Method for establishing fetal lung maturity recognition model based on ultrasonic imageomics, fetal lung maturity recognition model

Abstract

The application provides a method for establishing a fetal fetal lung maturity recognition model based on ultrasonic imaging. +6 The method comprises the following steps: collecting normal fetal lung ultrasonic image data of a gestational age of 20 to 40 weeks; obtaining corresponding image data according to the lung ultrasonic image data and marking a region of interest in each lung ultrasonic image in the image data; performing primary screening of imaging features of the marked region of interest; performing secondary screening of the primary screened imaging features by using ANOVA F test, and performing dimension reduction processing on the features obtained after the secondary screening; taking the imaging features used for establishing the fetal fetal lung maturity recognition model as input, taking the gestational age of a pregnant woman as label as output, training an AdeptiveXGBoost classifier, saving the trained AdeptiveXGBoost classifier, and obtaining the fetal fetal lung maturity recognition model. The application can achieve the purpose of recognizing and predicting the fetal lung maturity of a fetus by using a plurality of finally screened imaging features.

Description

Technical Field

[0001] This invention relates to the technical field of fetal lung maturity identification, and more specifically, to a method for establishing a fetal lung maturity identification model based on ultrasound radiomics, and a fetal lung maturity identification model. Background Technology

[0002] In recent years, the incidence of pregnancy complications has been increasing year by year, especially hypertensive disorders of pregnancy (HDP). The pathogenesis of HDP involves impaired remodeling of trophoblastic cells in the maternal spiral arteries, leading to placental ischemia, hypoxia, and dysfunction. The fetus cannot obtain sufficient oxygen and nutrients from the placenta, resulting in a chronic hypoxic state. This condition causes the fetal lungs to lag behind those of normal fetuses at the same gestational age, meaning the fetal lungs are immature. This can lead to various respiratory diseases in newborns and affect their long-term quality of life. Therefore, assessing fetal lung maturity is helpful for clinical decision-making and personalized care after birth. However, current prenatal ultrasound methods cannot accurately assess fetal lung maturity.

[0003] Radiomics can perform high-throughput, automated analysis of vast amounts of visually indistinguishable information from medical images, and machine learning methods can be used to extract features from these images for disease diagnosis and prediction. How to utilize ultrasound radiomics features to help assess fetal lung maturity is currently a key research focus.

[0004] In related technologies, Du Yanran, Jiao Jing, Ren Yunyun, Zhou Jianqiao, and others disclosed the "Application of Ultrasound Radiomics Technology in Assessing Fetal Lung Maturity." Each pregnant woman had one standard fetal lung ultrasound image collected within 72 hours before delivery. Based on ultrasound radiomics technology, fetal lung texture analysis was performed on the fetal lung ultrasound images. Combined with the presence or absence of pregnancy complications, high-throughput radiomics features were extracted. Using neonatal prognosis as the gold standard, fetal lung maturity assessment models applicable to different gestational weeks were established. Subsequently, in the corresponding validation set, the fetal lung maturity model was validated to predict the risk of neonatal respiratory diseases. However, this study mainly focused on the relationship between fetal lung maturity and neonatal respiratory diseases. Often, due to pregnancy complications, the fetal lung maturity does not correspond to the normal fetal lung maturity at the mother's gestational age. This study could not analyze and identify the gestational age corresponding to fetal lung maturity based on ultrasound radiomics features. For example, if the gestational age is 30 weeks, but the fetal lung maturity corresponds to the normal fetal lung maturity of a 27-week fetus, if the fetal lung maturity is not monitored in time during pregnancy, it may lead to respiratory diseases in the newborn and the inability to provide timely and effective personalized care. Summary of the Invention

[0005] To address the problems existing in the aforementioned related technologies, this invention aims to provide a method for establishing a fetal lung maturity identification model based on ultrasound radiomics. Specifically, this method for establishing a fetal lung maturity identification model based on ultrasound radiomics uses several radiomics features in ultrasound images as identification features for fetal lung maturity. This achieves the goal of identifying the gestational age of the mother corresponding to the fetal lung maturity based on ultrasound radiomics features, thereby providing an effective basis for formulating personalized care methods for the newborn.

[0006] Therefore, the first objective of this invention is to propose a method for establishing a fetal lung maturity identification model based on ultrasound radiomics.

[0007] The second objective of this invention is to propose a fetal lung maturity identification model.

[0008] To achieve the aforementioned first objective, the present invention provides a method for establishing a fetal lung maturity identification model based on ultrasound radiomics. This method for establishing a fetal lung maturity identification model based on ultrasound radiomics includes: Step S1: collecting data from fetuses aged 20 to 40 weeks of gestation. +6 Ultrasound images of the lungs of a normal fetus at 40 weeks of gestation; of which, 40 +6 The week represents 40 weeks plus 6 days; Step S2: Obtain the corresponding image data based on the lung ultrasound image data, and delineate the region of interest in each lung ultrasound image in the image data; Step S3: Perform initial screening of radiomics features on the delineated regions of interest; Step S4: Use ANOVA F test to perform secondary screening of the radiomics features obtained from the initial screening, and then perform dimensionality reduction processing on the features obtained after secondary screening to determine the radiomics features used for establishing the fetal lung maturity recognition model; Step S5: Use the radiomics features used for establishing the fetal lung maturity recognition model as input, and the gestational age of the pregnant woman as the label as the output, train the AdeptiveXGBoost classifier, and save the trained AdeptiveXGBoost classifier to obtain the fetal lung maturity recognition model; wherein, the fetal lung maturity recognition model realizes the recognition of the gestational age of the pregnant woman corresponding to the fetal lung maturity.

[0009] Preferably, the lung ultrasound image data in step S1 includes one or a combination of the following: fetal weight, fetal chest circumference, fetal abdominal circumference, fetal bilateral lung volume, fetal aortic pulmonary artery pulsatility index, fetal aortic pulmonary artery resistance index, fetal aortic pulmonary artery valve peak systolic velocity, fetal aortic pulmonary artery systolic blood flow acceleration time, and the ratio of fetal aortic pulmonary artery systolic blood flow acceleration time to ejection time.

[0010] Preferably, the specific operation of step S3 includes the following steps: Step S301: Perform norm normalization preprocessing on the delineated region of interest to perform preliminary extraction of radiomics features; Step S302: Construct an FEL autoencoder to perform initial screening on the preliminarily extracted radiomics features to obtain 128 radiomics features; wherein, the FEL autoencoder includes an input layer, an encoder, a decoder and an output layer connected in sequence.

[0011] Preferably, the encoder is a feedforward neural network, comprising a first fully connected layer, a first Dropout layer, a second fully connected layer, a second Dropout layer, and a third fully connected layer connected in sequence; the mathematical expression corresponding to the calculation process of the encoder is:

[0012] h = f(W) e x+b e (1)

[0013] In equation (1), x is the manually drawn region of interest, and the dimension of x is d; x is compressed into a latent space, which is represented as h, and the dimension of h is k, then k <d;W e This is the encoder's weight matrix; b e f is the encoder's bias vector; f is the activation function.

[0014] Preferably, the decoder is also a feedforward neural network, comprising a fourth fully connected layer, a third Dropout layer, a fifth fully connected layer, and a fourth Dropout layer connected in sequence; the mathematical expression corresponding to the calculation process of the decoder is:

[0015]

[0016] In equation (2), W d This is the weight matrix of the decoder; b d is the bias vector of the decoder; g is the activation function of the decoder; It is the reconstructed input data.

[0017] Preferably, the loss function of the FEL autoencoder is the cross-entropy loss function, which is used to measure the difference between the original input data x and the reconstructed input data. The differences between them.

[0018] Preferably, the specific operation of step S4 includes the following steps: Step S401: Use ANOVA F test to perform secondary screening on the radiomics features initially screened in step S3 to determine the features of the fetal lung region of interest that are linearly correlated with gestational age and select the top 5% of features; Step S402: Based on the recursive feature elimination method of Dynamic RFECV, perform dimensionality reduction processing on the obtained top 5% of features to determine the radiomics features used for establishing the fetal lung maturity recognition model.

[0019] To achieve the second objective mentioned above, the present invention also provides a fetal lung maturity identification model, which is a fetal lung maturity identification model established using the method for establishing a fetal lung maturity identification model based on ultrasound radiomics in any of the above technical solutions.

[0020] The beneficial effects of this invention are:

[0021] (1) The method for establishing a fetal lung maturity identification model based on ultrasound radiomics provided by the present invention is based on ultrasound images. Through several radiomics features that are finally selected, it is used to identify and predict the fetal lung maturity. When the gestational age corresponding to the identified fetal lung maturity is inconsistent with the actual gestational age of the pregnant woman, it is possible to predict the respiratory diseases that may occur after the birth of the fetus based on the fetal lung maturity. This helps to formulate personalized treatment and reasonable plans in advance, thereby reducing the various risks caused by respiratory diseases after the birth of the fetus.

[0022] (2) The method for establishing a fetal lung maturity recognition model based on ultrasound radiomics provided in this invention constructs an FEL autoencoder to map the features of manually drawn regions of interest (ROIs) to a lower-dimensional latent space, achieving dimensionality reduction, which helps to remove redundant features and reduce computational complexity. It also has a certain degree of resistance to noise and outliers, making the selected feature representations more reliable and suitable for subsequent fetal lung maturity analysis tasks.

[0023] Additional aspects and advantages of the invention will become apparent from the description which follows, or may be learned by practice of the invention. Attached Figure Description

[0024] Figure 1 A schematic flowchart of a method for establishing a fetal lung maturity identification model based on ultrasound radiomics according to an embodiment of the present invention is shown.

[0025] Figure 2 A schematic flowchart illustrating another embodiment of the present invention shows a method for establishing a fetal lung maturity identification model based on ultrasound radiomics.

[0026] Figure 3 This diagram illustrates radiomics features of 51 fetal lung ROIs that are linearly correlated with gestational age, as determined by the ANOVA F test according to an embodiment of the present invention.

[0027] Figure 4 A schematic diagram of 12 radiomics features for establishing a fetal lung maturity identification model according to an embodiment of the present invention is shown. Detailed Implementation

[0028] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0029] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0030] Figure 1 A schematic flowchart illustrating a method for establishing a fetal lung maturity identification model based on ultrasound radiomics, according to an embodiment of the present invention, is shown. Figure 1 As shown, the method for establishing the fetal lung maturity identification model based on ultrasound radiomics includes:

[0031] Step S1: Collect samples from individuals aged 20 to 40 weeks of gestation. +6 Ultrasound images of the lungs of a normal fetus at 40 weeks of gestation; of which, 40 +6 A week means 40 weeks plus 6 days;

[0032] Step S2: Obtain the corresponding image data based on the lung ultrasound image data, and delineate the region of interest in each lung ultrasound image in the image data;

[0033] Step S3: Perform initial screening of radiomics features on the outlined regions of interest;

[0034] Step S4: Use ANOVA F test to perform a second screening of the radiomics features selected in the first screening, and then perform dimensionality reduction on the features obtained after the second screening to determine the radiomics features used to establish the fetal lung maturity recognition model.

[0035] Step S5: Using the radiomics features used to establish the fetal lung maturity recognition model as input and the gestational age of the pregnant woman as the label as the output, train the AdeptiveXGBoost classifier and save the trained AdeptiveXGBoost classifier to obtain the fetal lung maturity recognition model; wherein, the fetal lung maturity recognition model realizes the recognition of the gestational age of the pregnant woman corresponding to the fetal lung maturity.

[0036] In this embodiment, the method for establishing a fetal lung maturity identification model based on ultrasound radiomics provided by the present invention, based on ultrasound images, achieves the purpose of identifying and predicting fetal lung maturity through several radiomics features that are ultimately selected. When the gestational age corresponding to the identified fetal lung maturity is inconsistent with the actual gestational age of the pregnant woman, it is possible to predict respiratory diseases that may occur in the fetus after birth based on the fetal lung maturity, which helps to formulate personalized treatment and reasonable plans in advance, thereby reducing various risks caused by respiratory diseases after the birth of the fetus.

[0037] The present invention provides a method for establishing a fetal lung maturity recognition model based on ultrasound radiomics. By constructing a FEL autoencoder, the features of manually delineated regions of interest (ROIs) are mapped to a lower-dimensional latent space, achieving dimensionality reduction. This helps to remove redundant features and reduce computational complexity. Simultaneously, it exhibits a certain degree of robustness against noise and outliers, resulting in more reliable feature representations that are more suitable for subsequent fetal lung maturity analysis tasks.

[0038] In one embodiment of the present invention, the lung ultrasound image data in step S1 includes one or a combination of the following: fetal weight, fetal chest circumference, fetal abdominal circumference, fetal bilateral lung volume, fetal aortic pulmonary artery pulsatility index, fetal aortic pulmonary artery resistance index, fetal aortic pulmonary artery valve peak systolic velocity, fetal aortic pulmonary artery systolic blood flow acceleration time, and the ratio of fetal aortic pulmonary artery systolic blood flow acceleration time to ejection time.

[0039] In one embodiment of the present invention, step S3 specifically includes the following steps: Step S301: Perform norm normalization preprocessing on the delineated region of interest to perform preliminary extraction of radiomics features; Step S302: Construct an FEL autoencoder to perform initial screening on the preliminarily extracted radiomics features to obtain 128 radiomics features; wherein, the FEL autoencoder includes an input layer, an encoder, a decoder and an output layer connected in sequence.

[0040] In one embodiment of the present invention, the encoder is a feedforward neural network, comprising a first fully connected layer, a first Dropout layer, a second fully connected layer, a second Dropout layer, and a third fully connected layer connected sequentially; the mathematical expression corresponding to the calculation process of the encoder is:

[0041] h = f(W) e x+b e (1)

[0042] In equation (1), x is the manually drawn region of interest, and the dimension of x is d; x is compressed into a latent space, which is represented as h, and the dimension of h is k, then k <d;W e This is the encoder's weight matrix; b e f is the encoder's bias vector; f is the activation function.

[0043] In one embodiment of the present invention, the decoder is also a feedforward neural network, comprising a fourth fully connected layer, a third Dropout layer, a fifth fully connected layer, and a fourth Dropout layer connected in sequence; the mathematical expression corresponding to the calculation process of the decoder is:

[0044]

[0045] In equation (2), W d This is the weight matrix of the decoder; b d is the bias vector of the decoder; g is the activation function of the decoder; It is the reconstructed input data.

[0046] In one embodiment of the present invention, the loss function of the FEL autoencoder is a cross-entropy loss function, which is used to measure the difference between the original input data x and the reconstructed input data. The differences between them.

[0047] In one embodiment of the present invention, step S4 specifically includes the following steps: Step S401: ANOVA F test is used to perform secondary screening on the radiomics features initially selected in step S3 to determine the features of the fetal lung region of interest that are linearly correlated with gestational age and select the top 5% of features; Step S402: Based on the recursive feature elimination method of DynamicRFECV, the top 5% of features are subjected to dimensionality reduction processing to determine the radiomics features used for establishing the fetal lung maturity recognition model.

[0048] In one embodiment of the present invention, the technical solution of the present invention further provides a fetal lung maturity recognition model, which is a fetal lung maturity recognition model established by using the method for establishing a fetal lung maturity recognition model based on ultrasonic imaging omics in any of the above embodiments.

[0049] The following will show the technical solution of the present invention with a specific embodiment. As Figure 2 shown, the implementation steps of the method for establishing a fetal lung maturity recognition model based on ultrasonic imaging omics in this specific embodiment of the present invention are as follows:

[0050] 1. Set inclusion and exclusion criteria

[0051] Instrument: Use a GE Voluson E10 ultrasonic diagnostic instrument, with a probe model C6-1 and a frequency of 1-6 MHZ, to collect two-dimensional ultrasonic images of the fetal lungs in the four-chamber heart cross-section of fetuses of normal pregnant women from 20 weeks to 40 +6 weeks. Among them, 40 +6 weeks means 40 weeks plus 6 days.

[0052] Inclusion criteria for pregnant women: ① After detailed inquiry of medical history, physical examination, routine electrocardiogram, blood sugar, and blood pressure, there are no abnormalities; ② Singleton pregnancy, with a clear last menstrual period; ③ The fetus has no known congenital malformations or chromosomal abnormalities; ④ Women have not taken steroids before childbirth. Exclusion criteria: ① Oligohydramnios; ② Respiratory system diseases after childbirth.

[0053] 2. Obtain clinical data

[0054] Measure the ultrasonic data of the lungs of fetuses from 20 weeks to 40 +6 weeks of normal fetuses. The measured ultrasonic image data of the lungs includes one or a combination of the following: including: fetal weight (abbreviated as EFW), fetal chest circumference (abbreviated as TC), fetal abdominal circumference (abbreviated as AC), fetal bilateral lung volume (abbreviated as FLV), fetal main pulmonary artery pulsatility index (abbreviated as PI), fetal main pulmonary artery resistance index (abbreviated as RI), fetal main pulmonary artery valve peak systolic velocity (abbreviated as PSV), fetal main pulmonary artery systolic blood flow acceleration time (abbreviated as AT), and the ratio of fetal main pulmonary artery systolic blood flow acceleration time to ejection time (abbreviated as AT / ET).

[0055] 3. Obtain image data (in practice, directly input the obtained images), draw the region of interest and perform feature screening;

[0056] (1) Acquiring image data: Select the transverse section of the fetal four-chamber view, observe the transverse section of the fetal chest, and adjust the probe to ensure that there is no obvious fetal rib acoustic shadow in at least one lung. To obtain the best image quality, adjust the settings according to the relevant characteristics of each pregnant woman and fetus, including depth, gain, frequency, time gain compensation, and harmonics for obese patients.

[0057] (2) Manual delineation of regions of interest (ROIs) from the acquired imaging data: Two sonographers manually delineate the outlines of the bilateral lung parenchyma (ROIs), avoiding the ribs and major blood vessels. If the physicians disagree, another senior sonographer (with over ten years of experience) participates in the discussion until a consensus is reached. After obtaining the ROIs, interpolation processing is performed on the ROI data; for missing data, the average value is used as the difference.

[0058] (3) For ROI, perform radiomics feature extraction and screening;

[0059] Specifically, ① the outlined ROI is preprocessed by norm normalization;

[0060] More specifically, the norm normalization process is as follows: divide each element of each feature dimension in all samples by the L2 norm of that feature to make it have a unit norm.

[0061] For a given feature X∈R N×d Where N is the total number of samples, d is the feature dimension, and norm normalization is applied to each column of features x∈R. N Perform the following operations:

[0062]

[0063] Where 1≤i≤N, norm(x) is the norm of x.

[0064] The L2 norm value of feature X is

[0065]

[0066] The delineated ROIs were divided into training and validation sets (8:2). Features were extracted from the ROIs using an open-source Python package, resulting in 1130 radiomics features (5 clinical features + 1125 radiomics features). These features were further filtered to obtain the most effective features for classification modeling, thereby optimizing model performance.

[0067] ② Perform feature filtering

[0068] In this specific embodiment, a FEL auto-encoder (FELAutoEncode, hereinafter referred to as FEL auto-encoder) was constructed to train features, so as to extract effective features from 1130 radiomics features, perform feature screening, and finally extract 128 features.

[0069] The FEL auto-encoder includes an input layer, an encoder, a decoder, and an output layer; assuming that the manually delineated ROI is x, the dimension of x is d, and it is desired to compress it into a latent space representation h, the dimension of h is k, then k < d; then an encoder Encoder needs to be constructed. The encoder Encoder is a feed-forward neural network, and the mathematical expression corresponding to the calculation process of this encoder Encoder is:

[0070] h = f(W e x + b e ) (1)

[0071] In formula (1), W e is the weight matrix of the encoder; b e is the bias vector of the encoder; f is the activation function.

[0072] The encoder in this specific embodiment includes a first fully connected layer, a first Dropout layer, a second fully connected layer, a second Dropout layer, and a third fully connected layer connected in sequence; the network features of each layer are shown in Table 1 below.

[0073] Corresponding to the encoder, a decoder is required, and the decoder is also a feed-forward neural network. The mathematical expression corresponding to the calculation process of this decoder is:

[0074]

[0075] In formula (2), W d is the weight matrix of the decoder; b d is the bias vector of the decoder; g is the activation function of the decoder; is the reconstructed input data.

[0076] The decoder in this specific embodiment includes a fourth fully connected layer, a third Dropout layer, a fifth fully connected layer, and a fourth Dropout layer connected in sequence; a fully connected layer is connected to the last Dropout layer (i.e., the fourth Dropout layer) of the decoder as the output layer, and the network features of each layer are shown in Table 1 below.

[0077] Table 1 Network features of the auto-encoder

[0078]

[0079] The goal of an autoencoder is to minimize the reconstruction error, which is the difference between the original input data x and the reconstructed data. The difference lies in the fact that the autoencoder uses the cross-entropy loss function in this specific embodiment.

[0080] After determining the encoder and decoder, the entire autoencoder needs to be trained. Training the autoencoder involves minimizing the aforementioned loss function. This is typically accomplished using gradient descent, with the following steps:

[0081] a. Initialize the parameters of the encoder and decoder;

[0082] b. For each manually drawn ROI, the implicit representation h is calculated by the encoder;

[0083] c. Use h as the input to the decoder, and use the decoder to calculate the reconstructed input data.

[0084] d. Calculate the loss function

[0085] e. Calculate the gradient of the loss with respect to all parameters using the backpropagation algorithm;

[0086] f. Update parameters to reduce loss, which can be done through gradient descent or other optimization algorithms (such as Adam);

[0087] g. Repeat steps b to f until convergence or the predetermined number of iterations is reached.

[0088] The advantage of using the FEL autoencoder constructed in this invention for feature selection lies in achieving dimensionality reduction by mapping the features of manually drawn ROIs to a lower-dimensional latent space. This helps remove redundant features and reduce computational complexity. It also has a certain degree of robustness to noise and outliers, resulting in more reliable feature representations that are more suitable for subsequent analysis tasks. This step yields 128 features.

[0089] 4. The ANOVA F test was used to further screen 128 radiomic features to identify fetal lung ROIs that were linearly correlated with gestational age and select the top 5% of features.

[0090] Specifically, using the ANOVA F-test, by calculating the variance F-value, features of fetal lung ROIs linearly correlated with gestational age were identified, and the top 5% of features were selected, totaling 51 features. For example... Figure 3 As shown, Figure 3The radiomics features of 51 fetal lung ROIs that showed a linear correlation with gestational age, as determined by the ANOVA F test, are shown. The names of the 51 selected radiomics features are also presented.

[0091] More specifically, the method for calculating the variance F-value is as follows:

[0092] Pregnant women at different gestational weeks are selected, and each gestational week category is considered a group. The pregnant women within each group are represented as individual samples. Let k be the number of groups, and nj be the sample size of the j-th group. It is the sample mean of the j-th group. It is the overall mean of all data, SS B SS represents the sum of squares between groups. w Let the sum of squares within a group be represented, then:

[0093]

[0094] Total Sum of Squares (SS) T Represented as:

[0095] SS T =SS B +SS W

[0096] The formula for calculating the ANOVA F-statistic is:

[0097]

[0098] in, It is the mean square between groups; is the within-group mean square, and N is the total number of all observations (all training samples).

[0099] 5. The top 5% of features were further dimensionality-reduced to determine the radiomics features used for fetal lung maturity identification and classification modeling.

[0100] Specifically, the top 5% of features are further dimensionality-reduced, and the recursive feature elimination method of Dynamic RFECV (Dynamic Recursive Feature Elimination with Cross-Validation) is applied to select the 51 features obtained in step 4: ① All features are evaluated and ranked using a gradient boosting tree (GBDT) model; ② Features are removed with a predetermined step size of 1; ③ After removing some features, the model is retrained using the remaining features, and the model's performance is evaluated using cross-validation; ④ The learning rate is dynamically adjusted based on the model's performance; ⑤ The above steps are iterated until a preset minimum number of features (12) is reached or the performance improvement is no longer significant, in order to obtain the optimal radiomics features.

[0101] The basic updates to the Gradient Boosting Tree (GBDT) model are as follows:

[0102] F m (x)=F m-1 (x)+η1.h m (x)

[0103] In the formula, F m (x) represents the model's prediction of input x after the m-th iteration; F m-1 (x) represents the predicted value at the (m-1)th iteration; η1 is the learning rate of the gradient boosting tree GBDT model, between 0 and 1, used to reduce the weights of each tree. m (x) is the output of the m-th tree (i.e. the m-th weak learner) for input x, which is constructed based on the residual (i.e. the negative gradient) of the current model.

[0104] The model's performance is evaluated using cross-validation, and the learning rate is dynamically adjusted. If the score fluctuation is below 0.1 for three consecutive iterations, the learning rate is divided by 5, and the next iteration is performed to accelerate model convergence. It should be noted that the recursive feature elimination method of Dynamic RFECV is an existing technology, and the basic update formula of the gradient boosting tree GBDT model described above only illustrates the principle of model update; the detailed process of selecting the 51 features is not elaborated here.

[0105] In this specific embodiment, the final model converges to 12 radiomics features used for establishing a fetal lung maturity recognition model, such as... Figure 4 As shown, it specifically includes:

[0106] wavelet-HL_firstorder_RobustMeanAbsoluteDeviation_us_paramsName1

[0107] gradient_firstorder_Energy_us_paramsName1

[0108] wavelet-HL_gldm_GrayLevelNonUniformity_us_paramsName1

[0109] wavelet-HH_firstorder_10Percentile_us_paramsNamel

[0110] lbp-2D_firstorder_Energy_us_paramsName1

[0111] wavelet-LL_glszm_LargeAreaHighGrayLevelEmphasis_us_paramsName1

[0112] original_glszm_LargeAreaHighGrayLevelEmphasis_us_paramsName1

[0113] wavelet-HL_gldm_DependenceNonUniformity_us_paramsNamel

[0114] wavelet-LL_glszm_GrayLevelNonUniformity_us_paramsNamel

[0115] squareroot_glszm_LargeAreaHighGrayLevelEmphasis_us_paramsName1

[0116] gradient_firstorder_TotalEnergy_us_paramsNamel

[0117] lbp-2D_firstorder_TotalEnergy_us_paramsNamel.

[0118] 6. Using the 12 radiomics features obtained in step 5 as input and the gestational age of the pregnant woman as the label as the output, train the AdeptiveXGBoost classifier, save the trained AdeptiveXGBoost classifier, and obtain the recognition model of fetal lung maturity corresponding to gestational age.

[0119] Specifically, the basic update formula for the AdeptiveXGBoost classifier can be expressed as:

[0120]

[0121] in, This represents the predicted value for the i-th sample after the t-th iteration; This represents the predicted value for the i-th sample after the (t-1)-th iteration, where η² is the learning rate and f t (x i ) is the decision tree (weak learner) added in the t-th iteration for sample x. i The predicted value.

[0122] Furthermore, in the optimization objective function of AdaptiveXGBoost, the learning rate also affects the rate of descent of the objective function, i.e.

[0123]

[0124] in, It is the loss function, Ω(f) t ) is the regularization term, and λ is the regularization strength.

[0125] It should be noted that the update principle of the AdeptiveXGBoost classifier is also an existing technology. In this specific embodiment, the detailed classification process of the 12 radiomics features will not be described in detail.

[0126] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for establishing a fetal lung maturity identification model based on ultrasound radiomics, characterized in that, include: Step S1: Collect samples from individuals aged 20 to 40 weeks of gestation. +6 Ultrasound images of the lungs of a normal fetus at 40 weeks of gestation; of which, 40 +6 A week means 40 weeks plus 6 days; Step S2: Obtain corresponding image data based on the lung ultrasound image data, and delineate the region of interest in each lung ultrasound image in the image data; Step S3: Perform initial screening of radiomics features on the outlined regions of interest; Step S4: Use the ANOVAF test to perform a second screening of the radiomics features selected in the first screening, and then perform dimensionality reduction on the features obtained after the second screening to determine the radiomics features used to establish the fetal lung maturity recognition model. Step S5: Using the radiomics features used to establish the fetal lung maturity recognition model as input and the gestational age of the pregnant woman as the label as the output, train the AdaptiveXGBoost classifier and save the trained AdaptiveXGBoost classifier to obtain the fetal lung maturity recognition model; wherein, the fetal lung maturity recognition model realizes the recognition of the gestational age of the pregnant woman corresponding to the fetal lung maturity. Step S3 includes the following steps: Step S301: Perform norm normalization preprocessing on the delineated region of interest to perform preliminary extraction of radiomics features; Step S302: Construct an FEL autoencoder to perform initial screening on the preliminarily extracted radiomics features, resulting in 128 radiomics features; The FEL automatic encoder includes an input layer, an encoder, a decoder, and an output layer connected in sequence. The encoder is a feedforward neural network, comprising a first fully connected layer, a first Dropout layer, a second fully connected layer, a second Dropout layer, and a third fully connected layer connected sequentially; the mathematical expression corresponding to the encoder's calculation process is: h=f(W e x+b e )(1) In equation (1), x is the manually drawn region of interest, and the dimension of x is d; x is compressed into a latent space, which is represented as h, and the dimension of h is k, then k <d;W e This is the encoder's weight matrix; b e is the encoder's bias vector; f is the activation function; The decoder is also a feedforward neural network, comprising a fourth fully connected layer, a third Dropout layer, a fifth fully connected layer, and a fourth Dropout layer connected in sequence; the mathematical expression corresponding to the calculation process of the decoder is: In equation (2), W d This is the weight matrix of the decoder; b d is the bias vector of the decoder; g is the activation function of the decoder; It is the reconstructed input data; The loss function of the FEL autoencoder is the cross-entropy loss function, which is used to measure the difference between the original input data x and the reconstructed input data. The differences between them; Step S4 includes the following steps: Step S401: Use the ANOVAF test to perform a second screening of the radiomics features initially selected in step S3, in order to identify features of the fetal lung region of interest that are linearly correlated with gestational age and select the top 5% of features. Step S402: Based on the recursive feature elimination method of Dynamic RFECV, the top 5% of the obtained features are dimensionality reduced to determine the radiomics features used for establishing the fetal lung maturity recognition model. Step S402 specifically includes: Step S4021: Evaluate and rank all the top 5% of features obtained using the Gradient Boosting Tree (GBDT) model; the update formula for the Gradient Boosting Tree (GBDT) model is as follows: F m (x)=F m-1 (x)+η1.h m (x) In the formula, F m (x) represents the prediction value of the gradient boosting tree GBDT model for input x after the m-th iteration; F m-1 (x) represents the predicted value at the (m-1)th iteration; η1 is the learning rate of the gradient boosting tree GBDT model, between 0 and 1, used to reduce the weights of each tree; h m (x) is the output of the m-th tree for input x; Step S4022: Remove part of the feature by a predetermined step size of 1; Step S4023: After removing some features, retrain the gradient boosting tree GBDT model using the remaining features, and evaluate the performance of the gradient boosting tree GBDT model through cross-validation. Step S4024: Dynamically adjust the learning rate η1 based on the performance of the gradient boosting tree GBDT model; Step S4025: Iterate through steps S4021 to S4024 until the radiomics features used to establish the fetal lung maturity recognition model reach the preset minimum number of features or the performance of the gradient boosting tree GBDT model no longer improves, so as to obtain the optimal radiomics features used to establish the fetal lung maturity recognition model. The update formula for the AdaptiveXGBoost classifier is as follows: in, This represents the predicted value for the i-th sample after the t-th iteration; This represents the predicted value for the i-th sample after the (t-1)-th iteration; η² is the learning rate; f t (x i ) is the decision tree pair for sample x added in the t-th iteration. i The predicted value; The objective function for optimizing the AdaptiveXGBoost classifier is: in, It is the loss function; Ω(f) t ) is the regularization term; λ is the regularization strength.

2. The method for establishing a fetal lung maturity identification model based on ultrasound radiomics according to claim 1, characterized in that, The lung ultrasound image data mentioned in step S1 includes one or a combination of the following: fetal weight, fetal chest circumference, fetal abdominal circumference, fetal bilateral lung volume, fetal aortic pulmonary artery pulsatility index, fetal aortic pulmonary artery resistance index, fetal aortic pulmonary artery valve peak systolic velocity, fetal aortic pulmonary artery systolic blood flow acceleration time, and the ratio of fetal aortic pulmonary artery systolic blood flow acceleration time to ejection time.

3. A fetal lung maturity recognition model, characterized in that, The identification model is a fetal lung maturity identification model established using the method for establishing a fetal lung maturity identification model based on ultrasound radiomics as described in any one of claims 1 or 2.

Images