A method for predicting fetal growth restriction, electronic device and storage medium

Abstract

The present invention discloses a method for predicting fetal growth restriction, an electronic device, and a storage medium. The method comprises: obtaining an ultrasonic image of the current placenta of a pregnant woman; inputting the ultrasonic image of the placenta of the pregnant woman into a placental function status determination model constructed based on a densely connected network to obtain a determination result of the current placental function status of the pregnant woman; wherein the determination result includes placental dysfunction and normal placental function; obtaining a prediction result of fetal growth restriction based on the determination result; if the determination result is placental dysfunction, outputting a prediction result of fetal growth restriction; if the determination result is normal placental function, outputting a prediction result of normal fetal growth. The placental function status determination model obtained through training in this application identifies the placental function status of the current placenta of a pregnant woman, improves the accuracy of the qualitative characterization of placental dysfunction, and thus improves the accuracy and timeliness of the prediction of fetal growth restriction.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method for predicting fetal growth restriction based on a densely connected network, an electronic device, and a computer-readable storage medium. Background Art

[0002] Fetal growth restriction (FGR) refers to the failure of a fetus to grow to its genetic potential due to maternal, fetal, and placental pathological factors. It is the second leading cause of perinatal mortality.

[0003] Placental factors are the primary cause of FGR. The placenta is a vital organ for maintaining pregnancy and fetal growth and development. Abnormal development of placental blood vessels and impaired perfusion lead to placental dysfunction (caused by uterine, placental, and fetal blood circulation disorders) and are the direct cause and primary factor in the loss of fetal growth potential and intrauterine growth restriction.

[0004] Clinically, the main challenges in the diagnosis and treatment of FGR are the difficulties in early prediction and diagnosis, as well as the difficulty in accurately assessing the degree of intrauterine ischemia and hypoxia. The placenta is a crucial organ for maintaining fetal growth and development in the uterus. Only by early assessment and accurate, real-time dynamic monitoring of placental dysfunction can we provide more opportunities for intervention to prevent the development and progression of fetal growth restriction and its adverse prognosis.

[0005] Lesions in different parts of the placenta can lead to placental dysfunction. The causes and pathological basis of lesions in different parts of the placenta are different, and lesions in each part of the placenta show unique ultrasound signs. Changes in the echogenicity of the placenta (i.e., abnormal echoes within the parenchyma) mainly include local parenchymal echogenicity changes such as placental infarction and hemorrhage caused by impaired placental blood perfusion, and diffuse parenchymal echogenicity changes caused by villous degeneration. The typical ultrasound manifestations of local parenchymal echogenicity changes are high echogenicity at the base of the placenta and echo-free areas where no blood flow signals are detected in the placenta; the main ultrasound manifestations of diffuse echogenicity changes are a significant decrease in the echogenicity of the placenta.

[0006] However, to date, from the perspective of ultrasound imaging, there is still a lack of technical means to accurately define abnormal echoes in the placental parenchyma that reflect placental dysfunction, and human intervention is required to determine them. There are subjective differences in human ultrasound image interpretation, and some abnormal echoes can only be identified by experienced high-level ultrasound physicians. Not only is it difficult to ensure the accuracy of the qualitative characterization of placental dysfunction, but the difficulty in qualitative determination also leads to missing the best opportunity for intervention for fetal growth restriction. Summary of the Invention

[0007] Embodiments of the present invention provide a method for predicting fetal growth restriction based on a densely connected network, an electronic device, and a computer-readable storage medium, which are used to solve the problem of difficulty and inaccuracy in qualitatively determining placental dysfunction due to subjective differences and abnormal echo changes that require identification by experienced high-level ultrasound physicians when manually identifying the functional status of the placenta from ultrasound images of the placenta.

[0008] The specific technical solutions are as follows:

[0009] In a first aspect, an embodiment of the present invention provides a method for predicting fetal growth restriction based on a densely connected network, comprising the following steps:

[0010] Obtain ultrasound images of the pregnant woman's current placenta;

[0011] The ultrasound image of the pregnant woman's placenta is input into a placental function status determination model constructed based on a densely connected network to obtain a determination result of the current placental function status of the pregnant woman; wherein the determination result includes placental dysfunction and normal placental function; the placental function status determination model uses a linear classification layer, and the activation function ReLU is as follows:

[0012]

[0013] Wherein, x is the pixel matrix value of the ultrasound image of the placenta;

[0014] A prediction result of fetal growth restriction is obtained based on the determination result; if the determination result is placental dysfunction, the output prediction result is fetal growth restriction; if the determination result is normal placental function, the output prediction result is normal fetal growth.

[0015] Optionally, training the placenta function status determination model constructed by the densely connected network comprises the following steps:

[0016] A large number of collected ultrasound images of pregnant women's placentas are labeled with information about the functional status of the placenta to obtain a dataset of placental ultrasound images; the labels include poor placental function and normal placental function;

[0017] Constructing a densely connected deep neural network model and setting an activation function ReLU for the deep neural network model using a linear classification layer;

[0018] The placental ultrasound image dataset is used to perform K-fold cross-validation training on the deep neural network model of the densely connected network to obtain the placental functional status determination model.

[0019] Optionally, the method of using a dataset of placental ultrasound images to perform cross-validation training on a densely connected network deep neural network model using K-fold to obtain the placental function status determination model includes:

[0020] The data set of placental ultrasound images is divided into K subsets, and then trained according to a preset batch size, number of training cycles, and learning rate; wherein, in each iteration, K-1 subsets are used as a training data set for training, and the remaining subset is used as a test data set for verification, and all the placental ultrasound images in the training data set are input into the deep neural network model for training to obtain a deep neural network model trained by the training data set of this iteration; the test data set is input into the deep neural network model obtained by this iterative training to obtain prediction results of data samples in the test data set; and a performance evaluation index is calculated based on the actual labels of the samples in the test data set and the prediction results;

[0021] Record the deep neural network model and corresponding performance evaluation indicators obtained by training the iterative training dataset obtained in each epoch;

[0022] According to the performance evaluation indicators of the deep neural network model obtained by each epoch training, the deep neural network model obtained by the epoch training with the best performance is obtained as the placenta functional status determination model.

[0023] Optionally, the performance evaluation index of the deep neural network model obtained by each epoch training is used as the deep neural network model for determining the placental function status, specifically:

[0024] After each epoch, each performance evaluation indicator of the epoch is calculated. When the performance evaluation indicator reaches the minimum threshold, and when the F1 score of the current epoch exceeds that of the previous epoch, the deep neural network model trained in the current epoch is saved to a specified file.

[0025] When all training cycles are completed, the deep neural network model saved in the specified file is used as the placenta function status determination model;

[0026] The performance evaluation indicators include test loss value, accuracy, precision, F1 score, recall rate and ROCAUC.

[0027] Optionally, each performance evaluation indicator of the deep neural network model obtained by each epoch training is the average value of the corresponding performance evaluation indicator of all iterations in the epoch.

[0028] Optionally, the ultrasound images of the placenta in the training dataset and the test dataset are loaded from a specified directory to the deep neural network model through the load_data function; and the total number of ultrasound images of the placenta in the training dataset, the number of ultrasound images of the placenta labeled as placental dysfunction, and the number of ultrasound images of the placenta labeled as normal placental function are counted, as well as the total number of ultrasound images of the placenta in the test dataset, the number of ultrasound images of the placenta predicted to be placental dysfunction, and the number of ultrasound images of the placenta predicted to be normal placental function are counted.

[0029] Optionally, the loss value is calculated using a cross entropy function, which is as follows:

[0030]

[0031] Where n is the number of ultrasound images of the placenta in the test dataset; p(x i ) is the label of the ultrasound image of the i-th placenta; q(x i ) is the predicted result of the ultrasound image label of the i-th placenta.

[0032] In a second aspect, an embodiment of the present invention further provides an electronic device, comprising a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus;

[0033] Memory for storing computer programs;

[0034] The processor is configured to implement the steps of the method for predicting fetal growth restriction based on a densely connected network as described in the first aspect when executing the program stored in the memory.

[0035] In a third aspect, an embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of a method for predicting fetal growth restriction based on a densely connected network as described in the first aspect above are implemented.

[0036] The embodiment of the present application provides a prediction scheme for fetal growth restriction based on a densely connected network, constructs a deep neural network model of the densely connected network, and the deep neural network model adopts a linear classification layer and an activation function ReLU, and obtains a placental function status determination model that can identify the placental function status from the ultrasound image of the prenatal placenta by training the deep neural network model, and identifies whether the current placental function status of the pregnant woman's placenta is placental dysfunction or normal placental function, avoiding errors caused by subjective differences in the determination of placental function status, and having low experience requirements for the prediction operator, thereby improving the accuracy of the qualitative characterization of placental dysfunction, thereby achieving highly accurate and timely prediction of fetal growth restriction through complete technical means, providing a basis for intervening in fetal growth restriction to avoid adverse outcomes of newborns and perinatal infants.

[0037] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purposes and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings.

[0038] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0040] Figure 1 A schematic flow chart of a method for predicting fetal growth restriction based on a densely connected network provided in an embodiment of the present application;

[0041] Figure 2 A schematic diagram of a process for training a placenta function status determination model in an embodiment of the present application;

[0042] Figure 3 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0044] This application uses an artificial intelligence model to classify ultrasound images of the placenta according to the placental functional status (including poor and normal placental functional status), to achieve accurate identification of the overall picture of the intrauterine placenta, especially the changes in the placental parenchymal echo, and to clarify abnormal placental echoes in ultrasound images, thereby achieving accurate prediction of fetal growth restriction, achieving timely clinical intervention for fetal growth restriction before birth and more rigorous monitoring effects, and reducing adverse perinatal outcomes.

[0045] This application provides a method for predicting fetal growth restriction based on a densely connected network, such as Figure 1 As shown, the following steps are included:

[0046] Step S100: Acquire an ultrasound image of the pregnant woman's current placenta.

[0047] Step S200: Input the ultrasound image of the pregnant woman's placenta into the placental function status determination model constructed based on the densely connected network to obtain the determination result of the current placental function status of the pregnant woman; wherein the determination result of the placental function status includes poor placental function and normal placental function.

[0048] Step S300: Obtain a prediction result of fetal growth restriction based on the determination result of the current placental function status of the pregnant woman; if the determination result of the placental function status is placental dysfunction, output the prediction result as fetal growth restriction; if the determination result of the placental function status is normal placental function, output the prediction result as normal fetal growth.

[0049] The embodiment of the present application constructs a deep neural network model of a densely connected network, and the deep neural network model adopts a linear classification layer and an activation function ReLU, and obtains a placenta function status determination model that can identify the placenta function status from the ultrasound image of the prenatal placenta by training the deep neural network model, and identify whether the placenta function status of the pregnant woman's current placenta is placenta dysfunction or placenta normal function, thereby avoiding errors caused by subjective differences in the determination of placenta function status, reducing the experience requirements for the prediction of the operator, and improving the accuracy of the qualitative characterization of placenta dysfunction, thereby achieving highly accurate and timely prediction of fetal growth restriction through complete technical means, providing a basis for intervening in fetal growth restriction to avoid adverse outcomes of newborns and perinatal infants.

[0050] In the process of training the deep neural network model of the densely connected network in the embodiment of the present application, a deterministic setting is used to avoid randomness in the training process, the stability of the model is evaluated by cross-validation, and overfitting is avoided by saving the best model of each fold. The model evaluation indicators are comprehensive, including precision, recall rate, F1 score and ROCAUC, thereby fully ensuring the performance of the trained placental functional status determination model.

[0051] In the embodiments of this application, Figure 2 As shown, training a placenta function status determination model includes the following steps:

[0052] Step S201: labeling a large number of collected ultrasound images of pregnant women's placentas with respect to the functional status of the placenta to obtain a data set of ultrasound images of the placenta; wherein the labeled labels include placental dysfunction and normal placental function; in this embodiment, the following steps are further included before step S201:

[0053] Collect ultrasound images of the placenta at different stages of antenatal care and the corresponding postpartum outcomes of a large number of pregnant women. Among them, the postpartum outcomes of these pregnant women with collected ultrasound images include adverse outcomes and normal outcomes; adverse outcomes include neonatal (including perinatal) death and neonatal weight far below normal levels;

[0054] Experienced medical experts identify abnormal echogenicity in each ultrasound image of the placenta at different prenatal stages based on the postpartum outcomes of pregnant women, determine the functional status of the placenta, and classify them according to the functional status of the placenta. Ultrasound images of a large number of pregnant women's placentas are collected. In this step, the functional status of the placenta in each ultrasound image of the placenta at different prenatal stages of pregnant women with adverse postpartum outcomes is not necessarily poor; some may have normal placental function.

[0055] Step S202: Construct a densely connected deep neural network model, and set an activation function ReLU for the deep neural network model using a linear classification layer; the present application implements a deep neural network model for the task of identifying abnormal (bad) placenta in ultrasound images of pregnant women. Specifically, based on the number and visual characteristics of ultrasound images, a densely connected network (DenseNet) is used to mine the semantic information of ultrasound images, and a linear classification layer and an activation function ReLU are used to implement the task of identifying abnormal placenta in ultrasound images. The activation function ReLU is as follows:

[0056]

[0057] Wherein, x is the pixel matrix value of the ultrasound image of the placenta;

[0058] Step S203: Using the data set of the placental ultrasound image, the deep neural network model of the densely connected network is trained using K-fold cross-validation (such as 5-fold cross-validation with K being 5) to obtain a placental function status determination model; wherein the input of the placental function status determination model is the placental ultrasound image of the pregnant woman, and the output is the placental function status of the placenta in the placental ultrasound image; in this embodiment, step S203 is specifically:

[0059] The data set of placental ultrasound images is divided into K subsets, and then training is performed according to a preset batch size (batch_size), number of training cycles (num_epochs) and learning rate (learning_rate). For example, the specific training parameters are a batch size (batch_size) of 4; a number of training cycles (num_epochs) of 40; and a learning rate (learning_rate) of 5e-6. In each iteration, K-1 subsets are used for training the training data set, and the remaining subset is used for verification of the test data set. First, all the placental ultrasound images in the training data set are input into the deep neural network model of the densely connected network for training. Then, the test data set is input into the deep neural network model of the densely connected network obtained by training the training data set in this iteration to obtain the prediction results of the data samples in the test data set (placental dysfunction or normal placental function). Performance evaluation indicators are calculated based on the actual labels of the samples in the test data set and the prediction results. The performance evaluation indicators of this embodiment include the test loss value, accuracy, precision, F1 score, recall rate and ROCAUC, etc. To ensure the repeatability of the training results, the random seed is set to 42, and the CuDNN benchmark is disabled to ensure the consistency of the training results.

[0060] Record the deep neural network model of the densely connected network obtained by the iterative training data set training obtained in each epoch training and the corresponding performance evaluation indicators. In order to obtain the model results with the best performance in a timely manner, this embodiment calculates each performance evaluation indicator of the epoch after each epoch, especially the F1 score of the epoch exceeds that of the previous epoch, and saves the state of the model obtained by the current epoch training. In this way, the deep neural network model of the densely connected network obtained by training is the model with the best performance, and is used as the placental function status determination model. Among them, each performance evaluation indicator of the epoch is the average value of the corresponding performance evaluation indicator of all iterations within the epoch.

[0061] The training environment of the placenta function status determination model trained by the deep neural network model of the densely connected network is based on the anaconda virtual environment for training. The deep neural network model of the densely connected network is constructed based on PyTorch and its related modules. The data loading method and the processing scripts for the steps of training the deep neural network model of the densely connected network to obtain the placenta function status determination model, the deep neural network model architecture function of the densely connected network, the function of the performance evaluation index, etc. are customized to automatically and quickly complete the training process.

[0062] In an embodiment of the present application, the ultrasound image data of the placenta in the training data set and the test data set are loaded into the deep neural network model of the densely connected network from the specified directory through the load_data function, and the total number of ultrasound images of the placenta in the training data set, the number of ultrasound images of the placenta labeled as placental dysfunction, and the number of ultrasound images of the placenta labeled as normal placental function are counted, and the total number of ultrasound images of the placenta in the test data set, the number of ultrasound images of the placenta predicted to be placental dysfunction, and the number of ultrasound images of the placenta predicted to be normal placental function are counted.

[0063] In order to better handle the task of classifying placental ultrasound images, a preprocessing process of the placental ultrasound images is also included before the training dataset and the test dataset are input into the deep neural network model of the densely connected network; the preprocessing process of the placental ultrasound images includes data enhancement techniques such as center cropping, conversion to tensor, normalization, random horizontal flipping and random rotation of the placental ultrasound images.

[0064] In this embodiment, the densely connected deep neural network model uses the cross entropy function to calculate the loss value that measures the training results and the true value of the data. The cross entropy function is as follows:

[0065]

[0066] Where n is the number of ultrasound images of the placenta in the test dataset; p(x i ) is the label of the ultrasound image of the i-th placenta; q(x i ) is the predicted result of the ultrasound image label of the i-th placenta.

[0067] In this embodiment, the AdamW optimizer is used in the process of training the placenta function status determination model.

[0068] and Figure 1 Corresponding to the embodiment shown, the embodiment of the present invention further provides an electronic device, see Figure 3 , including a processor 501, a communication interface 502, a memory 503 and a communication bus 504, wherein the processor 501, the communication interface 502, and the memory 503 communicate with each other via the communication bus 504;

[0069] Memory 503, used for storing computer programs;

[0070] The processor 501 is configured to implement any of the steps of the method for predicting fetal growth restriction based on a densely connected network in the above embodiments when executing the program stored in the memory.

[0071] and Figure 1Corresponding to the embodiment shown, an embodiment of the present invention further provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, any step of the method for predicting fetal growth restriction based on a densely connected network in the above embodiments is implemented.

[0072] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0073] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A method for predicting fetal growth restriction based on a densely connected network, characterized in that: The following steps are involved: Obtain ultrasound images of the pregnant woman's current placenta; The ultrasound image of the pregnant woman's placenta is input into a placental function status determination model constructed based on a densely connected network to obtain a determination result of the current placental function status of the pregnant woman; wherein the determination result includes placental dysfunction and normal placental function; the placental function status determination model uses a linear classification layer, and the activation function ReLU is as follows: ； Wherein, x is the pixel matrix value of the ultrasound image of the placenta; Obtaining a prediction result of fetal growth restriction according to the determination result; if the determination result is placental dysfunction, outputting a prediction result of fetal growth restriction; if the determination result is normal placental function, outputting a prediction result of normal fetal growth; Training the placenta function status determination model constructed by the densely connected network includes the following steps: A large number of collected ultrasound images of pregnant women's placentas are labeled with information about the functional status of the placenta to obtain a dataset of placental ultrasound images; the labels include poor placental function and normal placental function; Constructing a densely connected deep neural network model and setting an activation function ReLU for the deep neural network model using a linear classification layer; Using a dataset of placental ultrasound images, a densely connected network deep neural network model is trained using K-fold cross validation to obtain the placental functional status determination model; The method of using a data set of placental ultrasound images to perform cross-validation training on a deep neural network model of a densely connected network using K-fold to obtain the placental function status determination model includes: The data set of placental ultrasound images is divided into K subsets, and then trained according to a preset batch size, number of training cycles, and learning rate; wherein, in each iteration, K-1 subsets are used as a training data set for training, and the remaining subset is used as a test data set for verification, and all the placental ultrasound images in the training data set are input into the deep neural network model for training to obtain a deep neural network model trained by the training data set of this iteration; the test data set is input into the deep neural network model obtained by this iterative training to obtain prediction results of data samples in the test data set; and a performance evaluation index is calculated based on the actual labels of the samples in the test data set and the prediction results; Record the deep neural network model and corresponding performance evaluation indicators obtained by training the iterative training dataset obtained in each epoch; According to the performance evaluation index of the deep neural network model obtained by each epoch training, the deep neural network model obtained by the epoch training with the best performance is used as the placenta function status determination model; The performance evaluation index of the deep neural network model obtained by each epoch training is used to obtain the deep neural network model with the best performance as the placenta function status determination model, specifically: After each epoch, each performance evaluation indicator of the epoch is calculated. When the performance evaluation indicator reaches the minimum threshold, and when the F1 score of the current epoch exceeds that of the previous epoch, the deep neural network model trained in the current epoch is saved to a specified file. When all training cycles are completed, the deep neural network model saved in the specified file is used as the placenta function status determination model; The performance evaluation indicators include test loss value, accuracy, precision, F1 score, recall rate and ROCAUC.

2. The method according to claim 1, characterized in that Each performance evaluation indicator of the deep neural network model obtained by each epoch training is the average value of the corresponding performance evaluation indicator of all iterations in the epoch.

3. The method according to claim 1, characterized in that The ultrasound images of the placenta in the training data set and the test data set are loaded from the specified directory to the deep neural network model through the load_data function; and the total number of ultrasound images of the placenta in the training data set, the number of ultrasound images of the placenta labeled as placental dysfunction, and the number of ultrasound images of the placenta labeled as normal placental function are counted, as well as the total number of ultrasound images of the placenta in the test data set, the number of ultrasound images of the placenta predicted to be placental dysfunction, and the number of ultrasound images of the placenta predicted to be normal placental function.

4. The method according to claim 1, wherein The cross entropy function is used to calculate the loss value. The cross entropy function is as follows: ； Where n is the number of ultrasound images of the placenta in the test dataset; p(x i ) is the label of the ultrasound image of the i-th placenta; q(x i ) is the predicted result of the ultrasound image label of the i-th placenta.

5. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; The memory is used to store computer programs; The processor is configured to implement the method steps described in any one of claims 1 to 4 when executing the program stored in the memory.

6. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method steps according to any one of claims 1 to 4 are implemented.

Images