Clinical test item and case rapid matching system and method

Abstract

The invention relates to the technical field of medical information matching, in particular to a clinical test item and case rapid matching system and method, and the system comprises a data acquisition and storage layer which obtains clinical test item information and multi-dimensional electronic health record data of an enrollment subject; the data processing layer is used for preprocessing the data; the preliminary matching layer is used for eliminating enrollment subjects which do not conform to hard conditions of clinical test items based on a rule engine; the deep matching layer is used for calculating the matching probability of the registered subjects and the clinical test items in the preliminary matching result, and generating a candidate subject list according to a probability threshold value; the data mining layer is used for mining a residual subject list from a related medical platform when the number of people in the candidate subject list is insufficient; and the output layer is used for generating a matching report for each successfully matched subject and highlighting the key matching factors. The matching efficiency and accuracy of clinical test items and medical records can be improved.

Description

Technical Field

[0001] The present invention relates to the field of medical information matching technology, and more particularly to a system and method for quickly matching clinical trial projects and cases. Background Art

[0002] In clinical trials, subject recruitment is a critical step in launching a clinical trial. Recruiting suitable subjects and ensuring the clinical trial is completed within a specified timeframe is a challenge facing clinical trials.

[0003] The traditional subject recruitment method uses manual review to process clues of subject matching projects, which requires a lot of time and energy to extract effective information from subject materials and has low work efficiency.

[0004] In order to improve matching efficiency, some recruitment management systems have also emerged. For example, in patent document CN110415772A, a matching module is used to automatically match the qualifications of registered personnel, and then the medical record information is pushed to project personnel for further screening of qualifications. It can be seen that it can only perform a preliminary screening of the qualifications of registered personnel. Whether the medical record information can be applied to a certain clinical trial project still requires manual screening by project personnel, and its degree of automation is still relatively low.

[0005] Therefore, how to quickly match clinical trial projects with cases and improve matching efficiency and accuracy remains an urgent problem that technicians in this field need to solve. Summary of the Invention

[0006] In view of this, the present invention provides a system and method for quickly matching clinical trial projects and cases, which can improve the matching efficiency and accuracy of clinical trial projects and medical records.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] In a first aspect, the present invention provides a system for rapidly matching clinical trial projects and cases, comprising:

[0009] The data acquisition and storage layer is used to connect to relevant medical platforms, obtain clinical trial project information and multi-dimensional electronic health record data of registered subjects, and store them in different databases;

[0010] The data processing layer is used to extract features and align cross-modal features of the multi-dimensional electronic health record data of the registered subjects;

[0011] The preliminary matching layer is used to exclude registered subjects who do not meet the rigid conditions of the clinical trial project based on the rule engine, and obtain the preliminary matching results of the registered subjects under each clinical trial project;

[0012] The deep matching layer is used to calculate the matching probability between the registered subjects and the clinical trial projects in the preliminary matching results, and generate a list of candidate subjects based on the probability threshold;

[0013] The data mining layer is used to count the number of candidate subjects in the list at a preset time point before the deadline of the clinical trial project to see whether the number of subjects meets the data requirements of the current clinical trial project. If not, the remaining list of subjects matching the current clinical trial project is mined from the massive electronic health record database provided by the relevant medical platform;

[0014] The output layer is used to generate a matching report for each successfully matched subject and highlight the key matching factors.

[0015] Furthermore, the data acquisition and storage layer includes:

[0016] The first acquisition and storage module is used to connect to multiple medical platforms. When the number of candidate subjects in the list does not meet the data requirements of the current clinical trial project, the corresponding docking interface is activated to obtain massive electronic health record data from the relevant medical platforms, and the electronic health record data of the successfully matched subjects is stored in the first database;

[0017] The second acquisition module and storage module are used to obtain clinical trial project information and store the clinical trial project information in a second database according to department or doctor information;

[0018] The third acquisition module and storage module are used to obtain the multimodal electronic health record data of the registered subjects, and establish an individual model for each subject, divide the individual model of each subject into nodes according to the six dimensions of examination, testing, diagnosis, physical signs, surgery, and medication, and store the multimodal electronic health record data of the registered subjects in the third database according to the node order.

[0019] Furthermore, the data processing layer includes:

[0020] A feature extraction module is used to extract numerical indicator features, text features representing medical records and examination reports, and image features representing lesion appearance from the multi-dimensional electronic health record data of registered subjects to obtain multimodal features;

[0021] Cross-modal feature alignment module, used to align multimodal features across time.

[0022] The splicing module is used to splice the multimodal features of the same period to obtain the multimodal splicing features of different periods;

[0023] The time series feature extraction module is used to analyze the multimodal splicing features of different periods using a long short-term memory network to obtain dynamic time series data features used to characterize disease progression and treatment response trends.

[0024] Furthermore, the preliminary matching layer includes:

[0025] The initial screening rule setting module is used to set multiple hard conditions that match the current clinical trial project, including age conditions, numerical index conditions, keyword matching conditions, and imaging lesion appearance conditions;

[0026] The preliminary matching module is used to match the recent multimodal splicing features of the registered subjects with each hard condition one by one. When all are met, it enters the deep matching stage.

[0027] Furthermore, the depth matching layer includes:

[0028] The weighted fusion module uses a multi-head attention mechanism to calculate the correlation weights between the recent numerical indicator features, text features, and image features of the registered subjects, enhances the representation of multimodal features, and introduces time series data features as the final subject features;

[0029] The probability calculation module is used to set the probability threshold according to the urgency of the clinical trial project, calculate the matching probability between the subject characteristics and the clinical trial project, screen out candidate subjects according to the probability threshold, and update the candidate subject list in real time.

[0030] Furthermore, the process of the weighted fusion module processing multimodal features is expressed as follows:

[0031]

[0032] Among them, h final represents the final subject feature representation, h attn represents the enhanced representation of multimodal features based on the output of the multi-head attention mechanism, h time Represents the time series data feature, Concat represents the concatenation operation, W g represents the weight matrix, and σ represents the sigmoid function.

[0033] Furthermore, the process of the probability calculation module calculating the matching probability between the subject characteristics and the clinical trial items includes:

[0034] Calculate the subject characteristics h final Matching score S with clinical trial project characteristics:

[0035] S=cos(h final ,t trial )+λ·LLMscore (Medical record text, trial project description)

[0036] Map the matching score S to the probability value P:

[0037]

[0038] Among them, t trial Indicates the characteristics of clinical trial projects, LLM score represents the semantic matching score generated by the medical big model, λ is the weight coefficient. If the test item relies on text description, the value of λ is increased. If the test item is mainly based on numerical conditions, the value of λ is reduced. ω and b are trainable parameters, which are optimized through historical enrollment data.

[0039] Furthermore, the probability threshold is set as follows:

[0040] Set a default threshold as the basic threshold;

[0041] Clinical trial projects are divided into multiple levels according to their urgency. Each level corresponds to an urgency threshold, and the urgency threshold corresponding to a higher level is lower than that corresponding to a lower level.

[0042] The weighted sum of the urgency threshold and the basic threshold corresponding to the current clinical trial project is used as the final probability threshold. The calculation method is as follows:

[0043] T=β·T B +(1-β)·T U

[0044] Among them, T represents the final probability threshold of a clinical trial project, T B represents the basic threshold, T U It represents the urgency threshold, β represents the balance factor, and its value is [0,1]. β is dynamically adjusted according to the progress stage of the test project. The β values ​​used in the test start-up period, test execution period and test sprint period decrease accordingly.

[0045] Furthermore, when the data mining layer mines the electronic health record data, it preliminarily matches the subjects who meet the hard conditions of the current clinical trial project from the massive electronic health record database provided by the relevant medical platform according to the pre-set priority order, and then performs deep matching on the subjects who have successfully matched the preliminary match. When the number of deeply matched subjects is twice the number of remaining subjects required, the mining is stopped.

[0046] In a second aspect, the present invention provides a method for quickly matching clinical trial projects with cases, which is applicable to the system described above, comprising the following steps:

[0047] Obtain clinical trial project information and multi-dimensional electronic health record data of registered subjects, and store them in different databases;

[0048] Perform feature extraction and cross-modal feature alignment on multi-dimensional electronic health record data of registered subjects;

[0049] Based on the rule engine, the registered subjects who do not meet the rigid conditions of the clinical trial project are excluded to obtain the preliminary matching results of the registered subjects under each clinical trial project;

[0050] Calculate the matching probability between the registered subjects and clinical trial projects in the preliminary matching results, and generate a list of candidate subjects based on the probability threshold;

[0051] At a preset time point before the deadline of the clinical trial project, count the number of candidate subjects in the list to see if they meet the data requirements of the current clinical trial project. If not, mine the remaining list of subjects that match the current clinical trial project from the massive electronic health record database provided by the relevant medical platform;

[0052] Generate a matching report for each successfully matched subject, highlighting key matching factors.

[0053] It can be seen from the above technical solutions that compared with the prior art, the present invention has the following beneficial effects:

[0054] 1. The present invention conducts stratified screening of subjects through initial screening → deep matching → data mining. The entire screening process is automated, greatly improving matching efficiency. When the number of candidate subjects is insufficient, potential subject data from relevant medical platforms is automatically obtained to ensure that recruitment is not restricted by a single data source, allowing clinical trials to proceed as planned, avoiding project delays due to insufficient recruitment, and improving the flexibility of the system.

[0055] 2. The present invention extracts features from multiple dimensions such as numerical indicators, text, and images, covering the comprehensiveness of the subjects' health data, avoiding the limitations of traditional single-dimensional analysis, and simultaneously analyzing the temporal changes of multimodal splicing features, capturing the progression of the disease and the trend of treatment response, thus achieving highly accurate matching of clinical trials and cases.

[0056] 3. When performing deep matching, the present invention integrates cosine similarity and large model semantic scoring when calculating matching probability, taking into account the matching requirements of numerical conditions and text descriptions, thereby improving overall accuracy. At the same time, the probability threshold no longer sets a fixed basic threshold as in traditional methods. Instead, it sets an urgency threshold based on the urgency of clinical trial projects, making the threshold for urgent projects lower. The overall probability threshold is dynamically adjusted by weighted summation of the basic threshold and the urgency threshold, thus meeting the timeliness requirements of the project while avoiding excessive relaxation that may lead to a decline in matching quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0058] Figure 1 This is a structural framework diagram of the clinical trial project and case rapid matching system provided by the present invention;

[0059] Figure 2 This is a flow chart of the method for quickly matching clinical trial projects and cases provided by the present invention. DETAILED DESCRIPTION

[0060] 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 creative efforts are within the scope of protection of the present invention.

[0061] like Figure 1 As shown, the embodiment of the present invention discloses a system for quickly matching clinical trial projects and cases, including:

[0062] The data acquisition and storage layer is used to connect to relevant medical platforms, obtain clinical trial project information and multi-dimensional electronic health record data of registered subjects, and store them in different databases;

[0063] The data processing layer is used to extract features and align cross-modal features of the multi-dimensional electronic health record data of the registered subjects;

[0064] The preliminary matching layer is used to exclude registered subjects who do not meet the rigid conditions of the clinical trial project based on the rule engine, and obtain the preliminary matching results of the registered subjects under each clinical trial project;

[0065] The deep matching layer is used to calculate the matching probability between the registered subjects and the clinical trial projects in the preliminary matching results, and generate a list of candidate subjects based on the probability threshold;

[0066] The data mining layer is used to count the number of candidate subjects in the list at a preset time point before the deadline of the clinical trial project to see whether the number of subjects meets the data requirements of the current clinical trial project. If not, the remaining list of subjects matching the current clinical trial project is mined from the massive electronic health record database provided by the relevant medical platform;

[0067] The output layer is used to generate a matching report for each successfully matched subject and highlight the key matching factors.

[0068] The following is a further explanation of the various layers of the above system architecture.

[0069] 1. Data acquisition and storage layer, specifically including a first acquisition and storage module, a second acquisition module and storage module, and a third acquisition module and storage module.

[0070] The first acquisition and storage module is used to connect to multiple medical platforms. When the number of candidate subjects in the list does not meet the data requirements of the current clinical trial project, the corresponding docking interface is activated to obtain the massive electronic health record data of the relevant medical platform, and the electronic health record data of the successfully matched subjects is stored in the first database. The first acquisition and storage module connects to multiple medical platforms through the API interface, and encrypted communication is performed between each API interface and the medical platform to avoid the leakage of the subject's personal information. At the same time, the subject's health record data information is desensitized, and key information is only visible to authorized researchers. A confidentiality agreement is signed with the docked medical platform to ensure the legality of data acquisition.

[0071] The second acquisition module and storage module are used to obtain clinical trial project information, which is uploaded by the project leader, and store the clinical trial project information in the second database according to department or doctor information. Each doctor can only see his or her own clinical trial project information.

[0072] The third acquisition module and storage module are used to obtain the multimodal electronic health record data of the registered subjects, which are uploaded by the registered subjects, and establish an individual model of each subject. The individual model of each subject is divided into nodes according to the six dimensions of examination, testing, diagnosis, physical signs, surgery, and medication, and the multimodal electronic health record data of the registered subjects are stored in the third database according to the node order.

[0073] The present invention stores data in separate databases according to clinical trial project information, individual models of registered subjects, and mined subject profile information, which not only ensures the clarity of the data structure but also improves the query and calculation efficiency during subsequent processing.

[0074] 2. Data processing layer, which specifically includes feature extraction module, cross-modal feature alignment module, splicing module and temporal feature extraction module.

[0075] The feature extraction module is used to extract numerical indicator features, text features representing medical records and examination reports, and image features representing lesion appearances from the multi-dimensional electronic health record data of registered subjects to obtain multimodal features. The pre-trained medical model PubMedBERT can be used to extract entities from medical records and reports to generate text features. The numerical indicators in various examination reports (such as blood values ​​and urine routine values) are encoded into vectors and standardized to eliminate dimensionality effects. Convolutional neural networks are used to extract features such as lesion volume and texture from images.

[0076] The cross-modal feature alignment module is used to align multimodal features across time to perform correlation analysis on the various modal data of the subjects in the same period, thereby improving the consistency and accuracy of feature expression.

[0077] The splicing module is used to splice the multimodal features of the same period to obtain multimodal splicing features of different periods.

[0078] The time series feature extraction module is used to analyze the multimodal splicing features of different periods using a long short-term memory network to obtain dynamic time series data features used to characterize disease progression and treatment response trends.

[0079] 3. The preliminary matching layer includes the preliminary screening rule setting module and the preliminary matching module.

[0080] The initial screening rule setting module is used to set multiple hard conditions that match the current clinical trial project, including age conditions, numerical indicator conditions, keyword matching conditions, and imaging lesion appearance conditions. For example, the inclusion and exclusion conditions for a lung cancer targeted drug trial are: the age inclusion condition must meet 75 ≥ Age ≥ 18; the numerical indicator inclusion condition must meet EGFR mutation positive = 1, and the numerical indicator exclusion condition is: recent immunotherapy Mediation (Immunotherapy) = 1 or creatinine clearance < 30 mL / min; the keyword condition is: if the medical record text contains words such as "heart failure" and "severe cardiovascular disease history", it will be excluded; the imaging lesion appearance inclusion condition is: tumor diameter ≤ 5 cm.

[0081] The preliminary matching module is used to match the recent multimodal splicing features of the registered subjects with each hard condition one by one. When all are met, it enters the deep matching stage.

[0082] 4. The depth matching layer includes a weighted fusion module and a probability calculation module.

[0083] 1) The weighted fusion module uses a multi-head attention mechanism to calculate the correlation weights between the recent numerical indicator features, text features, and image features of the registered subjects, enhances the representation of multimodal features, and introduces time series data features as the final subject features. The process of processing multimodal features is expressed as follows:

[0084]

[0085] Among them, h final represents the final subject feature representation, h attn represents the enhanced representation of multimodal features based on the output of the multi-head attention mechanism, h time Represents the time series data feature, Concat represents the concatenation operation, W g represents the weight matrix, and σ represents the sigmoid function.

[0086] The dynamic attention weights between text features, numerical indicator features, and image features are calculated through a multi-head attention mechanism to enhance the association between key modalities. For example, in the matching of targeted drug trials, if the patient's gene mutation is strongly correlated with the trial drug target, the attention weight of the text features will be significantly improved.

[0087] 2) The probability calculation module is used to set the probability threshold according to the urgency of the clinical trial project, calculate the matching probability between the subject characteristics and the clinical trial project, screen out candidate subjects according to the probability threshold, and update the candidate subject list in real time.

[0088] The calculation process of the matching probability includes:

[0089] Calculate the subject characteristics h final Matching score S with clinical trial project characteristics:

[0090] S=cos(h final ,t trial )+λ·LLM score (Medical record text, trial project description)

[0091] Map the matching score S to the probability value P:

[0092]

[0093] Among them, t trial Indicates the characteristics of clinical trial projects, LLMscore represents the semantic matching score generated by the large medical model. If the large model has no direct probability output, the semantic similarity is calculated through the contrastive learning framework. λ is the weight coefficient, which is used to adjust the contribution of semantic matching and structured matching. If the test item relies on text descriptions (such as rare diseases), the value of λ is increased (for example, 0.7). If the test item is mainly based on numerical conditions (such as drug dosage trials), the value of λ is reduced (for example, 0.3). ω and b are trainable parameters, which are optimized through historical enrollment data.

[0094] The probability threshold is set as follows:

[0095] Set a default threshold as the basic threshold;

[0096] Clinical trial projects are divided into multiple levels according to their urgency. Each level corresponds to an urgency threshold, and the urgency threshold corresponding to a higher level is lower than that corresponding to a lower level.

[0097] The weighted sum of the urgency threshold and the basic threshold corresponding to the current clinical trial project is used as the final probability threshold. The calculation method is as follows:

[0098] T=β·T B +(1-β)·T U

[0099] Where T represents the final probability threshold of a clinical trial project; T B represents the basic threshold, T U β represents the urgency threshold; β represents the balance factor, ranging from 0 to 1. β is dynamically adjusted based on the trial project's progress, with decreasing β values ​​during the trial launch, execution, and sprint phases. Specifically, during the trial launch phase, β depends on the baseline threshold, and a possible value of 0.9 is acceptable. During the execution phase, β is dynamically adjusted based on the enrollment rate deviation rate, decreasing when there is a lag. During the sprint phase, the larger the remaining gap, the faster the β value decays.

[0100] For example, clinical trial projects can be divided into four levels according to their urgency, namely emergency, major and general. The emergency level can be projects with no available drugs, such as advanced tumors and rare diseases. The corresponding threshold T U It can be taken as 0.3 to reduce the overall probability threshold, appropriately relax the selection conditions, and quickly match the selected subjects; the major level can be a chronic disease optimization treatment trial project, and the corresponding threshold T U It can be 0.6; the general level can be a preventive vaccine verification project, and the corresponding threshold T U It can be 0.9, which uses a stricter value for screening.

[0101] 5. At the preset time point before the deadline of the clinical trial project, the data mining layer counts the number of candidate subjects in the list to see whether it meets the data requirements of the current clinical trial project. If not, the data mining layer mines the remaining list of subjects that match the current clinical trial project from the massive electronic health record database provided by the relevant medical platform.

[0102] When mining electronic health record data, subjects who meet the hard conditions of the current clinical trial project are preliminarily matched from the massive electronic health record database provided by the relevant medical platform according to the pre-set priority order. Then, deep matching is performed on the subjects who have successfully matched the preliminarily. When the number of deeply matched subjects is twice the number of remaining subjects required, mining is stopped.

[0103] For example, according to the matching degree between the characteristics of each hospital and each test project, the order of retrieving data from each medical platform is set for each test project. When conducting data mining, the API interface of the relevant medical platform is called in accordance with the priority order set for the current project to retrieve the health record data of potential subjects. Each time a patient's data is retrieved, a matching degree calculation is performed through the preliminary matching layer and the deep matching layer. If the matching degree is passed, the patient data is stored in the first database. If the matching degree is not passed, the data does not need to be stored. When the number of successfully matched subjects is twice the number of remaining subjects required, data reception is stopped. The reason for setting the number of potential subjects to twice the number of remaining subjects is that not all potential subjects are willing to accept the test, so a certain amount of data margin needs to be retained. This can not only ensure that the candidate pool is sufficient, but also avoid excessive mining and waste of computing resources.

[0104] When matching subjects for a certain trial project, priority is given to matching from subjects who have clearly registered for the trial project. If the number is insufficient, potential subjects who meet the needs of the current clinical trial project will be screened out from subjects who have registered but have not clearly registered for which clinical trial project, and the health record data of the screened subjects will be stored in a third database. If the number is still not met, the remaining subjects will be mined from relevant medical platforms.

[0105] 6. The output layer generates a matching report for each successfully matched subject, highlighting key matching factors. By highlighting key matching factors (such as specific indicator compliance and semantic description consistency), project personnel can quickly review and make decisions, reducing secondary analysis time and enhancing the readability and interpretability of the report.

[0106] In other embodiments, the system of the present invention also includes a feedback layer for obtaining real-time health status data of enrolled subjects during the progress of the clinical trial project. If one or more indicators of the enrolled subjects do not meet the hard matching conditions, a rematching mechanism is triggered to replace the subject.

[0107] In another embodiment, the present invention also provides a method for quickly matching clinical trial projects and cases, which is applicable to the above system, such as Figure 2 As shown, the following steps are included:

[0108] S1. Obtain clinical trial project information and multi-dimensional electronic health record data of registered subjects, and store them in different databases;

[0109] S2. Feature extraction and cross-modal feature alignment of multi-dimensional electronic health record data of registered subjects;

[0110] S3. Eliminate registered subjects who do not meet the rigid requirements of the clinical trial project based on the rule engine, and obtain preliminary matching results for registered subjects under each clinical trial project;

[0111] S4. Calculate the matching probability between the registered subjects and the clinical trial projects in the preliminary matching results, and generate a candidate subject list based on the probability threshold;

[0112] S5. At a preset time point before the deadline of the clinical trial project, count the number of candidate subjects in the list to see if they meet the data requirements of the current clinical trial project. If not, mine the remaining list of subjects that match the current clinical trial project from the massive electronic health record database provided by the relevant medical platform;

[0113] S6. Generate a matching report for each successfully matched subject and highlight the key matching factors.

[0114] For the specific implementation of S1 to S6, please refer to the above description.

[0115] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0116] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A rapid matching system for clinical trial projects and cases, characterized by: include: The data acquisition and storage layer is used to connect to relevant medical platforms, obtain clinical trial project information and multi-dimensional electronic health record data of registered subjects, and store them in different databases; The data processing layer is used to extract features and align cross-modal features of the multi-dimensional electronic health record data of the registered subjects; The preliminary matching layer is used to exclude registered subjects who do not meet the rigid conditions of the clinical trial project based on the rule engine, and obtain the preliminary matching results of the registered subjects under each clinical trial project; The deep matching layer is used to calculate the matching probability between the registered subjects and the clinical trial projects in the preliminary matching results, and generate a list of candidate subjects based on the probability threshold; The data mining layer is used to count the number of candidate subjects in the list at a preset time point before the deadline of the clinical trial project to see whether the number of subjects meets the data requirements of the current clinical trial project. If not, the remaining list of subjects matching the current clinical trial project is mined from the massive electronic health record database provided by the relevant medical platform; The output layer is used to generate a matching report for each successfully matched subject and highlight the key matching factors.

2. The clinical trial project and case rapid matching system according to claim 1 is characterized in that: The data acquisition and storage layer includes: The first acquisition and storage module is used to connect to multiple medical platforms. When the number of candidate subjects in the list does not meet the data requirements of the current clinical trial project, the corresponding docking interface is activated to obtain massive electronic health record data from the relevant medical platforms, and the electronic health record data of the successfully matched subjects is stored in the first database; The second acquisition module and storage module are used to obtain clinical trial project information and store the clinical trial project information in a second database according to department or doctor information; The third acquisition module and storage module are used to obtain the multimodal electronic health record data of the registered subjects, and establish an individual model for each subject, divide the individual model of each subject into nodes according to the six dimensions of examination, testing, diagnosis, physical signs, surgery, and medication, and store the multimodal electronic health record data of the registered subjects in the third database according to the node order.

3. The clinical trial project and case rapid matching system according to claim 1 is characterized in that: The data processing layer includes: A feature extraction module is used to extract numerical indicator features, text features representing medical records and examination reports, and image features representing lesion appearance from the multi-dimensional electronic health record data of registered subjects to obtain multimodal features; Cross-modal feature alignment module, used to align multimodal features across time. The splicing module is used to splice the multimodal features of the same period to obtain the multimodal splicing features of different periods; The time series feature extraction module is used to analyze the multimodal splicing features of different periods using a long short-term memory network to obtain dynamic time series data features used to characterize disease progression and treatment response trends.

4. The clinical trial project and case rapid matching system according to claim 3 is characterized in that: The preliminary matching layer includes: The initial screening rule setting module is used to set multiple hard conditions that match the current clinical trial project, including age conditions, numerical index conditions, keyword matching conditions, and imaging lesion appearance conditions; The preliminary matching module is used to match the recent multimodal splicing features of the registered subjects with each hard condition one by one. When all are met, it enters the deep matching stage.

5. The clinical trial project and case rapid matching system according to claim 3 is characterized in that: The depth matching layer includes: The weighted fusion module uses a multi-head attention mechanism to calculate the correlation weights between the recent numerical indicator features, text features, and image features of the registered subjects, enhances the representation of multimodal features, and introduces time series data features as the final subject features; The probability calculation module is used to set the probability threshold according to the urgency of the clinical trial project, calculate the matching probability between the subject characteristics and the clinical trial project, screen out candidate subjects according to the probability threshold, and update the candidate subject list in real time.

6. The clinical trial project and case rapid matching system according to claim 5 is characterized in that: The process of the weighted fusion module processing multimodal features is expressed as follows: Among them, h final represents the final subject feature representation, h attn represents the enhanced representation of multimodal features based on the output of the multi-head attention mechanism, h time Represents the time series data feature, Concat represents the concatenation operation, W g represents the weight matrix, and σ represents the sigmoid function.

7. The clinical trial project and case rapid matching system according to claim 6, characterized in that: The process of calculating the matching probability between the subject characteristics and the clinical trial items by the probability calculation module includes: Calculate the subject characteristics h final Matching score S with clinical trial project characteristics: S=cos(h final ,t trial )+λ·LLM score (Medical record text, trial project description) Map the matching score S to the probability value P: Among them, t trial Indicates the characteristics of clinical trial projects, LLM score represents the semantic matching score generated by the medical large model, λ is the weight coefficient. If the test item relies on text description, the value of λ is increased. If the test item is mainly based on numerical conditions, the value of λ is reduced. ω and b are trainable parameters.

8. The clinical trial project and case rapid matching system according to claim 5 is characterized in that: The probability threshold is set as follows: Set a default threshold as the basic threshold; Clinical trial projects are divided into multiple levels according to their urgency. Each level corresponds to an urgency threshold, and the urgency threshold corresponding to a higher level is lower than that corresponding to a lower level. The weighted sum of the urgency threshold and the basic threshold corresponding to the current clinical trial project is used as the final probability threshold. The calculation method is as follows: T=β·T B +(1-β)·T U Among them, T represents the final probability threshold of a clinical trial project, T B represents the basic threshold, T U It represents the urgency threshold, and β represents the balance factor, which takes a value of [0,1] and is dynamically adjusted according to the progress stage of the pilot project.

9. The clinical trial project and case rapid matching system according to claim 5, characterized in that: When the data mining layer mines the electronic health record data, it preliminarily matches subjects who meet the hard conditions of the current clinical trial project from the massive electronic health record database provided by the relevant medical platform according to the pre-set priority order, and then performs deep matching on the subjects who have successfully matched the preliminary results. When the number of deeply matched subjects is twice the number of remaining subjects required, the mining is stopped.

10. A method for rapid matching of clinical trial projects and cases, characterized in that: The system is applicable to any one of claims 1 to 9, comprising the following steps: Obtain clinical trial project information and multi-dimensional electronic health record data of registered subjects, and store them in different databases; Perform feature extraction and cross-modal feature alignment on multi-dimensional electronic health record data of registered subjects; Based on the rule engine, the registered subjects who do not meet the rigid conditions of the clinical trial project are excluded to obtain the preliminary matching results of the registered subjects under each clinical trial project; Calculate the matching probability between the registered subjects and clinical trial projects in the preliminary matching results, and generate a list of candidate subjects based on the probability threshold; At a preset time point before the deadline of the clinical trial project, count the number of candidate subjects in the list to see if they meet the data requirements of the current clinical trial project. If not, mine the remaining list of subjects that match the current clinical trial project from the massive electronic health record database provided by the relevant medical platform; Generate a matching report for each successfully matched subject, highlighting key matching factors.

Images