Fetal health assessment method and device based on detection data, terminal and medium
Through the cleaning of maternal and fetal detection data and the analysis of health relationship models, combined with the evaluation network, the problem of low accuracy of fetal premature birth risk assessment in the prior art is solved, and a more accurate fetal health assessment is achieved, including premature birth risk and birth defect prediction.
Patent Information
- Application Number
- CN202410992267.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-23
- Publication Date
- 2024-12-06
AI Technical Summary
The existing machine learning methods have low prediction accuracy in fetal premature birth risk assessment, and traditional methods are highly invasive and prone to infection and are affected by subjective judgment.
By obtaining maternal and fetal detection data for data cleaning, inputting health relationship models and evaluation networks, obtaining fetal health assessment information, including premature birth risk and birth defect prediction, using a circular structure to capture timing information, combining clustering and self-attention mechanisms to process data, perform Multiple correlation analysis.
It improves the accuracy of fetal health assessment results, combines analysis of premature birth and birth defects, learns multiple relationships, and reduces prediction errors.
Smart Images

Figure CN119092108A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of medical data processing technology, and in particular to a fetal health assessment method, device, terminal and medium based on detection data. Background Art
[0002] How to assess and predict the risk of premature birth has always been an important problem in the academic community. Domestic and foreign scholars have conducted a lot of research on the risk assessment and prediction of premature birth. Research using traditional methods includes: predicting premature birth based on ultrasound elastography, delivery pressure gauge, and intrauterine pressure gauge. These prediction methods are generally invasive, prone to infection, and the measured values are easily affected by subjective judgment. Therefore, predicting premature birth through machine learning methods has become a research hotspot. However, existing machine learning methods are limited to exploring the relationship between test data and premature birth, and the prediction accuracy is low.
[0003] Therefore, the existing technology still needs to be improved and developed. Summary of the invention
[0004] The technical problem to be solved by the present invention is to provide a fetal health assessment method, device, terminal and medium based on test data in response to the above-mentioned defects of the prior art, aiming to solve the problem that the machine learning method in the prior art is limited to exploring the relationship between test data and premature birth and has low prediction accuracy.
[0005] The technical solution adopted by the present invention to solve the problem is as follows:
[0006] In a first aspect, an embodiment of the present invention provides a method for fetal health assessment based on detection data, the method comprising:
[0007] Acquiring maternal and fetal detection data, and performing data cleaning on the maternal and fetal detection data;
[0008] Inputting the cleaned maternal-fetal detection data into a health relationship model to obtain perception information;
[0009] The cleaned maternal-fetal detection data and the perception information are input into an evaluation network to obtain fetal health assessment information, wherein the fetal health assessment information includes fetal premature birth risk and birth defect prediction information.
[0010] In one embodiment, the data cleaning method of the maternal-fetal detection data includes:
[0011] Performing entity recognition on each of the maternal and fetal detection data to obtain standard data fields, and generating a standardized data table based on all the standard data fields;
[0012] According to the standardized data table, duplicate data is removed:
[0013] After removing duplicate data, determine whether there are outliers in the remaining data. If so, process the outliers, which include outliers and / or missing values.
[0014] In one embodiment, removing duplicate data according to the standardized data table includes:
[0015] Perform data clustering according to the standardized data table to obtain several types of data;
[0016] For each category of data, calculate the similarity of the data in the category, and determine whether there is duplicate data based on the similarity. If so, remove the duplicate data;
[0017] After elimination, the search radius is adjusted, and the step of clustering the data according to the standardized data table is continued until there is no duplicate data in each category of data.
[0018] In one embodiment, the healthy relationship model is used to:
[0019] Capturing time series information according to the input maternal-fetal detection data through a loop structure, and obtaining contextual semantic information according to the time series information, wherein the loop structure retains a hidden state at each time step, each hidden state is determined based on the input of the current time step and the hidden state of the previous time step, and each time step corresponds to an element of the maternal-fetal detection data;
[0020] The perception information is obtained according to the contextual semantic information and the maternal-fetal detection data.
[0021] In one embodiment, the evaluation network includes a fetal disease classification model and a risk prediction model, and the cleaned maternal-fetal detection data and the perception information are input into the evaluation network to obtain fetal health assessment information, including:
[0022] Inputting the sensed information into the fetal disease classification model to obtain fetal birth defect prediction information;
[0023] The maternal and fetal detection data and the birth defect prediction information are input into the risk prediction model to obtain the risk of premature birth of the fetus.
[0024] In one embodiment, the fetal disease classification model is used to:
[0025] Taking the input data as samples to be predicted, and measuring the distance between the samples to be predicted and several samples of known categories respectively;
[0026] Filtering out a number of the nearest samples of the known category according to the distance measurement result, and determining the category information of the sample to be predicted according to all the nearest samples of the known category;
[0027] The birth defect prediction information is generated according to the category information of the prediction sample.
[0028] In one embodiment, the risk prediction model is used to:
[0029] Convert the input data into matrix form through the time series matrix layer;
[0030] Extracting phenotypes from the temporal matrix layer through a unilateral convolutional layer;
[0031] The important phenotypes are screened out from the extracted phenotypes through the maximum pooling layer;
[0032] The risk of premature birth is predicted based on the important phenotypes through a fully connected softmax prediction layer.
[0033] In a second aspect, an embodiment of the present invention further provides a fetal health assessment device based on detection data, the device comprising:
[0034] A cleaning module, used for acquiring maternal and fetal detection data and performing data cleaning on the maternal and fetal detection data;
[0035] A perception module, used for inputting the cleaned maternal and fetal detection data into a health relationship model to obtain perception information;
[0036] An evaluation module is used to input the cleaned maternal-fetal detection data and the perception information into an evaluation network to obtain fetal health evaluation information, wherein the fetal health evaluation information includes fetal premature birth risk and birth defect prediction information.
[0037] In a third aspect, an embodiment of the present invention further provides a terminal comprising a memory and one or more processors; the memory stores one or more programs; the program comprises instructions for executing any of the above-described fetal health assessment methods based on detection data; and the processor is used to execute the program.
[0038] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium on which a plurality of instructions are stored, wherein the instructions are suitable for being loaded and executed by a processor to implement any of the steps of the above-mentioned fetal health assessment method based on detection data.
[0039] Beneficial effects of the present invention: The embodiments of the present invention obtain maternal and fetal test data, clean the maternal and fetal test data; input the cleaned maternal and fetal test data into the health relationship model to obtain perception information; input the cleaned maternal and fetal test data and perception information into the evaluation network to obtain fetal health assessment information, wherein the fetal health assessment information includes the risk of premature birth and birth defect prediction information of the fetus. The present invention conducts a joint analysis of fetal premature birth and birth defects, and can learn the multiple correlations between maternal and fetal test data, fetal premature birth, and birth defects, thereby improving the accuracy of fetal health assessment results. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0041] Figure 1 It is a flowchart of a method for assessing fetal health based on detection data provided in an embodiment of the present invention.
[0042] Figure 2 It is a schematic diagram of the data cleaning process provided by an embodiment of the present invention.
[0043] Figure 3 It is a schematic diagram of the training process of the interpolation module provided in an embodiment of the present invention.
[0044] Figure 4 It is a schematic diagram of the relationship between the health relationship model and the evaluation network provided by the embodiment of the present invention.
[0045] Figure 5 It is a schematic diagram of the internal modules of the evaluation network provided by an embodiment of the present invention.
[0046] Figure 6 It is a schematic diagram of the structure of the risk prediction model provided by an embodiment of the present invention.
[0047] Figure 7 It is a module schematic diagram of a fetal health assessment device based on detection data provided by an embodiment of the present invention.
[0048] Figure 8 It is a principle block diagram of a terminal provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0049] The present invention discloses a fetal health assessment method, device, terminal and medium based on detection data. In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0050] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.
[0051] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as herein.
[0052] In view of the above-mentioned defects of the prior art, the present invention provides a fetal health assessment method based on test data, wherein the method obtains maternal and fetal test data, cleans the maternal and fetal test data; inputs the cleaned maternal and fetal test data into a health relationship model to obtain perception information; inputs the cleaned maternal and fetal test data and the perception information into an evaluation network to obtain fetal health assessment information, wherein the fetal health assessment information includes fetal premature birth risk and birth defect prediction information. The present invention conducts a joint analysis of fetal premature birth and birth defects, and can learn the multiple correlations between maternal and fetal test data, fetal premature birth, and birth defects, thereby improving the accuracy of fetal health assessment results.
[0053] like Figure 1 As shown, the method includes:
[0054] Step S100: Acquire maternal and fetal detection data, and perform data cleaning on the maternal and fetal detection data.
[0055] Specifically, maternal and fetal data refers to pregnant women and fetuses, and maternal and fetal test data refers to structured data recorded in the medical system, which is used to reflect the health information of pregnant women and fetuses, and can include multiple data types such as numerical values, text, and symbols. Since maternal and fetal test data may come from different testing equipment / hospitals, and the data obtained by different equipment / hospitals are complicated and messy, it is necessary to organize, process, proofread, and standardize the maternal and fetal test data, so as to automatically map it to a standard, standardized, and complete data space, and obtain more real and effective maternal and fetal test data, thereby improving the accuracy of subsequent predictions.
[0056] In one implementation, the data cleaning method of the maternal-fetal detection data includes:
[0057] Performing entity recognition on each of the maternal and fetal detection data to obtain standard data fields, and generating a standardized data table based on all the standard data fields;
[0058] According to the standardized data table, duplicate data is removed:
[0059] After removing duplicate data, determine whether there are outliers in the remaining data. If so, process the outliers, which include outliers and / or missing values.
[0060] Specifically, maternal and fetal test data is structured data expressed in data tables. The cleaning process of this type of structured data is mainly divided into two steps: data integration and outlier processing. Among them, data integration mainly includes three steps: field extraction, data table normalization, and duplicate elimination (such as Figure 2 As shown in the figure, outlier processing mainly includes two steps: outlier processing and missing value processing. For field extraction: Since the data expressions of different data sources may be different and mismatched, it is necessary to consider the problems of entity recognition and attribute redundancy. The purpose of entity recognition is to select the corresponding fields and merge their corresponding data to obtain the data fields required by the standard structured data table. For data table normalization: The process of forming a standardized data table by extracting fields from data tables of different sources and forms. The standardized data table is a pre-built standardized data table. Its construction method is determined based on the fetal health prediction task / premature birth and birth defect risk prediction task, and the algorithm efficiency is considered at the same time. For duplicate elimination: After field screening extraction and multi-table normalization integration, duplicate data needs to be eliminated. For outlier processing: The outliers of the integrated data set include outliers and missing values. The generation of outliers is mainly due to errors in information storage and conversion, or the emergence of certain difficult and complicated diseases that cause the data to deviate from normal values. The processing of outliers mainly adopts the method of correction or elimination. Missing data may be caused by human factors or system detection factors. The processing of missing values is mainly done by deletion or interpolation.
[0061] In one implementation, during entity recognition, fields with the same or different names but different meanings can be screened through prior knowledge and / or correlation analysis; and for multiple fields with different units, each field is normalized to a consistent value and unit range.
[0062] In one implementation, removing duplicate data according to the standardized data table includes:
[0063] Perform data clustering according to the standardized data table to obtain several types of data;
[0064] For each category of data, calculate the similarity of the data in the category, and determine whether there is duplicate data based on the similarity. If so, remove the duplicate data;
[0065] After elimination, the search radius is adjusted, and the step of clustering the data according to the standardized data table is continued until there is no duplicate data in each category of data.
[0066] Specifically, the elimination of duplicate data introduces clustering algorithm processing. First, the standardized data table is clustered, and then the similarity of data in the same class is calculated to determine whether the data is duplicated. If there is duplicate data, only one of the data is retained and the other duplicate data is eliminated. Then the search radius is adjusted to modify the density parameters of each type of data generated by clustering. The above steps are repeated iteratively until each type of data after clustering is approximate data, and the detection and elimination of duplicate data is completed.
[0067] In one implementation, the outlier detection method includes:
[0068] Clustering the remaining data after removing duplicate data is performed using a preset clustering algorithm to obtain several data clusters;
[0069] Determining whether there is a global outlier based on the distance information of each of the data clusters;
[0070] Whether there is a local outlier is determined by the density information of each of the data clusters.
[0071] Specifically, outliers can be divided into global outliers and local outliers. Global outliers are detected by distance-based methods, and local outliers are detected by density-based methods. Distance-based and density-based outlier detection methods require detecting the Euclidean distance and outlier factor of each object relative to other objects, which will bring unimaginable time consumption when the data scale increases. Clustering algorithms can cluster objects with similar characteristics to form multiple data clusters of different sizes through unsupervised learning. This embodiment combines clustering algorithms, distance-based outlier detection, and density-based outlier detection to find outliers, so as to reduce the scale of the problem and reduce the time complexity of the algorithm.
[0072] In one implementation, the method for processing missing values includes:
[0073] The data corresponding to the missing values are used as the data to be interpolated;
[0074] Iteratively sample the interpolation data to obtain a posterior distribution;
[0075] The missing values are filled by drawing values from the posterior distribution.
[0076] Specifically, maternal and fetal test data are mainly numerical data, so interpolation is used to process missing data. Missing values can be numerical, categorical, or count data. This embodiment uses a multivariate regression model in an iterative manner to interpolate missing values. The data corresponding to the missing values are used as the data to be interpolated. First, the data to be interpolated is iteratively sampled to obtain a posterior distribution, and values are extracted from the posterior distribution for filling. The interpolation method provided in this embodiment has a faster convergence speed and does not require the data set to be normalized to a normal distribution.
[0077] In another implementation, the maternal-fetal detection data is time series data, and the method for processing missing values includes:
[0078] Use time series data with missing values as data to be interpolated;
[0079] The missing values are interpolated using a self-attention mechanism through the time step correlation in the time series data, wherein the time step correlation is the correlation between observations at different time points within each variable.
[0080] Furthermore, in this embodiment, an interpolation module is pre-constructed based on the self-attention mechanism to perform the above-mentioned step of interpolating missing values by using the self-attention mechanism through the time step correlation in the time series data. The training method of the interpolation module includes:
[0081] Time align the input time series samples; (the time series can be an irregular time series);
[0082] Randomly remove a preset proportion of observations to obtain the indicator mask I that records the position of the removed values, the missing mask M that records the missing pattern, and the time series data obtained after removing some observations.
[0083] The time series data obtained after removing some observations Together with the missing mask M, it is input into the interpolation module to obtain the time series data after interpolation.
[0084] Based on time series data The loss value of the interpolation module is calculated using the indication mask I, and the parameters of the interpolation module are adjusted according to the loss value to achieve training of the interpolation module.
[0085] Specifically, Figure 3 As shown in Figure 2, the interpolation module based on the self-attention mechanism can use the time step correlation of each variable in the time series data to perform time series interpolation. Together with the missing mask M, it is input into the interpolation module, which enables the model to capture the distribution of observations to a certain extent and helps the model understand the time series.
[0086] In one implementation, since the lack of data itself can also provide information, this embodiment can input the time series of maternal and fetal test data without interpolation and the time series of maternal and fetal test data obtained after interpolation by the interpolation module into the health relationship model to obtain the final classification result, that is, the health status of the fetus. In addition, during training, the interpolation module and the subsequent health relationship model can be jointly optimized and trained together, and the multi-task joint optimization training method can be used to help the health relationship model better understand the time series, thereby obtaining better classification performance.
[0087] Step S200: input the cleaned maternal-fetal detection data into a health relationship model to obtain perception information.
[0088] Specifically, the perception information in this embodiment may be the fetal health status obtained by analyzing the currently input maternal and fetal test data by the health relationship model according to the pre-learned relationship between different test data and fetal health.
[0089] In one implementation, before inputting the cleaned maternal-fetal detection data into the health relationship model, the process further includes:
[0090] Extract features from the cleaned maternal and fetal test data, and perform relationship modeling based on the extracted feature data to obtain a relationship map;
[0091] Input the relationship map into the healthy relationship model.
[0092] Specifically, effectively utilizing the data features at different time nodes in maternal and fetal test data or the potential dependencies between different variable data at the same time node is crucial for the subsequent perception and prediction process. In order to effectively utilize the original data, it is necessary to first extract the features of the maternal and fetal test data, and then model the potential relationship between the extracted feature data. Through relationship modeling, the model can more accurately learn the potential relationship between variables, thereby enhancing the model's ability to analyze the features of maternal and fetal test data.
[0093] In one implementation, relationship modeling is performed based on the extracted feature data to obtain a relationship graph, including:
[0094] Generate a data node according to each feature data;
[0095] According to the potential time-varying dependencies between the feature data, the edges between the data nodes are generated, and the corresponding edge weights are generated according to the strength of the potential time-varying dependencies;
[0096] The graph structure constructed based on all data nodes, edges, and edge weights is used as the relationship graph.
[0097] Specifically, this embodiment can pre-set an adaptive graph structure data preprocessing module to generate a relationship map of maternal and fetal test data to achieve modeling of the potential relationship between variables in the maternal and fetal test data: first, assume a potential graph structure, take each feature data in the maternal and fetal test data as a pregnant woman data type node in the potential graph, and abstract the potential time-varying dependencies between different data nodes as edges between nodes in the potential graph, and the strength of the dependencies is abstracted as the weight of the edges. Finally, the graph structure constructed according to the above process is used as a relationship map to enhance the model's ability to analyze features of temporal and irregular maternal and fetal test data.
[0098] In one implementation, the healthy relationship model is used to:
[0099] Capturing time series information according to the input maternal-fetal detection data through a loop structure, and obtaining contextual semantic information according to the time series information, wherein the loop structure retains a hidden state at each time step, each hidden state is determined based on the input of the current time step and the hidden state of the previous time step, and each time step corresponds to an element of the maternal-fetal detection data;
[0100] The perception information is obtained according to the contextual semantic information and the maternal-fetal detection data.
[0101] Specifically, the maternal and fetal test data is time series data, and the health relationship model is constructed using a time series deep neural network, such as LSTM, GRU, etc. This embodiment uses a large amount of historical maternal and fetal test data collected for network learning to obtain a health relationship model, which implies the data relationship between fetal premature birth and birth defects and the health of pregnant women and fetuses. When the cleaned maternal and fetal test data is input, the quantitative risk of fetal premature birth and birth defects can be obtained. In actual application, the data is sent to the time series deep neural network. Since personal physical differences will affect the final result judgment, the contextual semantic information of different targets needs to be optimized. The length of maternal and fetal test data of different users may be different. This embodiment uses a loop structure to process the variable length sequence of historical measurements, and then merges the corresponding contextual semantic information with the sequence processed by the loop structure. The loop structure includes a hidden layer, which is expanded on the time step to form a chain structure. Each time step corresponds to an element of the sequence of input maternal and fetal test data. Therefore, the loop structure can process input sequences of any length and retain a hidden state at each time step to keep the memory of previous information.
[0102] In one implementation, the hidden layer processes the context data and extracts features of the context data through additional layers, assisted by activation functions to avoid data missing or gradient vanishing problems.
[0103] In one implementation, the health relationship model constructed using a time series deep neural network also includes a fully connected layer, which is used to: compress the results (such as maternal-fetal detection data / contextual information of maternal-fetal detection data) through the fully connected layer to obtain processed contextual semantic weights, fuse them with the sequence data after processing the source data to obtain the final joint representation data, and obtain perception information based on the joint representation data.
[0104] In one implementation, the health relationship model constructed using a time series deep neural network also uses an attention mechanism, which includes:
[0105] According to the maternal and fetal test data, each numerical value (for example, the test value of each test item) is predicted, the memory layer of the predicted data at the previous moment is spliced with the memory layer of the current moment, and then the Hadamard product is performed with the probability value obtained by the Softmax function of the source data at the same moment to obtain the attention weight coefficient (Attention). The weight / importance of each test item can be known through the attention weight coefficient.
[0106] In one implementation, this embodiment trains the health relationship model based on different massive data samples to improve the accuracy of the above-mentioned weight coefficients.
[0107] In one implementation, the prediction samples used in the training of the health relationship model are historically collected maternal and fetal test data, and the labeled data are corresponding maternal and fetal feature data.
[0108] Specifically, the maternal and fetal test data and maternal and fetal characteristic data used in the training are structured data, and are similarly stored in the designated data system in the form of a data table after being processed according to the above-mentioned data cleaning process. Among them, the maternal and fetal characteristic data is the unique data that describes the maternal and fetal characteristics, and records the information such as the pregnant woman, place of origin, nationality, occupation, and the fetal gender, number of days in the uterus, birth defects, etc. The characteristics of premature birth and birth defects are both fetal characteristic information and the disease risk caused by maternal and fetal health factors. Therefore, this embodiment uses it as the result of the model. The maternal and fetal test data belongs to a multi-recorded data sequence, which records the disease test data of the pregnant woman before and after pregnancy, and also records the disease data of the fetus in the womb and after birth. Therefore, the relationship between the characteristic data and the test data belongs to the relationship between a pair of maternal and fetal characteristic data and multiple test records. In this embodiment, the network parameters are adjusted through the maternal and fetal feature vector during model training, and then the maternal and fetal health test information machine learning based on the time series is realized, and the relationship model between the fetal premature birth and birth defects and the maternal and fetal test data perception is obtained, and finally the fetal premature birth and birth defect perception is realized.
[0109] Step S300: input the cleaned maternal-fetal detection data and the perception information into an evaluation network to obtain fetal health assessment information, wherein the fetal health assessment information includes fetal premature birth risk and birth defect prediction information.
[0110] Specifically, Figure 4 As shown, this embodiment pre-trains a large number of maternal and fetal test data samples to perform network learning to obtain a health relationship model and an evaluation network, which implies the data relationship between fetal premature birth and birth defects and the health of pregnant women and fetuses. Through the trained health relationship model and evaluation network, combined with the cleaned maternal and fetal test data, the quantitative risk of fetal premature birth and birth defects can be obtained. The prediction of birth defects can include the prediction of defects in the nervous system, digestive system, circulatory system, urogenital system, congenital heart disease, and head organs.
[0111] In one implementation, the evaluation network includes a fetal disease classification model and a risk prediction model, and the cleaned maternal-fetal detection data and the perception information are input into the evaluation network to obtain fetal health evaluation information, including:
[0112] Inputting the sensed information into the fetal disease classification model to obtain fetal birth defect prediction information;
[0113] The maternal and fetal detection data and the birth defect prediction information are input into the risk prediction model to obtain the risk of premature birth of the fetus.
[0114] Specifically, Figure 5 As shown, the fetal health risk assessment under maternal-fetal test data is mainly achieved by inputting the maternal-fetal test data into the health relationship model, sending the data to the disease prediction classification model after being processed by the health relationship model, and then using the fetal disease risk prediction model under the test data to predict premature birth and defects.
[0115] In one implementation, the fetal disease classification model is used to:
[0116] Taking the input data as samples to be predicted, and measuring the distance between the samples to be predicted and several samples of known categories respectively;
[0117] Filtering out a number of the nearest samples of the known category according to the distance measurement result, and determining the category information of the sample to be predicted according to all the nearest samples of the known category;
[0118] The birth defect prediction information is generated according to the category information of the prediction sample.
[0119] Specifically, this embodiment uses the nearest neighbor classification method to predict fetal birth defects. The fetal disease classification model pre-stores a large number of samples and marks the defect category corresponding to each sample. In actual application, the input data of the current fetal disease classification model is used as the sample to be predicted, the distance between the sample to be predicted and all the stored known category samples is measured, and a plurality of known category samples with the closest distance are selected. Finally, the birth defect prediction information of the current sample to be predicted can be determined through these selected known category samples. For example, for each birth defect, the proportion of samples with the birth defect in all the known category samples screened out is calculated. If the value of the sample proportion is greater than the preset threshold, it is judged that the current sample to be predicted also has the birth defect, otherwise the birth defect does not exist.
[0120] In one implementation, the risk prediction model is used to:
[0121] Convert the input data into matrix form through the time-series matrix layer;
[0122] Extracting phenotypes from the temporal matrix layer through a unilateral convolutional layer;
[0123] The important phenotypes are screened out from the extracted phenotypes through the maximum pooling layer;
[0124] The risk of premature birth is predicted based on the important phenotypes through a fully connected softmax prediction layer.
[0125] Specifically, Figure 6As shown, the risk prediction model is the core part of the disease prediction function. This embodiment uses a four-layer convolutional neural network model to establish a risk prediction model, and the risk prediction model is mainly predicted by extracting phenotypes. Four-layer convolutional neural network model: the first layer is composed of a time series matrix; the second layer is a unilateral convolution layer, which can extract phenotypes from the first layer; the third layer is a maximum pooling layer that introduces sparsity on the detected phenotype to screen and retain important phenotypes; the fourth layer is a fully connected softmax prediction layer, with the weight connection of neurons from input to output in front, and a corresponding mapping relationship in the middle. It can be trained in a preset mode and output the prediction results from the network. The four-layer convolutional neural network model used in this embodiment can effectively avoid the problem of too fast gradient loss speed during back propagation in the neural network. After training, an automatic evaluation method can be adopted to divide a part of the data set into an evaluation data set, and the model is evaluated using the evaluation data set in each round (epoch) to verify the effectiveness of the model and improve the accuracy of disease risk prediction.
[0126] In one implementation, the risk prediction model can combine the temporal smoothness of the time series matrix and choose to use different temporal fusion mechanisms, such as early fusion, late fusion, and slow fusion, so as to establish an advanced convolutional neural network architecture with temporal fusion to better predict the risk of premature birth / birth defects of the fetus.
[0127] In one implementation, the risk prediction model may also include a feature extraction layer and a feature mapping layer.
[0128] Based on the above embodiments, the present invention also provides a fetal health assessment device based on detection data, such as Figure 7 As shown, the device comprises:
[0129] A cleaning module 01 is used to obtain maternal and fetal detection data and perform data cleaning on the maternal and fetal detection data;
[0130] Perception module 02, used to input the cleaned maternal and fetal detection data into the health relationship model to obtain perception information;
[0131] The evaluation module 03 is used to input the cleaned maternal-fetal detection data and the perception information into the evaluation network to obtain fetal health evaluation information, wherein the fetal health evaluation information includes the fetal premature birth risk and birth defect prediction information.
[0132] Based on the above embodiment, the present invention further provides a terminal, whose principle block diagram can be shown as follows: Figure 8As shown. The terminal includes a processor, a memory, a network interface, and a display screen connected via a system bus. Among them, the processor of the terminal is used to provide computing and control capabilities. The memory of the terminal includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the terminal is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a fetal health assessment method based on detection data is implemented. The display screen of the terminal can be a liquid crystal display screen or an electronic ink display screen.
[0133] Those skilled in the art will understand that Figure 8 The principle block diagram shown in the figure is only a block diagram of a partial structure related to the scheme of the present invention, and does not constitute a limitation on the terminal to which the scheme of the present invention is applied. The specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0134] In one implementation, the memory of the terminal stores one or more programs, and is configured to be executed by one or more processors. The one or more programs include instructions for performing a fetal health assessment method based on detection data.
[0135] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
[0136] In summary, the present invention discloses a fetal health assessment method, device, terminal and medium based on test data. By acquiring maternal and fetal test data, the maternal and fetal test data are cleaned; the cleaned maternal and fetal test data are input into a health relationship model to obtain perception information; the cleaned maternal and fetal test data and the perception information are input into an evaluation network to obtain fetal health assessment information, wherein the fetal health assessment information includes fetal premature birth risk and birth defect prediction information. The present invention conducts a joint analysis of fetal premature birth and birth defects, and can learn the multiple correlations between maternal and fetal test data, fetal premature birth, and birth defects, thereby improving the accuracy of fetal health assessment results.
[0137] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.
Claims
1. A method for assessing fetal health based on test data, characterized in that: The method comprises: Acquiring maternal and fetal detection data, and performing data cleaning on the maternal and fetal detection data; Inputting the cleaned maternal-fetal detection data into a health relationship model to obtain perception information; The cleaned maternal-fetal detection data and the perception information are input into an evaluation network to obtain fetal health assessment information, wherein the fetal health assessment information includes fetal premature birth risk and birth defect prediction information.
2. The method for fetal health assessment based on detection data according to claim 1, characterized in that: The data cleaning method of the maternal-fetal detection data comprises: Performing entity recognition on each of the maternal and fetal detection data to obtain standard data fields, and generating a standardized data table based on all the standard data fields; According to the standardized data table, duplicate data is removed: After removing duplicate data, determine whether there are outliers in the remaining data. If so, process the outliers, which include outliers and / or missing values.
3. The method for fetal health assessment based on detection data according to claim 2, characterized in that: The step of eliminating duplicate data according to the standardized data table includes: Perform data clustering according to the standardized data table to obtain several types of data; For each category of data, calculate the similarity of the data in the category, and determine whether there is duplicate data based on the similarity. If so, remove the duplicate data; After elimination, the search radius is adjusted, and the step of clustering the data according to the standardized data table is continued until there is no duplicate data in each category of data.
4. The method for fetal health assessment based on detection data according to claim 1, characterized in that: The Healthy Relationships Model is used to: Capturing time series information according to the input maternal-fetal detection data through a loop structure, and obtaining contextual semantic information according to the time series information, wherein the loop structure retains a hidden state at each time step, each hidden state is determined based on the input of the current time step and the hidden state of the previous time step, and each time step corresponds to an element of the maternal-fetal detection data; The perception information is obtained according to the contextual semantic information and the maternal-fetal detection data.
5. The method for fetal health assessment based on detection data according to claim 1, characterized in that: The evaluation network includes a fetal disease classification model and a risk prediction model. The cleaned maternal and fetal detection data and the perception information are input into the evaluation network to obtain fetal health assessment information, including: Inputting the sensed information into the fetal disease classification model to obtain fetal birth defect prediction information; The maternal and fetal detection data and the birth defect prediction information are input into the risk prediction model to obtain the risk of premature birth of the fetus.
6. The method for fetal health assessment based on detection data according to claim 5, characterized in that: The fetal disease classification model is used to: Taking the input data as samples to be predicted, and measuring the distance between the samples to be predicted and several samples of known categories respectively; Filtering out a number of the nearest samples of the known category according to the distance measurement result, and determining the category information of the sample to be predicted according to all the nearest samples of the known category; The birth defect prediction information is generated according to the category information of the prediction sample.
7. The method for fetal health assessment based on detection data according to claim 5, characterized in that: The risk prediction model is used to: Convert the input data into matrix form through the time-series matrix layer; Extracting phenotypes from the temporal matrix layer through a unilateral convolutional layer; The important phenotypes are screened out from the extracted phenotypes through the maximum pooling layer; The risk of premature birth is predicted based on the important phenotypes through a fully connected softmax prediction layer.
8. A fetal health assessment device based on detection data, characterized in that: The device comprises: A cleaning module, used for acquiring maternal and fetal detection data and performing data cleaning on the maternal and fetal detection data; A perception module, used for inputting the cleaned maternal and fetal detection data into a health relationship model to obtain perception information; An evaluation module is used to input the cleaned maternal-fetal detection data and the perception information into an evaluation network to obtain fetal health evaluation information, wherein the fetal health evaluation information includes fetal premature birth risk and birth defect prediction information.
9. A terminal, characterized in that: The terminal includes a memory and one or more processors; the memory stores one or more programs; the program contains instructions for executing the fetal health assessment method based on detection data as described in any one of claims 1-7; and the processor is used to execute the program.
10. A computer-readable storage medium having a plurality of instructions stored thereon, characterized in that: The instructions are suitable for being loaded and executed by a processor to implement the steps of the method for assessing fetal health based on detection data as described in any one of claims 1 to 7 above.
Citation Information
Patent Citations
Birth defect prediction and risk assessment method and system based on machine learning and electronic equipment
CN111508603A
Premature delivery risk prediction method and device, computer equipment and storage medium
CN113921142A
Method for constructing bad outcome prediction model in perinatal period with restricted growth of fetus
CN116825341A
Fetal heart disease ultrasonic image quality detection method based on multi-task learning
CN117274667A
Internet-based pregnant woman health assessment method and system
CN118039143A
Cited By
Antenatal monitoring risk assessment system and method
CN119296788A
Prenatal monitoring risk assessment system and method
CN119296788B
