Fetal growth restriction risk assessment method and system, electronic equipment and product

Abstract

The invention is suitable for the field of data processing, and provides a risk assessment method and system for fetal growth limitation, electronic equipment and a product, and the method comprises the steps: obtaining multi-mode fetal growth data at the early pregnancy stage; the multi-modal fetal growth data comprises image modal data, clinical modal data and biochemical modal data; respectively extracting modal specific features of the image modal data, the clinical modal data and the biochemical modal data to obtain an image feature vector, a clinical feature vector and a biochemical feature vector; fusing the image feature vector, the clinical feature vector and the biochemical feature vector through super-dimensional calculation to obtain a target fusion feature vector; and inputting the target fusion feature vector into a risk assessment model to obtain risk assessment information of fetal growth limitation output by the risk assessment model. According to the scheme, the risk assessment method capable of accurately assessing the condition that the growth of the fetus is limited is provided.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of data processing, and particularly relates to a risk assessment method and system for fetal growth restriction, an electronic device and a product. BACKGROUND

[0002] Fetal Growth Restriction (FGR) is a common complication during pregnancy and is closely related to the health of newborns. Risk assessment of FGR helps clinical intervention decision-making. Current medical field mainly studies FGR risk assessment in the middle and late stages of pregnancy, and lacks FGR risk assessment methods in the early stages of pregnancy. SUMMARY

[0003] Embodiments of the present application provide a risk assessment method and system for fetal growth restriction, an electronic device and a product to solve the problem of lack of FGR risk assessment methods in the early stages of pregnancy in the prior art.

[0004] A first aspect of embodiments of the present application provides a risk assessment method for fetal growth restriction, comprising: obtaining multi-modal fetal growth data in the early stages of pregnancy; the multi-modal fetal growth data includes image modal data, clinical modal data and biochemical modal data; extracting modal-specific features of the image modal data, the clinical modal data and the biochemical modal data respectively to obtain an image feature vector, a clinical feature vector and a biochemical feature vector; fusing the image feature vector, the clinical feature vector and the biochemical feature vector through hyper-dimensional calculation to obtain a target fusion feature vector; inputting the target fusion feature vector into a risk assessment model to obtain risk assessment information for fetal growth restriction output by the risk assessment model.

[0005] A second aspect of embodiments of the present application provides a risk assessment system for fetal growth restriction, comprising: an acquisition module configured to obtain multi-modal fetal growth data in the early stages of pregnancy; the multi-modal fetal growth data includes image modal data, clinical modal data and biochemical modal data; an extraction module configured to extract modal-specific features of the image modal data, the clinical modal data and the biochemical modal data respectively to obtain an image feature vector, a clinical feature vector and a biochemical feature vector; a fusion module configured to fuse the image feature vector, the clinical feature vector and the biochemical feature vector through hyper-dimensional calculation to obtain a target fusion feature vector; an evaluation module configured to input the target fusion feature vector into a risk assessment model to obtain risk assessment information for fetal growth restriction output by the risk assessment model.

[0006] A third aspect of the embodiments of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method according to the first aspect when executing the computer program.

[0007] A fourth aspect of the embodiments of the present application provides a computer program product, comprising a computer program, wherein the computer program is executable by a processor to implement the steps of the method according to the first aspect.

[0008] A fifth aspect of the embodiments of the present application provides a computer-readable storage medium, which stores a computer program, wherein the computer program is executable by a processor to implement the steps of the method according to the first aspect.

[0009] As can be seen from the above, the present application extracts a feature vector from the obtained multi-modal fetal growth data (image modal data, clinical modal data and biochemical modal data) in the early pregnancy, combines the extracted feature vector with hyper-computing to obtain a target fusion feature vector for realizing risk assessment. Finally, the target fusion feature vector is input into a risk assessment model to obtain risk assessment information about fetal growth restriction output by the model. The representation ability of the target fusion feature vector is relatively strong, and the accuracy and reliability of the corresponding risk assessment information are also relatively high, that is, the present application provides a method capable of accurately assessing the risk of fetal growth restriction in the early pregnancy. BRIEF DESCRIPTION OF DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0011] Figure 1 is a flowchart of a fetal growth restriction risk assessment method provided by the embodiments of the present application Figure 1 ; Figure 2 is a flowchart of a fetal growth restriction risk assessment method provided by the embodiments of the present application Figure 2 ; Figure 3 is a structural diagram of a fetal growth restriction risk assessment system provided by the embodiments of the present application; Figure 4 is a structural diagram of an electronic device provided by the embodiments of the present application. DETAILED DESCRIPTION

[0012] In the following description, for purposes of explanation and not limitation, specific details are set forth such as particular architectures, techniques, etc. in order to provide a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known methods, devices, circuits, and

[0013] It is to be understood that the terminology "includes", "has", "holds", "contains" and / or "comprising", when used in this specification and in the following claims, indicates the presence of the stated features, integers, steps, operations, elements, and / or components but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0014] It is also to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting. As used in this specification and the appended claims, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.

[0015] It will be further understood that the terms "and / or", "including", "comprising" when used in this specification and in the following claims, specify the presence of features, integers, steps, operations, elements, and / or components with at least one of one or more associated possibilities, and do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0016] As used in this specification and the appended claims, the term "if' can, in some instances, be interpreted as meaning "when", or "once", or "in response to a determination" or "in response to detecting". Similarly, the phrase "if a determination" or "if detecting [described condition or event]" can, in some instances, be interpreted to mean "once a determination" or "in response to a determination" or "once detecting [described condition or event]" or "in response to detecting [described condition or event]".

[0017] In particular implementations, the terminals described in the embodiments of the present application include, but are not limited to, other portable devices such as mobile telephones, laptop computers, or tablet computers with touch-sensitive surfaces (e.g., touch screen displays and / or touch pads). It will be appreciated that, in certain embodiments, the devices are not portable communication devices, but rather desktop computers with touch-sensitive surfaces (e.g., touch screen displays and / or touch pads).

[0018] In the following discussion, a terminal including a display and a touch-sensitive surface is described. It will be appreciated, however, that a terminal can include one or more other physical user-interface devices, such as a physical keyboard, a mouse, and / or a joystick.

[0019] The terminal supports various applications, such as one or more of the following: a drawing application, a presentation application, a word processing application, a website creation application, a disk authoring application, a spreadsheet application, a game application, a telephone application, a video conferencing application, an e-mail application, an instant messaging application, a workout support application, a photo management application, a digital camera application, a digital camcorder application, a web browsing application, a digital music player application, and / or a digital video player application.

[0020] The various applications that can be executed on the terminal can use at least one common physical user interface device, such as a touch-sensitive surface. One or more functions of a touch-sensitive surface can be adjusted and / or changed between applications and / or within respective applications, and the corresponding information displayed on the terminal can be adjusted and / or changed. In this way, the common physical architecture (e.g., touch-sensitive surface) of the terminal can support a variety of applications with a user interface that is intuitive and transparent to the user.

[0021] It should be understood that the size of the serial number of each step in the embodiments does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0022] In order to illustrate the technical solutions described in the present application, the following will be described by specific embodiments.

[0023] Referring to Figure 1 , Figure 1 is a flow of a risk assessment method for fetal growth restriction provided by the embodiments of the present application Figure 1 . As shown in Figure 1 , a risk assessment method for fetal growth restriction, the method comprises the following steps: Step 101, obtaining multi-modal fetal growth data in early pregnancy; the multi-modal fetal growth data includes image modal data, clinical modal data and biochemical modal data.

[0024] Gestational age refers to the gestational period calculated from the first day of the last menstrual period, and each 7 days is recorded as a gestational week. Gestational age is a core time scale for evaluating the development stage of the fetus.

[0025] The risk assessment of the present application is based on multi-modal fetal growth data, therefore, the early pregnancy in the present application specifically refers to the 9th-14th gestational week, so as to obtain more complete multi-modal fetal growth data.

[0026] The image modality data is a placenta ultrasound image. The placenta ultrasound image is a preprocessed image with high image quality, which can improve the subsequent evaluation accuracy. The preprocessing methods include image enhancement and image segmentation. Through image enhancement, the image clarity and contrast are improved. Through image segmentation, the irrelevant area is cropped and the placenta area is accurately located.

[0027] In some embodiments, the collected original image is subjected to image segmentation processing to obtain a first image with a center cropped to an initial size (e.g., 448x448 pixels). The first image is subjected to pixel normalization processing to normalize to [0, 1] to obtain a second image. The second image is subjected to image enhancement processing (e.g., random flipping, brightness disturbance), and is adjusted to a set size (e.g., 512x512 pixels) to obtain the image modality data. The image size of the image modality data can be 512x512 pixels, and the image type is a grayscale image.

[0028] The clinical modality data is the clinical medical information of the mother of the fetus, i.e., the target pregnant woman, including age, body mass index (BMI), medical history (e.g., pregnancy complications), and obstetric history.

[0029] The biochemical modality data is the laboratory test values related to fetal growth and the image-derived parameters converted by the algorithm, including human chorionic gonadotropin (hCG), progesterone, crown-rump length (CRL), venous duct blood flow, umbilical artery blood flow, and nuchal translucency (NT). Among them, hCG and progesterone are laboratory test values, and CRL, venous duct blood flow, umbilical artery blood flow, and NT are image-derived parameters.

[0030] The image modality data can reflect potential abnormalities in placental structure and morphology, such as edge thickness and echo characteristics. Clinical modality data such as maternal age and medical history (e.g., hypertension and gestational diabetes) are closely related to the risk of FGR. Biochemical modality data is closely related to the speed of embryonic development.

[0031] After determining the target pregnant woman, the corresponding multi-modal fetal growth data is obtained, and the risk of fetal growth restriction of the fetus in the pregnant woman is evaluated based on the obtained data.

[0032] The present application uses multi-modal data to overcome the limitations of single-modal data, realizes multi-dimensional cross-validation, and improves the evaluation accuracy.

[0033] In step 102, the modality-specific features of the image modality data, the clinical modality data, and the biochemical modality data are extracted respectively to obtain image feature vectors, clinical feature vectors, and biochemical feature vectors.

[0034] The image feature vector is a low-dimensional dense vector extracted, quantitatively describing morphological structures (such as organ contours) and dynamic functions (such as blood flow textures).

[0035] The clinical feature vector is a structured semantic vector extracted, quantitatively describing key clinical information such as the state of the pregnant woman and the disease.

[0036] The biochemical feature vector is a low-dimensional statistical vector extracted, reflecting the metabolic growth state and trend changes of biochemical indicators.

[0037] Different modalities of data have significant differences in data structure, information density, and physical properties. This application selects a more suitable feature extraction method for each modality of data according to the heterogeneous characteristics of different modalities of data, independently extracts, and thus maximizes the discriminability of each modality, providing high-quality input for subsequent fusion.

[0038] In some embodiments, the modality-specific features of the image modality data, the clinical modality data, and the biochemical modality data are extracted respectively to obtain an image feature vector, a clinical feature vector, and a biochemical feature vector, including: extracting structural features and texture features in the image modality data to obtain the image feature vector; performing dimensionality reduction processing on the clinical modality data to obtain the clinical feature vector; and performing standardization processing on the biochemical modality data to obtain the biochemical feature vector.

[0039] Through independent feature extraction, high-quality feature vectors are extracted.

[0040] In some embodiments, the image features of the placenta ultrasound image, which is the image modality data, are extracted using a convolutional neural network (CNN) algorithm or other image processing algorithms (such as the Histogram of Oriented Gradients (HOG) algorithm, the Scale-Invariant Feature Transform (SIFT) algorithm), focusing on structural features and texture features, and then obtaining the image feature vector.

[0041] In some embodiments, a pre-trained convolutional neural network model is used to perform image feature extraction on the input image modality data, i.e., the placenta ultrasound image, to obtain the image feature vector. wherein, is the image modality data, i.e., the placenta ultrasound image, , For pre-trained convolutional neural networks, such as residual network 18 (Residual Network 18, ResNet-18), For the extracted image feature vector, the dimension is For example, .

[0042] In some embodiments, the clinical modality data is normalized to eliminate differences in data dimensions, and then key features are extracted through a dimension reduction algorithm to obtain a low-dimensional clinical feature vector that retains key information.

[0043] In some embodiments, the clinical modality data is normalized using a min-max normalization method to obtain clinical initial feature values. The normalization formula is , is a certain type of clinical data in the clinical modality data, such as age, BMI, etc. and are the minimum and maximum values in the clinical modality data, respectively is the normalized clinical initial feature value .

[0044] In some embodiments, the principal component analysis (PCA) algorithm is used to extract the orthogonal principal components in the clinical initial feature values to obtain the clinical feature vector. The PCA algorithm is an unsupervised dimension reduction algorithm that converts the original high-dimensional data into linearly independent variables (principal components) in a low-dimensional space through orthogonal transformation, retaining important information of the data.

[0045] For example, the normalized clinical initial feature value is input into the PCA model to obtain the clinical feature vector output by the PCA model , wherein is the PCA projection matrix , , , and are the data dimensions of the clinical modality data in different data forms .

[0046] In some embodiments, the biochemical numerical modality data is standardized, and the standardized data is used as the biochemical feature vector.

[0047] In some embodiments, the biochemical modality data is standardized using the zero-mean standardization (Z-score) algorithm to obtain the biochemical feature vector . Accordingly, the standardization formula is , . Wherein, is the th biochemical data, and are the mean and standard deviation corresponding to the th biochemical data, is the biochemical feature value after Z-score standardization processing corresponding to the th biochemical data, is the number of biochemical data, for example, 6, accordingly, .

[0048] Step 103, the image feature vector, the clinical feature vector and the biochemical feature vector are fused by hyperdimensional computing to obtain a target fusion feature vector.

[0049] Hyperdimensional Computing (HDC) is a brain-like computing paradigm that uses high-dimensional space (usually with a dimension of 10000) for information representation and processing. Its core operations include high-dimensional mapping, binding, and bundling. High-dimensional mapping refers to encoding original data into a hyper-dimensional random vector (called a hyper-vector). Binding refers to combining two hyper-vectors through multiplication operations (such as element-by-element multiplication) to form a new vector representing the association relationship. Bundling refers to adding multiple hyper-vectors to form a single vector through addition operations (such as element-by-element addition).

[0050] Feature fusion is the fusion of image, clinical, and biochemical feature information into a new vector.

[0051] Using hyperdimensional computing technology, multi-modal feature vectors, i.e., image feature vectors, clinical feature vectors, and biochemical feature vectors, are integrated to realize complex cross-modal interaction modeling and complementary information mining, effectively preserving key pathological and statistical information of each modality, significantly improving the expression ability of the fusion feature vector, and generating a target fusion feature vector with stronger discrimination.

[0052] In some embodiments, the fusion of the image feature vector, the clinical feature vector and the biochemical feature vector by hyperdimensional computing to obtain the target fusion feature vector comprises: mapping the image feature vector, the clinical feature vector and the biochemical feature vector to a set high-dimensional space to obtain a high-dimensional image feature vector, a high-dimensional clinical feature vector and a high-dimensional biochemical feature vector; according to the data association characteristics of the multi-modal fetal growth data, the high-dimensional image feature vector, the high-dimensional clinical feature vector and the high-dimensional biochemical feature vector are subjected to hierarchical feature fusion to obtain the target fusion feature vector.

[0053] In some embodiments, the set dimension corresponding to the high-dimensional space is set as needed, for example, set to 10000. According to the set dimension, each modal feature vector is mapped to a high-dimensional space of the same dimension, which effectively eliminates the modal heterogeneity of the original features, enhances the separability and complementarity between features, and provides a semantically aligned and information-complete unified representation for subsequent fusion.

[0054] In some embodiments, the high-dimensional mapping operation is to map low-dimensional feature vectors to a high-dimensional space using independent random projection matrices, which are used to preserve approximate geometric relationships.

[0055] In some embodiments, the high-dimensional mapping is represented as ,in, These represent images, clinical, and biochemistry, respectively. , for The random projection matrix corresponding to the modal data. To define the corresponding dimension for a higher-dimensional space (e.g.) ), yes The dimension of the feature vector corresponding to the modal data; for Feature vectors corresponding to modal data; This is a sign function that outputs a binary value of -1 or +1. , is the result of high-dimensional mapping D A binary eigenvector of dimension 1.

[0056] High-dimensional mapping transforms low-dimensional feature vectors (i.e., imaging, clinical, and biochemical feature vectors) from a low-dimensional space to a designated high-dimensional space, resulting in high-dimensional feature vectors (i.e., high-dimensional imaging, high-dimensional clinical, and high-dimensional biochemical feature vectors). This low-dimensional-to-high-dimensional transformation enhances the expressive power of the feature vectors, thereby improving the accuracy and reliability of FGR risk assessment.

[0057] After obtaining high-dimensional feature vectors through high-dimensional mapping, the high-dimensional feature vectors corresponding to these data are hierarchically fused based on the data association characteristics (such as complementarity and redundancy) among imaging modal data, clinical modal data, and biochemical modal data, ultimately yielding the target fused feature vector. This hierarchical fusion design addresses alignment, interaction, and integration issues, preventing information overload in early fusion or loss of association in late fusion.

[0058] Hyperdimensional computing technology efficiently processes high-dimensional heterogeneous data, reduces computational complexity, and improves the efficiency of risk assessment.

[0059] In some embodiments, the hierarchical feature fusion of the high-dimensional image feature vector, the high-dimensional clinical feature vector and the high-dimensional biochemical feature vector according to the data correlation characteristics of the multi-modal fetal growth data to obtain the target fusion feature vector comprises: fusing the high-dimensional clinical feature vector and the high-dimensional biochemical feature vector by a binding operation to obtain a first fusion feature vector; and fusing the first fusion feature vector and the high-dimensional image feature vector by a bundling operation to obtain the target fusion feature vector.

[0060] The clinical modal data and the biochemical modal data belong to structured, low-dimensional, semantically correlated modal data, and have strong co-occurrence and complementarity. The co-occurrence refers to the characteristic that the data of different modalities are generated synchronously at the same time point or event.

[0061] Therefore, the high-dimensional clinical feature vector corresponding to the clinical modal data and the high-dimensional biochemical feature vector corresponding to the biochemical modal data are tightly coupled by the binding operation to form a structured intermediate representation, i.e., the first fusion feature vector.

[0062] In some embodiments, the binding operation of the high-dimensional clinical feature vector and the high-dimensional biochemical feature vector is implemented by Hadamard multiplication, i.e., element-wise multiplication of the high-dimensional clinical feature vector and the high-dimensional biochemical feature vector.

[0063] In some embodiments, the binding operation is represented as , wherein ⊙ represents Hadamard product, i.e., element-wise multiplication; is the high-dimensional clinical feature vector, is the high-dimensional biochemical feature vector, is the clinical+biochemical fusion result, i.e., the first fusion feature vector.

[0064] In contrast, the image modal data belongs to visual modal data, and has great differences in structural characteristics such as representation, information source and noise characteristics from the clinical modal data and the biochemical modal data. Therefore, the high-dimensional image feature vector and the first fusion feature vector are fused by the bundling operation to retain the independent information contribution of the image modal data, while avoiding modal conflicts and information interference between different modal data.

[0065] In some embodiments, the bundling operation is represented as , wherein is the high-dimensional image feature vector, is the target fusion feature vector, The bundling operation realizes feature vector aggregation.

[0066] In some embodiments, the hierarchical feature fusion of the data correlation characteristics according to the multi-modal fetal growth data, the high-dimensional image feature vector, the high-dimensional clinical feature vector and the high-dimensional biochemical feature vector to obtain the target fusion feature vector comprises: fusing the high-dimensional clinical feature vector and the high-dimensional biochemical feature vector through a binding operation to obtain a first fusion feature vector; fusing the first fusion feature vector and the high-dimensional image feature vector through a bundling operation to obtain a second fusion feature vector; and performing binaryzation processing on the second fusion feature vector to obtain the target fusion feature vector.

[0067] The high-dimensional clinical feature vector and the high-dimensional biochemical feature vector are fused through a binding operation to obtain a first fusion feature vector, and then the first fusion feature vector and the high-dimensional image feature vector are fused through a bundling operation to obtain a second fusion feature vector. The binding operation and the bundling operation are described above and will not be described here. At this time, the second fusion feature vector is not determined as the target fusion feature vector, and binaryzation processing is performed on the second fusion feature vector.

[0068] In some embodiments, the bundling operation and the binaryzation processing are represented as . Wherein, .

[0069] The binaryzation processing is performed on the second fusion feature vector in the feature fusion process, the second fusion feature vector is converted into a binary representation, so as to simplify the data structure, reduce the calculation complexity, enhance the feature discrimination degree, and make the disease-related features more easily recognized by the classifier in the vector space. The binary feature is not sensitive to the numerical scale, and can effectively eliminate the distribution deviation caused by the dimension difference of different modal feature quantities.

[0070] The hierarchical feature fusion strategy above fuses the high-dimensional clinical feature vector and the high-dimensional biochemical feature vector through a binding operation (element-by-element multiplication), explicitly models the nonlinear interaction relationship between the two, and generates a strong discriminative first fusion feature vector. Then, the first fusion feature vector and the high-dimensional image feature vector are integrated through a bundling operation (element-by-element addition), the clinical-biochemical joint semantics are injected while the integrity of the image space structure is preserved, and the target fusion feature vector which comprehensively represents the fetal growth state is formed. This hierarchical fusion design takes into account the complementarity and difference between modalities, the binding captures the cross-modal dependence, avoids the information flooding of early fusion, the bundling balances the contribution, ensures that the key modalities (such as the structural texture details of the image) participate in the final decision, and improves the physiological interpretability and evaluation robustness of the feature vector.

[0071] In some embodiments, other feature fusion methods can also be used to integrate the multi-modal feature vectors into the target fusion feature vector.

[0072] In some embodiments, the attention mechanism-based Transformer model is used to fuse the image feature vector, the clinical feature vector and the biochemical feature vector by cascading fusion or parallel fusion to obtain a multi-dimensional fusion feature vector, and the multi-dimensional fusion feature vector is taken as a target fusion feature vector for FGR risk assessment.

[0073] In step 104, the target fusion feature vector is input into a risk assessment model to obtain risk assessment information of fetal growth restriction output by the risk assessment model.

[0074] In some embodiments, the risk assessment model is a processing model such as a Support Vector Machine (SVM) model, a random forest model or a deep neural network model for implementing FGR risk assessment.

[0075] The target fusion feature vector is input into a pre-trained risk assessment model, and the risk assessment model analyzes the received target fusion feature vector to output FGR risk assessment information corresponding to the target fusion feature vector.

[0076] The target fusion feature vector fuses data such as hCG and progesterone in the early stage of pregnancy, which has strong predictive ability for the growth status of the fetus. Based on the target fusion feature vector, the FGR risk assessment overcomes the limitations of the prior art and improves the accuracy and reliability of the FGR risk assessment in the early stage of pregnancy, so as to find high-risk individuals in the early stage of pregnancy and provide more sufficient time for subsequent fetal monitoring and intervention.

[0077] In some embodiments, a plurality of sample fusion feature vectors and risk assessment sample information corresponding to each sample fusion feature vector are obtained; each sample fusion feature vector is generated based on multi-modal fetal growth sample data, and the multi-modal fetal growth sample data includes image sample data, clinical sample data and biochemical sample data; model training is performed based on a plurality of sample fusion feature vectors and a plurality of risk assessment sample information to obtain the risk assessment model.

[0078] A plurality of sets of multi-modal sample data and fetal growth restriction conditions corresponding to each set of multi-modal sample data are pre-collected.

[0079] Each set of multi-modal sample data includes image sample data, clinical sample data and biochemical sample data. Feature extraction and feature fusion are performed on these multi-modal sample data. The feature extraction refers to the feature extraction process of the image modal data, the clinical modal data and the biochemical modal data described above. The feature fusion refers to the feature fusion process of the image feature vector, the clinical feature vector and the biochemical feature vector described above. Correspondingly, a sample fusion feature vector corresponding to each set of multi-modal sample data is obtained.

[0080] The fetal growth restriction condition is quantitatively evaluated to obtain risk assessment sample information corresponding to each sample fusion feature vector.

[0081] The plurality of fusion sample feature vectors and the corresponding risk assessment sample information are divided into a training set and a validation set according to a set proportion, for example, 4:1.

[0082] The sample fusion feature vectors in the training set are set as input data of an initial model, and the corresponding risk assessment sample information is set as output data of the initial model. The initial model is a processing model such as an SVM model, a random forest model, or a deep neural network model.

[0083] The initial model is trained based on the set input data and the set output data in the training set, and finally a risk assessment model for realizing FGR risk assessment is obtained.

[0084] The risk assessment performance of the risk assessment model is verified by the validation set, including the accuracy, sensitivity, and specificity of the risk assessment.

[0085] The accuracy calculation formula is .

[0086] The sensitivity calculation formula is .

[0087] The specificity calculation formula is .

[0088] Where TP represents that FGR actually exists and is correctly predicted as existing FGR (true positive); TN represents that FGR does not actually exist and is correctly predicted as not existing FGR (true negative); FP represents that FGR does not actually exist but is incorrectly predicted as existing FGR (false positive); and FN represents that FGR actually exists but is incorrectly predicted as not existing FGR (false negative).

[0089] The accuracy is the proportion that reflects the overall prediction correctness. The sensitivity is used to reflect the proportion of FGR individuals that are correctly identified, and measures the missed diagnosis. The specificity is the proportion of normal fetuses that are correctly identified as normal fetuses, and measures the misjudgment.

[0090] Based on the validation set and the above calculation formula, the accuracy, sensitivity, and specificity of the risk assessment model are verified. The risk assessment model has high accuracy, sensitivity, and specificity. High accuracy indicates that the risk assessment information output by the risk assessment model has high reliability; high sensitivity indicates that the risk assessment model can identify FGR high-risk individuals as early as possible and reduce the risk of missed diagnosis; and high specificity indicates that the risk assessment model can avoid misjudgment and unnecessary intervention on normal fetuses.

[0091] The method can effectively improve the fusion quality and evaluation accuracy of multi-modal data, and has good clinical application prospects.

[0092] In some embodiments, the fusion sample feature vector used for training the risk assessment model is a feature vector obtained by using hyper-dimensional calculation fusion or a feature vector obtained by using other feature fusion methods such as an attention mechanism-based Transformer model.

[0093] In some embodiments, the risk assessment model is a comprehensive model, which can analyze the generation type of the target fusion feature vector input into the risk assessment model. The generation types include generation based on hyper-dimensional calculation technology, generation based on an attention mechanism-based Transformer model, and the like. The target fusion feature vectors of different generation types correspond to different evaluation sub-models. After determining the target generation type of the target fusion feature vector, the target evaluation sub-model corresponding to the target generation type is switched, and risk assessment is performed based on the target fusion feature vector and the target evaluation sub-model to obtain risk assessment information about fetal growth restriction.

[0094] In some embodiments, the risk assessment information is a risk probability value.

[0095] It should be noted that the data used in the present application are all ex vivo data, and the present application does not intervene or operate on the living body in any form.

[0096] In some embodiments, the high-dimensional vector is suitable for inputting into a classifier of a radial basis function kernel mapping (RBF kernel) architecture (such as RBF kernel + SVM) to implement nonlinear classification. The RBF kernel can capture complex nonlinear relationships, and the risk assessment of fetal growth restriction is exactly nonlinear classification. Therefore, a classifier of an RBF kernel architecture can be constructed as the risk assessment model of the present application, and the input parameter of the model is the target fusion feature vector.

[0097] In some embodiments, the risk assessment model is an RBF kernel architecture SVM model for implementing FGR risk assessment. The risk assessment model includes multiple important parameters, and parameter learning is performed through the sample fusion feature vector and risk assessment sample information in the training set to implement model training, and thus the risk assessment model is obtained. The risk assessment sample information is a sample classification decision value, which indicates whether FGR exists or not. For example, -1 indicates that FGR does not exist, and 1 indicates that FGR exists.

[0098] In some embodiments, model training is divided into two training phases. The first model training phase trains a first SVM model that can output classification decision values. The second model training phase trains the first SVM model that outputs classification decision values ​​into a second SVM model that outputs risk probability values, thus obtaining the risk assessment model.

[0099] In some embodiments, model training in the first model training phase is achieved through an objective function. The objective function is: .

[0100] in, The hyperplane normal vector; For bias terms; Let be a slack variable, representing the th Error per sample; The square of the Euclidean distance; The regularization coefficient controls the size of the margin, achieving a trade-off between misclassification rates; This is an RBF kernel mapping, containing kernel function parameters. ; For the first Each sample is fused into a feature vector.

[0101] The learning process of the objective function is an optimization problem of SVM with soft margin, which allows for a certain degree of misclassification to ensure the robustness of the model.

[0102] Parameter learning is achieved through cross-validation of the objective function, i.e., adjusting the kernel function parameters. and regularization coefficient The RBF kernel can enhance the nonlinear fitting ability and further improve the generalization ability of the risk assessment model to obtain the optimal assessment performance.

[0103] After the first model training phase, the first SVM model that outputs classification decision values ​​is obtained. By inputting the target fusion feature vector into the first SVM model, the classification decision value output by the first SVM model (such as -1 or 1) can be obtained, which is a single classification.

[0104] This application introduces the sigmoid function, which maps classification decision values ​​to risk probability values, i.e., outputs values ​​in the range [0,1].

[0105] In some embodiments, model training in the second model training phase is implemented based on the sigmoid function. The sigmoid function has a slope parameter A and a position parameter B. Parameters A and B are obtained by minimizing the cross-entropy loss using the log-likelihood minimization method. The fitting function set for the log-likelihood minimization method is shown below: ; ; .

[0106] wherein, is the sample risk probability corresponding to the th sample fusion feature vector.

[0107] After the model training of the second model training stage, a second SVM model outputting risk probability values is obtained, , .

[0108] Compared with single classification, the risk probability values have higher clinical reference significance, and a stratified management strategy can be set to achieve fine risk assessment.

[0109] In some embodiments, the risk assessment model is represented as , is the risk probability value. The method described in the present application can be formally represented as , . Wherein, is a feature extractor of image modal data, outputting an image feature vector ; is a dimension reduction function corresponding to the clinical modal data, generating a clinical feature vector ; is a biochemical feature vector; represents a high-dimensional mapping and fusion function, which corresponds to generating a target fusion feature vector ; is a first SVM model outputting a decision classification value, i.e., a SVM decision function, is a sigmoid function.

[0110] The risk assessment information of the present application has important clinical guidance significance.

[0111] In some embodiments, the present application can be applied to the following medical scenarios: (1) The method described in the present application is used as an auxiliary tool for obstetric clinical medical diagnosis. The multi-modal fetal growth data input by the doctor is subjected to feature extraction and feature fusion, and then FGR risk assessment is realized, FGR risk assessment information is output, early FGR screening and individualized monitoring scheme are formulated, and the risk of adverse outcomes of newborns is reduced; (2) Develop the method described in this application as a cloud service and integrate it into a remote medical platform or a mobile application for pregnancy management. Pregnant women can collect image modal data using a portable ultrasound device, upload the image modal data, clinical modal data, and biochemical modal data to the cloud through a terminal device such as a mobile phone for FGR risk assessment, and the remote medical team can provide consultation or referral recommendations based on the FGR risk assessment information. This is particularly suitable for areas with insufficient medical resources, enabling the universal access of FGR screening and improving the accessibility of maternal health management; (3) Embed the method described in this application into intelligent ultrasound equipment or medical imaging systems to develop intelligent medical devices. Intelligent medical devices can analyze image modal data, clinical modal data, and biochemical modal data through built-in algorithms such as convolutional neural network feature extraction and hyper-computing technology, and display FGR risk assessment information in real-time during the examination process. Intelligent medical devices can be connected to hospital information systems to improve the intelligent level of ultrasound equipment and simplify the operation process for doctors; (4) Use the method described in this application as a research tool for obstetrics and fetal medicine to analyze the correlation between multi-modal fetal growth data and FGR etiology, explore new biochemical markers and image features, and research institutions can use the method described in this application to build an FGR risk assessment dataset, support multi-center clinical trials, verify the reliability of the risk assessment model, provide data-driven support for FGR pathogenesis research, and promote medical development; (5) Apply the method described in this application to the health insurance industry for risk assessment of maternal insurance products. Insurance companies can adjust premiums or develop personalized insurance plans based on FGR risk assessment information, cooperate with medical institutions to promote early screening services, reduce medical costs for high-risk pregnancies, optimize insurance calculation models, and improve customer health management experience; (6) The processing framework of the method described in this application is universal and can be extended to early screening and evaluation of other fetal abnormalities (such as Down syndrome and neural tube defects). By adjusting the feature extraction strategy and classification model, it can adapt to multi-modal data of different diseases, such as using placental ultrasound images, NT, and genetic biomarkers to assess chromosome abnormality risk, and expand the application range of risk assessment technology in fetal health screening; (7) Integrate the method described in this application into smart wearable devices or pregnancy management platforms, combine real-time monitoring data (such as maternal heart rate and activity level) with multi-modal fetal growth data, and provide personalized pregnancy health recommendations, such as recommending dietary adjustments, exercise plans, or medical interventions based on FGR risk assessment information, optimizing the fetal development environment, achieving precise management during pregnancy, improving maternal compliance, and improving fetal health.

[0112] In summary, the method provided in the application comprehensively utilizes multi-modal fetal growth data, comprehensively captures the diversified causes of FGR, improves the accuracy of early pregnancy FGR risk assessment, and fills the gap in the prior art in early pregnancy FGR risk assessment.

[0113] In the embodiment of the application, a feature vector is extracted from the acquired multi-modal fetal growth data (image modal data, clinical modal data and biochemical modal data) in the early pregnancy, the extracted feature vector is fused by hyper-dimensional calculation, and then a target fusion feature vector for realizing risk assessment is obtained. Finally, the target fusion feature vector is input into a risk assessment model, and risk assessment information about fetal growth restriction output by the model is obtained. The representation ability of the target fusion feature vector is relatively strong, and the accuracy and reliability of the risk assessment information obtained accordingly are also relatively high, that is, the application provides a method capable of accurately assessing the risk of fetal growth restriction in early pregnancy.

[0114] Referring to Figure 2 , Figure 2 is a flow of a fetal growth restriction risk assessment method provided in the embodiment of the application Figure 2 . As shown in Figure 2 , a fetal growth restriction risk assessment method comprises the following steps: Step 201, acquiring multi-modal fetal growth data in early pregnancy; the multi-modal fetal growth data comprises image modal data, clinical modal data and biochemical modal data.

[0115] The implementation process of this step is the same as that of step 101 in the foregoing embodiment, which will not be described here again.

[0116] Step 202, respectively extracting modal-specific features of the image modal data, the clinical modal data and the biochemical modal data to obtain an image feature vector, a clinical feature vector and a biochemical feature vector.

[0117] The implementation process of this step is the same as that of step 102 in the foregoing embodiment, which will not be described here again.

[0118] Step 203, fusing the image feature vector, the clinical feature vector and the biochemical feature vector by hyper-dimensional calculation to obtain a target fusion feature vector.

[0119] The implementation process of this step is the same as that of step 103 in the foregoing embodiment, which will not be described here again.

[0120] Step 204, inputting the target fusion feature vector into a risk assessment model to obtain risk assessment information about fetal growth restriction output by the risk assessment model; the risk assessment information is a risk probability value.

[0121] The implementation process of this step is the same as that of step 104 in the foregoing embodiments, and will not be repeated here.

[0122] Step 205: Determine the detection gestational age corresponding to the multi-modal fetal growth data; different gestational ages correspond to different risk level threshold intervals.

[0123] The detection gestational age refers to the gestational age corresponding to the collection period of the multi-modal fetal growth data.

[0124] Different risk level threshold intervals are set for different gestational ages to adapt to the stage characteristics of fetal growth and development and the dynamics of disease pathophysiological changes, and to realize precise stratified management and optimization of intervention timing.

[0125] Among them, the risk level threshold interval is set according to needs. For example, the medium risk threshold interval for 10 weeks of pregnancy is [0.02, 0.03], and the medium risk threshold interval for 14 weeks of pregnancy is [0.0168, 0.0240].

[0126] Through gestational age calibration and dynamic threshold matching mechanism, precise stratified management of fetal growth restriction is realized, the early identification rate of high-risk fetuses is significantly improved, and the occupation of medical resources by low-risk groups is reduced, providing time-based decision support for perinatal intervention.

[0127] Step 206: According to the risk level threshold interval corresponding to the detection gestational age, determine the risk level corresponding to the risk probability value and the risk prompt information corresponding to the risk level.

[0128] The risk level threshold interval corresponding to the detection gestational age is determined, and the risk level of the risk probability value is determined under the corresponding risk level threshold interval. The risk level includes, for example, low risk, medium risk, and high risk, which are set according to needs.

[0129] Correspondingly, the risk prompt information corresponding to the risk level is determined and used as a clinical auxiliary suggestion.

[0130] In some embodiments, the risk probability value is represented as If , it is determined that the risk level is low risk, and the risk prompt information is routine follow-up; if , it is determined that the risk level is medium risk, and the risk prompt information is that there is potential risk, and it is suggested to strengthen nutritional intervention and shorten the examination cycle; if , it is determined that the risk level is high risk, and the risk prompt information is that further evaluation of placental function and monitoring of blood flow are suggested, and if necessary, high-risk pregnancy management is included.

[0131] Step 207: Output the risk probability value, the risk level, and the risk prompt information.

[0132] In some embodiments, the risk probability value, the risk level and the risk prompt information are output through the interactive interface, or are integrated into a standardized printed report for output.

[0133] The risk probability value, the risk level and the risk prompt information are visually displayed, thereby providing decision support for clinical diagnosis of fetal growth restriction.

[0134] In the embodiments of the present application, different risk level threshold intervals are set for different gestational weeks. After obtaining the risk assessment information of the risk probability value of fetal growth restriction through the risk assessment model, the detection gestational week corresponding to the multi-modal fetal growth data is determined, the risk level corresponding to the risk probability value is determined according to the risk level threshold interval corresponding to the detection gestational week, and the risk prompt information corresponding to the risk level is determined. By using the gestational week-specific threshold interval, the misjudgment caused by the static standard is avoided. Meanwhile, the risk probability value, the risk level and the risk prompt information are output, thereby realizing the visual display of information.

[0135] Referring to Figure 3 , Figure 3 is a structural diagram of a risk assessment system for fetal growth restriction provided by the embodiments of the present application. Only parts related to the embodiments of the present application are shown for ease of description.

[0136] The risk assessment system 300 for fetal growth restriction comprises an acquisition module 301, an extraction module 302, a fusion module 303 and an evaluation module 304.

[0137] The acquisition module 301 is configured to acquire multi-modal fetal growth data in the early stage of pregnancy. The multi-modal fetal growth data comprises image modal data, clinical modal data and biochemical modal data.

[0138] The extraction module 302 is configured to extract modal-specific features of the image modal data, the clinical modal data and the biochemical modal data respectively, thereby obtaining an image feature vector, a clinical feature vector and a biochemical feature vector.

[0139] The fusion module 303 is configured to fuse the image feature vector, the clinical feature vector and the biochemical feature vector through hyper-dimensional calculation, thereby obtaining a target fusion feature vector.

[0140] The evaluation module 304 is configured to input the target fusion feature vector into a risk assessment model, thereby obtaining risk assessment information of fetal growth restriction output by the risk assessment model.

[0141] In some embodiments, the extraction module is specifically configured to: extract structural features and texture features in the image modal data, thereby obtaining the image feature vector; dimensionally reduce the clinical modality data to obtain the clinical feature vector; standardize the biochemical modality data to obtain the biochemical feature vector.

[0142] In some embodiments, the fusion module is configured to: map the image feature vector, the clinical feature vector and the biochemical feature vector to a set high-dimensional space to obtain a high-dimensional image feature vector, a high-dimensional clinical feature vector and a high-dimensional biochemical feature vector; perform hierarchical feature fusion on the high-dimensional image feature vector, the high-dimensional clinical feature vector and the high-dimensional biochemical feature vector according to data correlation characteristics of the multi-modality fetal growth data to obtain the target fusion feature vector.

[0143] In some embodiments, the fusion module is specifically configured to: fuse the high-dimensional clinical feature vector and the high-dimensional biochemical feature vector through a binding operation to obtain a first fusion feature vector; fuse the first fusion feature vector and the high-dimensional image feature vector through a bundling operation to obtain the target fusion feature vector.

[0144] In some embodiments, the fusion module is specifically configured to: fuse the high-dimensional clinical feature vector and the high-dimensional biochemical feature vector through a binding operation to obtain a first fusion feature vector; fuse the first fusion feature vector and the high-dimensional image feature vector through a bundling operation to obtain a second fusion feature vector; perform binaryzation processing on the second fusion feature vector to obtain the target fusion feature vector.

[0145] In some embodiments, the risk assessment information is a risk probability value, and the system further comprises an information processing output module configured to: determine a detection gestational age corresponding to the multi-modality fetal growth data, wherein different gestational ages correspond to different risk level threshold intervals; determine, according to the risk level threshold interval corresponding to the detection gestational age, the risk level corresponding to the risk probability value and risk prompt information corresponding to the risk level; output the risk probability value, the risk level and the risk prompt information.

[0146] In some embodiments, the system further comprises a model training module configured to: obtaining a plurality of sample fusion feature vectors and risk assessment sample information corresponding to each of the sample fusion feature vectors, wherein each of the sample fusion feature vectors is generated based on multi-modal fetal growth sample data, and the multi-modal fetal growth sample data comprises image sample data, clinical sample data and biochemical sample data; training a model based on the plurality of sample fusion feature vectors and the plurality of risk assessment sample information to obtain the risk assessment model.

[0147] The risk assessment system for fetal growth restriction provided by the embodiments of the present application can implement each process of the embodiments of the risk assessment method for fetal growth restriction and achieve the same technical effects. To avoid repetition, no further description is given here.

[0148] Figure 4 is a structural diagram of an electronic device provided by an embodiment of the present application. As shown in the diagram, the electronic device 4 of the embodiment comprises at least one processor 40 (only one is shown in the diagram), a memory 41, and a computer program 42 stored in the memory 41 and executable on the at least one processor 40, wherein the processor 40 implements the steps in any of the method embodiments described above when executing the computer program 42. Figure 4

[0149] The electronic device 4 can be a desktop computer, a notebook computer, a palm computer, a cloud server and the like. The electronic device 4 can include, but is not limited to, the processor 40 and the memory 41. Those skilled in the art can understand that the electronic device 4 shown in the diagram is merely an example of the electronic device 4 and does not constitute a limitation on the electronic device 4, and the electronic device 4 can include more or fewer components than those shown in the diagram, or combine certain components or different components, for example, the electronic device can also include an input / output device, a network access device, a bus and the like. Figure 4

[0150] The processor 40 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0151] ​​The memory 41 can be an internal storage unit of the electronic device 4, such as a hard disk or a memory of the electronic device 4. The memory 41 can also be an external storage device of the electronic device 4, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. equipped on the electronic device 4. Further, the memory 41 can also include both the internal storage unit and the external storage device of the electronic device 4. The memory 41 is used to store the computer program and other programs and data required by the electronic device. The memory 41 can also be used to temporarily store data that has been output or will be output.

[0152] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above functional units and modules is exemplified, and in actual application, the above functions can be completed by different functional units and modules according to needs, that is, the internal structure of the system is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit, and the integrated unit can be realized in the form of hardware or software. In addition, the specific names of each functional unit and module are only for easy distinction, and do not limit the protection scope of the application. The specific working process of the units and modules in the system can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0153] In the above embodiments, the description of each embodiment has its own emphasis, and the parts not described or recorded in detail in a certain embodiment can be referred to the relevant description of other embodiments.

[0154] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the application.

[0155] In the embodiments of the present application, it should be understood that the disclosed system / electronic device and method can be implemented in other manners. For example, the embodiments of the system / electronic device described above are merely schematic. For example, the division of the modules or units is only a logical function division. There can be another division manner for the actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between different units, can be indirect couplings or communication connections through some interfaces, and electrical, mechanical or other forms.

[0156] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., can be located in one place, or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiments.

[0157] In addition, each functional unit in the various embodiments of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0158] The integrated module / unit, if realized in the form of a software functional unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. When the processor executes the computer program, the steps of the above-mentioned various method embodiments can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable file or some intermediate form. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the computer readable medium can include appropriate contents according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.

[0159] The application can realize all or part of the processes in the above-mentioned embodiment methods, and can also be realized by a computer program product. When the computer program product runs on an electronic device, the electronic device is caused to perform the steps in the above-mentioned various method embodiments.

[0160] The above-mentioned embodiments are only used to illustrate the technical solutions of the present application, rather than limit them. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalents. The modifications or replacements do not change the essence of the corresponding technical solutions, and should be included in the protection scope of the present application.

Claims

1. A method of risk assessment for fetal growth restriction, characterized in that, The method comprises the following steps: acquiring multi-modal fetal growth data in early pregnancy; the multi-modal fetal growth data comprises image modal data, clinical modal data and biochemical modal data; extracting modal-specific features of the image modal data, the clinical modal data and the biochemical modal data respectively to obtain an image feature vector, a clinical feature vector and a biochemical feature vector; fusing the image feature vector, the clinical feature vector and the biochemical feature vector through hyper-dimensional calculation to obtain a target fusion feature vector; inputting the target fusion feature vector into a risk assessment model to obtain risk assessment information of fetal growth restriction output by the risk assessment model.

2. The method of claim 1, wherein, The step of extracting modal-specific features of the image modal data, the clinical modal data and the biochemical modal data respectively to obtain an image feature vector, a clinical feature vector and a biochemical feature vector comprises the following steps: extracting structural features and texture features in the image modal data to obtain the image feature vector; performing dimension reduction processing on the clinical modal data to obtain the clinical feature vector; performing standardization processing on the biochemical modal data to obtain the biochemical feature vector.

3. The method of claim 1, wherein, The step of fusing the image feature vector, the clinical feature vector and the biochemical feature vector through hyper-dimensional calculation to obtain a target fusion feature vector comprises the following steps: mapping the image feature vector, the clinical feature vector and the biochemical feature vector to a set high-dimensional space to obtain a high-dimensional image feature vector, a high-dimensional clinical feature vector and a high-dimensional biochemical feature vector; performing hierarchical feature fusion on the high-dimensional image feature vector, the high-dimensional clinical feature vector and the high-dimensional biochemical feature vector according to data correlation characteristics of the multi-modal fetal growth data to obtain the target fusion feature vector.

4. The method of claim 3, wherein, The step of performing hierarchical feature fusion on the high-dimensional image feature vector, the high-dimensional clinical feature vector and the high-dimensional biochemical feature vector according to data correlation characteristics of the multi-modal fetal growth data to obtain the target fusion feature vector comprises the following steps: fusing the high-dimensional clinical feature vector and the high-dimensional biochemical feature vector through binding operation to obtain a first fusion feature vector; fusing the first fusion feature vector and the high-dimensional image feature vector through bundling operation to obtain the target fusion feature vector.

5. The method of claim 3, wherein, The step of performing hierarchical feature fusion on the high-dimensional image feature vector, the high-dimensional clinical feature vector and the high-dimensional biochemical feature vector according to data correlation characteristics of the multi-modal fetal growth data to obtain the target fusion feature vector comprises the following steps: fusing the high-dimensional clinical feature vector and the high-dimensional biochemical feature vector through binding operation to obtain a first fusion feature vector; fusing the first fusion feature vector and the high-dimensional image feature vector through bundling operation to obtain a second fusion feature vector; performing binaryzation processing on the second fusion feature vector to obtain the target fusion feature vector.

6. The method of claim 1, wherein, The risk assessment information is a risk probability value, and after the step of inputting the target fusion feature vector into a risk assessment model to obtain risk assessment information of fetal growth restriction output by the risk assessment model, the method further comprises the following steps: determine a detection gestational age corresponding to the multi-modal fetal growth data; different gestational ages correspond to different risk level threshold intervals; determine, according to the risk level threshold interval corresponding to the detection gestational age, the risk level corresponding to the risk probability value and risk prompt information corresponding to the risk level; output the risk probability value, the risk level and the risk prompt information.

7. The method of claim 1, wherein, The method further comprises: obtaining a plurality of sample fusion feature vectors and risk assessment sample information corresponding to each of the sample fusion feature vectors; each of the sample fusion feature vectors is generated based on multi-modal fetal growth sample data, and the multi-modal fetal growth sample data includes image sample data, clinical sample data and biochemical sample data; performing model training based on a plurality of the sample fusion feature vectors and a plurality of the risk assessment sample information to obtain the risk assessment model.

8. A system for assessing the risk of fetal growth restriction, characterized in that, comprises: an obtaining module configured to obtain multi-modal fetal growth data in an early gestation period; the multi-modal fetal growth data includes image modal data, clinical modal data and biochemical modal data; an extracting module configured to extract modal-specific features of the image modal data, the clinical modal data and the biochemical modal data respectively to obtain an image feature vector, a clinical feature vector and a biochemical feature vector; a fusion module configured to fuse the image feature vector, the clinical feature vector and the biochemical feature vector through hyper-dimensional calculation to obtain a target fusion feature vector; an evaluation module configured to input the target fusion feature vector into a risk assessment model to obtain risk assessment information of fetal growth restriction output by the risk assessment model.

9. An electronic device, comprising: comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to enable the electronic device to implement the method in any one of claims 1 to 7.

10. A computer program product, characterised in that, comprises a computer program which, when executed, causes the method in any one of claims 1 to 7 to be performed.

Images