Maternal-fetal cohort dataset generation method and apparatus, terminal, and storage medium

Abstract

Provided are a maternal-fetal cohort dataset generation method and apparatus, a terminal, and a storage medium. The method comprises: acquiring health data of a plurality of users from different medical institutions, wherein the health data comprises a plurality of pieces of original text data and original image data covering pre-pregnancy, pregnancy and postpartum stages of the users and representing health conditions of the users and / or their fetuses; processing the original text data of all the users to obtain structured text data of the users; processing the original image data of all the users to obtain enhanced image data of the users; and combining the structured text data and the enhanced image data of the users to form a maternal-fetal cohort dataset. By integrating the health data provided by the different medical institutions that covers the pre-pregnancy, pregnancy and postpartum stages and represents the health conditions of the users and their fetuses, a comprehensive dataset can be formed for maternal-fetal health research.

Description

Methods, devices, terminals, and storage media for generating maternal-fetal population cohort datasets Technical Field

[0001] This invention relates to the field of medical data acquisition technology, and in particular to a method, apparatus, terminal and storage medium for generating maternal-fetal population cohort datasets. Background Technology

[0002] In current technologies, the data construction model for maternal-fetal health research is still imperfect. Data on pregnant women and fetuses during medical procedures are usually archived separately by different medical institutions. This fragmented data situation limits sample size and diversity. In addition, existing data systems often focus only on the health of a single aspect of the pregnant woman or fetus, lacking continuous records of pregnant women and fetuses before, during, and after pregnancy, making it difficult to form a comprehensive dataset for maternal-fetal health research.

[0003] Therefore, existing technologies have shortcomings and need to be improved and developed.

[0004] Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method, device, terminal and storage medium for generating maternal-fetal population cohort datasets, in order to address the above-mentioned deficiencies of the prior art, and to solve the problem that it is difficult to form a comprehensive dataset for maternal-fetal health research in the prior art.

[0006] The technical solution adopted by this invention to solve the technical problem is as follows:

[0007] In a first aspect, embodiments of the present invention provide a method for generating a maternal-fetal population cohort dataset, the method comprising:

[0008] Health data from multiple users obtained from different medical institutions, covering the user's pre-pregnancy, pregnancy and postpartum stages, including multiple raw text data and raw image data characterizing the health status of the user and / or their fetus;

[0009] The original text data of all users is processed to obtain the structured text data of each user;

[0010] The original image data of all users is processed to obtain the enhanced image data of each user;

[0011] The structured text data and the enhanced image data of each user are merged to form a maternal-fetal population cohort dataset.

[0012] In one implementation, the raw text data includes clinical text data, feature text data, and intervention text data; the processing of the raw text data of all users to obtain structured text data for each user includes:

[0013] Input the clinical text data, feature text data, and intervention text data of all users into the trained text cleaning model;

[0014] The clinical text data, feature text data, and intervention text data of all users are processed by the text data cleaning model to obtain physiological indicator data and clinical event description data of all users. Both the physiological indicator data and the clinical event description data contain user identifiers.

[0015] The physiological indicator data and clinical event description data of the same user are associated to obtain the structured text data of each user.

[0016] In one implementation, the text data cleaning model processes the clinical text data, feature text data, and intervention text data of all users to obtain physiological indicator data and clinical event description data of all users, including:

[0017] The BERT model in the text data cleaning model encodes the clinical text data, feature text data, and intervention text data of all users to obtain basic semantic information. The basic semantic information is then input into the bidirectional LSTM network in the text data cleaning model and processed by the bidirectional LSTM network to obtain contextual semantic features.

[0018] The contextual semantic features are input into the Transformer module in the text data cleaning model. The Transformer module performs self-attention calculation to obtain global semantic information. The global semantic information is then input into the multilayer perceptron in the text data cleaning model. The multilayer perceptron distinguishes between numbers and text to obtain numerical information and textual information.

[0019] The K-means clustering algorithm was used to process the digital information to obtain the physiological index data of all users;

[0020] The Transformer module and the multilayer perceptron are used to process the text information to obtain clinical event description data for all users.

[0021] In one implementation, the original image data is medical image data; the processing of the original image data of all users to obtain enhanced image data for each user includes:

[0022] Each of the aforementioned medical images undergoes an image integrity check to obtain the check results;

[0023] If the examination result is passed, the size of the medical image is adjusted to a preset size to obtain a second medical image;

[0024] All the second medical images are input into the trained image enhancement model, and the image enhancement model performs image enhancement processing to obtain multiple enhanced image data;

[0025] Each enhanced image data is associated with its corresponding user identifier to obtain the enhanced image data for each user.

[0026] In one implementation, performing an image integrity check on each of the medical images to obtain the check results includes:

[0027] The medical images were subjected to image damage checks and image content loss checks, respectively.

[0028] If the medical image passes the image damage check and image content missing check, the check result is determined to be "pass";

[0029] If the medical image fails either the image damage check or the image content missing check, the check result is determined to be failed.

[0030] In one embodiment, the step of inputting all the second medical images into a trained image enhancement model, and performing image enhancement processing through the image enhancement model to obtain multiple enhanced image data, including:

[0031] Input all the second medical images into the trained image enhancement model;

[0032] After the downsampling convolutional neural network module in the image enhancement model extracts features from all the second medical images, the extracted features are input into the enhancement layer for enhancement to obtain enhanced features. The enhanced features are then input into the upsampling convolutional neural network module in the image enhancement model.

[0033] The upsampling convolutional neural network module performs image restoration on the enhanced features to generate multiple enhanced image data.

[0034] In one implementation, after performing an image integrity check on each of the medical images and obtaining the check results, the method further includes:

[0035] If the test result is negative, the medical image with the negative test result is deleted.

[0036] Secondly, embodiments of the present invention also provide an apparatus for generating a maternal-fetal population cohort dataset, comprising:

[0037] The data acquisition module is used to acquire health data of multiple users from different medical institutions. The health data covers the user's pre-pregnancy, pregnancy and postpartum stages, including multiple raw text data and raw image data that characterize the health status of the user and / or their fetus.

[0038] The text data generation module is used to process the original text data of all users to obtain the structured text data of each user;

[0039] The image data generation module is used to process the original image data of all users to obtain enhanced image data for each user;

[0040] The dataset generation module is used to merge the structured text data and the enhanced image data of each user to form a maternal-fetal population cohort dataset.

[0041] Thirdly, embodiments of the present invention also provide a terminal, the terminal comprising: a memory, a processor, and a maternal-fetal population cohort dataset generation program stored in the memory and executable on the processor, wherein when the maternal-fetal population cohort dataset generation program is executed by the processor, it implements the steps of the maternal-fetal population cohort dataset generation method as described above.

[0042] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a maternal-fetal population cohort dataset generation program, which can be executed to implement the steps of the maternal-fetal population cohort dataset generation method as described above.

[0043] The beneficial effects of this invention are as follows: This invention acquires health data from multiple users across different medical institutions. This health data covers the users' pre-pregnancy, pregnancy, and postpartum stages, including multiple raw text and image data representing the health status of the user and / or their fetus. The raw text data of all users is processed to obtain structured text data for each user. The raw image data of all users is processed to obtain enhanced image data for each user. The structured text data and enhanced image data of each user are then merged to form a maternal-fetal population cohort dataset. This invention, by integrating health data from different medical institutions covering the user's pre-pregnancy, pregnancy, and postpartum stages and the health status of their fetus, can form a relatively comprehensive dataset for maternal-fetal health research. Attached Figure Description

[0044] Figure 1 is a flowchart of a preferred embodiment of the method for generating a maternal-fetal population cohort dataset in this invention.

[0045] Figure 2 is a schematic diagram of the original text data processing flow in this invention.

[0046] Figure 3 is a schematic diagram of the original image data processing flow in this invention.

[0047] Figure 4 is a schematic diagram of a preferred embodiment of the maternal-fetal population cohort dataset generation device of the present invention.

[0048] Figure 5 is a block diagram of the terminal principle of the present invention. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0050] In current technologies, the data construction model for maternal-fetal health research is still imperfect. Data on pregnant women and fetuses during medical procedures are usually archived separately by different medical institutions. This fragmented data situation limits sample size and diversity. In addition, existing data systems often focus only on the health of a single aspect of the pregnant woman or fetus, lacking continuous records of pregnant women and fetuses before, during, and after pregnancy, making it difficult to form a comprehensive dataset for maternal-fetal health research.

[0051] To address the aforementioned deficiencies in existing technologies, this invention provides a method, apparatus, terminal, and storage medium for generating a maternal-fetal population cohort dataset. The method includes: acquiring health data from multiple users across different medical institutions, the health data covering the users' pre-pregnancy, pregnancy, and postpartum stages, including multiple raw text and image data characterizing the health status of the user and / or their fetus; processing the raw text data of all users to obtain structured text data for each user; processing the raw image data of all users to obtain enhanced image data for each user; and merging the structured text data and enhanced image data of each user to form a maternal-fetal population cohort dataset. This invention, by integrating health data from different medical institutions covering the user's pre-pregnancy, pregnancy, and postpartum stages and the health status of their fetus, can form a relatively comprehensive dataset for maternal-fetal health research.

[0052] Please refer to Figure 1. The method for generating a maternal-fetal population cohort dataset according to an embodiment of the present invention includes the following steps:

[0053] Step S100: Obtain health data from multiple users from different medical institutions. The health data covers the user's pre-pregnancy, pregnancy and postpartum stages, including multiple raw text data and raw image data characterizing the health status of the user and / or their fetus.

[0054] Specifically, the user's health data comes from multiple different medical institutions. To cover the user's health status before pregnancy, during pregnancy, and postpartum, and to ensure the health of the user and / or their fetus, the user's consent was obtained before the data was acquired. Pre-pregnancy specifically refers to the 12 months prior to pregnancy, during which the user's own health data is collected, including both text and images. During pregnancy, the data includes the health data of both the user and the fetus. Postpartum specifically refers to the 12 months after delivery, including the health data of both the user and the fetus, in both text and images.

[0055] Please refer to Figure 1. The method for generating maternal-fetal population cohort datasets according to this embodiment of the invention further includes the following steps:

[0056] Step S200: Process the original text data of all users to obtain the structured text data of each user.

[0057] Specifically, the raw text data consists of textual representations of the user's own and / or the fetus's health status. Because the raw text data originates from different medical institutions and is diverse in source, it contains redundant information and inconsistent formats. Therefore, it needs to be processed to generate structured text data in a unified format for subsequent formation of a maternal-fetal population cohort dataset.

[0058] In one implementation, the raw text data includes clinical text data, feature text data, and intervention text data; the processing of the raw text data of all users to obtain structured text data for each user includes:

[0059] Input the clinical text data, feature text data, and intervention text data of all users into the trained text cleaning model;

[0060] The text data cleaning model processes all the clinical text data, the feature text data, and the intervention text data to obtain physiological indicator data and clinical event description data for all users. Both the physiological indicator data and the clinical event description data contain user identifiers.

[0061] The physiological indicator data and clinical event description data of the same user are associated and stored in a preset data format to obtain the structured text data of each user.

[0062] Specifically, clinical text data refers to clinical data obtained through examinations and diagnoses conducted by medical institutions during the maternal-fetal clinical stage, and is in text format. This includes medical records, diagnostic reports, treatment plans, etc. Feature text data reflects the user's personal characteristics and / or the fetus's personal characteristics. User personal characteristics include age, genetic diseases, hobbies, etc., while fetal personal characteristics include heart rate, fetal length, fetal weight, and types of birth defects. Types of birth defects include defects of the nervous system, digestive system, circulatory system, and genitourinary system. Intervention text data comprises data on drug therapy, nutritional therapy, exercise therapy, and other supportive treatments provided to the user and / or fetus through medical institutions.

[0063] In one implementation, the text data cleaning model processes the clinical text data, feature text data, and intervention text data of all users to obtain physiological indicator data and clinical event description data of all users, including:

[0064] The BERT model in the text data cleaning model encodes the clinical text data, feature text data, and intervention text data of all users to obtain basic semantic information. The basic semantic information is then input into the bidirectional LSTM network in the text data cleaning model and processed by the bidirectional LSTM network to obtain contextual semantic features.

[0065] The contextual semantic features are input into the Transformer module in the text data cleaning model. The Transformer module performs self-attention calculation to obtain global semantic information. The global semantic information is then input into the multilayer perceptron in the text data cleaning model. The multilayer perceptron distinguishes between numbers and text to obtain numerical information and textual information.

[0066] The K-means clustering algorithm was used to process the digital information to obtain the physiological index data of all users;

[0067] The Transformer module and the multilayer perceptron are used to process the text information to obtain clinical event description data for all users.

[0068] Specifically, the flowchart of the original text data processing in this invention is shown in Figure 2. When the original text data is input into the text data cleaning model, the pre-trained BERT model first extracts the basic semantics. The BERT model (Bidirectional Encoder Representations from Transformers) is a pre-trained deep bidirectional encoder model that can better understand the semantic information in the text data. Even when faced with complex medical terms and grammatical structures, it can accurately extract key information, providing a foundation for subsequent data cleaning. Then, the basic semantic information is input into the bidirectional LSTM network (i.e., bidirectional long short-term memory network) in the text data cleaning model. The bidirectional LSTM network performs semantic extraction and classification to obtain contextual semantic features. The bidirectional LSTM network can capture the contextual semantic relationships in the text data, such as identifying which words describe the same physiological indicator, thereby improving the accuracy of data extraction. The contextual semantic features are input into the Transformer module of the text data cleaning model. The Transformer module performs self-attention calculation to obtain global semantic information. This global semantic information is then input into the multilayer perceptron of the text data cleaning model. The multilayer perceptron distinguishes between numbers and text, obtaining numerical information and text information. Numerical information refers to data containing numbers, while text information refers to data containing strings. Using the K-means clustering algorithm to process the numerical information can effectively identify and remove outliers. For example, physiological indicator data that significantly exceeds the normal range can be identified as noise data and removed, thereby improving data reliability. The physiological indicator data in this invention includes blood pressure, blood glucose, white blood cell count, etc., derived from test results. Clinical event description data includes symptom descriptions, treatment processes, and diagnostic conclusions. Symptom descriptions include the patient's subjective feelings, such as "dry mouth" and "dizziness." Diagnostic conclusions are conclusions reached by doctors based on the patient's symptoms and examination results, such as "gestational hyperglycemia" and "gestational hypertension." After generating physiological indicator data and clinical event description data, the physiological indicator data and clinical event description data of the same user are associated and stored in a preset data format to obtain structured text data for each user.

[0069] Please refer to Figure 1. The method for generating maternal-fetal population cohort datasets according to this embodiment of the invention further includes the following steps:

[0070] Step S300: Process the original image data of all users to obtain enhanced image data for each user.

[0071] Specifically, the original image data comes from different medical institutions, and enhancing it can effectively improve the image quality so that a maternal-fetal cohort dataset can be formed to predict birth defects in the fetus.

[0072] In one implementation, the original image data is medical image data; the processing of the original image data of all users to obtain enhanced image data for each user includes:

[0073] Each of the aforementioned medical images undergoes an image integrity check to obtain the check results;

[0074] If the examination result is passed, the size of the medical image is adjusted to a preset size to obtain a second medical image;

[0075] All the second medical images are input into the trained image enhancement model, and the image enhancement model performs image enhancement processing to obtain multiple enhanced image data;

[0076] Each enhanced image data is associated with its corresponding user identifier to obtain the enhanced image data for each user.

[0077] Specifically, medical images include CT images and ultrasound images. A flowchart of the raw image data processing process is shown in Figure 3. After acquiring the raw image data, an image integrity check is performed. Only medical images that pass the integrity check are enhanced. The integrity check eliminates incomplete or damaged medical images, ensuring the quality of subsequent processing and analysis. When the check passes, the corresponding medical image is adjusted to a preset size to improve the efficiency of image enhancement processing. After adjusting the image size, it is input into a trained image enhancement model, which performs image enhancement processing to obtain multiple enhanced image data. Then, each user's enhanced image is associated with their user identifier for data management. The user identifier is a unique identifier for the user, which can be a national ID number. In this invention, the user refers to a pregnant woman, and the health data of the pregnant woman's fetus is also attributed to the pregnant woman. This setup aims to ensure that the health information of the pregnant woman and her fetus is tracked and recorded uniformly and continuously.

[0078] In one implementation, performing an image integrity check on each of the medical images to obtain the check result includes:

[0079] The medical images were subjected to image damage checks and image content loss checks, respectively.

[0080] If the medical image passes the image damage check and image content missing check, the check result is determined to be "pass";

[0081] If the medical image fails either the image damage check or the image content missing check, the check result is determined to be failed.

[0082] Specifically, image integrity checking includes image damage checking and image content missing checking. The image damage checking process includes: calculating the average pixel value and standard deviation of the medical image; if the average pixel value exceeds a first preset value or falls below a second preset value, the image damage checking result is confirmed as failing; if the standard deviation exceeds a third preset value, the image damage checking result is confirmed as failing. The image content missing checking process includes: performing edge detection on the medical image to identify pixels located at the image edges; calculating the proportion of non-zero pixels among the pixels at the image edges; if the proportion of non-zero pixels among the pixels at the image edges is lower than a first preset threshold, the image content missing checking result is determined to be failing. Edge detection can be performed using the Canny Edge Detection algorithm. The first preset threshold ranges from 5% to 10%.

[0083] In one implementation, the step of inputting all the second medical images into a trained image enhancement model, and performing image enhancement processing through the image enhancement model to obtain multiple enhanced image data, including:

[0084] Input all the second medical images into the trained image enhancement model;

[0085] After the downsampling convolutional neural network module in the image enhancement model extracts features from all the second medical images, the extracted features are input into the enhancement layer for enhancement to obtain enhanced features. The enhanced features are then input into the upsampling convolutional neural network module in the image enhancement model.

[0086] The upsampling convolutional neural network module performs image restoration on the enhanced features to generate multiple enhanced image data.

[0087] Specifically, the downsampling convolutional neural network module (i.e., downsampling CNN) employs a three-layer convolutional neural network (CNN) structure. Each convolutional layer is followed by a ReLU (Rectified Linear Unit) activation function and a batch normalization layer to progressively compress image dimensions and extract features. The enhancement layer contains three convolutional neural network layers that do not change the spatial dimensions of the image, used to enhance features. The upsampling convolutional neural network module (i.e., upsampling CNN) includes three deconvolutional layers, each followed by a ReLU activation function and a batch normalization layer, used to progressively restore the image to its original dimensions. During the upsampling process, by performing residual connections with the feature maps of the corresponding layers in the downsampling stage, the upsampling convolutional neural network module can fuse feature information at different scales, thereby generating a more refined and accurate enhanced image.

[0088] In one implementation, after performing an image integrity check on each of the medical images and obtaining the check results, the method further includes:

[0089] If the test result is negative, the medical image with the negative test result is deleted.

[0090] Specifically, when the inspection result is unsatisfactory, the corresponding medical image is deleted, thereby improving the efficiency of data management.

[0091] Please refer to Figure 1. The method for generating maternal-fetal population cohort datasets according to this embodiment of the invention further includes the following steps:

[0092] Step S400: Merge the structured text data and the enhanced image data of each user to form a maternal-fetal population cohort dataset.

[0093] Specifically, the structured text data and the enhanced image data of each user are combined to form a maternal-fetal population cohort dataset.

[0094] In summary, this invention addresses the problems of heterogeneity, incompleteness, asymmetry, and limited coverage in existing maternal-fetal health data by proposing a novel method for constructing a high-quality maternal-fetal population cohort dataset. Considering that birth defects are influenced by multiple factors such as maternal illness, genetics, nutrition, exercise, and sleep, this invention integrates maternal and fetal health data from different medical institutions, covering the entire pre-pregnancy, pregnancy, and postpartum stages, to construct a comprehensive maternal-fetal population cohort dataset. During the construction process, text data is cleaned, and image data is enhanced. Combining text and image data forms the maternal-fetal population cohort dataset, which provides strong data support for in-depth research into the causes of birth defects and for more accurate prediction and tracing.

[0095] In one embodiment, as shown in FIG4, based on the above-described method for generating maternal-fetal population cohort datasets, the present invention also provides a corresponding apparatus for generating maternal-fetal population cohort datasets, comprising:

[0096] The data acquisition module 100 is used to acquire health data of multiple users from different medical institutions. The health data covers the user's pre-pregnancy, pregnancy and postpartum stages, including multiple raw text data and raw image data that characterize the health status of the user and / or their fetus.

[0097] The text data generation module 200 is used to process the original text data of all users to obtain the structured text data of each user;

[0098] The image data generation module 300 is used to process the original image data of all users to obtain enhanced image data for each user;

[0099] The dataset generation module 400 is used to merge the structured text data and the enhanced image data of each user to form a maternal-fetal population cohort dataset.

[0100] In one implementation, the raw text data includes clinical text data, feature text data, and intervention text data; the text data generation module includes:

[0101] The data input unit is used to input all the clinical text data, the feature text data and the intervention text data into the trained text data cleaning model;

[0102] The data processing unit is used to process the clinical text data, feature text data and intervention text data of all users through the text data cleaning model to obtain physiological indicator data and clinical event description data of all users. The physiological indicator data and clinical event description data both contain user identifiers.

[0103] The first association unit is used to associate the physiological indicator data and the clinical event description data of the same user identifier to obtain the structured text data of each user.

[0104] In one implementation, the data processing unit includes:

[0105] The context semantic feature generation unit is used to encode the clinical text data, feature text data and intervention text data of all users through the BERT model in the text data cleaning model to obtain basic semantic information, and input the basic semantic information into the bidirectional LSTM network in the text data cleaning model. After processing by the bidirectional LSTM network, the context semantic features are obtained.

[0106] The digital information and text information generation unit is used to input the contextual semantic features into the Transformer module in the text data cleaning model, perform self-attention calculation through the Transformer module to obtain global semantic information, input the global semantic information into the multilayer perceptron in the text data cleaning model, and distinguish between digital and text through the multilayer perceptron to obtain digital information and text information.

[0107] The physiological indicator data generation unit is used to process the digital information using the K-means clustering algorithm to obtain the physiological indicator data of all users;

[0108] The clinical event description data generation unit is used to process the text information using the Transformer module and the multilayer perceptron to obtain clinical event description data for all users.

[0109] In one implementation, the original image data is a medical image; the enhanced image data generation unit includes:

[0110] The inspection result generation unit is used to perform an image integrity check on each of the medical images and obtain the inspection result.

[0111] The second medical image generation unit is used to adjust the size of the medical image to a preset size if the examination result is passed, so as to obtain a second medical image.

[0112] An enhanced image data generation unit is used to input all the second medical images into a trained image enhancement model, and perform image enhancement processing through the image enhancement model to obtain multiple enhanced image data.

[0113] The second association unit is used to associate each of the enhanced image data with the corresponding user identifier to obtain the enhanced image data of each user.

[0114] In one implementation, the inspection result generation unit includes:

[0115] The inspection unit is used to perform image damage inspection and image content missing inspection on the medical images respectively;

[0116] The first inspection result determination unit is used to determine that the inspection result is passed if the medical image passes the image damage inspection and the image content missing inspection.

[0117] The second inspection result determination unit is used to determine that the inspection result is failed if the medical image fails either the image damage inspection or the image content missing inspection.

[0118] In one implementation, the enhanced image data generation unit includes:

[0119] An image input unit is used to input all the second medical images into the trained image enhancement model;

[0120] The enhanced feature generation unit is used to extract features from all the second medical images through the downsampling convolutional neural network module in the image enhancement model, input the extracted features into the enhancement layer for enhancement, and then input the enhanced features into the upsampling convolutional neural network module in the image enhancement model.

[0121] The image enhancement unit is used to restore the image by the upsampling convolutional neural network module to generate multiple enhanced image data.

[0122] In one implementation, the enhanced image data generation unit further includes:

[0123] An image deletion unit is used to delete medical images whose inspection results are not passed if the inspection result is not passed.

[0124] Based on the above embodiments, the present invention also provides a terminal, the principle block diagram of which is shown in Figure 5. The terminal includes a processor, a memory, a network interface, and a display screen connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and a maternal-to-fetal population cohort dataset generation device. The internal memory provides an environment for the operation of the operating system and the maternal-to-fetal population cohort dataset generation device in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When executed by the processor, the maternal-to-fetal population cohort dataset generation device implements the steps of any of the above-described maternal-to-fetal population cohort dataset generation methods. The display screen of the terminal can be a liquid crystal display (LCD) or an electronic ink display.

[0125] Those skilled in the art will understand that the principle block diagram shown in Figure 5 is only a block diagram of a part of the structure related to the present invention and does not constitute a limitation on the terminal to which the present invention is applied. The specific terminal may include more or fewer components than shown in the figure, or combine some components, or have different component arrangements.

[0126] In one embodiment, a terminal is provided, the terminal including a memory, a processor, and a maternal-fetal population cohort dataset generation program stored in the memory and executable on the processor. When the maternal-fetal population cohort dataset generation program is executed by the processor, it implements the steps of any of the maternal-fetal population cohort dataset generation methods provided in the embodiments of the present invention.

[0127] This invention also provides a computer-readable storage medium storing a maternal-fetal population cohort dataset generation program. When the maternal-fetal population cohort dataset generation program is executed by a processor, it implements the steps of any of the maternal-fetal population cohort dataset generation methods provided in this invention.

[0128] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0129] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the above device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this invention. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0130] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0131] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0132] In the embodiments provided by this invention, it should be understood that the disclosed apparatus / terminal devices and methods can be implemented in other ways. For example, the apparatus / terminal device embodiments described above are merely illustrative. For instance, the division of modules or units described above is only a logical functional division, and in actual implementation, it can be divided in other ways. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0133] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not mean that the essence of the corresponding technical solutions deviates from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for generating a maternal-fetal population cohort dataset, characterized in that, The method comprises: obtaining health data of a plurality of users from different medical institutions, the health data covering the pre-pregnancy, pregnancy and postpartum stages of the users, including a plurality of original text data and original image data characterizing the health status of the users and / or their fetuses; processing the original text data of all users to obtain structured text data of each user; processing the original image data of all users to obtain enhanced image data of each user; merging the structured text data and the enhanced image data of each user to form a maternal-fetal population cohort data set.

2. The maternal population cohort dataset generation method of claim 1, wherein, The original text data includes clinical text data, feature text data and intervention text data; the processing of the original text data of all users to obtain structured text data of each user comprises: inputting the clinical text data, feature text data and intervention text data of all users into a trained text data cleaning model; processing the clinical text data, feature text data and intervention text data of all users through the text data cleaning model to obtain physiological index data and clinical event description data of all users, wherein the physiological index data and the clinical event description data both contain user identifiers; associating the physiological index data and the clinical event description data of the same user identifier to obtain structured text data of each user.

3. The maternal population cohort dataset generation method of claim 2, wherein, The processing of the clinical text data, feature text data and intervention text data of all users through the text data cleaning model to obtain physiological index data and clinical event description data of all users comprises: encoding the clinical text data, feature text data and intervention text data of all users through a BERT model in the text data cleaning model to obtain basic semantic information, inputting the basic semantic information into a bidirectional LSTM network in the text data cleaning model, processing through the bidirectional LSTM network to obtain context semantic features; inputting the context semantic features into a Transformer module in the text data cleaning model, performing self-attention calculation through the Transformer module to obtain global semantic information, inputting the global semantic information into a multi-layer perceptron in the text data cleaning model, performing differentiation between numbers and text through the multi-layer perceptron to obtain numerical information and text information; processing the numerical information through a K-means clustering algorithm to obtain physiological index data of all users; processing the text information through the Transformer module and the multi-layer perceptron to obtain clinical event description data of all users.

4. The maternal population cohort dataset generation method of claim 1, wherein, The original image data is medical images; the processing of the original image data of all users to obtain enhanced image data of each user comprises: performing image integrity checking on each medical image to obtain a checking result; if the checking result is passed, adjusting the size of the medical image to a preset size to obtain a second medical image; inputting all the second medical images into a trained image enhancement model, performing image enhancement processing on the image enhancement model to obtain a plurality of enhanced image data; associating each of the enhanced image data with a corresponding user identifier to obtain enhanced image data for each user.

5. The maternal population cohort dataset generation method of claim 4, wherein, The image integrity check on each of the medical images obtains a check result, including: respectively performing image damage inspection and image content missing inspection on the medical images; if the medical images pass the image damage inspection and the image content missing inspection, determining that the check result is passed; if the medical images fail any of the image damage inspection or the image content missing inspection, determining that the check result is failed.

6. The maternal population cohort dataset generation method of claim 4, wherein, The inputting all the second medical images into a trained image enhancement model, performing image enhancement processing on the image enhancement model to obtain a plurality of enhanced image data, includes: inputting all the second medical images into a trained image enhancement model; extracting features from all the second medical images through a down-sampling convolutional neural network module in the image enhancement model, inputting the extracted features into an enhancement layer for enhancement, and inputting the enhanced features into an up-sampling convolutional neural network module in the image enhancement model; performing image restoration on the enhanced features through the up-sampling convolutional neural network module to generate a plurality of enhanced image data.

7. The maternal population cohort dataset generation method of claim 4, wherein, After the image integrity check on each of the medical images obtains a check result, it further includes: if the check result is failed, deleting the medical image whose check result is failed.

8. A maternal-fetal human cohort dataset generation apparatus, characterized by, It includes: a data acquisition module configured to acquire health data of a plurality of users from different medical institutions, the health data covering pre-pregnancy, pregnancy and postpartum stages of the users, including a plurality of original text data and original image data representing health conditions of the users and / or their fetuses; a text data generation module configured to process the original text data of all users to obtain structured text data for each user; an image data generation module configured to process the original image data of all users to obtain enhanced image data for each user; a data set generation module configured to merge the structured text data and the enhanced image data of each user to form a maternal-fetal population cohort data set.

9. A terminal, characterized by comprising: The terminal includes a memory, a processor, and a maternal-fetal population cohort data set generation program stored in the memory and executable on the processor. When the processor executes the maternal-fetal population cohort data set generation program, it implements the steps of the maternal-fetal population cohort data set generation method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a maternal-fetal population cohort data set generation program that can be executed to implement the steps of the maternal-fetal population cohort data set generation method according to any one of claims 1-7.

Images