Method and device for analyzing medical records of patients with abnormal serum PSA based on big data

Abstract

The present invention discloses a method and device for analyzing the medical records of patients with abnormal serum PSA based on big data, and relates to the field of medical care information technology. The method includes: extracting and identifying multiple user feature information in the user's medical record; obtaining multiple category attention parameters and multiple information attention parameters; obtaining multiple data change parameters; obtaining multiple modified information attention parameters; obtaining user access plans; obtaining access achievement probability, and combining the user access plan as the user medical record analysis result. The method solves the technical problem that it is difficult to accurately and comprehensively extract key information from massive medical record data and quantify its influence on PSA test results in the existing medical record analysis of patients with abnormal serum PSA, which leads to low efficiency of medical record analysis and low accuracy and reliability of medical record decision access, and achieves the technical effect of improving the efficiency of medical record analysis and the accuracy of access plan prediction.

Description

Technical Field

[0001] The present application relates to the technical field related to medical care information, and specifically to a method and device for analyzing the medical records of patients with abnormal serum PSA based on big data. Background Art

[0002] With the rapid development of big data technology and the continuous advancement of medical technology, the analysis and mining of large amounts of medical record data has become a key means to improve medical efficiency and the accuracy of subsequent access plan decisions. Especially in the screening and diagnosis of prostate cancer, serum prostate-specific antigen (PSA) test is a common biomarker detection method. Whether the results are abnormal or not has extremely important guiding significance for the patient's further treatment and follow-up. However, it is often difficult to fully evaluate the patient's condition based on the PSA test results alone. Therefore, it is necessary to conduct a comprehensive analysis in combination with the patient's medical record information. Traditional medical record analysis is difficult to deeply mine the effective information in the medical records from multiple dimensions and levels, and it is impossible to quickly and accurately quantify and evaluate the impact of different information on the PSA test results, affecting the accuracy of subsequent analysis and prediction of user access plans.

[0003] Therefore, in the current technologies related to medical record analysis of patients with abnormal serum PSA, it is difficult to accurately and comprehensively extract key information from massive medical record data and quantify its impact on PSA test results, which in turn leads to technical problems such as low efficiency of medical record analysis and low accuracy and reliability of medical record decision access. Summary of the invention

[0004] The present application provides a method and device for analyzing the medical records of patients with abnormal serum PSA based on big data, and adopts technical means such as feature information extraction, influence attention analysis, parameter correction calculation and training to achieve probability prediction model. It solves the technical problem that it is difficult to accurately and comprehensively extract key information from massive medical record data and quantify its influence on PSA test results, which leads to low efficiency of medical record analysis and low accuracy and reliability of medical record decision access, thereby achieving the technical effect of improving the efficiency of medical record analysis and the accuracy of access plan prediction.

[0005] The present application provides a method for analyzing the medical records of patients with abnormal serum PSA based on big data, the method comprising: collecting user medical records of users who undergo serum PSA tests and have abnormal test results, extracting and identifying multiple user feature information of multiple information categories in the user medical records; performing attention analysis on the impact of PSA test results on the multiple information categories and multiple user feature information, respectively, to obtain multiple category attention parameters and multiple information attention parameters; extracting timestamp information of multiple user feature information, performing data change probability analysis, and obtaining multiple data change parameters; performing correction calculations on the multiple information attention parameters according to the multiple data change parameters, and obtaining multiple corrected information attention parameters; calculating multiple attention parameters according to the multiple category attention parameters and the multiple corrected information attention parameters, and performing user access plan analysis according to the multiple attention parameters, using the multiple user feature information, and combining the PSA test results to obtain user access plans; performing access achievement probability prediction according to the user access plan, combining the multiple user feature information and the PSA test results, and obtaining access achievement probability, which is combined with the user access plan as the user medical record analysis result.

[0006] In a possible implementation, the user medical records of users who undergo serum PSA tests and have abnormal test results are collected, and multiple user feature information of multiple information categories in the user medical records are extracted and identified, and the following processing is also performed: the user medical records of users who undergo serum PSA tests and have abnormal test results are collected, and the user medical records include user medical record images; based on image segmentation, the images of multiple information areas in the user medical record images are segmented to obtain multiple information area images of multiple information categories, and information identification is performed to obtain multiple user feature information.

[0007] In a possible implementation, based on image segmentation, the images of multiple information areas in the user medical record image are segmented to obtain multiple information area images, and information recognition is performed, and the following processing is performed: based on the historical recognition data of multiple user medical records, a sample user medical record image set is collected, and the images of the information areas of multiple information categories in each user medical record image are segmented and marked to obtain a sample medical record image segmentation result set; based on image segmentation, a medical record image segmentation path is constructed, and the sample user medical record image set and the sample medical record image segmentation result set are used for training until completion; the user feature information in each type of segmented information area image in the sample medical record image segmentation result set is recognized and marked to obtain multiple sample user feature information sets; based on a convolutional neural network, a user information recognition path including multiple area information recognition branches is constructed, and the sample medical record image segmentation result set and the multiple sample user feature information sets are used for training until completion; the medical record image segmentation path and the user information recognition path are connected to segment and identify the user medical record image to obtain multiple user feature information of multiple information categories.

[0008] In a possible implementation, the influence attention analysis of the PSA test results is performed on the multiple information categories and the multiple user characteristic information respectively, and the following processing is performed: according to the multiple information categories, based on the user medical records of multiple sample users, multiple sample user characteristic information sequences and PSA test result sequences are collected; the correlation between the changes of user characteristic information in the multiple sample user characteristic information sequences and the changes of PSA test results in the PSA test result sequences is analyzed to obtain multiple category attention parameters; according to the multiple sample user characteristic information sequences and the PSA test result sequences, the proportion of the multiple user characteristic information among users with abnormal test results is extracted to obtain multiple information attention parameters.

[0009] In a possible implementation, timestamp information of multiple user feature information is extracted, data change probability analysis is performed, and multiple data change parameters are obtained. The following processing is also performed: timestamp information of multiple user feature information is extracted, and multiple cumulative durations are calculated based on the current time node; sample user feature information sets and sample cumulative duration sets of multiple information categories are collected, and a sample data change parameter set is collected based on the change amplitude of the sample user feature information at the current time node; the sample user feature information set and the sample cumulative duration set are used as training inputs, and the sample data change parameter set is used as training supervision to train a user information change analyzer; the user information change analyzer is used to perform data change analysis on the multiple user feature information and multiple cumulative durations to obtain multiple data change parameters.

[0010] In a possible implementation, multiple attention parameters are calculated based on the multiple category attention parameters and the multiple correction information attention parameters, and user access plan analysis is performed according to the multiple attention parameters using the multiple user feature information in combination with the PSA test results. The following processing is also performed: multiple attention parameters for the multiple user feature information are calculated based on the multiple category attention parameters and the multiple correction information attention parameters; based on the multiple information categories, multiple access plan generators for access plan analysis are constructed through ensemble learning, wherein each access plan generator includes M access plan generation paths, and M is an integer greater than 1; multiple analysis quantities are calculated based on the multiple attention parameters in combination with M; the multiple user feature information are respectively input into the access plan generation paths of multiple analysis quantities in the multiple access plan generators to generate multiple access plan sets; voting is performed on the multiple access plan sets to obtain user access plans, wherein voting weights are allocated in the voting process according to the multiple attention parameters, and the user access plan includes access method, access time, and access frequency.

[0011] In a possible implementation, based on the multiple information categories, multiple access plan generators for access plan analysis are constructed through ensemble learning, and the following processing is also performed: based on the multiple information categories, multiple sample user feature information sets and sample PSA test result sets, as well as a sample access plan set are collected according to the user medical records of a sample user group that has successfully accessed; within the first sample user feature information set, the sample PSA test result set, and the sample access plan set of the first information category, part of the sample data is randomly extracted to construct a first access plan generation path of the first access plan generator; within the first sample user feature information set, the sample PSA test result set, and the sample access plan set, part of the sample data is continued to be randomly extracted to construct a second access plan generation path of the first access plan generator until the Mth access plan generation path is constructed; and construction based on ensemble learning is continued to be performed to obtain multiple access plan generators each including M access plan generation paths.

[0012] In a possible implementation, according to the user access plan, combined with the multiple user feature information and PSA test results, the access achievement probability is predicted to obtain the access achievement probability, and the following processing is also performed: based on the user access data in the historical time, multiple sample user feature information sets, sample PSA test result sets, and sample user access plan sets are collected, and according to the number of accesses achieved, the sample access achievement probability set is calculated to obtain the sample access achievement probability set; the multiple sample user feature information sets, sample PSA test result sets, and sample user access plan sets are used as training inputs, and the sample access achievement probability set is used as training supervision to train the access achievement probability predictor; the access achievement probability predictor is used to predict the access achievement probability of the multiple user feature information, PSA test results, and user access plans to obtain the access achievement probability.

[0013] The present application also provides a device for analyzing the medical records of patients with abnormal serum PSA based on big data, including: a user characteristic information identification module, the user characteristic information identification module is used to collect user medical records of users who have undergone serum PSA tests and have abnormal test results, extract and identify multiple user characteristic information of multiple information categories in the user medical records; an influence attention analysis module, the influence attention analysis module is used to perform influence attention analysis of PSA test results on the multiple information categories and multiple user characteristic information respectively, and obtain multiple category attention parameters and multiple information attention parameters; a data change parameter acquisition module, the data change parameter acquisition module is used to extract timestamp information of multiple user characteristic information, perform data change probability analysis, and obtain multiple data change parameters; a correction information attention parameter acquisition module, the correction information attention parameter acquisition module The information attention parameter acquisition module is used to modify and calculate the multiple information attention parameters according to the multiple data change parameters to obtain multiple modified information attention parameters; the user access plan acquisition module is used to calculate and obtain multiple attention parameters according to the multiple category attention parameters and the multiple modified information attention parameters, and according to the multiple attention parameters, use the multiple user feature information and respectively combine the PSA test results to perform user access plan analysis to obtain a user access plan; the access achievement probability acquisition module is used to predict the access achievement probability according to the user access plan, combined with the multiple user feature information and the PSA test results, to obtain the access achievement probability, and combine the user access plan as the user medical record analysis result.

[0014] Through the big data-based serum PSA abnormal patient medical record analysis method and device proposed in this application, multiple user feature information in the user's medical record is extracted and identified; multiple category attention parameters and multiple information attention parameters are obtained; multiple data change parameters are obtained; multiple modified information attention parameters are obtained; user access plans are obtained; access achievement probability is obtained, and combined with the user access plan, it is used as the user medical record analysis result. It solves the technical problem that it is difficult to accurately and comprehensively extract key information from massive medical record data and quantify its impact on PSA test results in the existing serum PSA abnormal patient medical record analysis, which leads to low efficiency of medical record analysis and low accuracy and reliability of medical record decision access, and achieves the technical effect of improving the efficiency of medical record analysis and the accuracy of access plan prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solution of the embodiment of the present disclosure, the accompanying drawings of the embodiment of the present disclosure will be briefly introduced below. A flow chart is used in the present application to illustrate the operations performed by the device according to the embodiment of the present application. It should be understood that the previous or following operations are not necessarily performed precisely in order. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or more operations can be removed from these processes.

[0016] Figure 1 A schematic diagram of the process of analyzing the medical records of patients with abnormal serum PSA based on big data provided in an embodiment of the present application;

[0017] Figure 2 This is a schematic diagram of the structure of a device for analyzing medical records of patients with abnormal serum PSA based on big data provided in an embodiment of the present application.

[0018] Explanation of the reference numerals: user characteristic information identification module 10, attention influence analysis module 20, data change parameter acquisition module 30, correction information attention parameter acquisition module 40, user access plan acquisition module 50, access achievement probability acquisition module 60. DETAILED DESCRIPTION

[0019] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0020] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0021] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict, and the terms "first\second" involved are merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "including" and "having" and any variations are intended to cover non-exclusive inclusions, for example, a process, method, device, product, or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by technicians in the technical field of this application. The terms used herein are for the purpose of describing the embodiments of the present application only.

[0022] The present application embodiment provides a method for analyzing the medical records of patients with abnormal serum PSA based on big data, such as Figure 1 As shown, the method includes:

[0023] Step S100, collect user medical records of users who have undergone serum PSA tests and have abnormal test results, extract and identify multiple user feature information of multiple information categories in the user medical records. In the medical field of prostate health management, serum PSA (prostate-specific antigen) tests are often used as a screening tool for prostate cancer. PSA (prostate-specific antigen) is a protein produced by prostate cells, mainly present in male semen, and very small amounts are present in the blood. PSA may be abnormally elevated in diseases such as prostate cancer, benign prostatic hyperplasia (BPH) and prostatitis. The PSA test measures PSA levels through blood tests. When interpreting the PSA test results, the doctor comprehensively considers the patient's age, medical history, physical examination and other relevant examination results. For example, if the PSA test results are abnormal (i.e., outside the normal range), and by reviewing and analyzing the patient's medical records, other risk factors for prostate cancer (such as age over 50 years old, family history, etc.) are obtained, the doctor will usually further evaluate the patient's health status. Specifically, the medical records of all patients who have undergone serum PSA tests and have abnormal test results are retrieved from the medical system, that is, the patient's historical diagnosis, treatment records, Various information such as imaging examinations includes multiple different information categories, which may include but are not limited to basic information (such as patient name, age, gender, contact information, etc.), medical history information (such as previous medical history, surgical history, family medical history, etc.), PSA test results, other blood or urine test results, imaging examination (such as B-ultrasound, CT, MRI, etc.) results, treatment information (such as treatment plans already taken, medication status, treatment effects, etc.) and lifestyle habits (such as eating habits, exercise habits, smoking and drinking status, etc.); for each information category, key user feature information is extracted from it. For example, in the medical history information, feature information such as whether the patient has a family history of prostate cancer or a history of other urinary system diseases is extracted.

[0024] In a possible implementation, step S100 further includes step S110, collecting a user medical record of a user who has undergone a serum PSA test and has an abnormal test result, wherein the user medical record includes a user medical record image. A user medical record refers to the medical record information of a patient who has undergone a serum PSA (prostate-specific antigen) test and has an abnormal result, including a user medical record image, wherein the user medical record image generally refers to a scanned copy or electronic version of the medical record, including a doctor's diagnosis record, treatment plan, prescription, examination report, etc., and may include: the patient's medical history, chief complaint, current medical history, past history, physical examination record, laboratory test results (such as PSA test results), imaging examination reports (such as MRI, B-ultrasound, etc.), pathological diagnosis results, etc.

[0025] Step S100 also includes step S120, based on image segmentation, segmenting the images of multiple information areas in the user medical record image, obtaining multiple information area images of multiple information categories, and performing information recognition to obtain multiple user feature information. The user medical record image is preprocessed, including denoising, enhancement, binarization and other steps, and an image segmentation algorithm is applied to distinguish different information areas in the medical record image based on color, texture, shape and other features in the image, such as image segmentation based on threshold segmentation, edge detection, etc., to obtain multiple information area images corresponding to multiple information categories. After the image segmentation is completed, the information category to which each segmented area belongs is identified, and user feature information is extracted from each information area, which may include the patient's name, age, gender, medical history, diagnosis results, treatment plan, etc.

[0026] In a possible implementation, step S120 further includes step S121, collecting a set of sample user medical record images based on historical identification data of multiple user medical records, segmenting and marking images of information areas of multiple information categories in each user medical record image, and obtaining a set of sample medical record image segmentation results.

[0027] Collect historical identification data of multiple user medical records, select representative medical record images as the sample user medical record image set according to the diversity of medical record information, such as different disease types, medical record formats, text and image layouts, etc., use image segmentation technology to segment multiple information categories in each sample medical record image, identify the segmented information areas, assign a unique label category to each area, and the label or category corresponds to the actual information category in the medical record, such as patient name, age, gender, diagnosis results, treatment plan, etc., obtain multiple sample medical record image segmentation results, and organize the segmentation results of all sample medical record images into a set, namely, the sample medical record image segmentation result set.

[0028] Step S120 also includes step S122, based on image segmentation, constructing a medical record image segmentation path, and using the sample user medical record image set and the sample medical record image segmentation result set to perform training until completion. The medical record image segmentation path is constructed to separate different information areas (such as patient information, diagnosis results, treatment plans, etc.) in the medical record image, and the sample user medical record image set and the corresponding sample medical record image segmentation result set are used as training data for training, so that it can accurately identify and segment different information areas in the medical record image. When the performance of the model on the validation set reaches the expected level, the training is stopped, and the model is saved as the medical record image segmentation path.

[0029] Step S120 also includes step S123, identifying and marking the user characteristic information in each type of segmented information region image in the sample medical record image segmentation result set, and obtaining multiple sample user characteristic information sets. For each information region image, image processing and recognition technology is used to identify the user characteristic information therein, and the identified user characteristic information is marked. For each information region image, the identified and marked user characteristic information is organized into a set, and multiple sample user characteristic information sets are obtained.

[0030] Step S120 also includes step S124, based on the convolutional neural network, constructing a user information recognition path including multiple regional information recognition branches, and using the sample medical record image segmentation result set and multiple sample user feature information sets to perform training until completion. A user information recognition model is constructed based on a convolutional neural network, including multiple regional information recognition branches, each branch is responsible for identifying user feature information in a specific information area, using the sample medical record image segmentation result set and the corresponding multiple sample user feature information sets as training data, and training the user information recognition model based on the convolutional neural network so that it can accurately identify and extract user feature information in each information area. When the performance of the model on the validation set reaches the expected level, the training is stopped and the model is saved as a user information recognition path.

[0031] Step S120 also includes step S125, connecting the medical record image segmentation path and the user information identification path, segmenting and identifying the user medical record image, and obtaining multiple user feature information of multiple information categories. The trained medical record image segmentation path and the user information identification path are connected, and the user medical record image to be processed is used as the input medical record image segmentation path. The input medical record image is segmented through the medical record image segmentation path, and then each information area image obtained after segmentation is passed to the corresponding user information identification path for identification, and the user feature information is extracted from each information area image through the user information identification path.

[0032] Step S200, performing attention analysis on the PSA test results for the multiple information categories and the multiple user feature information, respectively, to obtain multiple category attention parameters and multiple information attention parameters. The attention analysis of the impact of PSA test results refers to quantifying the impact of different information categories and user characteristic information on the PSA test results, that is, analyzing the correlation between multiple information types and PSA test abnormalities (such as past medical history information, especially medical history related to prostate cancer, prostatitis, etc., which has a high correlation with PSA test abnormalities). For each category, its correlation with the PSA test results is analyzed, and the corresponding multiple category attention parameters are calculated to reflect the impact of the category information on the PSA test results; multiple user characteristic information and the probability of occurrence of PSA test abnormalities are analyzed. For example, with the increase of age, the volume of the prostate gradually increases, and the PSA level may also increase accordingly, especially for elderly men over 60 years old, who have a higher risk of increased PSA levels; within a short period of time after ejaculation, the PSA level may increase, which is usually temporary, but may affect the accuracy of the PSA test; for each user characteristic information, its correlation with the PSA test results is also analyzed, and multiple information attention parameters are calculated to reflect the impact of the characteristic information on the PSA test results.

[0033] In a possible implementation, step S200 further includes step S210, collecting multiple sample user feature information sequences and PSA test result sequences based on the user medical records of multiple sample users according to the multiple information categories. Based on multiple information categories (such as age, medical history, living habits, family history, etc.), collect these categories of user feature information from the user medical records of each sample user, and arrange them into sequences in chronological order (if available). For example, a user feature information sequence may include the patient's age, blood pressure, blood sugar and other data at different time points; collect their PSA test results from the user medical records of the sample users, and arrange them into sequences in chronological order. This sequence will show the PSA level of each user at different time points.

[0034] Step S200 also includes step S220, analyzing the correlation between the changes in the user feature information in the plurality of sample user feature information sequences and the changes in the PSA test results in the PSA test result sequence, and obtaining a plurality of category attention parameters. For each sample user, the changes in each feature information in the user feature information sequence are compared with the changes in the PSA test results in the PSA test result sequence. Specifically, the correlation between the two changes is quantified by correlation coefficients, covariances, etc., and one or more category attention parameters are generated for each information category accordingly, reflecting the contribution or influence of different categories of user feature information on the changes in the PSA test results.

[0035] Step S200 also includes step S230, extracting the proportion of the multiple user feature information among users with abnormal test results according to the multiple sample user feature information sequences and PSA test result sequences, and obtaining multiple information attention parameters. According to the multiple sample user feature information sequences and PSA test result sequences, those users with abnormal PSA test results (such as exceeding the normal range) are screened out from all sample users. For these users with abnormal test results, the proportion of each feature information in their user feature information sequence is counted. For example, the proportion of users with a history of hypertension among all users with abnormal test results is calculated. These proportion values ​​will be used as information attention parameters, reflecting which feature information is more important in the user group with abnormal test results.

[0036] Step S300, extract the timestamp information of multiple user feature information, perform data change probability analysis, and obtain multiple data change parameters. The timestamp information is the time information corresponding to when the user fills in the data. Specifically, some user feature information may change if it is data in the previous time, while some data will not change, such as family medical history will not change, and prostate medical history may change, such as recovery or new related medical history. Data change probability analysis refers to analyzing the trend of multiple user feature information changing over time, so as to evaluate the possible changes and magnitudes of these user feature information in the future. Specifically, the observed time series samples are used as input, the time series patterns are identified, a mathematical model is established to describe the change trend of the feature information, and the change rate of the feature information between adjacent timestamps is calculated to understand the fluctuation of the feature information in a short period of time, and then check whether the feature information shows periodic changes, such as seasonal changes or periodic fluctuations, and finally obtain multiple data change parameters for quantitatively describing the changes of user feature information over time.

[0037] In a possible implementation, step S300 further includes step S310, extracting timestamp information of multiple user feature information, and calculating multiple cumulative durations according to the current time node. Extract timestamp information of multiple user feature information from user medical records or related data sources, and calculate multiple cumulative durations according to the current time node (i.e., the time point during analysis) and the timestamp of each user feature information, wherein the cumulative duration refers to the total duration from the latest timestamp of each feature information to the current time node.

[0038] Step S300 also includes step S320, collecting a set of sample user characteristic information of multiple information categories and a set of sample cumulative durations, and collecting a set of sample data change parameters according to the change amplitude of the sample user characteristic information at the current time node. The set of sample user characteristic information of multiple information categories is collected, including various characteristic information from different users, such as age, weight, blood pressure, blood sugar, etc. At the same time, a set of sample cumulative durations corresponding to these characteristic information is collected, each cumulative duration corresponds to specific user characteristic information, and a set of sample data change parameters is collected according to the change amplitude of the sample user characteristic information at the current time node (i.e., the difference from the previous time point).

[0039] Step S300 also includes step S330, using the sample user feature information set and the sample cumulative duration set as training inputs, using the sample data change parameter set as training supervision, to train a user information change analyzer. A change analysis model is constructed based on a neural network or a regression model, using the sample user feature information set and the sample cumulative duration set as training inputs, using the sample data change parameter set as training supervision, to train the change analysis model, so that the analysis model can accurately predict or evaluate the change amplitude of the feature information based on the feature information and its duration, and finally obtain a user information change analyzer.

[0040] Step S300 also includes step S340, using the user information change analyzer to perform data change analysis on the multiple user feature information and the multiple accumulated durations to obtain multiple data change parameters. The multiple user feature information and the corresponding multiple accumulated durations are input into the trained user information change analyzer for data change analysis, and the user information change analyzer will output multiple data change parameters, which represent the possible change range or trend of the user feature information at the current time node.

[0041] Step S400, based on the multiple data change parameters, the multiple information attention parameters are corrected and calculated to obtain multiple corrected information attention parameters. The influence of the data change parameters on the information attention parameters is analyzed. For example, if the change rate of a certain user characteristic information is very high, then its influence on the PSA test results is also constantly changing, and its information attention parameters need to be adjusted, that is, based on the data change parameters, the information attention parameters are corrected, for example, using a linear or nonlinear model to simulate the relationship between the data change parameters and the information attention parameters, and correcting them accordingly.

[0042] Step S500, calculate and obtain multiple attention parameters according to the multiple category attention parameters and the multiple modified information attention parameters, and according to the multiple attention parameters, use the multiple user feature information, respectively combine the PSA test results to perform user access plan analysis, and obtain a user access plan. Combine the multiple category attention parameters with the multiple modified information attention parameters (for example, weighted average, linear combination, etc. may be used) to calculate and obtain multiple final attention parameters. The attention parameters will more accurately reflect the degree of influence of different information categories and user feature information on the PSA test results. Then, based on the calculated multiple attention parameters, use multiple user feature information (such as the user's medical history information, preference interests, historical behavior, etc.) respectively combined with the PSA test results to perform user access plan analysis, that is, train multiple base learners in parallel based on decision trees or neural networks. For each base learner, use user feature information and PSA test results as input to analyze the user access path. In each base learner, predictions of multiple user access paths may be generated. By combining the results of multiple base learners, the possible user access paths and access plans are analyzed. When it is necessary to select a specific user access plan, the majority voting strategy can be adopted. That is, for each possible access path, the number of votes supporting the path in all base learners is counted, and the path with the most votes is selected as the final recommended access plan. This user access plan is based on the prediction results of multiple base learners and the majority voting strategy, and has higher robustness and accuracy. If there are multiple paths with the same and highest votes, further screening can be carried out based on other indicators (such as path complexity, expected user satisfaction, etc.).

[0043] In a possible implementation, step S500 further includes step S510, calculating multiple attention parameters of the multiple user feature information according to the multiple category attention parameters and the multiple correction information attention parameters. Using multiple category attention parameters (reflecting the contribution of different categories of user feature information to the change of PSA test results) and correction information attention parameters, the attention parameters of each user feature information are calculated, and these attention parameters represent the importance or weight of each user feature information in subsequent analysis.

[0044] Step S500 also includes step S520, based on the multiple information categories, multiple access plan generators for performing access plan analysis are constructed through ensemble learning, wherein each access plan generator includes M access plan generation paths, and M is an integer greater than 1. Based on multiple information categories, multiple access plan generators for performing access plan analysis are constructed through ensemble learning (such as random forest, gradient boosting tree, etc.), each access plan generator includes M access plan generation paths, and these paths may have different algorithms and models for generating access plans according to user feature information.

[0045] Step S500 also includes step S530, calculating and obtaining multiple analysis quantities according to the multiple attention parameters combined with M. Multiple analysis quantities are calculated according to the calculated multiple attention parameters and M (the number of access plan generation paths), and these analysis quantities may determine how many access plan generation paths each user feature information should be input into for processing, and a higher attention parameter may correspond to a larger number of analyses.

[0046] Step S500 also includes step S540, inputting the plurality of user feature information into the plurality of analyzed access plan generation paths in the plurality of access plan generators to generate a plurality of access plan sets. The plurality of user feature information is respectively input into the plurality of analyzed access plan generation paths in the plurality of access plan generators, each path generates a plurality of access plans according to the input user feature information, and the access plans generated by all the paths are combined to form a plurality of access plan sets, each set may contain a different number of access plans, reflecting the different understanding and processing methods of the user feature information by different paths.

[0047] Step S500 also includes step S550, voting on the multiple access plan sets to obtain a user access plan, wherein voting weights are assigned according to the multiple attention parameters in the voting process, and the user access plan includes access mode, access time, and access frequency. Voting is performed on multiple access plan sets to determine the final user access plan. During the voting process, voting weights are assigned to each set or access plan within the set according to the multiple attention parameters calculated previously. The access plan corresponding to the user characteristic information with higher attention parameters will obtain a greater weight. The final user access plan includes access mode (such as telephone, email, door-to-door visit, etc.), access time (such as a specific time point or time period) and access frequency (such as once a week, once a month, etc.).

[0048] In a possible implementation, step S520 further includes step S521, based on the multiple information categories, collecting multiple sample user feature information sets and sample PSA test result sets, and sample access scheme sets according to the user medical records of the sample user group that has successfully accessed. According to multiple information categories (such as user demographic information, health status, historical access records, etc.), data is collected from the user medical records of the sample user group that has successfully accessed, including multiple sample user feature information sets (representing different characteristics of users, such as age, gender, health status, etc.), sample PSA test result sets (prostate-specific antigen (PSA) test results), and sample access scheme sets (representing different access schemes that lead to successful access, such as access time, access method, etc.).

[0049] Step S520 also includes step S522, randomly extracting part of the sample data from the first sample user feature information set, the sample PSA test result set, and the sample access solution set of the first information category, and constructing a first access solution generation path of the first access solution generator. Based on the sample data of the first information category (a subset randomly selected from the user feature information), the first access solution generator is constructed. Specifically, part of the sample data is randomly extracted from the first sample user feature information set, the sample PSA test result set, and the sample access solution set, and a prediction model is constructed based on a decision tree or a neural network. The model is trained using these randomly extracted sample data to obtain the first access solution generation path of the first access solution generator, which is used to predict or generate possible access solutions based on the user feature information.

[0050] Step S520 also includes step S523, continuing to randomly extract part of the sample data in the first sample user feature information set, the sample PSA test result set, and the sample access plan set, and constructing a second access plan generation path of the first access plan generator until the Mth access plan generation path is constructed. In the same first sample user feature information set, the sample PSA test result set, and the sample access plan set, continue to randomly extract part of the sample data, use the newly extracted sample data to construct the second access plan generation path of the first access plan generator, and repeat this process until M different access plan generation paths are constructed, each of which is constructed based on different random sample data, providing different access plan prediction or generation methods.

[0051] Step S520 also includes step S524, continuing to construct based on ensemble learning to obtain multiple access plan generators that each include M access plan generation paths. After completing the construction of the M access plan generation paths of the first access plan generator, continue to construct more access plan generators based on other information categories and corresponding sample data, each access plan generator contains M access plan generation paths, and these paths are all constructed based on different information categories and random sample data. Using the idea of ​​ensemble learning, by constructing multiple different access plan generators and generation paths, and using random sampling to increase the diversity and generalization ability of the model, eventually, these generators can work together to provide users with more accurate and personalized access plans.

[0052] Step S600: According to the user access plan, combined with the multiple user feature information and the PSA test results, the access achievement probability is predicted, and the access achievement probability is obtained, combined with the user access plan, as the user medical record analysis result. Use a machine learning model (such as random forest, neural network, etc.) to build and train the model, learn the relationship between the user feature information and the PSA test results and the access achievement probability, combine the user feature information and the PSA test results, and predict the access achievement probability of the user according to the access plan. The prediction result is a probability value, which indicates the possibility of the user visiting according to the recommended access plan. Then, the predicted access achievement probability is combined with the user access plan to form the user medical record analysis result. The user medical record analysis result may be used in medical decision support, health risk assessment, patient education and other aspects, thereby providing accurate medical services to improve medical quality.

[0053] In a possible implementation, step S600 further includes step S610, collecting multiple sample user feature information sets, sample PSA test result sets, and sample user access plan sets based on user access data in historical time, and calculating and obtaining a sample access achievement probability set based on the number of accesses achieved. From the user access data in historical time, multiple sample user feature information sets are collected, sample PSA test result sets related to these users are collected, and a sample user access plan set is collected, that is, the access plans actually adopted by users in history, such as access time, access method, etc., and then according to the access achievement status in the historical data (for example, whether the user has accessed according to the access plan, or whether the access has achieved the expected goal), a sample access achievement probability set is calculated, and this probability set represents the probability of each sample access plan actually being achieved.

[0054] Step S600 also includes step S620, using the multiple sample user feature information sets, sample PSA test result sets, and sample user access plan sets as training inputs, and using the sample access achievement probability set as training supervision, to train an access achievement probability predictor. A prediction model is constructed based on machine learning (such as logistic regression, random forest, etc.), using the collected multiple sample user feature information sets, sample PSA test result sets, and sample user access plan sets as training inputs, and using the sample access achievement probability set as training supervision, the prediction model is trained to obtain an access achievement probability predictor for predicting the probability of access achievement from user feature information, PSA test results, and user access plans.

[0055] Step S600 also includes step S630, using the access achievement probability predictor to predict the access achievement probability for the multiple user feature information, PSA test results and user access plans to obtain the access achievement probability. After the training is completed, the trained access achievement probability predictor can be used to predict new user data, and the given multiple user feature information, PSA test results and user access plans are input into the access achievement probability predictor, which analyzes and processes these input data and finally outputs an access achievement probability, which indicates the possibility of access achievement under the given user feature information, PSA test results and user access plan.

[0056] In the above, refer to Figure 1 The method for analyzing the medical records of patients with abnormal serum PSA based on big data according to an embodiment of the present invention is described in detail. Figure 2 A device for analyzing medical records of patients with abnormal serum PSA based on big data according to an embodiment of the present invention is described.

[0057] The device for analyzing the medical records of patients with abnormal serum PSA based on big data according to an embodiment of the present invention is used to solve the technical problems that the existing analysis of the medical records of patients with abnormal serum PSA is difficult to accurately and comprehensively extract key information from massive medical record data and quantify its influence on the PSA test results, which leads to low efficiency of medical record analysis and low accuracy and reliability of medical record decision access, thereby achieving the technical effect of improving the efficiency of medical record analysis and the accuracy of access plan prediction. The device for analyzing the medical records of patients with abnormal serum PSA based on big data includes: a user feature information identification module 10, an influence attention analysis module 20, a data change parameter acquisition module 30, a correction information attention parameter acquisition module 40, a user access plan acquisition module 50, and an access achievement probability acquisition module 60.

[0058] A user characteristic information identification module 10, the user characteristic information identification module 10 is used to collect user medical records of users who have undergone serum PSA tests and have abnormal test results, and extract and identify multiple user characteristic information of multiple information categories in the user medical records;

[0059] An attention-affecting analysis module 20, wherein the attention-affecting analysis module 20 is used to perform attention-affecting analysis of the PSA test results on the multiple information categories and the multiple user characteristic information, respectively, to obtain multiple category attention parameters and multiple information attention parameters;

[0060] A data change parameter acquisition module 30, wherein the data change parameter acquisition module 30 is used to extract timestamp information of multiple user feature information, perform data change probability analysis, and obtain multiple data change parameters;

[0061] A modified information attention parameter obtaining module 40, wherein the modified information attention parameter obtaining module 40 is used to perform a modified calculation on the multiple information attention parameters according to the multiple data change parameters to obtain multiple modified information attention parameters;

[0062] A user access scheme obtaining module 50, the user access scheme obtaining module 50 is used to calculate and obtain multiple attention parameters according to the multiple category attention parameters and the multiple correction information attention parameters, and to perform user access scheme analysis according to the multiple attention parameters and the multiple user feature information in combination with the PSA test results to obtain a user access scheme;

[0063] The access achievement probability acquisition module 60 is used to predict the access achievement probability based on the user access plan, combined with the multiple user feature information and the PSA test results, to obtain the access achievement probability, combined with the user access plan, as the user medical record analysis result.

[0064] The specific configuration of the user characteristic information identification module 10 will be described in detail below. The user characteristic information identification module 10 may further include: collecting user medical records of users who undergo serum PSA tests and have abnormal test results, wherein the user medical records include user medical record images; based on image segmentation, segmenting images of multiple information regions in the user medical record images to obtain multiple information region images of multiple information categories, and performing information identification to obtain multiple user characteristic information.

[0065] The specific configuration of the user characteristic information identification module 10 will be described in detail below. The user characteristic information identification module 10 further includes: according to the historical identification data of multiple user medical records, a sample user medical record image set is collected, and the images of the information areas of multiple information categories in each user medical record image are segmented and marked to obtain a sample medical record image segmentation result set; based on image segmentation, a medical record image segmentation path is constructed, and the sample user medical record image set and the sample medical record image segmentation result set are used for training until completion; the user characteristic information in each type of segmented information area image in the sample medical record image segmentation result set is identified and marked to obtain multiple sample user characteristic information sets; based on a convolutional neural network, a user information identification path including multiple area information identification branches is constructed, and the sample medical record image segmentation result set and the multiple sample user characteristic information sets are used for training until completion; the medical record image segmentation path and the user information identification path are connected to segment and identify the user medical record image to obtain multiple user characteristic information of multiple information categories.

[0066] The specific configuration of the attention-affecting analysis module 20 will be described in detail below. The attention-affecting analysis module 20 may further include: collecting multiple sample user feature information sequences and PSA test result sequences based on the user medical records of multiple sample users according to the multiple information categories; analyzing the correlation between the changes in the user feature information in the multiple sample user feature information sequences and the changes in the PSA test results in the PSA test result sequences to obtain multiple category attention parameters; extracting the proportion of the multiple user feature information in users with abnormal test results according to the multiple sample user feature information sequences and the PSA test result sequences to obtain multiple information attention parameters.

[0067] The specific configuration of the data change parameter acquisition module 30 will be described in detail below. The data change parameter acquisition module 30 may further include: extracting timestamp information of multiple user feature information, and calculating multiple cumulative durations according to the current time node; collecting sample user feature information sets and sample cumulative duration sets of multiple information categories, and collecting sample data change parameter sets according to the change amplitude of the sample user feature information at the current time node; using the sample user feature information set and the sample cumulative duration set as training input, using the sample data change parameter set as training supervision, training a user information change analyzer; using the user information change analyzer, performing data change analysis on the multiple user feature information and multiple cumulative durations to obtain multiple data change parameters.

[0068] The specific configuration of the user access plan acquisition module 50 will be described in detail below. The user access plan acquisition module 50 may further include: calculating and obtaining multiple attention parameters of the multiple user feature information according to the multiple category attention parameters and the multiple correction information attention parameters; constructing multiple access plan generators for access plan analysis through ensemble learning based on the multiple information categories, wherein each access plan generator includes M access plan generation paths, and M is an integer greater than 1; obtaining multiple analysis quantities according to the multiple attention parameters combined with M calculations; inputting the multiple user feature information into the access plan generation paths of the multiple analysis quantities in the multiple access plan generators respectively, and generating multiple access plan sets; performing voting processing on the multiple access plan sets to obtain user access plans, wherein voting weights are allocated in the voting processing according to the multiple attention parameters, and the user access plan includes access mode, access time, and access frequency.

[0069] The specific configuration of the user access plan acquisition module 50 will be described in detail below. The user access plan acquisition module 50 further includes: based on the multiple information categories, according to the user medical records of the sample user group that has been successfully accessed, collecting multiple sample user feature information sets and sample PSA test result sets, as well as sample access plan sets; randomly extracting part of the sample data from the first sample user feature information set, sample PSA test result set and sample access plan set of the first information category, and constructing the first access plan generation path of the first access plan generator; continuing to randomly extract part of the sample data from the first sample user feature information set, sample PSA test result set and sample access plan set, and constructing the second access plan generation path of the first access plan generator, until the Mth access plan generation path is constructed; continuing to construct based on ensemble learning, and obtaining multiple access plan generators including M access plan generation paths respectively.

[0070] The specific configuration of the access achievement probability acquisition module 60 will be described in detail below. The access achievement probability acquisition module 60 further includes: collecting multiple sample user feature information sets, sample PSA test result sets, and sample user access plan sets based on the user access data in the historical time, and calculating and obtaining the sample access achievement probability set based on the number of accesses achieved; using the multiple sample user feature information sets, sample PSA test result sets, and sample user access plan sets as training inputs, using the sample access achievement probability set as training supervision, and training the access achievement probability predictor; using the access achievement probability predictor to predict the access achievement probability of the multiple user feature information, PSA test results, and user access plans to obtain the access achievement probability.

[0071] The apparatus for analyzing medical records of patients with abnormal serum PSA based on big data provided in an embodiment of the present invention can execute the method for analyzing medical records of patients with abnormal serum PSA based on big data provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0072] Although the present application makes various references to certain modules in the device according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0073] The above specific implementations do not constitute a limitation on the protection scope of this application. It should be understood by those skilled in the art that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of this application should be included in the protection scope of this application.

Claims

1. A method for analyzing the medical records of patients with abnormal serum PSA based on big data, characterized in that: The method comprises: Collecting user medical records of users who undergo serum PSA tests and have abnormal test results, extracting and identifying multiple user feature information of multiple information categories in the user medical records; Performing influence attention analysis of the PSA test results on the multiple information categories and the multiple user feature information respectively to obtain multiple category attention parameters and multiple information attention parameters; Extract timestamp information of multiple user feature information, perform data change probability analysis, and obtain multiple data change parameters; According to the multiple data change parameters, the multiple information attention parameters are corrected and calculated to obtain multiple corrected information attention parameters; Calculate multiple attention parameters according to the multiple category attention parameters and the multiple correction information attention parameters, and perform user access plan analysis according to the multiple attention parameters and the multiple user feature information in combination with the PSA test results to obtain a user access plan; According to the user access plan, combined with the plurality of user feature information and the PSA test results, the access achievement probability is predicted to obtain the access achievement probability, which is combined with the user access plan as the user medical record analysis result; The attention-affecting analysis of the PSA test results is performed on the multiple information categories and the multiple user feature information respectively, including: According to the multiple information categories, based on the user medical records of the multiple sample users, collecting multiple sample user feature information sequences and PSA test result sequences; Analyzing the correlation between changes in user feature information in the plurality of sample user feature information sequences and changes in PSA test results in the PSA test result sequences to obtain a plurality of category attention parameters; According to the plurality of sample user feature information sequences and the PSA test result sequence, extracting the proportion of the plurality of user feature information appearing among users with abnormal test results, and obtaining a plurality of information attention parameters; The method further comprises: calculating a plurality of attention parameters according to the plurality of category attention parameters and the plurality of correction information attention parameters, and performing a user access scheme analysis according to the plurality of attention parameters and the plurality of user feature information in combination with the PSA test results, including: Calculate and obtain multiple attention parameters of the multiple user feature information according to the multiple category attention parameters and the multiple correction information attention parameters; Based on the multiple information categories, multiple access plan generators for performing access plan analysis are constructed through ensemble learning, wherein each access plan generator includes M access plan generation paths, M is an integer greater than 1, and the ensemble learning is a random forest or a gradient boosting tree; Obtain multiple analysis quantities according to the multiple attention parameters combined with M calculation; Inputting the plurality of user feature information into the access plan generation paths of the plurality of analyzed quantities in the plurality of access plan generators respectively to generate a plurality of access plan sets; Voting is performed on the multiple access plan sets to obtain a user access plan, wherein voting weights are allocated according to the multiple attention parameters in the voting process, and the user access plan includes access mode, access time, and access frequency.

2. The method for analyzing medical records of patients with abnormal serum PSA based on big data according to claim 1, characterized in that: Collect user medical records of users who undergo serum PSA tests and have abnormal test results, extract and identify multiple user feature information of multiple information categories in the user medical records, including: Collecting a user medical record of a user who undergoes a serum PSA test and has an abnormal test result, wherein the user medical record includes a user medical record image; Based on image segmentation, the images of multiple information areas in the user medical record image are segmented to obtain multiple information area images of multiple information categories, and information recognition is performed to obtain multiple user feature information.

3. The method for analyzing medical records of patients with abnormal serum PSA based on big data according to claim 2, characterized in that: Based on image segmentation, images of multiple information regions in the user medical record image are segmented to obtain multiple information region images, and information recognition is performed, including: According to the historical identification data of multiple user medical records, a sample user medical record image set is collected, and images of information areas of multiple information categories in each user medical record image are segmented and marked to obtain a sample medical record image segmentation result set; Based on image segmentation, a medical record image segmentation path is constructed, and training is performed using the sample user medical record image set and the sample medical record image segmentation result set until completion; Identifying and marking user feature information in each type of segmented information region image in the sample medical record image segmentation result set to obtain multiple sample user feature information sets; Based on a convolutional neural network, a user information recognition path including multiple regional information recognition branches is constructed, and training is performed using the sample medical record image segmentation result set and multiple sample user feature information sets until completion; The medical record image segmentation path and the user information identification path are connected to segment the user medical record image and identify the information to obtain multiple user feature information of multiple information categories.

4. The method for analyzing medical records of patients with abnormal serum PSA based on big data according to claim 1, characterized in that: Extract timestamp information of multiple user feature information, perform data change probability analysis, and obtain multiple data change parameters, including: Extracting timestamp information of multiple user feature information, and calculating multiple cumulative durations according to the current time node; Collecting a set of sample user characteristic information and a set of sample cumulative durations of multiple information categories, and collecting a set of sample data change parameters according to the change amplitude of the sample user characteristic information at the current time node; Using the sample user feature information set and the sample cumulative duration set as training inputs, using the sample data change parameter set as training supervision, to train a user information change analyzer; The user information change analyzer is used to perform data change analysis on the multiple user feature information and multiple accumulated durations to obtain multiple data change parameters.

5. The method for analyzing medical records of patients with abnormal serum PSA based on big data according to claim 1, characterized in that: Based on the multiple information categories, multiple access plan generators for performing access plan analysis are constructed through integrated learning, including: Based on the multiple information categories, collecting multiple sample user feature information sets and sample PSA test result sets, as well as a sample access plan set according to user medical records of the sample user group that has been successfully accessed; Randomly extract part of the sample data from the first sample user feature information set, the sample PSA test result set and the sample access plan set of the first information category, and construct a first access plan generation path of the first access plan generator; Continue to randomly extract part of the sample data from the first sample user feature information set, the sample PSA test result set, and the sample access plan set to construct a second access plan generation path of the first access plan generator until an Mth access plan generation path is constructed; Continue to build based on ensemble learning to obtain multiple access plan generators that each include M access plan generation paths.

6. The method for analyzing medical records of patients with abnormal serum PSA based on big data according to claim 1, characterized in that: According to the user access scheme, combined with the plurality of user feature information and the PSA test results, the access achievement probability is predicted to obtain the access achievement probability, including: According to the user access data in the historical time, multiple sample user feature information sets, sample PSA test result sets, and sample user access plan sets are collected, and according to the number of accesses achieved, the sample access achievement probability set is calculated; Using the plurality of sample user feature information sets, sample PSA test result sets, and sample user access plan sets as training inputs, and using the sample access achievement probability set as training supervision, to train an access achievement probability predictor; The access achievement probability predictor is used to predict the access achievement probability of the multiple user feature information, PSA test results and user access plans to obtain the access achievement probability.

7. A device for analyzing the medical records of patients with abnormal serum PSA based on big data, characterized in that: The device is used to implement the method for analyzing the medical records of patients with abnormal serum PSA based on big data according to any one of claims 1 to 6, and the device comprises: A user characteristic information identification module, the user characteristic information identification module is used to collect user medical records of users who have undergone serum PSA tests and have abnormal test results, extract and identify multiple user characteristic information of multiple information categories in the user medical records; An attention-affecting analysis module, wherein the attention-affecting analysis module is used to perform attention-affecting analysis on the PSA test results of the multiple information categories and the multiple user characteristic information, respectively, to obtain multiple category attention parameters and multiple information attention parameters; A data change parameter acquisition module, the data change parameter acquisition module is used to extract timestamp information of multiple user feature information, perform data change probability analysis, and obtain multiple data change parameters; A modified information attention parameter acquisition module, wherein the modified information attention parameter acquisition module is used to perform a modified calculation on the multiple information attention parameters according to the multiple data change parameters to obtain multiple modified information attention parameters; A user access scheme acquisition module, the user access scheme acquisition module is used to calculate and obtain multiple attention parameters according to the multiple category attention parameters and the multiple correction information attention parameters, and perform user access scheme analysis according to the multiple attention parameters, using the multiple user feature information, and respectively combining the PSA test results to obtain a user access scheme; A module for obtaining the probability of access achievement is used to predict the probability of access achievement based on the user access plan, combined with the multiple user feature information and the PSA test results, to obtain the probability of access achievement, and combine it with the user access plan as the user medical record analysis result.

Images