedback, adjust the sampling frequency of the detection device and the combination of detection items to increase the collection density of the structured parameters;

[0032] The report correction module is used to correct the structured parameters and unstructured feedback according to the time deviation value and the logical consistency analysis results, and generate a final clinical trial report containing correction marks.

[0033] Beneficial effects of the present invention:

[0034] The present invention aligns and matches the event timestamps of structured parameters with the semantic time tags in unstructured feedback, calculates the time deviation value, and extracts the sentiment tendency value and symptom description vocabulary in the unstructured feedback based on natural language processing technology, and performs logical consistency matching detection with the structured parameters, effectively solving the problem of independent processing of the two types of data and lack of verification mechanism. When it is detected that the missing rate of structured parameters exceeds the threshold, a supplementary question request for related symptoms is automatically sent to the subject terminal; for related events omitted in unstructured feedback, the sampling frequency and project combination of the detection equipment are dynamically adjusted to ensure that the objective parameters are collected completely and the subjective feedback is clear. On this basis, the data is corrected according to the time deviation value and the logical consistency analysis results, and a final clinical trial report with correction marks is generated. In the process, data traceability is ensured by creating a time calibration log, a list of conflicting intervals, etc., which significantly reduces reporting deviations, improves the accuracy of test result judgments, and effectively promotes the progress of biomedical research. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The present invention will be further described below with reference to the accompanying drawings.

[0036] Figure 1 This is a flow chart of an artificial intelligence-based clinical trial CRC management method of the present invention;

[0037] Figure 2 This is a module diagram of an artificial intelligence-based clinical trial CRC management system of the present invention. DETAILED DESCRIPTION

[0038] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0039] See also Figure 1 As shown, the present invention is a clinical trial CRC management method based on artificial intelligence, comprising the following steps:

[0040] The first step is data acquisition and time alignment. Through professional medical monitoring equipment and recording tools, structured parameters and unstructured feedback in clinical trials are obtained. Structured parameters include physiological indicators, drug metabolism parameters, and equipment detection values ​​recorded in chronological order; unstructured feedback includes researchers' voice records, subjects' electronic diary texts, and imaging assessment annotations. The system will align and match the event timestamps of structured parameters with the semantic time stamps in unstructured feedback. For example, it will identify "felt unwell yesterday afternoon" from the unstructured feedback text, convert "yesterday afternoon" into a standardized time interval, and then map it with the timestamp on the structured parameter timeline to calculate the time deviation between the two.

[0041] The second step is logical consistency matching. Natural language processing techniques are used to extract sentiment values ​​and symptom descriptions from unstructured feedback. For example, the sentiment of "anxiety" and symptom descriptions such as "cough" and "fever" can be identified from electronic diaries. These contents are then matched with structured parameters within the corresponding time window to check whether the body temperature parameter increased during the time period when the "fever" symptom was recorded. This helps determine the logical consistency between the description and the parameter value.

[0042] Then comes data supplementation and adjustment. When the system detects that the missing rate of structured parameters exceeds the preset threshold, it will automatically send a supplementary question request to the subject's terminal. This request will include symptom inquiries related to the missing parameters. For example, if blood pressure parameters are not collected enough, the subject will be asked whether he or she has symptoms such as dizziness and palpitations. If an event related to any structured parameter is missed in the unstructured feedback, the system will adjust the sampling frequency of the detection equipment and the combination of detection items to increase the collection density of the structured parameters.

[0043] Finally, data correction and report generation are performed. Structured parameters and unstructured feedback are corrected based on time deviation values ​​and logical consistency analysis results. The original records are retained during the correction process, while a revised version is generated. Finally, a final clinical trial report with correction markers is generated, clearly demonstrating the data corrections and providing a reliable basis for clinical trial evaluation.

[0044] In a preferred embodiment of the present invention, structured parameters are data that can be stored and analyzed in a standardized format, including physiological indicator time series data, pharmacokinetic parameters, and device detection values. Physiological indicator time series data include heart rate, blood pressure, body temperature, and other data recorded in chronological order, which can reflect the subject's basic physical condition at different time points. Pharmacokinetic parameters include values ​​related to drug absorption, distribution, metabolism, and excretion in the body, which are crucial for evaluating drug efficacy and safety. Device detection values ​​are precise data collected by various medical devices such as blood glucose meters and electrocardiographs. Unstructured feedback, on the other hand, is data that cannot be expressed in a unified structure and includes researcher voice recordings, subject electronic diary text, and imaging assessment annotations. Researcher voice recordings may include analysis and judgment of the subject's condition; subject electronic diary text contains information such as symptoms and emotions recorded based on the subject's own experience; and imaging assessment annotations describe and interpret the results of imaging examinations such as X-rays and CT scans.

[0045] In another preferred embodiment of the present invention, the alignment and matching process is:

[0046] All sentences containing time descriptions are identified from the textual content of subjective feedback, such as "I felt unwell yesterday morning." Time expressions such as "yesterday morning" are extracted and converted into standardized time intervals, such as converting "yesterday morning" into a specific date and morning time period. Simultaneously, the precise timestamps recorded by the medical device are arranged in the order of acquisition to generate a structured parameter timeline. Next, a mapping relationship is established between each standardized time interval and its nearest neighbor timestamp on the structured parameter timeline, and the length of overlap between the two is calculated. For example, if a standardized time interval is "April 28, 2025, 9:00-10:00," and the nearest neighbor timestamp on the structured parameter timeline is 9:15, the system calculates the overlap between the two. If the overlap is less than a preset percentage (e.g., 30%), a virtual time node is inserted into the structured parameter timeline, and the average values ​​of each parameter within that time interval are recalculated as supplementary data. For example, if heart rate data is missing within that time interval, the heart rate values ​​for the preceding and following times are calculated to supplement the data for that interval. In addition, all events where the time deviation exceeds the allowable range (e.g., the deviation exceeds 30 minutes) are recorded, and a time calibration log file containing the original timestamp, corrected timestamp, and deviation value is generated for easy subsequent review and analysis.

[0047] In another preferred embodiment of the present invention, the logical consistency identification process includes:

[0048] First, the present invention constructs a comprehensive and professional medical vocabulary library, which contains a wealth of information such as symptom names, physical signs, and drug reactions. For each word in the vocabulary library, a corresponding physiological indicator type and an expected numerical range are associated. Taking the symptom of "palpitations" as an example, the physiological indicator type associated with it is heart rate, and the expected numerical range may be set to be higher than the normal heart rate range (such as the normal heart rate of an adult at rest is 60-100 beats / minute. If palpitations occur, the heart rate may be higher than 100 beats / minute). At the same time, the physical sign of "rash" may be associated with examination indicators related to skin appearance, and the expected numerical range will be defined differently according to the different types and severity of the rash. For drug reactions, such as "drug allergy", the physiological indicators that may be associated include the eosinophil ratio in blood routine, the level of immunoglobulin E, etc., and there will be corresponding normal reference ranges and abnormal fluctuation intervals corresponding to them.

[0049] After completing the construction of the medical vocabulary database, the system will perform detailed word segmentation on the subjective feedback text. This process is similar to splitting an article into meaningful words. Through a specific word segmentation algorithm, the subjective feedback text is decomposed into independent vocabulary units. Then, the system will filter out the keywords that exist in the medical vocabulary database from these vocabulary units. For example, if the subject records "I feel dizzy and a little nauseous today" in the electronic diary text, the system will filter out the two keywords "dizzy" and "nausea" and record their frequency of occurrence. Recording the frequency of keyword occurrence helps analyze the severity and duration of symptoms. If a keyword appears frequently, it may mean that the symptom is more serious or lasts for a long time.

[0050] Next, the system will conduct a precise search in the structured parameter database to find the numerical records of the physiological indicators corresponding to the filtered keywords in the same time period. Assuming that the keyword is "chest pain", the system will search for the specific values ​​of relevant physiological indicators such as electrocardiogram and myocardial enzymes during the time period when the "chest pain" symptoms are recorded. After finding these numerical records, the system will draw a time series change curve of the indicator. Taking the ST segment changes in the electrocardiogram as an example, by drawing a curve of the ST segment changes over a period of time, its changing trend can be intuitively observed.

[0051] After obtaining the temporal change curve of the physiological indicators, the system will compare and analyze the physiological indicator values ​​corresponding to the keywords with the expected value range associated with the medical vocabulary. When it is found that the physiological indicator values ​​corresponding to the keywords do not fall within the expected range, or their changing trends do not match the description, the system will mark the time interval as a contradictory interval to be verified. For example, the subjective feedback text describes "shortness of breath", but in the corresponding time period, the collected respiratory rate values ​​are within the normal range and there is no obvious upward trend. This situation will be marked as a contradictory interval.

[0052] Finally, the present invention generates a list of conflicting intervals, and each entry in the list contains detailed information. The conflict type will clearly indicate whether the symptoms do not match the indicator values, or whether the indicator change trend is contrary to the description; the associated parameter names will accurately list the physiological indicators involved, such as blood pressure, blood sugar, etc.; the original text fragment of the subjective feedback will fully present the original description content of the subject or researcher, which is convenient for subsequent further verification; the recommended verification method will provide specific operational suggestions based on different conflicting situations, such as asking the subject again about his or her specific feelings at the time, re-checking the accuracy of the detection equipment, and conducting additional special inspections. By generating such a detailed list of conflicting intervals, it is possible to provide clear guidance for subsequent data verification and correction work, ensuring the logical consistency and reliability of clinical trial data.

[0053] In another preferred embodiment of the present invention, the process of generating the supplementary question is:

[0054] First, the present invention will search from a preset medical knowledge base based on the type of missing structured parameters. This medical knowledge base integrates a large amount of medical expertise, covering various disease symptoms, examination indicators and corresponding relationships. For example, if the missing parameters are related to blood sugar, the present invention will retrieve symptom descriptions related to abnormal blood sugar, such as "thirst, polyuria, fatigue", etc., as well as common question templates, such as "Have you always felt thirsty recently? Do you drink more water than usual?"

[0055] Next, the present invention sorts the retrieved question templates by symptom probability. Question templates corresponding to symptoms that are closely related to the missing parameter and have a high probability of occurring are given a high priority. For example, in the case of dysglycemia, the symptom "thirst" has a high probability of occurring, and related questions will be selected first. This ensures that key questions that are strongly related to the missing parameter are asked first.

[0056] To help participants better understand the questions, the present invention converts professional medical terms into easily understandable everyday expressions and adds examples. For example, "large fluctuations in blood sugar" can be expressed as "Have you ever felt suddenly anxious or sweaty, but then recovered after a while?" and the example "It's like that uncomfortable feeling you get after being hungry for a long time."

[0057] The present invention will also analyze the operation records and usage habits of the subject's terminal to determine the window for sending questions. Through analysis, it is found that the subject usually uses the terminal device more frequently between 8 and 10 pm, so this time period will be selected to send questions. After the question is sent, the present invention will continue to monitor the response status of the subject's terminal. If no reply is received within a preset time (such as 24 hours), the present invention will automatically trigger a reminder notification, such as sending a text message or push message to remind the subject to reply in time.

[0058] In another preferred embodiment of the present invention, adjusting the sampling frequency of the test equipment specifically includes:

[0059] First, the present invention will identify the type of structured parameter that needs to be supplemented and the corresponding medical device model. For example, if blood pressure parameters need to be supplemented, the present invention will determine which brand and model of blood pressure monitor is used.

[0060] Then, the present invention will query the current working status, remaining storage capacity and battery life of the target device. Understand whether the device is in normal working condition, whether the storage capacity is sufficient to store the new data, and whether the battery power can support the work after increasing the sampling frequency. At the same time, the present invention will analyze the historical usage data of the device to predict the subject's cooperation and activity pattern in the future time period. If the historical data shows that the subject is relatively quiet and has a high degree of cooperation from 3 to 5 pm every day, then the present invention will select this time period to adjust the sampling frequency

[0061] While the subject is at rest and the device is idle, the present invention gradually increases the sampling frequency. The magnitude of each increase is determined by the device's performance and stability, until the device's maximum capacity is reached. For example, if a blood pressure monitor originally samples once an hour, the present invention will first adjust the sampling rate to once every half hour, observe the device's performance, and if it is normal, continue to adjust the sampling rate to once every 15 minutes until the device reaches the maximum sampling frequency it can withstand.

[0062] Finally, the present invention will perform time stamp alignment on the additionally collected data and the original data to ensure that the newly added data accurately corresponds to the original data, and then update the structured parameter database to ensure the integrity and accuracy of the data in the database.

[0063] In another preferred embodiment of the present invention, the modifying of structured parameters and unstructured feedback specifically includes:

[0064] When an event with a time deviation exceeding a threshold is detected, the present invention creates a temporary data copy. This preserves the original record while also generating a revised version. For example, if a physiological metric recording has a time deviation exceeding 30 minutes, the present invention will copy the record and perform the correction on the copy.

[0065] This invention converts the descriptions in subjective feedback into numerical scores. Scoring criteria include descriptive detail, word accuracy, and logical coherence. For example, a subjective feedback with detailed descriptions, accurate wording, and clear logic may receive a higher score; conversely, a vague description with inaccurate wording may receive a lower score.

[0066] Next, the present invention establishes a linear regression model between the structured parameter values ​​and the subjective ratings and calculates the correlation coefficient between the two. This coefficient can be used to determine the degree of association between subjective feedback and the structured parameter. When the correlation coefficient is lower than the preset standard, it indicates that the association between the two is weak and the parameter value needs to be adjusted. In this case, the present invention uses a sliding window algorithm to smooth the parameter values. The adjustment range is proportional to the subjective rating, that is, the higher the subjective rating, the larger the adjustment range.

[0067] In the final report, the present invention will highlight all corrected data points and display a comparison chart of the values ​​before and after the adjustment. This allows users to clearly see the data correction status and facilitates data review and analysis.

[0068] In another preferred embodiment of the present invention, the generation of the final clinical trial report includes:

[0069] The present invention constructs a hierarchical display structure. The first layer displays the original, uncorrected data. This data is directly obtained from clinical trials and remains unmodified, accurately reflecting the initial state at the time of data collection. The second layer overlays the corrected data points. This hierarchical structure allows users to clearly compare the original and corrected data, intuitively understanding data changes.

[0070] In order to further provide detailed information, the present invention will add a detailed information panel next to each corrected data point. When the user needs to understand the specific situation of a corrected data point, he only needs to check the detailed information panel. The panel will display the original value, corrected value, adjustment basis and associated subjective feedback segment of the data point. For example, if the original value of a physiological indicator is 50 and it is 55 after correction, the adjustment basis may be a correction based on the logical relationship between the symptoms described in the subjective feedback and the indicator. The associated subjective feedback segment will display a specific description of the relevant symptoms, such as "The patient feels physically weak and it is speculated that the indicator may be low, so it is corrected."

[0071] The present invention will also generate a complete verification process record file. This file contains all time calibration logs, a list of conflicting intervals, and correction operation records. The time calibration log records all events in which the time deviation value exceeds the allowable range when the time information in the structured parameters and unstructured feedback is aligned and matched, as well as the corresponding original timestamp, corrected timestamp, and deviation value. The conflicting interval list lists the time intervals in which the subjective feedback description content is inconsistent with the numerical logic of the structured parameters, including the conflict type, the name of the associated parameter, the original text fragment of the subjective feedback, and the recommended verification method. The correction operation record records in detail the specific operation steps and process for correcting the data.

[0072] To facilitate management and traceability, the present invention assigns a unique version identifier to each report generation. This identifier records the report's generation time, operator, and number of revisions. This version identifier allows users to clearly understand the report's generation and modification history, ensuring traceability and accuracy.

[0073] In another preferred embodiment of the present invention, a separate event record is created for each data correction operation. This record includes the triggering condition, operation time, and impact scope. The triggering condition specifies the reason for the data correction, such as when the structured parameter loss rate exceeds a preset threshold or when subjective feedback is inconsistent with the structured parameter logic. The operation time records the specific time when the correction operation occurred, facilitating temporal analysis and tracing. The impact scope indicates the extent of the correction operation's impact on other data and parameters, for example, whether it affects the statistical analysis results of related indicators.

[0074] The present invention establishes a data modification association network. Within this network, the nodes of other related parameters affected by the modified data point are marked. For example, if a correction to a physiological indicator may affect the calculation results of other related indicators, these affected indicator nodes will be clearly marked in the association network. This way, the chain reaction of data modification operations can be clearly understood, ensuring data consistency and accuracy.

[0075] When a correction error is discovered, the present invention traces back along the associated network to the original data collection node. This means that starting from the final correction result, the original data collection source can be gradually found and the verification process can be retriggered. This re-verification ensures the accuracy and reliability of the data.

[0076] The present invention will generate a correction audit report. The report records in detail the time when the error was discovered, the impact assessment results and the correct values ​​after re-verification. The error discovery time clarifies when the correction error was discovered, which helps to analyze the time point and possible causes of the problem. The impact assessment results will analyze the extent of the impact of the correction error on the entire clinical trial data and results, such as whether it affects the statistical analysis results of key indicators, whether it affects the judgment of drug efficacy, etc. The correct values ​​after re-verification provide accurate data after re-verification, providing a reliable basis for subsequent analysis and decision-making. Finally, the present invention will store the audit report as an independent attachment bound to the final test report for easy access and reference by users.

[0077] Please refer to Figure 2 As shown, the present invention also includes an artificial intelligence-based clinical trial CRC management system for implementing the above-mentioned artificial intelligence-based clinical trial CRC management method, including:

[0078] The time matching module is used to obtain structured parameters and unstructured feedback in clinical trials, align the event timestamps of structured parameters with the semantic time stamps in unstructured feedback, and calculate the time deviation between the two;

[0079] The symptom verification module is used to extract sentiment values ​​and symptom description words from unstructured feedback based on natural language processing technology, and match them with structured parameters within the corresponding time window to identify the logical consistency between the description content and the parameter values;

[0080] The test adjustment module is configured to, when it is detected that the missing rate of structured parameters exceeds a preset threshold, send a supplementary question request to the subject terminal, wherein the question request includes symptom inquiry content associated with the missing parameters; and when it is detected that an event related to any structured parameter is omitted in the unstructured feedback, adjust the sampling frequency of the detection device and the combination of detection items to increase the collection density of the structured parameters;

[0081] The report correction module is used to correct the structured parameters and unstructured feedback according to the time deviation value and the logical consistency analysis results, and generate a final clinical trial report containing correction marks.

[0082] The above is a detailed description of an embodiment of the present invention. However, the content described is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A clinical trial CRC management method based on artificial intelligence, characterized in that: The following steps are involved: Obtain structured parameters and unstructured feedback from clinical trials, align the event timestamps of the structured parameters with the semantic time stamps in the unstructured feedback, and calculate the time offset between the two; Based on natural language processing technology, emotional tendency values ​​and symptom description words are extracted from unstructured feedback, and matched with structured parameters within the corresponding time window to identify the logical consistency between the description content and the parameter values; When it is detected that the structured parameter missing rate exceeds a preset threshold, a supplementary question request is sent to the subject terminal, wherein the question request includes symptom inquiry content associated with the missing parameter; When it is detected that an event related to any structured parameter is omitted in the unstructured feedback, the sampling frequency of the detection equipment and the combination of detection items are adjusted to increase the collection density of the structured parameters; According to the time deviation value and logical consistency analysis results, the structured parameters and unstructured feedback are revised and the final clinical trial report containing the revised marks is generated.

2. The artificial intelligence-based clinical trial CRC management method according to claim 1, characterized in that: The structured parameters include physiological indicator time series data, pharmacokinetic parameters and equipment detection values, and the unstructured feedback includes researcher voice records, subject electronic diary text and image evaluation annotations.

3. The artificial intelligence-based clinical trial CRC management method according to claim 1, characterized in that: The alignment and matching process is as follows: Identify all sentences containing time descriptions from the text content of subjective feedback, extract the time expressions involved and convert them into standardized time intervals; arrange the precise timestamps recorded by medical equipment in the order of acquisition to generate a structured parameter timeline; establish a mapping relationship between each standardized time interval and the nearest neighbor timestamp on the structured parameter timeline, and calculate the overlapping time length between the two; when the overlapping time length is less than the preset ratio, insert a virtual time node on the structured parameter timeline, and recalculate the average value of each parameter in the time interval as supplementary data; record all events in which the time deviation value exceeds the allowable range, and generate a time calibration log file containing the original timestamp, the corrected timestamp and the deviation value.

4. The artificial intelligence-based clinical trial CRC management method according to claim 1, characterized in that: The logical consistency identification process includes: A medical vocabulary database containing symptom names, physical signs, and drug reactions was constructed, with each word associated with a corresponding physiological indicator type and expected numerical range. The subjective feedback text was segmented, and keywords existing in the medical vocabulary database were screened out and their frequency of occurrence was recorded. The numerical records of the physiological indicators corresponding to the keywords within the same time interval were searched in the structured parameter database, and a time series change curve of the indicator was drawn. When the value of the physiological indicator corresponding to the keyword did not fall within the expected range or the change trend did not match the description, the time interval was marked as a contradictory interval to be verified. A list of contradictory intervals was generated, and each entry in the list contained the contradiction type, the associated parameter name, the original subjective feedback fragment, and the suggested verification method.

5. The artificial intelligence-based clinical trial CRC management method according to claim 1, characterized in that: The generation process of the supplementary questions is as follows: According to the type of missing structured parameters, related symptom descriptions and common question templates are retrieved from the preset medical knowledge base; question templates are sorted according to the probability of symptom occurrence, and high-priority questions that are strongly associated with missing parameters are given priority; professional medical terms are converted into everyday expressions that are easy for subjects to understand, and examples are added to improve question comprehension; the operation records and usage habits of the subject's terminal are analyzed, and its active time period is selected as the question sending window; the response status of the subject's terminal is continuously monitored after the question is sent, and a reminder notification is automatically triggered when no reply is received after the timeout.

6. The artificial intelligence-based clinical trial CRC management method according to claim 1, characterized in that: Adjusting the sampling frequency of the test equipment specifically includes: Identify the types of structured parameters that need to be additionally collected and their corresponding medical device models; query the current working status, remaining storage capacity, and battery life of the target device; analyze the historical usage data of the device to predict the subject's degree of cooperation and activity patterns in the future time period; when the subject is in a stationary state and the device is idle, gradually increase the sampling frequency until the maximum carrying capacity of the device is reached; align the timestamps of the additionally collected data with the original data, and update them to the structured parameter database.

7. The artificial intelligence-based clinical trial CRC management method according to claim 1, characterized in that: The modification of the structured parameters and the unstructured feedback specifically includes: Create a temporary data copy for events where the time deviation value exceeds the threshold, retain the original record and generate a revised version; convert the descriptive content in the subjective feedback into a numerical score, and the scoring criteria include the degree of descriptive detail, word accuracy and logical coherence; establish a linear regression model between the structured parameter values ​​and the subjective scores, and calculate the correlation coefficient between the two; when the correlation coefficient is lower than the preset standard, use a sliding window algorithm to smooth the parameter values, and the adjustment range is proportional to the subjective score; highlight all revised data points in the final report, and display a comparison chart of the values ​​before and after the adjustment.

8. The artificial intelligence-based clinical trial CRC management method according to claim 1, characterized in that: The generation of the final clinical trial report includes: Construct a hierarchical display structure, with the first layer showing the original collected uncorrected data and the second layer overlaying the corrected data points; add a detailed information panel next to each corrected data point; display the original value, corrected value, adjustment basis, and associated subjective feedback fragments of the data point in the detailed information panel; generate a complete verification process record file, including all time calibration logs, conflict interval lists, and correction operation records; set a unique version identifier for each report generation, recording the generation time, operator, and number of corrections.

9. The artificial intelligence-based clinical trial CRC management method according to claim 1, characterized in that: Create an independent event record for each data correction operation, including the triggering conditions, operation time and impact range; establish a data modification association network, and mark other related parameter nodes affected by the corrected data point; When a correction error is found, the associated network is traced back to the original data collection node to re-trigger the verification process; Generate a revised audit report, detailing the time of error discovery, impact assessment results, and corrected values ​​after re-verification; store the audit report as an independent attachment bound to the final test report.

10. An artificial intelligence-based clinical trial CRC management system, used to implement the artificial intelligence-based clinical trial CRC management method according to any one of claims 1 to 9, characterized in that: include: The time matching module is used to obtain structured parameters and unstructured feedback in clinical trials, align the event timestamps of structured parameters with the semantic time stamps in unstructured feedback, and calculate the time deviation between the two; The symptom verification module is used to extract sentiment values ​​and symptom description words from unstructured feedback based on natural language processing technology, and match them with structured parameters within the corresponding time window to identify the logical consistency between the description content and the parameter values; The test adjustment module is configured to, when it is detected that the missing rate of structured parameters exceeds a preset threshold, send a supplementary question request to the subject terminal, wherein the question request includes symptom inquiry content associated with the missing parameters; and when it is detected that an event related to any structured parameter is omitted in the unstructured feedback, adjust the sampling frequency of the detection device and the combination of detection items to increase the collection density of the structured parameters; The report correction module is used to correct the structured parameters and unstructured feedback according to the time deviation value and the logical consistency analysis results, and generate a final clinical trial report containing correction marks.

Images