Infertility ultrasonic image auxiliary analysis method, system and equipment and storage medium

Abstract

The invention discloses an infertility ultrasonic image auxiliary analysis method, system and device and a storage medium, and belongs to the technical field of auxiliary analysis, and the method comprises the steps: obtaining and preprocessing a pelvic cavity ultrasonic image data set, and obtaining a preprocessed image data set; performing image optimization processing on the preprocessed image data set to obtain a target image data set; constructing an infertility auxiliary analysis network model, and pre-training the infertility auxiliary analysis network model to obtain a pre-trained auxiliary analysis network model; optimizing the pre-trained auxiliary analysis network model to obtain an optimized auxiliary analysis network model; performing iterative training on the optimization auxiliary analysis network model according to the target image data set to obtain a target auxiliary analysis network model; inputting a pelvic ultrasonic image to be detected into the target auxiliary analysis network model, and outputting an infertility detection result; compared with a traditional method, whether the infertility exists or not can be analyzed more quickly and accurately.

Description

Technical Field

[0001] The present application belongs to the field of auxiliary analysis, and specifically relates to an infertility ultrasound image auxiliary analysis method, system, device and storage medium. Background Art

[0002] As a major challenge in the field of reproductive health, infertility has a complex diagnostic process and involves multiple examination methods. In recent years, with the rapid development of medical imaging technology, especially the popularization of ultrasound imaging technology, its application in the diagnosis of infertility has become increasingly widespread. There are many factors that lead to infertility. For male factors, it is usually a problem with male sperm quality, while for female factors, it may be ovarian ovulation disorders, fallopian tube damage, endometriosis, and pelvic effusion. The cause of female infertility is usually diagnosed through ultrasound.

[0003] At present, traditional ultrasound diagnosis relies on the doctor's clinical experience and visual judgment. This process is easily affected by subjective factors. The doctor's experience and ability have a great influence on the judgment, and there is room for improvement in diagnostic efficiency and accuracy.

[0004] Therefore, developing an auxiliary analysis method based on ultrasound imaging to achieve objective, rapid and accurate analysis of infertility has become a hot topic in current research. Summary of the invention

[0005] The purpose of the embodiments of the present application is to provide an infertility ultrasound image-assisted analysis method, system, device and storage medium, which can solve the problem of low accuracy of infertility analysis in the prior art due to reliance on the doctor's clinical experience and visual judgment.

[0006] In order to solve the above technical problems, this application is implemented as follows:

[0007] In a first aspect, an embodiment of the present application provides an ultrasound image-assisted analysis method for infertility, the method comprising:

[0008] Acquiring a pelvic ultrasound image dataset, and preprocessing the pelvic ultrasound image dataset to obtain a preprocessed image dataset;

[0009] Performing image optimization processing on the preprocessed image data set to obtain a target image data set;

[0010] Constructing an infertility auxiliary analysis network model, and pre-training the infertility auxiliary analysis network model according to the target image data set to obtain a pre-trained auxiliary analysis network model;

[0011] According to the predicted output of the pre-trained auxiliary analysis network model, the pre-trained auxiliary analysis network model is optimized to obtain an optimized auxiliary analysis network model;

[0012] Iteratively training the optimized auxiliary analysis network model according to the target image data set to obtain a target auxiliary analysis network model;

[0013] The pelvic ultrasound image to be detected after the preprocessing and the image optimization processing is input into the target auxiliary analysis network model for processing, and the infertility detection result is output.

[0014] As an optional implementation of the first aspect of the present application, the preprocessing of the pelvic ultrasound image dataset to obtain the preprocessed image dataset is specifically:

[0015] Performing Wiener filtering on the pelvic ultrasound image data set according to a Wiener filter to remove noise data in the pelvic ultrasound image data set;

[0016] Performing image annotation processing on the pelvic ultrasound image dataset from which noise data is removed according to the trained model to highlight the uterus, ovaries, fallopian tubes and pelvic effusion in the pelvic ultrasound image dataset;

[0017] Image correction is performed on the pelvic ultrasound image data set after image annotation processing to eliminate geometric distortion generated during image acquisition to obtain the preprocessed image data set.

[0018] As an optional implementation of the first aspect of the present application, the image optimization processing is performed on the pre-processed image data set to obtain the target image data set, specifically:

[0019] Performing feature extraction on each of the preprocessed images in the preprocessed image data set to obtain corresponding uterus extraction images, ovary extraction images, fallopian tube extraction images, and pelvic effusion extraction images;

[0020] Respectively enhancing the uterus extracted image, the ovary extracted image, the fallopian tube extracted image and the pelvic effusion extracted image to obtain a uterus enhanced image, an ovary enhanced image, a fallopian tube enhanced image and a pelvic effusion enhanced image;

[0021] Each of the preprocessed images is fused with the corresponding uterus enhanced image, ovary enhanced image, fallopian tube enhanced image and pelvic effusion enhanced image to obtain the target image data set.

[0022] As an optional implementation of the first aspect of the present application, the uterus extracted image, the ovary extracted image, the fallopian tube extracted image and the pelvic effusion extracted image are enhanced respectively to obtain a uterus enhanced image, an ovary enhanced image, a fallopian tube enhanced image and a pelvic effusion enhanced image, specifically:

[0023] Performing the enhancement processing on the uterus extraction image includes:

[0024] Performing multi-layer downsampling on the uterus extraction image to obtain a plurality of uterus sampling sub-images with successively decreasing resolutions;

[0025] Taking the uterus sampling sub-image with the highest resolution as the uterus initial image, performing edge enhancement processing on each of the uterus sampling sub-images other than the initial image to obtain a corresponding uterus enhanced sub-image;

[0026] Performing image fusion processing on all the uterus enhanced sub-images and the uterus initial image to obtain the uterus enhanced image;

[0027] The enhancement processing is performed on the ovary extracted image, the fallopian tube extracted image and the pelvic effusion extracted image respectively to obtain the corresponding ovary enhanced image, fallopian tube enhanced image and pelvic effusion enhanced image.

[0028] As an optional implementation of the first aspect of the present application, the pre-trained auxiliary analysis network model is optimized according to the predicted output of the pre-trained auxiliary analysis network model to obtain an optimized auxiliary analysis network model, specifically:

[0029] Establishing a BCE loss function according to the difference between the predicted output and the target image data set;

[0030] Constructing a target loss function according to the data distribution characteristics of the predicted output and the BCE loss function;

[0031] The pre-trained auxiliary analysis network model is optimized according to the target image data set and the target loss function to obtain the optimized auxiliary analysis network model.

[0032] As an optional implementation of the first aspect of the present application, the iterative training of the optimized auxiliary analysis network model according to the target image data set to obtain the target auxiliary analysis network model is specifically:

[0033] The target image data set is divided into S equal samples, and S-1 samples are randomly selected as training sets;

[0034] Sampling the target image data set N times to obtain N training sets;

[0035] The optimized auxiliary analysis network model is iteratively trained in turn according to the N training sets to obtain the target auxiliary analysis network model.

[0036] As an optional implementation of the first aspect of the present application, the target-assisted analysis network model includes a local feature extraction module, an adaptive feature fusion module and a classifier; the target-assisted analysis network model processes the pelvic ultrasound image to be detected, specifically:

[0037] The local feature extraction module performs feature extraction on the input pelvic ultrasound image to be detected through a basic convolution block to extract the initial features of the uterus, the initial features of the fallopian tube, the initial features of the ovary and the initial features of the pelvic effusion in the pelvic ultrasound image to be detected;

[0038] Performing deep extraction on the pelvic ultrasound image to be detected according to the residual convolution block in the local feature extraction module to extract uterine depth features, fallopian tube depth features, ovarian depth features and pelvic effusion depth features in the pelvic ultrasound image to be detected;

[0039] The local feature extraction module performs weighted fusion on the uterus initial feature and the uterus depth feature, the fallopian tube initial feature and the fallopian tube depth feature, the ovary initial feature and the ovary depth feature, and the pelvic effusion initial feature and the pelvic effusion depth feature according to multi-scale processing to obtain uterus target feature, fallopian tube target feature, ovary target feature and pelvic effusion target feature;

[0040] The adaptive feature fusion module assigns weights to the uterus target feature, the fallopian tube target feature, the ovary target feature and the pelvic effusion target feature respectively according to an adaptive attention mechanism;

[0041] The adaptive feature fusion module splices the weighted uterus target feature, fallopian tube target feature, ovary target feature and pelvic effusion target feature according to a splicing algorithm to obtain an ultrasound image target feature;

[0042] The classifier classifies the ultrasonic image target features to obtain the infertility test result.

[0043] In a second aspect, an embodiment of the present application provides an infertility ultrasound image-assisted analysis system, the system comprising:

[0044] An image acquisition module acquires a pelvic ultrasound image data set, and preprocesses the pelvic ultrasound image data set to obtain a preprocessed image data set;

[0045] An image optimization module, used for optimizing the pre-processed image data set output by the image acquisition module to obtain a target image data set;

[0046] Model building module, used to build an infertility auxiliary analysis network model;

[0047] A first training module is used to pre-train the infertility auxiliary analysis network model established by the model establishment module to obtain a pre-trained auxiliary analysis network model;

[0048] A model optimization module, which optimizes the pre-trained auxiliary analysis network model according to the predicted output of the pre-trained auxiliary analysis network model to obtain an optimized auxiliary analysis network model;

[0049] A second training module is used to iteratively train the optimized auxiliary analysis network model according to a preset training method to obtain a target auxiliary analysis network model;

[0050] The result output module is used to output the analysis result of the pelvic ultrasound image to be detected input by the user.

[0051] In a third aspect, an embodiment of the present application provides an electronic device, which includes a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps of the method described in the first aspect.

[0052] In a fourth aspect, an embodiment of the present application provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the method described in the first aspect are implemented.

[0053] In the embodiments of the present application, compared with the prior art, the following technical effects are achieved: by acquiring a pelvic ultrasound image data set and performing preprocessing (such as Wiener filter denoising, image annotation, and image correction), noise and geometric distortion can be significantly reduced, image quality can be improved, and thus the accuracy of analysis can be enhanced; by image optimization processing (such as feature extraction and enhancement), key areas such as the uterus, ovaries, fallopian tubes, and pelvic effusion can be made clearer, which helps doctors or artificial intelligence models to identify and analyze more accurately; by constructing an infertility auxiliary analysis network model and performing pre-training, optimization, and iterative training, the generalization ability and analysis accuracy of the model can be continuously improved; by optimizing the model using a target loss function, it helps to reduce prediction errors and improve the reliability of analysis results; the local feature extraction module in the target auxiliary analysis network model can Accurately extract the initial features and deep features of the uterus, fallopian tubes, ovaries and pelvic effusion; the adaptive feature fusion module can effectively fuse these features through the adaptive attention mechanism and splicing algorithm to form a more comprehensive and accurate ultrasound image target feature; by dividing the target image data set into multiple samples and performing random sampling training, it can simulate different data set conditions, improve the model's adaptability to different data, and enhance the model's robustness; the automated and intelligent infertility ultrasound image-assisted analysis method can reduce the time and cost of manual analysis and improve analysis efficiency; doctors can use the analysis results of the artificial intelligence model to make accurate analysis and treatment decisions more quickly; this technical solution provides a new method and idea for ultrasound image-assisted analysis of infertility, which will help promote the innovation and development of medical technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 is a flow chart of an infertility ultrasound image-assisted analysis method provided by some embodiments of the present application;

[0055] Figure 2 This is a structural diagram of a target-assisted analysis network model in an infertility ultrasound image-assisted analysis method provided in some embodiments of the present application. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0057] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here. In addition, the "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally represents that the objects associated with each other are in an "or" relationship.

[0058] In conjunction with the accompanying drawings, the following describes in detail an infertility ultrasound image-assisted analysis method, system, device and storage medium provided by the embodiments of the present application through specific embodiments and their application scenarios.

[0059] Example

[0060] An infertility ultrasound image-assisted analysis method, comprising:

[0061] S100: Acquire a pelvic ultrasound image dataset, and preprocess the pelvic ultrasound image dataset to obtain a preprocessed image dataset;

[0062] Furthermore, first, pelvic ultrasound images of a large number of patients are acquired through medical equipment as a pelvic ultrasound image dataset; these original image data may contain noise, blur or inconsistent information, and therefore need to be preprocessed to improve the quality and consistency of the images, thereby obtaining a preprocessed image dataset.

[0063] It should be noted that, in S100, the pelvic ultrasound image dataset is preprocessed to obtain a preprocessed image dataset, specifically:

[0064] S110: performing Wiener filtering on the pelvic ultrasound image dataset according to the Wiener filter to remove noise data in the pelvic ultrasound image dataset;

[0065] S120: performing image annotation processing on the pelvic ultrasound image dataset from which noise data is removed according to the trained model, so as to highlight the uterus, ovary, fallopian tube and pelvic effusion in the pelvic ultrasound image dataset;

[0066] S130: performing image correction on the pelvic ultrasound image dataset after image annotation processing to eliminate geometric distortion generated during the image acquisition process, thereby obtaining a preprocessed image dataset.

[0067] Furthermore, the Wiener filter is used to process the pelvic ultrasound image dataset; the Wiener filter is an adaptive filter that can filter according to the statistical characteristics of the signal and noise, effectively remove the noise in the image, and retain the detailed information of the image as much as possible; through the Wiener filter processing, the clarity of the pelvic ultrasound image can be significantly improved, providing better image quality for subsequent analysis and analysis. The trained model is used to perform image annotation processing on the pelvic ultrasound image dataset after noise removal; this model can be a segmentation model based on deep learning, which can automatically identify and annotate the key structures in the image; through image annotation processing, the key structures in the pelvis can be clearly displayed, providing strong support for subsequent analysis. Image correction is performed on the pelvic ultrasound image dataset after image annotation processing. Image correction can include operations such as rotation, scaling, and translation of the image to ensure that the structure in the image is consistent with the actual anatomical structure; through image correction, the geometric distortion in the image can be eliminated, and the accuracy and readability of the image can be improved; this is crucial for subsequent image analysis and analysis. Through this series of preprocessing steps, a high-quality preprocessed image dataset can be obtained, providing strong support for subsequent auxiliary analysis of infertility.

[0068] S200: performing image optimization processing on the preprocessed image data set to obtain a target image data set;

[0069] Furthermore, in order to further improve the quality of the image and make it more suitable for subsequent model training and analysis, the preprocessed image data set is subjected to image optimization processing.

[0070] It should be noted that, in S200, the pre-processed image data set is subjected to image optimization processing to obtain the target image data set, specifically:

[0071] S210: performing feature extraction on each preprocessed image in the preprocessed image data set to obtain corresponding uterus extraction images, ovary extraction images, fallopian tube extraction images, and pelvic effusion extraction images;

[0072] S220: performing enhancement processing on the uterus extracted image, the ovary extracted image, the fallopian tube extracted image, and the pelvic effusion extracted image respectively, so as to obtain a uterus enhanced image, an ovary enhanced image, a fallopian tube enhanced image, and a pelvic effusion enhanced image;

[0073] S230: Each pre-processed image is fused with the corresponding uterus enhanced image, ovary enhanced image, fallopian tube enhanced image and pelvic effusion enhanced image to obtain a target image data set.

[0074] Furthermore, each pre-processed image in the pre-processed image data set is processed using techniques such as image segmentation, edge detection or deep learning to extract the corresponding uterus extraction image, ovary extraction image, fallopian tube extraction image and pelvic effusion extraction image; through feature extraction, clear and accurate feature images of each key structure can be obtained, providing a basis for subsequent enhancement processing. Adaptive histogram equalization, contrast stretching, sharpening, image scaling or deep learning methods are used to enhance the uterus extraction image, ovary extraction image, fallopian tube extraction image and pelvic effusion extraction image respectively; these methods can be adaptively adjusted according to the local features of the image to achieve the best enhancement effect; through enhancement processing, clearer and higher contrast uterus enhancement images, ovary enhancement images, fallopian tube enhancement images and pelvic effusion enhancement images can be obtained. Image fusion techniques, such as weighted averaging, maximum fusion or deep learning, are used to fuse each preprocessed image with the corresponding uterine enhancement image, ovarian enhancement image, fallopian tube enhancement image and pelvic effusion enhancement image. During the fusion process, it is necessary to ensure the coordination between the enhancement effect of the key structures and the overall image. Through image fusion, a target image data set can be obtained that retains the original image information and highlights the key structures. These target image data sets will be more suitable for training and testing the infertility auxiliary analysis network model.

[0075] It should be noted that in S220, the uterus extracted image, the ovary extracted image, the fallopian tube extracted image and the pelvic effusion extracted image are enhanced respectively to obtain the uterus enhanced image, the ovary enhanced image, the fallopian tube enhanced image and the pelvic effusion enhanced image, specifically:

[0076] S221: performing multi-layer downsampling on the uterus extraction image to obtain a plurality of uterus sampling sub-images with successively decreasing resolutions;

[0077] S222: taking the uterus sampling sub-image with the highest resolution as the uterus initial image, and performing edge enhancement processing on each uterus sampling sub-image other than the initial image to obtain a corresponding uterus enhanced sub-image;

[0078] S223: performing image fusion processing on all uterine enhanced sub-images and the uterine initial image to obtain a uterine enhanced image;

[0079] S224: Similarly, the ovary extracted image, the fallopian tube extracted image and the pelvic effusion extracted image are respectively subjected to the enhancement processing steps of S221-S223 to obtain the corresponding ovary enhanced image, fallopian tube enhanced image and pelvic effusion enhanced image.

[0080] Furthermore, the uterus extraction image is subjected to multi-layer downsampling processing; downsampling can be achieved through image scaling technology, such as bilinear interpolation, bicubic interpolation or nearest neighbor interpolation; each layer of downsampling will generate a sampling sub-image with lower resolution; through multi-layer downsampling, a series of uterus sampling sub-images with different resolutions can be obtained, providing a basis for subsequent edge enhancement processing. The uterus sampling sub-image with the highest resolution is used as the uterus initial image. For each uterus sampling sub-image other than the initial image, an edge detection algorithm (such as Sobel operator, Canny edge detector, etc.) is used for edge enhancement processing; the edge detection algorithm can identify and highlight the edge information in the image; through edge enhancement processing, a series of uterus enhancement sub-images with more obvious edge information can be obtained. Image fusion technology (such as weighted average, maximum fusion, etc.) is used to fuse all uterus enhancement sub-images with the uterus initial image; during the fusion process, it is necessary to ensure that the image information of different resolutions can be seamlessly connected to generate a high-quality uterus enhancement image; through image fusion processing, a uterus enhancement image that retains the original image information and enhances the edge information can be obtained. Similarly, the above method is applied to the ovarian extracted image, the fallopian tube extracted image and the pelvic effusion extracted image; for the ovarian extracted image, the fallopian tube extracted image and the pelvic effusion extracted image, steps S221 to S223 are repeated respectively; that is, the feature image of each key structure is subjected to multi-layer downsampling, edge enhancement processing and image fusion processing to generate the corresponding ovarian enhanced image, fallopian tube enhanced image and pelvic effusion enhanced image.

[0081] S300: constructing an infertility auxiliary analysis network model, and pre-training the infertility auxiliary analysis network model according to a target image data set to obtain a pre-trained auxiliary analysis network model;

[0082] Furthermore, an infertility auxiliary analysis network model is constructed, which is a deep learning model; then, the model is pre-trained using the target image dataset so that the obtained pre-trained auxiliary analysis network model can initially learn the image features related to infertility.

[0083] S400: Optimizing the pre-trained auxiliary analysis network model according to the predicted output of the pre-trained auxiliary analysis network model to obtain an optimized auxiliary analysis network model;

[0084] Furthermore, after pre-training, the model is optimized according to the prediction output of the pre-training auxiliary analysis network model; the optimization step may include adjusting the hyperparameters of the model (such as learning rate, batch size, etc.), modifying the model structure, or adding regularization terms, etc., to improve the generalization ability and accuracy of the model.

[0085] It should be noted that, in S400, the pre-trained auxiliary analysis network model is optimized according to the predicted output of the pre-trained auxiliary analysis network model to obtain an optimized auxiliary analysis network model, specifically:

[0086] S410: Establishing a BCE loss function according to the difference between the predicted output and the target image data set;

[0087] S420: constructing a target loss function according to the data distribution characteristics of the predicted output and the BCE loss function;

[0088] S430: Optimizing the pre-trained auxiliary analysis network model according to the target image data set and the target loss function to obtain an optimized auxiliary analysis network model.

[0089] Furthermore, Binary Cross-Entropy (BCE) is used as the loss function; BCE loss function is often used for binary classification problems and can measure the difference between the predicted value and the true value; in this stage, the predicted output of the pre-trained auxiliary analysis network model is compared with the target image dataset (i.e., the processed real image features), the difference between them is calculated, and the BCE loss function is established accordingly; through the BCE loss function, the difference between the model prediction results and the target image dataset can be quantified, providing clear guidance for subsequent optimization. On the basis of establishing the BCE loss function, the data distribution characteristics of the predicted output of the pre-trained auxiliary analysis network model are further considered. This includes statistical characteristics such as the mean, variance, and distribution morphology of the data; according to these data distribution characteristics, the BCE loss function can be weighted or adjusted to construct a target loss function that is more in line with the actual situation; by constructing the target loss function, the difference between the model prediction results and the target image dataset can be more comprehensively measured, while considering the distribution characteristics of the data, improving the pertinence and effectiveness of the optimization. Optimization algorithms such as backpropagation and gradient descent are used to adjust the model parameters according to the gradient information of the target loss function. During the optimization process, the model parameters are continuously updated iteratively until the target loss function converges or reaches the preset number of iterations. Through optimization processing, a more accurate and robust optimization-assisted analysis network model can be obtained. This model can better adapt to the characteristics of the target imaging data set and improve the accuracy and reliability of infertility analysis.

[0090] Specifically, the target loss function in S420 is expressed by the following formula:

[0091]

[0092] f1(m i , ni )=m i log(n i )

[0093] f2(m i , n i )=(1-m i )log(1-n i )

[0094] Among them, g(m,n) represents the target loss function, K represents the total number of samples, i represents the number of i-th samples, m represents the true label of the sample (the true disease state of the patient corresponding to the target ultrasound image), n represents the predicted probability of the sample, and m i represents the true label of the i-th sample, n i represents the predicted probability of the i-th sample, α represents the first adjustment factor, and β represents the second adjustment factor;

[0095] Furthermore, according to different errors in the model prediction process, the values ​​of α and β can be adjusted in time; for example, if an error occurs in predicting a diseased patient as a non-disease patient (false negative) during the model prediction process, α can be set greater than β; if an error occurs in predicting a non-disease patient as a diseased patient (false positive) during the model prediction process, α can be set less than β; and the ratio between α and β can be determined based on the proportion of false positives and false negatives; by using such a loss function, the model can learn how to better predict the state of infertility during the training process; the first adjustment factor α and the second adjustment factor β allow different penalties for different types of errors, which helps to improve the performance of the model, especially on unbalanced data sets.

[0096] S500: iteratively training the optimized auxiliary analysis network model according to the target image data set to obtain a target auxiliary analysis network model;

[0097] Furthermore, after optimization, the model is iteratively trained using the target image dataset; during the iteration process, the model will continuously learn and adjust its internal parameters to better fit the imaging characteristics of infertility. Through multiple iterations of training, the model gradually converges and the final target-assisted analysis network model is obtained.

[0098] It should be noted that, in S500, the optimized auxiliary analysis network model is iteratively trained according to the target image data set to obtain the target auxiliary analysis network model; specifically:

[0099] S510: Divide the target image data set into S equal samples, and randomly select S-1 samples as training sets;

[0100] S520: Sampling the target image data set N times to obtain N training sets;

[0101] S530: Iteratively train the optimized auxiliary analysis network model according to the N training sets in turn to obtain a target auxiliary analysis network model.

[0102] Further, first, the target image dataset is divided into S equal samples. Then, in each iteration, S-1 samples are randomly selected from the S equal samples as training sets, and the remaining sample is used as a validation set (or test set, but at this stage, the focus is more on training, so it is usually used as a validation set to adjust model parameters); through sample division, it can be ensured that different data subsets can be used for each iterative training, thereby increasing the generalization ability of the model and reducing the risk of overfitting; on this basis, the target image dataset is sampled N times independently; each sampling is performed according to the method in S510 to obtain S-1 training samples and 1 validation sample, but the specific sample combination of each sampling is different; in this way, N different training sets can be obtained; through multiple samplings, different data sets can be provided for each iterative training, thereby further increasing the generalization ability of the model and reducing the dependence on data distribution. In each iteration, a training set is used to train the optimized auxiliary analysis network model; during the training process, the loss function is calculated based on the model's prediction results and the true labels of the pelvic ultrasound images, and the model parameters are updated through the back propagation algorithm; after the training is completed, the performance of the model is evaluated using the validation set, and the model parameters or hyperparameters (such as learning rate, batch size, etc.) are adjusted according to the evaluation results; then, the next iterative training is performed until the preset number of iterations N is reached or the model performance converges; through iterative training, the model parameters can be gradually optimized and its performance can be improved; the final target auxiliary analysis network model will have better generalization ability and higher analysis accuracy.

[0103] S600: inputting the pelvic ultrasound image to be tested after preprocessing and image optimization processing into the target auxiliary analysis network model for processing, and outputting the infertility test result;

[0104] Furthermore, the pelvic ultrasound image to be detected after preprocessing and image optimization is input into the target-assisted analysis network model. The target-assisted analysis network model processes the pelvic ultrasound image to be detected and outputs the analysis result of the pelvic ultrasound image to be detected, so as to determine whether there is infertility.

[0105] It should be noted that in S600, the target-assisted analysis network model includes a local feature extraction module, an adaptive feature fusion module and a classifier; the target-assisted analysis network model processes the pelvic ultrasound image to be detected, specifically:

[0106] S610: The local feature extraction module performs feature extraction on the input pelvic ultrasound image to be detected through a basic convolution block to extract the initial features of the uterus, the initial features of the fallopian tube, the initial features of the ovary, and the initial features of the pelvic effusion in the pelvic ultrasound image to be detected;

[0107] S620: performing deep extraction on the pelvic ultrasound image to be detected according to the residual convolution block in the local feature extraction module to extract the uterus depth feature, fallopian tube depth feature, ovary depth feature and pelvic effusion depth feature in the pelvic ultrasound image to be detected;

[0108] S630: The local feature extraction module performs weighted fusion on the uterus initial feature and the uterus depth feature, the fallopian tube initial feature and the fallopian tube depth feature, the ovary initial feature and the ovary depth feature, and the pelvic effusion initial feature and the pelvic effusion depth feature according to the multi-scale processing, so as to obtain the uterus target feature, the fallopian tube target feature, the ovary target feature and the pelvic effusion target feature;

[0109] S640: The adaptive feature fusion module assigns weights to the uterus target feature, the fallopian tube target feature, the ovary target feature, and the pelvic effusion target feature respectively according to the adaptive attention mechanism;

[0110] S650: The adaptive feature fusion module splices the weighted uterus target feature, fallopian tube target feature, ovary target feature and pelvic effusion target feature according to the splicing algorithm to obtain the ultrasound image target feature;

[0111] S660: The classifier classifies the target features of the ultrasound image to obtain an infertility detection result.

[0112] Specifically, in S610, the local feature extraction module extracts features from the pelvic ultrasound image to be detected by the following formula:

[0113] F(X)={f(x1),f(x2),f(x3),f(x4)}

[0114] Among them, F represents the initial feature extraction function, X represents the pelvic ultrasound image to be detected, f(x1) represents the initial feature of the uterus, f(x2) represents the initial feature of the fallopian tube, f(x3) represents the initial feature of the ovary, and f(x4) represents the initial feature of the pelvic effusion;

[0115] In S620, the local feature extraction module performs deep extraction of the pelvic ultrasound image to be detected, which is expressed by the following formula:

[0116] G(X)={g(x1),g(x2),g(x3),g(x4)}

[0117] Among them, G represents the initial feature extraction function, X represents the pelvic ultrasound image to be detected, g(x1) represents the uterus depth feature, g(x2) represents the fallopian tube depth feature, g(x3) represents the ovary depth feature, and g(x4) represents the pelvic fluid depth feature;

[0118] In S630, the multi-scale processing process is expressed by the following formula:

[0119] h(x1)=af(x1)+g(x1)

[0120] h(x2)=bf(x2)+g(x2)

[0121] h(x3)=cf(x3)+g(x3)

[0122] h(x4)=df(x4)+g(x4),

[0123] Among them, H(x1) represents the target feature of uterus, H(x2) represents the target feature of fallopian tube, H(x3) represents the target feature of ovary, h(x4) represents the target feature of pelvic effusion, a represents the weight of the initial feature of uterus, b represents the weight of the initial feature of fallopian tube, c represents the weight of the initial feature of ovary, and d represents the weight of the initial feature of pelvic effusion;

[0124] In S640 and S650, weights are assigned to the uterus target feature, the fallopian tube target feature, the ovary target feature, and the pelvic effusion target feature, respectively, and spliced ​​together using a splicing algorithm, which is expressed by the following formula:

[0125]

[0126] Among them, H(x) represents the target feature of ultrasound image, δ represents the weight of uterus target feature, ε represents the weight of fallopian tube target feature, θ represents the weight of ovary target feature, and γ represents the weight of pelvic effusion target feature; Indicates the feature concatenation symbol.

[0127] Furthermore, the pelvic ultrasound image to be detected is input into the local feature extraction module, and the basic convolution blocks are used to extract the initial features of the uterus, fallopian tubes, ovaries and pelvic effusion respectively; these initial features represent the basic information of each key structure in the image; the initial features of the uterus, fallopian tubes, ovaries and pelvic effusion are obtained, providing a basis for subsequent deep feature extraction. In the local feature extraction module, the residual convolution block is used to further process the initial features to extract the deep features of the uterus, fallopian tubes, ovaries and pelvic effusion; these deep features contain richer structural information and details; the deep features of the uterus, fallopian tubes, ovaries and pelvic effusion are obtained, providing material for subsequent feature fusion. In the local feature extraction module, a multi-scale processing method (such as convolution kernels or pooling layers of different scales) is used to perform weighted fusion of initial features and deep features; specifically, the initial features and deep features of the uterus, fallopian tubes, ovaries, and pelvic effusion are weighted respectively to obtain the target features of the uterus, fallopian tubes, ovaries, and pelvic effusion; through feature fusion, the model's ability to capture key structural features is improved, providing a more reliable feature representation for subsequent classification tasks. In the adaptive feature fusion module, an adaptive attention mechanism is used to assign weights to the target features of the uterus, fallopian tubes, ovaries, and pelvic effusion; these weights reflect the importance of different features in the analysis process; through weight assignment, the model can pay more attention to features that have an important impact on the analysis results, thereby improving the accuracy of the analysis. In the adaptive feature fusion module, a splicing algorithm (such as Concatenation) is used to splice the uterus target features, fallopian tube target features, ovary target features and pelvic effusion target features; the spliced ​​features contain complete information of all key structures; through feature splicing, an ultrasound image target feature containing all key structural features is obtained, providing a comprehensive feature representation for subsequent classification tasks; the ultrasound image target feature is input into the classifier (such as fully connected layer + Softmax function) for classification processing. The classifier determines whether the input image belongs to the infertility category based on the feature representation; outputs the infertility test results to provide auxiliary analysis information for doctors.

[0128] Compared with the prior art, the infertility ultrasound image-assisted analysis method according to the present embodiment has the following technical effects: by acquiring a pelvic ultrasound image data set and performing preprocessing (such as Wiener filter denoising, image annotation and image correction), noise and geometric distortion can be significantly reduced, image quality can be improved, and thus the accuracy of analysis can be enhanced; by image optimization processing (such as feature extraction and enhancement), key areas such as the uterus, ovaries, fallopian tubes and pelvic effusion can be made clearer, which helps doctors or artificial intelligence models to identify and analyze more accurately; by constructing an infertility auxiliary analysis network model and performing pre-training, optimization and iterative training, the generalization ability and analysis accuracy of the model can be continuously improved; by optimizing the model using a target loss function, it is helpful to reduce prediction errors and improve the reliability of analysis results; by using local features in the target-assisted analysis network model, the accuracy of the analysis can be improved. The feature extraction module can accurately extract the initial features and deep features of the uterus, fallopian tubes, ovaries and pelvic effusion; the adaptive feature fusion module can effectively fuse these features through the adaptive attention mechanism and splicing algorithm to form a more comprehensive and accurate ultrasound image target feature; by dividing the target image data set into multiple samples and performing random sampling training, different data set conditions can be simulated, the model's adaptability to different data can be improved, and the model's robustness can be enhanced; the automated and intelligent infertility ultrasound image-assisted analysis method can reduce the time and cost of manual analysis and improve analysis efficiency; doctors can use the analysis results of the artificial intelligence model to make accurate analysis and treatment decisions more quickly; this technical solution provides a new method and idea for ultrasound image-assisted analysis of infertility, which helps promote the innovation and development of medical technology.

[0129] It should be noted that the infertility ultrasound image-assisted analysis method provided in the embodiment of the present application can be executed by an infertility ultrasound image-assisted analysis system, or a control module in the infertility ultrasound image-assisted analysis system for executing and loading an infertility ultrasound image-assisted analysis method. In the embodiment of the present application, an infertility ultrasound image-assisted analysis system is used to execute and load an infertility ultrasound image-assisted analysis method as an example to illustrate an infertility ultrasound image-assisted analysis method provided in the embodiment of the present application.

[0130] An infertility ultrasound image-assisted analysis system, comprising:

[0131] An image acquisition module acquires a pelvic ultrasound image data set, and preprocesses the pelvic ultrasound image data set to obtain a preprocessed image data set;

[0132] An image optimization module is used to optimize the pre-processed image data set output by the image acquisition module to obtain a target image data set;

[0133] Model building module, used to build an infertility auxiliary analysis network model;

[0134] The first training module is used to pre-train the infertility auxiliary analysis network model established by the model establishment module to obtain a pre-trained auxiliary analysis network model;

[0135] A model optimization module optimizes the pre-trained auxiliary analysis network model according to the predicted output of the pre-trained auxiliary analysis network model to obtain an optimized auxiliary analysis network model;

[0136] The second training module is used to iteratively train the optimized auxiliary analysis network model according to a preset training method to obtain a target auxiliary analysis network model;

[0137] The result output module is used to output the analysis result of the pelvic ultrasound image to be detected input by the user.

[0138] According to an infertility ultrasound image-assisted analysis system of the present embodiment, a pelvic ultrasound image data set is acquired through an image acquisition module and preprocessed, ensuring that the acquired ultrasound image data is of high quality and low in noise, providing a reliable basis for subsequent analysis. Preprocessing may include operations such as denoising, contrast enhancement, and standardization to make the image more suitable for machine learning and analysis; the preprocessed image data set is optimized through an image optimization module; the image clarity, detail expression, and contrast can be further improved, which helps doctors or models to more accurately identify and analyze pelvic structures, thereby improving the accuracy of the analysis; an infertility auxiliary analysis network model is established through a model building module, and the first training module performs pre-training to obtain a pre-trained auxiliary analysis network model; the pre-training process helps the model to initially learn image features, laying the foundation for subsequent optimization and iterative training; the model is optimized according to the training results through a model optimization module; the analysis accuracy and generalization ability of the model can be further improved, so that It can more accurately identify and analyze different types of pelvic ultrasound images, thereby providing more reliable auxiliary analysis results; the optimized model can be iteratively trained according to the preset training method through the second training module; the parameters and structure of the model can be continuously fine-tuned to make it more adaptable to actual analysis needs and improve the accuracy and efficiency of the analysis; the user inputs the pelvic ultrasound image to be detected through the image acquisition module, and after preprocessing by the image acquisition module and optimization by the image optimization module, the result is predicted by the target auxiliary analysis network model, and finally the result is output through the data output module; the system can automatically and quickly output the analysis results, provide doctors with timely auxiliary analysis information, and help doctors make more accurate analysis and treatment decisions.

[0139] An infertility ultrasound image-assisted analysis system in the embodiment of the present application may be a device, or a component, integrated circuit, or chip in a terminal. The device may be a mobile electronic device or a non-mobile electronic device. Exemplarily, the mobile electronic device may be a mobile phone, a tablet computer, a laptop computer, a PDA, a netbook, or a personal digital assistant (PDA), etc., and the non-mobile electronic device may be a server, a network attached storage (NAS), a personal computer (PC), etc., which is not specifically limited in the embodiment of the present application.

[0140] An infertility ultrasound image-assisted analysis system in the embodiment of the present application may be a device having an operating system. The operating system may be an Android operating system, an iOS operating system, or other possible operating systems, which are not specifically limited in the embodiment of the present application.

[0141] The infertility ultrasound image-assisted analysis system provided in the embodiment of the present application can achieve Figure 1 to Figure 2 In the method embodiment, each process of implementing the method for ultrasound image-assisted analysis of infertility is not described here to avoid repetition.

[0142] Optionally, an embodiment of the present application also provides an electronic device, including a processor, a memory, and a program or instruction stored in the memory and executable on the processor. When the program or instruction is executed by the processor, each process of the above-mentioned infertility ultrasound image-assisted analysis method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0143] An embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, each process of the above-mentioned infertility ultrasound image-assisted analysis method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0144] The processor is a processor in the electronic device described in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0145] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0146] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0147] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. An ultrasound image-assisted analysis method for infertility, characterized in that: The method comprises: Acquiring a pelvic ultrasound image dataset, and preprocessing the pelvic ultrasound image dataset to obtain a preprocessed image dataset; Performing image optimization processing on the preprocessed image data set to obtain a target image data set; Constructing an infertility auxiliary analysis network model, and pre-training the infertility auxiliary analysis network model according to the target image data set to obtain a pre-trained auxiliary analysis network model; According to the predicted output of the pre-trained auxiliary analysis network model, the pre-trained auxiliary analysis network model is optimized to obtain an optimized auxiliary analysis network model; Iteratively training the optimized auxiliary analysis network model according to the target image data set to obtain a target auxiliary analysis network model; The pelvic ultrasound image to be detected after the preprocessing and the image optimization processing is input into the target auxiliary analysis network model for processing, and the infertility detection result is output.

2. The method for analyzing infertility using ultrasound images according to claim 1, characterized in that: The preprocessing of the pelvic ultrasound image dataset to obtain the preprocessed image dataset is specifically: Performing Wiener filtering on the pelvic ultrasound image data set according to a Wiener filter to remove noise data in the pelvic ultrasound image data set; Performing image annotation processing on the pelvic ultrasound image dataset from which noise data is removed according to the trained model to highlight the uterus, ovaries, fallopian tubes and pelvic effusion in the pelvic ultrasound image dataset; Image correction is performed on the pelvic ultrasound image data set after image annotation processing to eliminate geometric distortion generated during image acquisition to obtain the preprocessed image data set.

3. The method for analyzing infertility using ultrasound images according to claim 1, characterized in that: The image optimization processing is performed on the pre-processed image data set to obtain the target image data set, specifically: Performing feature extraction on each of the preprocessed images in the preprocessed image data set to obtain corresponding uterus extraction images, ovary extraction images, fallopian tube extraction images, and pelvic effusion extraction images; Respectively enhancing the uterus extracted image, the ovary extracted image, the fallopian tube extracted image and the pelvic effusion extracted image to obtain a uterus enhanced image, an ovary enhanced image, a fallopian tube enhanced image and a pelvic effusion enhanced image; Each of the preprocessed images is fused with the corresponding uterus enhanced image, ovary enhanced image, fallopian tube enhanced image and pelvic effusion enhanced image to obtain the target image data set.

4. The method for analyzing infertility using ultrasound images according to claim 3, characterized in that: The step of respectively performing enhancement processing on the uterus extracted image, the ovary extracted image, the fallopian tube extracted image and the pelvic effusion extracted image to obtain a uterus enhanced image, an ovary enhanced image, a fallopian tube enhanced image and a pelvic effusion enhanced image is specifically as follows: Performing the enhancement processing on the uterus extraction image includes: Performing multi-layer downsampling on the uterus extraction image to obtain a plurality of uterus sampling sub-images with successively decreasing resolutions; Taking the uterus sampling sub-image with the highest resolution as the uterus initial image, performing edge enhancement processing on each of the uterus sampling sub-images other than the initial image to obtain a corresponding uterus enhanced sub-image; Performing image fusion processing on all the uterus enhanced sub-images and the uterus initial image to obtain the uterus enhanced image; The enhancement processing is performed on the ovary extracted image, the fallopian tube extracted image and the pelvic effusion extracted image respectively to obtain the corresponding ovary enhanced image, fallopian tube enhanced image and pelvic effusion enhanced image.

5. The method for analyzing infertility using ultrasound images according to claim 1, characterized in that: The step of optimizing the pre-trained auxiliary analysis network model according to the predicted output of the pre-trained auxiliary analysis network model to obtain an optimized auxiliary analysis network model is specifically as follows: Establishing a BCE loss function according to the difference between the predicted output and the target image data set; Constructing a target loss function according to the data distribution characteristics of the predicted output and the BCE loss function; The pre-trained auxiliary analysis network model is optimized according to the target image data set and the target loss function to obtain the optimized auxiliary analysis network model.

6. The method for analyzing infertility using ultrasound images according to claim 1, characterized in that: The iterative training of the optimized auxiliary analysis network model according to the target image data set to obtain the target auxiliary analysis network model is specifically as follows: The target image data set is divided into S equal samples, and S-1 samples are randomly selected as training sets; Sampling the target image data set N times to obtain N training sets; The optimized auxiliary analysis network model is iteratively trained in turn according to the N training sets to obtain the target auxiliary analysis network model.

7. The infertility ultrasound image-assisted analysis method according to claim 1, characterized in that: The target-assisted analysis network model includes a local feature extraction module, an adaptive feature fusion module and a classifier; the target-assisted analysis network model processes the pelvic ultrasound image to be detected, specifically: The local feature extraction module performs feature extraction on the input pelvic ultrasound image to be detected through a basic convolution block to extract the initial features of the uterus, the initial features of the fallopian tube, the initial features of the ovary and the initial features of the pelvic effusion in the pelvic ultrasound image to be detected; Performing deep extraction on the pelvic ultrasound image to be detected according to the residual convolution block in the local feature extraction module to extract uterine depth features, fallopian tube depth features, ovarian depth features and pelvic effusion depth features in the pelvic ultrasound image to be detected; The local feature extraction module performs weighted fusion on the uterus initial feature and the uterus depth feature, the fallopian tube initial feature and the fallopian tube depth feature, the ovary initial feature and the ovary depth feature, and the pelvic effusion initial feature and the pelvic effusion depth feature according to multi-scale processing to obtain uterus target feature, fallopian tube target feature, ovary target feature and pelvic effusion target feature; The adaptive feature fusion module assigns weights to the uterus target feature, the fallopian tube target feature, the ovary target feature and the pelvic effusion target feature respectively according to an adaptive attention mechanism; The adaptive feature fusion module splices the weighted uterus target feature, fallopian tube target feature, ovary target feature and pelvic effusion target feature according to a splicing algorithm to obtain an ultrasound image target feature; The classifier classifies the ultrasonic image target features to obtain the infertility test result.

8. An infertility ultrasound image-assisted analysis system, implementing an infertility ultrasound image-assisted analysis method as claimed in any one of claims 1 to 7, characterized in that: The system comprises: An image acquisition module acquires a pelvic ultrasound image data set, and preprocesses the pelvic ultrasound image data set to obtain a preprocessed image data set; An image optimization module, used for optimizing the pre-processed image data set output by the image acquisition module to obtain a target image data set; Model building module, used to build an infertility auxiliary analysis network model; A first training module is used to pre-train the infertility auxiliary analysis network model established by the model establishment module to obtain a pre-trained auxiliary analysis network model; A model optimization module, which optimizes the pre-trained auxiliary analysis network model according to the predicted output of the pre-trained auxiliary analysis network model to obtain an optimized auxiliary analysis network model; A second training module is used to iteratively train the optimized auxiliary analysis network model according to a preset training method to obtain a target auxiliary analysis network model; The result output module is used to output the analysis result of the pelvic ultrasound image to be detected input by the user.

9. An electronic device, characterized in that: The electronic device includes a processor, a memory, and a program or instruction stored in the memory and executable on the processor. When the program or instruction is executed by the processor, the steps of an ultrasound image-assisted analysis method for infertility as described in any one of claims 1 to 7 are implemented.

10. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by the processor, the steps of the infertility ultrasound image-assisted analysis method as described in any one of claims 1-7 are implemented.

Images