Thyroid nodule ultrasound image segmentation method, system and device, medium and product

Abstract

The invention discloses a thyroid nodule ultrasound image segmentation method, system and device, a medium and a product, and relates to the technical field of medical image processing.The method comprises the steps that after a thyroid nodule data set is obtained, the thyroid nodule data set is preprocessed, and the preprocessed data set comprises a training set, a verification set and a test set; respectively training and verifying the U-net network by using the training set and the verification set to obtain a thyroid nodule image segmentation model; testing the thyroid nodule image segmentation model by using the test set to obtain a thyroid nodule image segmentation result; the problems that the omission ratio of small nodules is high and nodule boundaries are difficult to accurately segment can be solved, and the thyroid nodule detection precision is improved.

Description

Technical Field

[0001] This application relates to the technical field of medical image processing, and particularly to a method, system, device, medium and product for segmenting ultrasound images of thyroid nodules. Background Art

[0002] Thyroid cancer is one of the common malignant tumors of the endocrine system. If not detected and treated in time, it will lead to the deterioration of the condition and even metastasis to other organs. Accurately detecting and evaluating thyroid nodules helps to detect the lesions of patients early and formulate a reasonable treatment plan to prevent thyroid cancer. At present, due to its non-invasive, safe nature and advantages in soft tissue imaging, ultrasound technology has become the preferred method for thyroid nodule screening. However, existing thyroid nodule segmentation methods, such as U-net and its variants, although performing well in many medical image segmentation tasks, still have problems of high missed detection rate for small nodules and difficulty in accurately segmenting the nodule boundary when facing the low contrast, speckle noise and diversity of nodule morphology in ultrasound imaging data.

[0003] Therefore, there is an urgent need to provide a method and system for segmenting ultrasound images of thyroid nodules to overcome the problems of high missed detection rate for small nodules and difficulty in accurately segmenting the nodule boundary of existing thyroid nodule segmentation methods for ultrasound images, improve the detection accuracy of thyroid nodules, and accurately segment the boundary of thyroid nodules. Summary of the Invention

[0004] The purpose of this application is to provide a method, system, device, medium and product for segmenting ultrasound images of thyroid nodules, which can overcome the problems of high missed detection rate for small nodules and difficulty in accurately segmenting the nodule boundary of existing thyroid nodule segmentation methods for ultrasound images, improve the detection accuracy of thyroid nodules, and accurately segment the boundary of thyroid nodules.

[0005] To achieve the above purpose, this application provides the following solutions:

[0006] In the first aspect, this application provides a method for segmenting ultrasound images of thyroid nodules, and the method for segmenting ultrasound images of thyroid nodules includes:

[0007] Obtain a thyroid nodule data set; the thyroid nodule data set includes: image samples and segmentation labels;

[0008] Preprocess the thyroid nodule data set to obtain a preprocessed data set; the preprocessed data set includes: a training set, a validation set and a test set;

[0009] The U-net network is trained and verified using the training set and the validation set respectively to obtain a thyroid nodule image segmentation model; the thyroid nodule image segmentation model takes an image as input and outputs the segmentation result of the thyroid nodule; the thyroid nodule image segmentation model includes: an encoding network and a decoding network; the encoding network includes a first encoding layer, a second encoding layer, a third encoding layer, a fourth encoding layer, a fifth encoding layer and 4 downsampling layers; the first encoding layer, the second encoding layer, the third encoding layer and the fourth encoding layer each include 2 multi-branch convolution modules; the fifth encoding layer includes a scale selection atrous pyramid module; the multi-branch convolution module and the scale selection atrous pyramid module are used to capture multi-scale context information; the decoding network includes a first fusion layer, a second fusion layer, a third fusion layer, a fourth fusion layer, 4 decoding layers and 4 upsampling layers; the first fusion layer includes 3 cross-level fusion modules and 1 convolution layer; the second fusion layer includes 2 cross-level fusion modules and 1 convolution layer; the third fusion layer includes 1 cross-level fusion module and 1 convolution layer;

[0010] The thyroid nodule image segmentation model is tested using the test set to obtain the thyroid nodule image segmentation result.

[0011] Optionally, the thyroid nodule data set is preprocessed to obtain a preprocessed data set, specifically including:

[0012] The thyroid nodule data set is scaled to obtain a scaled data set;

[0013] The scaled data set is data-augmented to obtain a preprocessed data set.

[0014] Optionally, the multi-branch convolution module includes: 3 two-dimensional convolution units, 3 batch normalization units and ReLU activation function units.

[0015] Optionally, the scale selection atrous pyramid module includes: a pooling path, a scale selection path and a feature reconstruction path;

[0016] The pooling path is used to process the input feature F in through three parallel adaptive pooling layers, reduce the two-dimensional space of the channels, perform feature mapping using the ReLU function and two-dimensional convolution, and restore it to the input feature size through bilinear interpolation operation to obtain the feature F pool,i ;

[0017] The scale selection path is used to process the input feature F in through three parallel two-dimensional atrous convolutions with different dilation rates to obtain three features F atrous,n with different scales, and concatenate them along the channel dimension to obtain a multi-scale feature F atrous ;

[0018] The feature reconstruction path is used to transform the input feature F in to increase the number of channels through a two-dimensional convolution to obtain the feature F in '; and then the feature F pool,i , the multi-scale feature F atrous and the feature F in ' are concatenated along the channel dimension and integrated using a two-dimensional convolution to obtain the scale selection atrous pyramid output feature.

[0019] Optionally, the cross-level fusion module includes: a transposed convolution operation layer and two parallel depthwise separable convolution layers;

[0020] The transposed convolution operation layer is used to map the feature F i to the same dimension as the feature F i-1 and perform concatenation to obtain the fused feature F c ; where the feature F i and the feature F i-1 are two adjacent-level features;

[0021] The depthwise separable convolution layer is used to map the fused feature F c and perform concatenation along the channel dimension to obtain the cross-level fusion output feature.

[0022] Optionally, the U-net network is trained and verified using the training set and the verification set respectively to obtain a thyroid nodule image segmentation model, specifically including:

[0023] Using the formula to determine the loss between the thyroid nodule image segmentation result and the segmentation label; where Loss is the loss function, N is the whole of all pixels in the image, X(i,j) is the prediction result of the pixel (i,j), and Y(i,j) is the ground truth label.

[0024] Determine the thyroid nodule image segmentation model according to the loss between the thyroid nodule image segmentation result and the segmentation label.

[0025] In a second aspect, the present application provides a thyroid nodule ultrasound image segmentation system, and the thyroid nodule ultrasound image segmentation system includes:

[0026] A dataset acquisition module for acquiring a thyroid nodule dataset; the thyroid nodule dataset includes: image samples and segmentation labels;

[0027] A preprocessing module for preprocessing the thyroid nodule dataset to obtain a preprocessed dataset; the preprocessed dataset includes: a training set, a verification set, and a test set;

[0028] A thyroid nodule image segmentation model building module, which is used to train and validate the U-net network by using the training set and the validation set respectively to obtain a thyroid nodule image segmentation model; the thyroid nodule image segmentation model takes an image as input and outputs the segmentation result of the thyroid nodule; the thyroid nodule image segmentation model includes: an encoding network and a decoding network; the encoding network includes a first encoding layer, a second encoding layer, a third encoding layer, a fourth encoding layer, a fifth encoding layer and 4 downsampling layers; the first encoding layer, the second encoding layer, the third encoding layer and the fourth encoding layer all include 2 multi-branch convolution modules; the fifth encoding layer includes a scale selection atrous pyramid module; the multi-branch convolution module and the scale selection atrous pyramid module are used to capture multi-scale context information; the decoding network includes a first fusion layer, a second fusion layer, a third fusion layer, a fourth fusion layer, 4 decoding layers and 4 upsampling layers; the first fusion layer includes 3 cross-level fusion modules and 1 convolution layer; the second fusion layer includes 2 cross-level fusion modules and 1 convolution layer; the third fusion layer includes 1 cross-level fusion module and 1 convolution layer;

[0029] A segmentation result determination module, which is used to test the thyroid nodule image segmentation model by using the test set to obtain a thyroid nodule image segmentation result.

[0030] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the computer program to implement the steps of a thyroid nodule ultrasound image segmentation method as described in any one of the above.

[0031] In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of a thyroid nodule ultrasound image segmentation method as described in any one of the above are implemented.

[0032] In a fifth aspect, the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, the steps of a thyroid nodule ultrasound image segmentation method as described in any one of the above are implemented.

[0033] According to the specific embodiments provided by the present application, the present application has the following technical effects:

[0034] The present application provides a method, system, device, medium and product for segmenting thyroid nodule ultrasound images. Based on the encoding network and decoding network of the U-net network, a multi-scale representation strategy is adopted in the encoding network part to obtain more information in the image and improve the accuracy of the image segmentation result. In the decoding network part, the encoding features are fused level by level through cross-level fusion operations and supplemented to the upsampling process to gradually restore the image, improving the granularity information of the image edge, and further improving the detection accuracy of thyroid nodules, and being able to accurately segment the boundary of thyroid nodules. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0036] Figure 1 It is a flowchart of a method for segmenting thyroid nodule ultrasound images in an embodiment of the present application;

[0037] Figure 2 It is a schematic structural diagram of the U-net network in an embodiment of the present application;

[0038] Figure 3 It is a schematic structural diagram of a multi-branch convolution module in an embodiment of the present application;

[0039] Figure 4 It is a schematic structural diagram of a scale selection atrous pyramid module in an embodiment of the present application;

[0040] Figure 5 It is a schematic structural diagram of a fusion layer in an embodiment of the present application;

[0041] Figure 6 It is a schematic structural diagram of a cross-level fusion module in an embodiment of the present application;

[0042] Figure 7 It is a schematic structural diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] 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 embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0044] To make the above objects, features, and advantages of the present application more apparent and understandable, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0045] In an exemplary embodiment, as Figure 1 shown, a method for segmenting thyroid nodule ultrasound images is provided, including the following S1 - S4, where:

[0046] S1: Obtain a thyroid nodule dataset; the thyroid nodule dataset includes: image samples and segmentation labels.

[0047] S2: Preprocess the thyroid nodule dataset to obtain a preprocessed dataset.

[0048] Scale the sizes of all image samples and segmentation labels in the thyroid nodule dataset to 256×256 pixel points, and use data augmentation methods such as normalization and random flipping to obtain preprocessed thyroid nodule images and segmentation labels. Divide 64% of the thyroid nodule dataset into a training set, 16% into a validation set, and 20% into a test set.

[0049] S3: Use the training set and the validation set to train and validate the U - net network respectively to obtain a thyroid nodule image segmentation model.

[0050] During the training and validation process of the thyroid nodule feature extraction model, the training set and the validation set are alternately input, and the number of iterations is set. The process of each iteration is as follows:

[0051] Suppose a pair of thyroid nodule images of a sample is defined as Input∈R 3×h×w and the corresponding segmentation label is defined as Iabel∈R 1×h×w . Input the training set into the U - net network for forward propagation and backpropagation to update the network parameters. During the forward propagation process, the thyroid nodule image Input passes through the encoding network and the decoding network to obtain the output Output; during the backpropagation process, use the loss function formula to calculate the loss Loss between the network output Output and the segmentation label Iabel, and then calculate the backpropagation gradient and update the network parameters. The loss function formula is as follows:

[0052]

[0053] where the variable N encapsulates the entirety of all pixels within the image and refers to the model's prediction of the pixel value at position (i, j). Among them, X(i, j) represents the prediction result of pixel (i, j), and Y(i, j) represents the true label.

[0054] Subsequently, the validation set is input into the U-net network for forward propagation, and the thyroid nodule segmentation result is output and the accuracy index is calculated to verify the performance of the model at this time. If the performance of the model at this time is better than that of the previous optimal model, the network parameters of the model at this time are saved. The training process is repeated, and the network parameters are updated multiple times to obtain the final thyroid nodule image segmentation model.

[0055] The U-net network includes an encoding network and a decoding network. Among them, the encoding network includes 5 encoding layers and 4 downsampling layers; the decoding network includes 4 decoding layers, 4 upsampling layers and 3 fusion layers. The input end of the first encoding layer is used to input the thyroid nodule image, and the output end of the first decoding layer is used to output the thyroid nodule segmentation result.

[0056] In an exemplary embodiment, as Figure 2 shown, the output end of each encoding layer is connected to the input end of the corresponding downsampling layer, the input end of the fifth encoding layer is connected to the output end of the fourth downsampling layer, the output end of the fifth encoding layer is connected to the input end of the fourth upsampling layer, and the input end of the fourth decoding layer is connected to the output end of the fourth upsampling layer; the input end of the first fusion layer is connected to the output ends of the second encoding layer, the third encoding layer, the fourth encoding layer and the fifth encoding layer, and the output of the first fusion layer is added element-wise to the output of the first upsampling layer and then connected to the input end of the first decoding layer; the input end of the second fusion layer is connected to the output ends of the third encoding layer, the fourth encoding layer and the fifth encoding layer, and the output of the second fusion layer is added element-wise to the output of the second upsampling layer and then connected to the input end of the second decoding layer; the input end of the third fusion layer is connected to the output ends of the fourth encoding layer and the fifth encoding layer, and the output of the third fusion layer is added element-wise to the output of the third upsampling layer and then connected to the input end of the third decoding layer.

[0057] In the encoding network, the encoding layer is used to encode the input image Input to obtain five different levels of feature outputs F i e (i = 1, 2, 3, 4, 5), which respectively correspond to the feature levels of the network.

[0058] The first to fourth encoding layers are all composed of two multi-branch convolution modules. In an exemplary embodiment, as Figure 3 shown, the multi-branch convolution module is composed of 3 parallel two-dimensional convolution units, 3 batch normalization units and ReLU activation function units. Given the input feature map F in ∈R c′×h×w , three different-scale feature outputs F 1 , F 2 , F 3Feature aggregation is performed through element addition, and the ReLU activation function is used to obtain the final multi-scale feature output F out ∈R c×h×w The calculation formula is as follows:

[0059]

[0060] F 3 = B(Conv 1×1 (F in ))

[0061] F out = δ(F 1 + F 2 + F 3 )

[0062] Among them, Conv 1×1 (·), and are 1×1 convolution, 3×3 convolution with a dilation rate of 1, and 3×3 convolution with a dilation rate of 2 respectively. B represents the batch normalization operation, and δ represents the ReLU activation function

[0063] The fifth encoding layer is composed of a scale selection atrous pyramid module. In an exemplary embodiment, as Figure 4 shown, the scale selection atrous pyramid module consists of a pooling path, a scale selection path, and a feature reconstruction path

[0064] On the pooling path, the input feature F in ∈R c×h×w reduces the two-dimensional space h×w of each channel to i×i (i = 1, 2, 4) through three parallel adaptive pooling layers. Then, the ReLU function and 1×1 two-dimensional convolution are used for feature mapping. Finally, it is restored to the input feature size through bilinear interpolation operation. The calculation formula is as follows

[0065] F pool,i = interpol(Conv 1×1 (δ(APool(F in ), i)), i = 1, 2, 4

[0066] Among them, interpol(·) is bilinear interpolation, and APool(·) is an adaptive pooling operation

[0067] On the scale path, the input feature F in obtains three features F atrous,n of different scales through two-dimensional atrous convolution with three different dilation rates n (n = 1, 2, 4) in parallel, and they are concatenated along the channel dimension to obtain the multi-scale feature F atrous The calculation formula is as follows

[0068]

[0069] F atrous = Concat[F atrous,1 , F atrous,2 , F atrous,4 。

[0070] Among them, Concat is to concatenate different features in the channel dimension.

[0071] Then, scale selection features are obtained through a scale selection-based spatial attention mechanism. It uses conventional convolution operations to adjust the channels of F atrous and obtains spatial weights at three different scales through a Softmax classifier F attention,n . Multiply the spatial weight map by the feature map corresponding to each scale F atrous,n and finally add the obtained results to get the output feature F conv on the scale selection path. The calculation formula is as follows:

[0072] F attention,n = Softmax(Conv * (F atrous ))。

[0073] F conv = F atrous,1 × F attention,1 + F atrous,2 × F attention,2 + F atrous,4 × F attention,4 。

[0074] Among them, Conv * (·) represents a conventional convolution combination using 1×1 convolution, 3×3 convolution, and ReLU activation function.

[0075] On the feature reconstruction path, a 1×1 two-dimensional convolution is used to increase the channel dimension of the input feature F in , and the calculation formula is as follows:

[0076] F in ′ = Conv 1×1 (F in )。

[0077] Finally, the output features of the three paths are concatenated along the channel dimension and integrated using a 1×1 two-dimensional convolution to obtain the output feature, and the calculation formula is as follows:

[0078] F out = δ(B(Conv 1×1 (Concat[F conv , Fin ′,F pool,i )))。

[0079] In the decoding network, the decoding layer and the upsampling layer are used to decode the input encoded features to obtain four different levels of feature outputs F from shallow to deep i d (i = 1, 2, 3, 4), corresponding to the feature levels of the network respectively. The encoded features and are used as the inputs of the fusion layer to gradually integrate high-level features and low-level features. Then, the output of the fusion layer is added to the decoding process of the corresponding level to better restore the fine-grained information of the image.

[0080] The 4-layer decoding layer is composed of two convolutional layers. Given the input feature map F in ∈R c′×h×w and the decoding output F out ∈R c×h×w . The calculation method of each convolutional layer is as follows:

[0081] F out = δ(B(Conv 3×3 (F in ))).

[0082] As Figure 5 shown, the first fusion layer is composed of 3 cross-level fusion modules and 1 convolutional layer, the second fusion layer is composed of 2 cross-level fusion modules and 1 convolutional layer, and the third fusion layer is composed of 1 cross-level fusion module and 1 convolutional layer.

[0083] In the first fusion layer, the feature maps output by each encoding layer are gradually aggregated through two cross-level fusion units to obtain the prior knowledge rich in high-level semantic information

[0084]

[0085] where CLFU(·) represents the cross-level fusion unit, and represent the inputs of the cross-level fusion unit.

[0086] Secondly, and are fused through a cross-level fusion unit and passed through a 3×3 two-dimensional convolutional layer to obtain the output F fusion1 of the fusion layer.

[0087]

[0088] Finally, the output F fusion1Perform an element-wise addition operation with the output of the first upsampling layer, and then connect it to the input end of the first decoding layer.

[0089] Specifically, the cross-level fusion unit receives the features of two adjacent levels and F i-1 ∈R c×h×w , as Figure 6 shown. First, map F i-1 to the same dimension as F i through a transposed convolution operation and concatenate them to obtain the fused feature F c ∈R 2c×h×w . The calculation formula is as follows:

[0090] F c = Concat[U(F i ), F i-1 .

[0091] Among them, U(·) represents the transposed convolution.

[0092] Then, perform feature mapping through two parallel depthwise separable convolutional layers and concatenate them along the channel dimension to obtain the output feature.

[0093]

[0094] F out = Concat[F 1 , F 2 .

[0095] Among them, represents a depthwise separable convolution with a dilation rate set to n and a kernel size of 3×3.

[0096] The cross-level fusion unit receives the features of two adjacent levels. First, use a transposed convolution to unify the sizes of the two features and concatenate them along the channel to obtain the fused feature. Subsequently, perform feature mapping through two parallel depthwise separable convolutions and concatenate them along the channel dimension to obtain the output feature.

[0097] The specific steps of the second fusion layer and the third fusion layer are similar to those of the first fusion layer 1.

[0098] S4: Use the test set to test the thyroid nodule image segmentation model to obtain the thyroid nodule image segmentation result.

[0099] Substitute the test set of thyroid nodule images into the trained thyroid nodule graphic segmentation model, and forward propagation to obtain the segmentation result of the thyroid nodule images.

[0100] After obtaining the thyroid nodule dataset, this application preprocesses the thyroid nodule dataset, and uses the training set and the validation set to train and validate the U-net network respectively to obtain a thyroid nodule image segmentation model. The U-net network includes an encoding network and a decoding network. This application adopts a multi-scale representation strategy in the encoding network part to obtain more information in the image and improve the accuracy of the image segmentation result. In the decoding network part, through cross-level fusion operations, the encoded features are fused level by level and supplemented to the upsampling process to gradually restore the image, improving the granularity information of the image edge, and thus improving the detection accuracy of thyroid nodules and being able to accurately segment the boundary of thyroid nodules. It overcomes the problems of high missed detection rate for small nodules and difficulty in accurately segmenting the nodule boundary.

[0101] In an exemplary embodiment, a thyroid nodule ultrasound image segmentation system is provided, including:

[0102] A dataset acquisition module for acquiring a thyroid nodule dataset; the thyroid nodule dataset includes: image samples and segmentation labels.

[0103] A preprocessing module for preprocessing the thyroid nodule dataset to obtain a preprocessed dataset; the preprocessed dataset includes: a training set, a validation set, and a test set.

[0104] A thyroid nodule image segmentation model establishment module for training and validating the U-net network using the training set and the validation set respectively to obtain a thyroid nodule image segmentation model; the thyroid nodule image segmentation model takes an image as input and outputs the segmentation result of thyroid nodules; the thyroid nodule image segmentation model includes: an encoding network and a decoding network; the encoding network includes a first encoding layer, a second encoding layer, a third encoding layer, a fourth encoding layer, a fifth encoding layer, and 4 downsampling layers; the first encoding layer, the second encoding layer, the third encoding layer, and the fourth encoding layer each include 2 multi-branch convolution modules; the fifth encoding layer includes a scale selection atrous pyramid module; the multi-branch convolution module and the scale selection atrous pyramid module are used to capture multi-scale context information; the decoding network includes a first fusion layer, a second fusion layer, a third fusion layer, a fourth fusion layer, 4 decoding layers, and 4 upsampling layers; the first fusion layer includes 3 cross-level fusion modules and 1 convolution layer; the second fusion layer includes 2 cross-level fusion modules and 1 convolution layer; the third fusion layer includes 1 cross-level fusion module and 1 convolution layer.

[0105] A segmentation result determination module for testing the thyroid nodule image segmentation model using the test set to obtain a thyroid nodule image segmentation result.

[0106] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structure diagram can be as shown in Figure 7 . The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store a thyroid nodule image segmentation model. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a method for segmenting thyroid nodule ultrasound images.

[0107] Those skilled in the art can understand that Figure 7 the structure shown in is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in Figure 7 , or combine some components, or have different component arrangements.

[0108] In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, it implements the steps in the above method embodiments.

[0109] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by the processor, it implements the steps in the above method embodiments.

[0110] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by the processor, it implements the steps in the above method embodiments.

[0111] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random access memories (ReRAM), magnetoresistive random access memories (MRAM), ferroelectric random access memories (FRAM), phase change memories (PCM), graphene memories, etc. Volatile memories can include random access memory (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0112] The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logics, data processing logics based on quantum computing, etc., without limitation.

[0113] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered to be within the scope described in this specification.

[0114] Specific examples are used in this article to elaborate on the principles and implementation manners of the present application. The descriptions of the above embodiments are only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A thyroid nodule ultrasound image segmentation method, characterized in that: The thyroid nodule ultrasound image segmentation method comprises: Acquire a thyroid nodule dataset; the thyroid nodule dataset includes: image samples and segmentation labels; Preprocessing the thyroid nodule data set to obtain a preprocessed data set; the preprocessed data set includes: a training set, a validation set, and a test set; The U-net network is trained and verified using the training set and the validation set respectively to obtain a thyroid nodule image segmentation model; the thyroid nodule image segmentation model takes the image as input and takes the thyroid nodule segmentation result as output; the thyroid nodule image segmentation model includes: an encoding network and a decoding network; the encoding network includes a first encoding layer, a second encoding layer, a third encoding layer, a fourth encoding layer, a fifth encoding layer and four downsampling layers; the first encoding layer, the second encoding layer, the third encoding layer and the fourth encoding layer each include two multi-branch convolution modules; the fifth encoding layer includes a scale-selective dilated pyramid module; the multi-branch convolution module and the scale-selective dilated pyramid module are used to capture multi-scale contextual information; the decoding network includes a first fusion layer, a second fusion layer, a third fusion layer, a fourth fusion layer, four decoding layers and four upsampling layers; the first fusion layer includes three cross-level fusion modules and one convolution layer; the second fusion layer includes two cross-level fusion modules and one convolution layer; the third fusion layer includes one cross-level fusion module and one convolution layer; The thyroid nodule image segmentation model is tested using the test set to obtain a thyroid nodule image segmentation result.

2. The thyroid nodule ultrasonic image segmentation method according to claim 1, characterized in that: Preprocessing the thyroid nodule data set to obtain a preprocessed data set specifically includes: Scaling the size of the thyroid nodule dataset to obtain a scaled dataset; Data enhancement is performed on the scaled data set to obtain a preprocessed data set.

3. The thyroid nodule ultrasonic image segmentation method according to claim 1, characterized in that: The multi-branch convolution module includes: 3 two-dimensional convolution units, 3 batch normalization units and a ReLU activation function unit.

4. The thyroid nodule ultrasonic image segmentation method according to claim 1, characterized in that: The scale selection hollow pyramid module includes: a pooling path, a scale selection path and a feature reconstruction path; The pooling path is used to transform the input feature F in Through three parallel adaptive pooling layers, the two-dimensional space of the channel is reduced, the ReLU function and two-dimensional convolution are used for feature mapping, and the input feature size is restored through bilinear interpolation operation to obtain the feature F pool,i ; The scale selection path is used to transform the input feature F in Three features of different scales F are obtained by parallelizing three two-dimensional dilated convolutions with different dilation rates. atrous,n , and concatenate along the channel dimension to obtain the multi-scale feature F atrous ; The feature reconstruction path is used to transform the input feature F in Through two-dimensional convolution, the channel dimension is increased to obtain the feature F in '; and then the feature F pool,i , the multi-scale feature F atrous and the feature F in ′ is concatenated along the channel dimension and integrated using two-dimensional convolution to obtain the scale-selective hollow pyramid output features.

5. The thyroid nodule ultrasonic image segmentation method according to claim 1, characterized in that: The cross-level fusion module includes: a transposed convolution operation layer and two parallel depth-separable convolution layers; The transposed convolution operation layer is used to transform the feature F i Mapped to the feature F i-1 The same dimension, and concatenated to obtain the fusion feature F c ; Among them, feature F i and feature F i-1 are two adjacent level features; The depth-wise separable convolutional layer is used to transform the fusion feature F c Mapping is performed and concatenated along the channel dimension to obtain cross-level fusion output features.

6. The thyroid nodule ultrasonic image segmentation method according to claim 1, characterized in that: The U-net network is trained and verified using the training set and the verification set respectively to obtain a thyroid nodule image segmentation model, which specifically includes: Using the formula Determine the loss of thyroid nodule image segmentation results and segmentation labels; where Loss is the loss function, N is the total number of pixels in the image, X(i,j) is the predicted result of pixel (i,j), and Y(i,j) is the true label. A thyroid nodule image segmentation model is determined according to the thyroid nodule image segmentation result and the loss of the segmentation label.

7. A thyroid nodule ultrasound image segmentation system, characterized in that: The thyroid nodule ultrasound image segmentation system comprises: A data set acquisition module is used to acquire a thyroid nodule data set; the thyroid nodule data set includes: image samples and segmentation labels; A preprocessing module, used to preprocess the thyroid nodule data set to obtain a preprocessed data set; the preprocessed data set includes: a training set, a validation set and a test set; A thyroid nodule image segmentation model establishment module is used to use the training set and the verification set to train and verify the U-net network respectively to obtain a thyroid nodule image segmentation model; the thyroid nodule image segmentation model takes the image as input and takes the thyroid nodule segmentation result as output; the thyroid nodule image segmentation model includes: an encoding network and a decoding network; the encoding network includes a first encoding layer, a second encoding layer, a third encoding layer, a fourth encoding layer, a fifth encoding layer and four downsampling layers; the first encoding layer, the second encoding layer, the third encoding layer and the fourth encoding layer each include two multi-branch convolution modules; the fifth encoding layer includes a scale-selective dilated pyramid module; the multi-branch convolution module and the scale-selective dilated pyramid module are used to capture multi-scale context information; the decoding network includes a first fusion layer, a second fusion layer, a third fusion layer, a fourth fusion layer, four decoding layers and four upsampling layers; the first fusion layer includes three cross-level fusion modules and one convolution layer; the second fusion layer includes two cross-level fusion modules and one convolution layer; the third fusion layer includes one cross-level fusion module and one convolution layer; The segmentation result determination module is used to test the thyroid nodule image segmentation model using the test set to obtain a thyroid nodule image segmentation result.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the thyroid nodule ultrasound image segmentation method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for segmenting thyroid nodules ultrasonic images according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for segmenting thyroid nodules ultrasonic images according to any one of claims 1 to 6 is implemented.

Images