A pelvic floor dysfunction evaluation method based on self-supervised transfer learning

Abstract

The application provides a pelvic floor dysfunction evaluation method based on self-supervised transfer learning, relates to the technical field of medical image diagnosis, and comprises the following steps: acquiring a pelvic floor ultrasound image and an ImageNet-1K data set, and performing data preprocessing; a neural network model is established according to the ImageNet-1K data set after data preprocessing, and self-supervised pre-training is performed to obtain a pre-trained neural network model; the pelvic floor ultrasound image after data preprocessing is subjected to self-supervised pre-training based on the pre-trained neural network model; an encoder in the pre-trained neural network model is extracted to obtain image features of the ultrasound image, a pelvic floor dysfunction evaluation classification model is established through a binary classification full connection neural network, and the pelvic floor dysfunction evaluation classification model is used for pelvic floor dysfunction evaluation; the application solves the detection problem of pelvic floor dysfunction through transfer learning, and the obtained model has higher classification accuracy and better generalization performance.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image diagnosis, in particular to a pelvic floor dysfunction evaluation method based on self-supervised transfer learning. BACKGROUND

[0002] Pelvic floor dysfunction is a common disease in middle-aged and elderly women, with an incidence of about 30% to 50%, generally accompanied by displacement of pelvic organs, and seriously affects the health and quality of life of women. In recent years, with the rapid development of artificial intelligence technology, deep learning technology has also been used in the detection of pelvic floor dysfunction.

[0003] The existing deep learning method for detecting pelvic floor dysfunction mainly uses a convolutional neural network, and then directly uses the pelvic floor ultrasound image for training. Such a method has two problems: first, the previous network has low accuracy and weak generalization ability; second, pelvic floor dysfunction evaluation generally relies on detecting the relative position relationship of pelvic organs, which is a challenging task for neural networks.

[0004] Based on this, the present application provides a pelvic floor dysfunction evaluation method based on self-supervised transfer learning to solve the above problems. SUMMARY

[0005] The purpose of the present application is to provide a pelvic floor dysfunction evaluation method based on self-supervised transfer learning, which can improve the detection accuracy and generalization performance of the model.

[0006] The technical scheme of the present application is:

[0007] In a first aspect, the present application provides a pelvic floor dysfunction evaluation method based on self-supervised transfer learning, which includes the following steps:

[0008] S1, obtaining a pelvic floor ultrasound image and an ImageNet-1K dataset, and performing data preprocessing;

[0009] S2, establishing a neural network model according to the ImageNet-1K dataset after data preprocessing, and performing self-supervised pre-training to obtain a pre-trained neural network model;

[0010] S3, performing self-supervised pre-training on the pelvic floor ultrasound image after data preprocessing based on the pre-trained neural network model;

[0011] S4, extracting the encoder in the pre-trained neural network model to obtain the image features of the ultrasound image, and establishing a pelvic floor dysfunction evaluation classification model through a binary classification fully connected neural network, and using the pelvic floor dysfunction evaluation classification model to evaluate pelvic floor dysfunction.

[0012] In step S2, the pre-processed ImageNet-1K dataset is used to pre-train the neural network model in a self-supervised manner, which is mainly MLM pre-training. During the pre-training, the encoder is responsible for randomly masking a part of the image and extracting the features of the masked image, and the decoder is responsible for restoring the masked image using the features extracted by the encoder. After the training, the trained neural network model is saved and used in step S3 to pre-train the pre-processed pelvic floor ultrasound images in a self-supervised manner.

[0013] It should be noted that the neural network model of the present application adopts a self-attention architecture.

[0014] Further, in step S1, the data preprocessing includes data edge processing and data normalization processing. The calculation formula of the data normalization processing is:

[0015]

[0016] wherein, represents the normalized data, x c represents the data of the cth channel of the input image, μ(x c ) represents the mean of all data in the cth channel, and σ(x c ) represents the variance of all data in the cth channel.

[0017] Further, step S2 includes the following sub-steps:

[0018] S21, based on the pre-processed ImageNet-1K dataset, a neural network model is established, and a ViT model is used as an encoder to put the pre-processed ImageNet-1K dataset into the encoder to split the image into multiple image blocks;

[0019] S22, the feature representation of each image block is obtained through the embedding layer of the encoder;

[0020] S23, the image is masked according to a pre-set masking ratio;

[0021] S24, an auxiliary cls feature is added to the masked image to obtain an input feature;

[0022] S25, the input feature is input into the self-attention layer of the encoder for interaction calculation to obtain the output feature of the encoder;

[0023] S26, a random noise vector is added to the output feature of the encoder, and the attention layer of the decoder is input for processing to obtain a feature vector;

[0024] S27, obtaining the feature vector corresponding to the covered picture block from the feature vector obtained by the decoder, and predicting the original image of the picture block by using a full connection layer to obtain the pixel value of the predicted image block;

[0025] S28, performing self-supervised pre-training on the neural network model, constantly optimizing the loss function of the neural network model by using a back propagation algorithm, adjusting the parameters of the neural network model, obtaining an encoder for extracting model features and a decoder for restoring pictures, and thus obtaining the self-supervised pre-trained neural network model.

[0026] wherein the feature vector corresponding to the covered picture block is the feature predicted by the model.

[0027] Further, the calculation process of the above interaction calculation includes:

[0028] q i ,k i ,v i =z i L q ,z i L k ,z i L v ,L q,k,v ∈R D×D ,

[0029]

[0030]

[0031] wherein q, k, and v are three feature vectors, D is the feature dimension, i represents the number of the picture block, L q , L k , L v respectively represent the linear mapping of q, k, and v, R D×D represents the dimension size of L q , L k , L v is DxD, j represents the jth attention layer, softmax represents the softmax function, T represents the transpose, Att() represents the attention value, and MLP() represents the neural network, represents the output feature.

[0032] Further, the calculation formula of the loss function of the above neural network model includes:

[0033]

[0034] wherein MSELoss represents the loss function of the neural network model, n represents the number of samples, r represents the covering ratio, H and W respectively represent the length and width of the input picture, and pi represents a pixel value of an i-th pixel point of a picture, represents a model prediction pixel value of the i-th pixel point.

[0035] Further, the step S4 comprises the following sub-steps:

[0036] S41, extract an encoder in the pre-trained neural network model to obtain image features of the ultrasound image;

[0037] S42, perform classification training on the pelvic floor ultrasound image with the image features through a binary classification fully connected neural network to establish a pelvic floor dysfunction evaluation classification model, and continuously optimize a loss function of the pelvic floor dysfunction evaluation classification model through a back propagation algorithm to adjust classification network parameters of the model;

[0038] S43, perform pelvic floor dysfunction evaluation using the pelvic floor dysfunction evaluation classification model with the adjusted classification network parameters to obtain a pelvic floor dysfunction evaluation result.

[0039] Further, in the step S42, the loss function of the pelvic floor dysfunction evaluation classification model uses a cross-entropy loss, and a calculation formula thereof is:

[0040]

[0041] wherein, CEloss represents the loss function of the pelvic floor dysfunction evaluation classification model, n represents a sample quantity, y i represents a true label value of the sample, y i ′ represents a prediction value of the sample.

[0042] In a second aspect, the present application provides an electronic device, characterized in that comprising:

[0043] a memory for storing one or more programs;

[0044] a processor;

[0045] When the one or more programs are executed by the processor, a pelvic floor dysfunction evaluation method based on self-supervised transfer learning as in any one of the above first aspects is implemented.

[0046] In a third aspect, the present application provides a computer readable storage medium having a computer program stored thereon, which is executed by a processor to implement a pelvic floor dysfunction evaluation method based on self-supervised transfer learning as in any one of the above first aspects.

[0047] Compared with the prior art, the present application has at least the following advantages or beneficial effects:

[0048] (1) The application discloses a pelvic floor dysfunction evaluation method based on self-supervised transfer learning, which solves the detection problem of pelvic floor dysfunction through transfer learning, that is, a method of pre-training on a large natural image dataset (ImageNet-1k), then continuing pre-training on ultrasound images, and finally performing classification fine-tuning, so that the obtained model has higher classification accuracy and better generalization performance;

[0049] (2) The application uses an MLM pre-training task for pre-training, and compared with other pre-training tasks, the MLM pre-training task needs to use adjacent pictures to predict the covered pictures during pre-training, divides the pictures into different regions, and makes each region have different covering ratios according to importance, so that the probability of covering the core region is improved, the model can better extract the picture features of the core region, the model can more easily extract the adjacent position relationship in the picture, and the detection accuracy of the model is improved;

[0050] (3) The application uses a neural network model based on a self-attention architecture to detect pelvic floor dysfunction, and the neural network model based on the self-attention architecture of the interaction between picture blocks can better detect the relative position relationship between pelvic organs, and is beneficial to improving the detection accuracy of the final model. BRIEF DESCRIPTION OF DRAWINGS

[0051] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiments, and it should be understood that the following drawings only show some embodiments of the application, and therefore should not be regarded as a limitation to the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0052] Figure 1 A step diagram of the application of a pelvic floor dysfunction evaluation method based on self-supervised transfer learning is shown in the figure.

[0053] Figure 2 An ultrasound image covering schematic diagram is shown in the figure.

[0054] Figure 3 A schematic structural block diagram of an electronic device is shown in the figure.

[0055] Figure legend: 101, memory; 102, processor; 103, communication interface. DETAILED DESCRIPTION

[0056] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some but not all of the embodiments of the present application. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations.

[0057] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed application, but only represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work are within the scope of protection of the present application.

[0058] It should be noted that similar reference numerals and letters in the following drawings represent similar items, and therefore, once an item is defined in one drawing, it need not be further defined and explained in subsequent drawings.

[0059] It should be noted that, in this document, the term "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or apparatus that includes a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent in such a process, method, article or apparatus. Without more limitations, the elements defined by the statement "comprising" do not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the elements.

[0060] In the description of the present application, it should also be noted that, unless otherwise explicitly specified and limited, the terms "arranged", "connected" should be understood broadly, for example, can be fixedly connected, or detachably connected, or integrally connected; can be mechanically connected, or electrically connected; can be directly connected, or indirectly connected through an intermediate medium, or the communication inside two elements. For those of ordinary skill in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0061] The following will describe some embodiments of the present application in detail with reference to the drawings. In the case of no conflict, each of the following embodiments and each feature in the embodiments can be combined with each other.

[0062] Embodiment 1

[0063] Please refer to Figure 1 , Figure 1 The figure shows the steps of a pelvic floor dysfunction evaluation method based on self-supervised transfer learning provided by the embodiments of the present application.

[0064] The application provides a pelvic floor dysfunction evaluation method based on self-supervised transfer learning, which comprises the following steps:

[0065] S1, obtain a pelvic floor ultrasound image and an ImageNet-1K data set, and perform data preprocessing;

[0066] S2, establish a neural network model according to the ImageNet-1K data set after data preprocessing, and perform self-supervised pre-training to obtain a pre-trained neural network model;

[0067] S3, based on the pre-trained neural network model, the pelvic floor ultrasound image after data preprocessing is self-supervised pre-trained;

[0068] S4, extract the encoder in the pre-trained neural network model to obtain the image features of the ultrasound image, and establish a pelvic floor dysfunction evaluation classification model through a binary classification full connection neural network, and use the pelvic floor dysfunction evaluation classification model to evaluate the pelvic floor dysfunction.

[0069] As a preferred embodiment, in step S1, the data preprocessing includes data edge processing and data normalization processing, and the calculation formula of the data normalization processing is:

[0070]

[0071] wherein, represents the data after normalization processing, x c represents the data of the cth channel of the input image, μ(x c represents the mean of all data in the cth channel, and σ(x c represents the variance of all data in the cth channel.

[0072] As a preferred embodiment, step S2 comprises the following sub-steps:

[0073] S21, based on the ImageNet-1K data set after data preprocessing, a neural network model is established, and a ViT model is used as an encoder to put the pictures of the ImageNet-1K data set after data preprocessing into the encoder to split into multiple picture blocks;

[0074] It should be noted that in this embodiment 1, after a 224x224 size picture in the ImageNet-1k data set enters the ViT model, the ViT model will regard each 14x14 size picture block as a picture block, so a 224x224 size picture block will be split into 16x16=196 picture blocks.

[0075] S22, obtain the feature representation of each picture block through the embedding layer of the encoder;

[0076] S23, the picture is covered according to the preset covering ratio;

[0077] It should be noted that the covering manner is to randomly delete the features of 196xr picture blocks, and the feature length is 768, wherein r represents the covering ratio;

[0078] S24, the auxiliary cls feature is added to the covered picture to obtain an input feature;

[0079] It should be noted that the auxiliary cls feature added at this time is a 768-dimensional feature vector, and finally a (196-196xr+1) x 768-dimensional input feature can be obtained;

[0080] S25, the input feature is input into the self-attention layer of the encoder for interaction calculation to obtain the output feature of the encoder;

[0081] It should be noted that the self-attention layer of the encoder is a 12-layer self-attention layer, and the output of the encoder is still a feature of (196-196xr+1) patches, wherein r represents the covering ratio;

[0082] S26, a random noise vector is added to the output feature of the encoder, and is input into the attention layer of the decoder for processing to obtain a feature vector;

[0083] It should be noted that before entering the decoder, the model will add a random noise vector to the position of the deleted patch, so that the number of features becomes 196+1 again; the decoder is composed of four self-attention layers, the attention layer architecture is consistent with the encoder, and the feature is still obtained after the attention layer processing. 196+1 feature vectors;

[0084] S27, the feature vector corresponding to the covered picture block is taken out from the feature vector obtained by the decoder, and a fully connected layer is used to predict the original image of the picture block to obtain the pixel value of the predicted image block;

[0085] It should be noted that among the 197 feature vectors, the feature vector corresponding to the covered patch is taken out, that is, the model prediction feature, and a fully connected layer is used to predict the original image of the patch, and finally the pixel value of the predicted patch is obtained;

[0086] S28, the neural network model is pre-trained in a self-supervised manner, the loss function of the neural network model is continuously optimized through a back propagation algorithm, the parameters of the neural network model are adjusted, the encoder for extracting model features and the decoder for restoring pictures are obtained, and thus the self-supervised pre-trained neural network model is obtained.

[0087] wherein the feature vector corresponding to the covered picture block is the feature predicted by the model.

[0088] As a preferred implementation, the calculation process of the interactive calculation includes:

[0089] q i ,k i ,v i =z i L q ,z i L k ,z i L v ,L q,k,v ∈R D×D ,

[0090]

[0091]

[0092] wherein q, k, v are three feature vectors, D is the feature dimension, i represents the number of the picture block, L q , L k , L v respectively represent the linear mapping of q, k, v, R D×D represents the dimension size of L q , L k , L v is DxD, j represents the jth attention layer, softmax represents the softmax function, T represents the transpose, Att() represents the attention value, MLP() represents the neural network, represents the output feature.

[0093] As a preferred implementation, the calculation formula of the loss function of the neural network model includes:

[0094]

[0095] wherein MSELoss represents the loss function of the neural network model, n represents the sample number, r represents the covering ratio, H, W respectively represent the length and width of the input picture, p i represents the pixel value of the ith pixel point of the picture, represents the model predicted pixel value of the ith pixel point.

[0096] As a preferred embodiment, in step S3, the pre-processed pelvic floor ultrasound image is self-supervised pre-trained based on the pre-trained neural network model. The pre-training model of step B is mainly copied, and the input picture is replaced with the ultrasound picture to continue pre-training. After the picture passes through the embedding layer of ViT, the feature representation of 196 picture blocks can be obtained. Then, the picture is covered according to the covering ratio r. The covering method is different from the random deletion of step S2. This pre-training changes the covering probability of the picture according to the importance of the picture area, such as Figure 2 As shown in FIG. 6, an ultrasound image covering diagram is shown. After a picture is divided into 196 picture blocks, it will be divided into three parts. The first part is the region of 6x6 blocks in the middle of the picture, and the probability of covering this part is increased. The second part is the remaining area of the middle 10x10 blocks excluding the first part, and the probability of covering this part is unchanged. The third part is the remaining area of the total 16x16 blocks excluding the first and second parts, and the probability of covering this part will be reduced. The covering effect is shown in FIG. 7. Figure 2 As shown in FIG. 7, it can be seen that the probability of covering the central region is significantly increased.

[0097] As a preferred embodiment, step S4 includes the following steps:

[0098] S41, extracting the encoder in the pre-trained neural network model to obtain the image features of the ultrasound image;

[0099] S42, classifying and training the pelvic floor ultrasound image with image features through a binary classification fully connected neural network to establish a pelvic floor dysfunction assessment classification model, and continuously optimizing the loss function of the pelvic floor dysfunction assessment classification model through a back propagation algorithm to adjust the classification network parameters of the model;

[0100] S43, using the pelvic floor dysfunction assessment classification model with adjusted classification network parameters to perform pelvic floor dysfunction assessment to obtain the pelvic floor dysfunction assessment result.

[0101] As a preferred embodiment, in step S42, the loss function of the pelvic floor dysfunction assessment classification model uses cross-entropy loss, and its calculation formula is:

[0102]

[0103] Wherein, CEloss represents the loss function of the pelvic floor dysfunction assessment classification model, n represents the number of samples, y i represents the true label value of the sample, y i ′ represents the predicted value of the sample.

[0104] Example 2

[0105] Referring to Figure 3 , Figure 3 A schematic structural block diagram of an electronic device provided for Embodiment 2 of the present application.

[0106] An electronic device includes a memory 101, a processor 102 and a communication interface 103, which are electrically connected to each other directly or indirectly to realize the transmission or interaction of data. For example, these elements can be electrically connected to each other through one or more communication buses or signal lines. The memory 101 can be used to store software programs and modules, and the processor 102 performs various functional applications and data processing by executing the software programs and modules stored in the memory 101. The communication interface 103 can be used for signaling or data communication with other node devices.

[0107] Among them, the memory 101 can be, but is not limited to, a random access memory (RAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM) and the like.

[0108] The processor 102 can be an integrated circuit chip with signal processing capability. The processor 102 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP) and the like; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.

[0109] It can be understood that the structure shown in the figure is only schematic, and the pelvic floor dysfunction evaluation method based on self-supervised transfer learning can also include more or fewer components than those shown in the figure, or have a different configuration from that shown in the figure. Each component shown in the figure can be realized by hardware, software or a combination thereof.

[0110] In the embodiments of the present application, it should be understood that the disclosed method can also be implemented by other means. The above-described embodiments are only illustrative, for example, the flowchart or block diagram in the drawings shows the possible implementation architecture, function and operation of the method and computer program product according to the embodiments of the present application. In this regard, each block in the flowchart or block diagram can represent a module, a program segment or a part of code, which contains one or more executable instructions for implementing the specified logic function. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur in different order from that shown in the drawings. For example, two consecutive blocks can actually be executed substantially in parallel, and sometimes they can be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0111] In addition, the functional modules in the various embodiments of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0112] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0113] In summary, the method for evaluating pelvic floor dysfunction based on self-supervised transfer learning provided by the embodiments of the present application solves the detection problem of pelvic floor dysfunction through transfer learning, i.e. a method of first pre-training on a large natural image dataset (ImageNet-1k), then continuing pre-training on ultrasound images, and finally performing classification fine-tuning. The obtained model has higher classification accuracy and better generalization performance.

[0114] The above description is only preferred embodiments of the present application and is not used to limit the present application. The present application can have various modifications and changes for those skilled in the art. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0115] It is obvious for those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, but can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application. Therefore, the embodiments should be regarded as exemplary and non-limiting, the scope of the present application is defined by the appended claims rather than the above description, and all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present application. Any reference signs in the claims should not be regarded as limiting the involved claims.

Claims

1. An electronic device, comprising: The method comprises the following steps: A memory for storing one or more programs; A processor; When the one or more programs are executed by the processor, a pelvic floor dysfunction evaluation method based on self-supervised transfer learning is implemented; wherein the pelvic floor dysfunction evaluation method based on self-supervised transfer learning comprises the following steps: S1, obtaining a pelvic floor ultrasound image and an ImageNet-1K dataset, and performing data preprocessing; S2, establishing a neural network model according to the ImageNet-1K dataset after data preprocessing, and performing self-supervised pre-training to obtain a pre-trained neural network model; S3, performing self-supervised pre-training on the pelvic floor ultrasound image after data preprocessing based on the pre-trained neural network model; S4, extracting the encoder in the pre-trained neural network model to obtain the image features of the ultrasound image, and establishing a pelvic floor dysfunction evaluation classification model through a binary classification fully connected neural network, and using the pelvic floor dysfunction evaluation classification model to evaluate the pelvic floor dysfunction; Step S2 comprises the following steps: S21, based on the ImageNet-1K dataset after data preprocessing, a neural network model is established, and a ViT model is used as an encoder to put the pictures of the ImageNet-1K dataset after data preprocessing into the encoder to split them into multiple picture blocks; S22, obtain the feature representation of each picture block through the embedding layer of the encoder; S23, according to the pre-set covering ratio, cover the picture; S24, add auxiliary cls features to the covered picture to obtain input features; S25, input the input features into the self-attention layer of the encoder for interaction calculation to obtain the output features of the encoder; S26, add a random noise vector to the output features of the encoder, and input it into the attention layer of the decoder for processing to obtain a feature vector; S27, take out the feature vector corresponding to the covered picture block from the feature vector obtained by the decoder, and use a fully connected layer to predict the original image of the picture block to obtain the pixel value of the predicted image block; S28, self-supervised pre-training of the neural network model, continuously optimizing the loss function of the neural network model through the back propagation algorithm, adjusting the parameters of the neural network model, obtaining the encoder for extracting model features and the decoder for restoring pictures, thereby obtaining the self-supervised pre-trained neural network model; The calculation process of the interaction calculation comprises: ， ， ， wherein q, k, v are three feature vectors, D is a feature dimension, i denotes a number of a picture block, , , denote linear mappings of q, k, v respectively, denotes , has a dimension size of D x D, j denotes a j-th attention layer, denotes a function, T denotes a transpose, denotes an attention value, denotes a neural network, denotes an output feature; Step S4 comprises the following steps: S41, extract the encoder in the pre-trained neural network model to obtain the image features of the ultrasound image; S42, classify and train the pelvic floor ultrasound image with image features through a binary classification fully connected neural network to establish a pelvic floor dysfunction evaluation classification model, and continuously optimize the loss function of the pelvic floor dysfunction evaluation classification model through the back propagation algorithm to adjust the classification network parameters of the model; S43, using the pelvic floor dysfunction evaluation classification model with adjusted classification network parameters to evaluate the pelvic floor dysfunction to obtain the pelvic floor dysfunction evaluation result.

2. An electronic device as claimed in claim 1, characterized in that In step S1, the data preprocessing comprises data edge processing and data normalization processing, and the calculation formula of the data normalization processing is: ， wherein, denotes the data after normalization processing, denotes the data of the c-th channel of the input image, denotes the mean of all data under the c-th channel, denotes the variance of all data under the c-th channel.

3. An electronic device as claimed in claim 1, characterized in that A calculation formula of a loss function of the neural network model comprises: ， wherein, denotes a loss function of the neural network model, n denotes a number of samples, denotes a covering ratio, , W are the length and width of the input picture, respectively, denotes a pixel value of the i-th pixel of the picture, denotes a model predicted pixel value of the i-th pixel.

4. An electronic device as in claim 1, wherein, In step S42, a loss function of the pelvic floor dysfunction assessment classification model uses a cross-entropy loss, and a calculation formula thereof is: ， wherein, represents a loss function of the pelvic floor dysfunction assessment classification model, n represents a sample number, represents a true label value of a sample, represents a predicted value of a sample.

Images