Ultrasonic elastic imaging segmentation method and system for thyroid nodules and breast nodules

Abstract

The present invention discloses an ultrasonic elastic imaging segmentation method and system for thyroid nodules and breast nodules, and relates to the technical field of ultrasonic elastic imaging segmentation. The ultrasonic elastic imaging segmentation method and system for thyroid nodules and breast nodules, by setting an image acquisition module, a feature extraction module, a model construction module, an image segmentation module, and a model evaluation module, combines a decoder optimized by a channel attention mechanism and a jump connection to construct an image segmentation model, and pays more attention to the positioning accuracy of the segmented area in the bottom stage of the model, thereby increasing the precision of the model segmentation. In the shallow stage of the model, more attention is paid to the recognition of the edge of the segmented area, thereby increasing the segmentation and fullness of the model. Accurate segmentation of thyroid nodules and breast nodules is of great significance for judging whether thyroid cancer and breast cancer have metastasized, and good thyroid nodule and breast nodule segmentation effects are obtained in ultrasonic elastic images, which has a high clinical application value.

Description

Technical Field

[0001] The present invention relates to the technical field of ultrasonic elastic imaging segmentation, and in particular to an ultrasonic elastic imaging segmentation method and system for thyroid nodules and breast nodules. Background Art

[0002] Thyroid cancer and breast cancer are the two most common malignant tumors in the world. The choice of treatment is closely related to the stage, and determining whether thyroid nodules and breast nodules are involved is one of the key factors in determining the stage of cancer. Ultrasound elastography is a useful means of early diagnosis of benign and malignant nodules. Since the shapes and appearances of thyroid nodules and breast nodules are similar in imaging, and the diagnostic discrimination methods are similar, the ultrasound images of thyroid nodules and breast nodules can be segmented using the same technical means. A large number of studies have shown that compared with traditional ultrasound imaging, ultrasound elastography can more accurately reflect the information of thyroid nodules and breast nodules, and uses different colors to reflect the hardness information of thyroid nodules and breast nodules, which has a higher diagnostic value.

[0003] Ultrasound elastography is a new type of ultrasound diagnostic technology that has developed rapidly in recent years. It uses dynamic imaging technology to measure tissue hardness, and by analyzing the hardness differences of different tissues, it can achieve non-invasive diagnosis of lesion tissues. At present, most ultrasound elastography examinations of thyroid nodules and breast nodules use real-time elastography technology. The working principle of this technology is: the doctor manually applies pressure to the skin on the probe to deform the nodule and measure the strain and hardness of the tissue. In real-time elastography, the hardness of the nodule is displayed on the ultrasound image through color coding, and different colors represent different hardness levels. Usually, malignant nodules appear darker in the image due to their harder structure, while benign nodules appear lighter in color. Therefore, accurate positioning and segmentation of thyroid nodules and breast nodules based on ultrasound elastic images has become one of the important steps in the diagnosis and treatment of thyroid cancer and breast cancer.

[0004] At present, in the study of ultrasound elastic images, professional doctors are usually required to manually segment nodules. This process is not only time-consuming and laborious, but also due to individual differences between doctors, the analysis and evaluation of the same image is prone to subjective bias or omission of important features, resulting in certain segmentation errors. Therefore, it is of great significance to study the method of automatic segmentation of nodules in ultrasound elastic images. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides a method and system for ultrasonic elastic imaging segmentation of thyroid nodules and breast nodules, which solves the problems of the above-mentioned background technology.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: an ultrasonic elastic imaging segmentation system for thyroid nodules and breast nodules, comprising the following modules: an image acquisition module, a feature extraction module, a model building module, an image segmentation module, and a model evaluation module; the image acquisition module is used to acquire thyroid nodule images and breast nodule images of target patients; the feature extraction module is used to extract image features of thyroid nodule image data and breast nodule images of target patients based on the convolutional neural network DenseNet in deep learning as an encoder, and acquire feature information in thyroid nodule images and breast nodule images; the model building ... The model construction module is used to design a multi-scale fusion enhancement decoder based on the feature information in the thyroid nodule images and breast nodule images, combining the attention mechanism and the hole convolution, modeling the boundaries and textures of the nodules from multiple scales and ranges, and constructing an image segmentation model; the image segmentation module is used to segment the thyroid nodule image data and breast nodule image data of the target patient based on the image segmentation model, and obtain the segmented image data of the thyroid nodule area and the breast nodule area; the model evaluation module is used to evaluate the image segmentation model, obtain the segmentation index of the image segmentation model, analyze the segmentation index of the image segmentation model, and correct the image segmentation model.

[0007] Furthermore, the specific process of image feature extraction of the target patient's thyroid nodule image data and breast nodule image is as follows: based on the convolutional neural network DenseNet in deep learning as an encoder, the target patient's thyroid nodule image and breast nodule image are preprocessed; the preprocessed thyroid nodule image and breast nodule image are input into DenseNet, and through the forward propagation process, the feature information in the thyroid nodule image and the breast nodule image is extracted, and the feature information includes the shape, texture, and edge of the nodule; the features extracted from DenseNet are represented in vector or matrix form, and the fully connected layer is used to map the features to a fixed-length vector space and store them in the image feature database.

[0008] Furthermore, the specific process of constructing the image segmentation model is as follows: upsampling the feature information in the thyroid nodule images and breast nodule images output by the encoder, and dynamically adjusting the weight coefficients of features of different scales through the attention mechanism; modeling the boundaries and textures of nodules from multiple scales and ranges, and designing jump connections through the selective kernel network to fully integrate the intermediate features of the encoder with the output features of the decoder to construct an image segmentation model.

[0009] Furthermore, the thyroid nodule region and breast nodule region segmentation image data specifically include thyroid nodule segmentation data and breast nodule segmentation data; the thyroid nodule segmentation data include the area of ​​the thyroid nodule region segmented by the image segmentation model, the real area of ​​the thyroid nodule, and the area of ​​the thyroid nodule overlapping area; the breast nodule segmentation data include the area of ​​the breast nodule region segmented by the image segmentation model, the real area of ​​the breast nodule, and the area of ​​the breast nodule overlapping area.

[0010] Furthermore, the specific process of evaluating the image segmentation model is as follows: based on the thyroid nodule segmentation data, the accuracy of the image segmentation model in segmenting thyroid nodules is evaluated, and the thyroid nodule segmentation index of the image segmentation model is obtained; based on the breast nodule segmentation data, the accuracy of the image segmentation model in segmenting breast nodules is evaluated, and the breast nodule segmentation index of the image segmentation model is obtained.

[0011] Furthermore, the specific process of obtaining the segmentation index of the image segmentation model is as follows: compare the image segmentation model segmentation thyroid nodule index and the image segmentation model segmentation breast nodule index with the set standard value of the image segmentation model segmentation thyroid nodule index and the standard value of the image segmentation model segmentation breast nodule index, respectively, and perform ratio operation with the allowable deviation value of the image segmentation model segmentation thyroid nodule index and the allowable deviation value of the image segmentation model segmentation breast nodule index; assign weights to the image segmentation model segmentation thyroid nodule index and the image segmentation model segmentation breast nodule index, and perform comprehensive operation processing to obtain the segmentation index of the image segmentation model.

[0012] Furthermore, it is characterized in that: the specific formula of the segmentation index of the image segmentation model is as follows:

[0013]

[0014] In the formula, ωgh represents the segmentation index of the image segmentation model, TNS represents the thyroid nodule index segmented by the image segmentation model, BGH represents the breast nodule index segmented by the image segmentation model, JKS represents the standard value of the thyroid nodule index segmented by the image segmentation model, aTR represents the standard value of the breast nodule index segmented by the image segmentation model, ΔBS represents the allowable deviation value of the thyroid nodule index segmented by the image segmentation model, ΔVA represents the allowable deviation value of the breast nodule index segmented by the image segmentation model, φ1 represents the weight factor of the thyroid nodule index segmented by the image segmentation model, and φ2 represents the weight factor of the breast nodule index segmented by the image segmentation model.

[0015] Furthermore, the specific process of analyzing the segmentation index of the image segmentation model is as follows: based on the segmentation index of the image segmentation model, a comparative analysis is performed with the threshold of the segmentation index of the image segmentation model to obtain the segmentation compliance index of the image segmentation model; the segmentation compliance index of the image segmentation model is compared with the set segmentation compliance index range of the image segmentation model, and when the segmentation compliance index of the image segmentation model exceeds the segmentation compliance index range of the image segmentation model, the image segmentation model is corrected.

[0016] Furthermore, the specific process of obtaining the segmentation conformity index of the image segmentation model is as follows: the segmentation index of the image segmentation model, the image segmentation model segmentation thyroid nodule index and the image segmentation model segmentation breast nodule index are respectively compared with the threshold of the segmentation index of the image segmentation model, the threshold of the image segmentation model segmentation thyroid nodule index and the threshold of the image segmentation model segmentation breast nodule index, and comprehensive processing is performed to obtain the segmentation conformity index of the image segmentation model.

[0017] The ultrasonic elastic imaging segmentation method for thyroid nodules and breast nodules comprises the following steps: S1. Acquire thyroid nodule images and breast nodule images of target patients; S2. Based on the convolutional neural network DenseNet in deep learning as an encoder, perform image feature extraction on the thyroid nodule image data and breast nodule image of the target patient, and obtain feature information in the thyroid nodule image and the breast nodule image; S3. Based on the feature information in the thyroid nodule image and the breast nodule image, design a multi-scale fusion enhancement decoder in combination with the attention mechanism and the hole convolution, model the boundary and texture of the nodule from multiple scales and ranges, and construct an image segmentation model; S4. Based on the constructed segmentation model, segment the thyroid nodule image data and breast nodule image data of the target patient, and obtain segmented image data of the thyroid nodule area and the breast nodule area; S5. Evaluate the image segmentation model, obtain the segmentation index of the image segmentation model, analyze the segmentation index of the image segmentation model, and correct the image segmentation model.

[0018] The present invention has the following beneficial effects:

[0019] (1) The ultrasound elastic imaging segmentation system for thyroid nodules and breast nodules, by setting up an image acquisition module, a feature extraction module, a model construction module, an image segmentation module, and a model evaluation module, combines a decoder optimized by a channel attention mechanism and a jump connection to construct an image segmentation model. In the underlying stage of the model, more attention is paid to the positioning accuracy of the segmented area, which increases the precision of the model segmentation. In the shallow stage of the model, more attention is paid to the recognition of the edge of the segmented area, which increases the segmentation and fullness of the model. Accurate segmentation of thyroid nodules and breast nodules is of great significance for judging whether thyroid cancer and breast cancer have metastasized, and also plays an important role in the staging and treatment of cancer.

[0020] (2) The ultrasound elastic imaging segmentation method for thyroid nodules and breast nodules can provide important support and guidance for the research and development and application of ultrasound elastic imaging segmentation methods and systems by segmenting the thyroid nodule image data and breast nodule image data of the target patient, segmenting the thyroid nodule and breast nodule image data, evaluating the image segmentation model, and analyzing and correcting it.

[0021] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 The present invention is a flow chart of the ultrasonic elastic imaging segmentation system for thyroid nodules and breast nodules;

[0023] Figure 2 The present invention is a flow chart of the ultrasonic elastic imaging segmentation method for thyroid nodules and breast nodules. DETAILED DESCRIPTION

[0024] The embodiments of the present application use an ultrasonic elastic imaging segmentation method and system for thyroid nodules and breast nodules to reduce the time and labor cost of manual segmentation, improve segmentation accuracy and consistency, and promote clinical application and research progress.

[0025] The overall idea of ​​the problem in the embodiment of this application is as follows:

[0026] Acquire thyroid nodule images and breast nodule images of the target patient.

[0027] Based on the convolutional neural network DenseNet in deep learning as an encoder, image features are extracted from the thyroid nodule image data and breast nodule images of the target patients to obtain the feature information in the thyroid nodule images and breast nodule images.

[0028] Based on the feature information in thyroid nodule images and breast nodule images, a multi-scale fusion enhanced decoder is designed with attention mechanism and dilated convolution to model the boundaries and textures of nodules from multiple scales and ranges, and construct an image segmentation model.

[0029] Based on the constructed segmentation model, the thyroid nodule image data and breast nodule image data of the target patient are segmented, the segmented image data of the thyroid nodule area and the breast nodule area are obtained, the image segmentation model is evaluated, the segmentation index of the image segmentation model is obtained, the segmentation index of the image segmentation model is analyzed, and the image segmentation model is corrected.

[0030] See also Figure 1The embodiment of the present invention provides a technical solution: an ultrasonic elastic imaging segmentation system for thyroid nodules and breast nodules, comprising the following modules: an image acquisition module, a feature extraction module, a model construction module, an image segmentation module, and a model evaluation module; the image acquisition module is used to acquire the thyroid nodule image and the breast nodule image of the target patient; the feature extraction module is used to extract image features of the thyroid nodule image data and the breast nodule image of the target patient based on the convolutional neural network DenseNet in deep learning as an encoder, and acquire feature information in the thyroid nodule image and the breast nodule image; the model construction module is used to design a multi-scale fusion enhancement decoder based on the feature information in the thyroid nodule image and the breast nodule image, combine the attention mechanism and the hole convolution, model the boundary and texture of the nodule from multiple scales and ranges, and construct an image segmentation model; the image segmentation module is used to segment the thyroid nodule image data and the breast nodule image data of the target patient based on the image segmentation model, and acquire segmented image data of the thyroid nodule area and the breast nodule area; the model evaluation module is used to evaluate the image segmentation model, acquire the segmentation index of the image segmentation model, analyze the segmentation index of the image segmentation model, and correct the image segmentation model.

[0031] In this embodiment, DenseNet is a deep learning architecture designed to overcome the gradient vanishing problem in deep neural networks and promote feature reuse. Unlike traditional convolutional neural networks, DenseNet introduces the concept of dense connections, and the output of each layer is passed to all subsequent layers of the next layer. This densely connected structure allows each layer to directly access the feature maps from all previous layers, thereby enhancing the propagation and reuse of features, helping to improve the performance and training efficiency of the network. The convolutional neural network DenseNet in deep learning is used as an encoder to extract features from thyroid nodule images and breast nodule images. This helps to capture key feature information in the image and provide data support for subsequent image segmentation. Based on the feature information extracted from thyroid nodule images and breast nodule images, a multi-scale fusion enhancement decoder is designed by combining attention mechanism and void convolution. Such a design can better model the boundary and texture features of the nodules, thereby building a more accurate image segmentation model.

[0032] Specifically, the specific process of image feature extraction of thyroid nodule image data and breast nodule images of target patients is as follows: based on the convolutional neural network DenseNet in deep learning as an encoder, the thyroid nodule images and breast nodule images of the target patients are preprocessed; the preprocessed thyroid nodule images and breast nodule images are input into DenseNet, and through the forward propagation process, the feature information in the thyroid nodule images and breast nodule images is extracted, and the feature information includes the shape, texture, and edge of the nodules; the features extracted from DenseNet are represented as vectors or matrices, and the fully connected layer is used to map the features to a fixed-length vector space and store them in the image feature database.

[0033] In this embodiment, Dense-UNet uses DenseNet as the encoder of U-Net. Compared with the encoder structure of the original U-Net, the network has a densely connected deep convolution operation, and its feature extraction ability is stronger and more sensitive to channel features. Considering that ultrasonic elastic images can locate the position of thyroid nodules and breast nodules and provide rich channel information, the present invention designs an encoder structure of an attention multi-scale segmentation network based on the unique advantages of DenseNet for channel features. The encoder consists of a Stem feature extractor, 4 DenseBlocks and downsampling operations. Specifically, Stem includes a convolution layer with a convolution kernel size of 7×7, a batch normalization layer, a rectified linear activation function and a maximum pooling layer. DenseBlock is used to achieve more detailed feature extraction, and the number of densely connected layers contained in it varies with the deepening of the network. The 4 DenseBlocks contain 6, 12, 24, and 16 layers respectively. Each densely connected layer is combined using a skip connection method to fuse the features of each stage through channels to ensure that the model weighs the global features. The downsampling operation consists of a convolution layer with a convolution kernel of 1×1 and an average pooling layer. After each downsampling operation, the size and number of channels of the feature map are reduced by half to achieve feature dimension reduction and reduce the number of network parameters. The input ultrasound elasticity image size in this paper is 3×256×256. After passing through the designed DenseNet encoder, the final output feature is 1024×8×8.

[0034] Specifically, the specific process of constructing the image segmentation model is as follows: up-sample the feature information in the thyroid nodule images and breast nodule images output by the encoder, and dynamically adjust the weight coefficients of features of different scales through the attention mechanism; model the boundaries and textures of nodules from multiple scales and ranges, and design jump connections through the selective kernel network to fully integrate the intermediate features of the encoder with the output features of the decoder to construct an image segmentation model.

[0035] In this embodiment, the upsampling operation is used to restore the feature map output by the encoder to the same size as the original input image. This is usually achieved using techniques such as deconvolution or interpolation. The attention mechanism is used to dynamically adjust the weight coefficients of features of different scales so that the model can focus on areas or features that are more important to the image segmentation task. This can improve the model's perception of details such as nodule boundaries and textures. Through multi-scale modeling, the model is able to model the boundaries and textures of nodules considering different scales and ranges. This can help the model better understand the structural information at different scales in the image, thereby improving the accuracy and robustness of segmentation. Selective kernel network design usually involves the use of specific convolution kernel structures or network structures to capture different levels of feature information at different scales. This helps the model better adapt to the needs of image segmentation tasks and improves the performance of the model. Jump connection is a technique for connecting the intermediate features of the encoder with the output features of the decoder. Doing so can fully integrate feature information of different scales, help the decoder better restore detail information, and alleviate the problem of information loss.

[0036] Specifically, the thyroid nodule region and breast nodule region segmentation image data specifically include thyroid nodule segmentation data and breast nodule segmentation data; the thyroid nodule segmentation data include the thyroid nodule region area segmented by the image segmentation model, the real region area of ​​the thyroid nodule, and the thyroid nodule overlapping region area; the breast nodule segmentation data include the breast nodule region area segmented by the image segmentation model, the real region area of ​​the breast nodule, and the breast nodule overlapping region area.

[0037] In this embodiment, based on the image segmentation model, the thyroid nodule image data and breast nodule image data of the target patient are segmented to obtain the segmented image data of the thyroid nodule region and the breast nodule region. The segmented image data of the thyroid nodule region and the breast nodule region include the area of ​​the thyroid nodule region segmented by the image segmentation model: the number of pixels or area of ​​the thyroid nodule region predicted by the model. The area of ​​the real area of ​​the thyroid nodule: the number of pixels or area of ​​the thyroid nodule region in the real annotated data. The area of ​​the overlapping area of ​​the thyroid nodule: the number of pixels or area of ​​the overlapping between the model segmentation result and the real annotated result. This can be used to measure the accuracy and overlap of the model segmentation. The area of ​​the breast nodule region segmented by the image segmentation model: the number of pixels or area of ​​the breast nodule region predicted by the model. The area of ​​the real area of ​​the breast nodule: the number of pixels or area of ​​the breast nodule region in the real annotated data. The area of ​​the overlapping area of ​​the breast nodule: the number of pixels or area of ​​the overlapping between the model segmentation result and the real annotated result. This can be used to evaluate the performance of the model on the breast nodule segmentation task.

[0038] Specifically, the specific process of evaluating the image segmentation model is as follows: based on thyroid nodule segmentation data, the accuracy of the image segmentation model in segmenting thyroid nodules is evaluated, and the thyroid nodule segmentation index of the image segmentation model is obtained; based on breast nodule segmentation data, the accuracy of the image segmentation model in segmenting breast nodules is evaluated, and the breast nodule segmentation index of the image segmentation model is obtained.

[0039] In this implementation scheme, in addition to visually inspecting the segmentation results of the model by professional medical imaging experts and scoring the segmentation results according to actual conditions to obtain the image segmentation model segmentation thyroid nodule index, the image segmentation model segmentation thyroid nodule index can also be calculated by the following calculation formula. The specific formula of the image segmentation model segmentation thyroid nodule index is as follows:

[0040]

[0041] In the formula, TNS represents the index of thyroid nodules segmented by the image segmentation model, mod represents the area of ​​thyroid nodules segmented by the image segmentation model, tue represents the area of ​​the true area of ​​thyroid nodules, kpl represents the area of ​​thyroid nodule overlap, α1 represents the weight factor of the area of ​​thyroid nodule overlap to the area of ​​thyroid nodule segmented by the image segmentation model, and α2 represents the weight factor of the area of ​​thyroid nodule overlap to the area of ​​the true area of ​​thyroid nodules. The index of thyroid nodule segmentation by the image segmentation model is used to evaluate the performance of the image segmentation model. Generally, the higher the index, the better the segmentation effect of the model. The area of ​​thyroid nodule segmentation by the image segmentation model measures the segmentation effect of the model on thyroid nodules, that is, the size of the area occupied by the nodule that the model believes. The weight factor of the area of ​​thyroid nodule overlap to the area of ​​thyroid nodule segmented by the image segmentation model. This parameter may be used to adjust the importance of the area of ​​overlap in the evaluation to better reflect the accuracy of the model. The weight factor of the area of ​​thyroid nodule overlap to the area of ​​the true area of ​​thyroid nodules is used to adjust the importance of the area of ​​overlap relative to the area of ​​the true area to more accurately evaluate the performance of the model. The weight factor of the area of ​​the thyroid nodule overlap region to the area of ​​the thyroid nodule segmented by the image segmentation model and the weight factor of the area of ​​the thyroid nodule overlap region to the area of ​​the real thyroid nodule region are set according to the experience and knowledge of medical experts and clinical data. For example, if the area of ​​the thyroid nodule overlap region to the area of ​​the thyroid nodule segmented by the image segmentation model is more important in image segmentation, then the weight factor α1 may be assigned a larger value.

[0042] In addition to using the cross-validation method to evaluate the performance of the model in the breast nodule segmentation task, the data set is divided into multiple subsets, and then each subset is used as a validation set in turn, and the rest is used as a training set. The model is trained multiple times and evaluated to obtain the image segmentation model segmentation breast nodule index. In addition, the image segmentation model segmentation breast nodule index can also be calculated using the following calculation formula. The specific formula for the image segmentation model segmentation breast nodule index is as follows:

[0043]

[0044] In the formula, BGH represents the index of breast nodules segmented by the image segmentation model, Gbk represents the area of ​​the breast nodules segmented by the image segmentation model, Fhj represents the area of ​​the real area of ​​the breast nodules, Chm represents the area of ​​the overlapping area of ​​the breast nodules, β1 represents the weight factor of the area of ​​the overlapping area of ​​the breast nodules to the area of ​​the breast nodules segmented by the image segmentation model, and β2 represents the weight factor of the area of ​​the overlapping area of ​​the breast nodules to the area of ​​the real area of ​​the breast nodules. The index of breast nodules segmented by the image segmentation model is used to evaluate the accuracy of the model in segmenting breast nodules. The weight factor of the area of ​​the overlapping area of ​​the breast nodules to the area of ​​the breast nodules segmented by the image segmentation model and the weight factor of the area of ​​the overlapping area of ​​the breast nodules to the area of ​​the real area of ​​the breast nodules are set according to the experience and knowledge of medical experts and clinical data, respectively. The weight factor of the area of ​​the overlapping area of ​​the breast nodules to the area of ​​the breast nodules segmented by the image segmentation model is used to adjust the importance of the overlapping area in the overall performance evaluation. The weight factor of the area of ​​the overlapping area of ​​the breast nodule to the area of ​​the true area of ​​the breast nodule is used to balance the importance between the area of ​​the overlapping area and the area of ​​the true area. The breast nodule segmentation index of the image segmentation model can provide a quantitative evaluation of the performance of the model in the breast nodule segmentation task. By comparing the index values ​​of different models or the same model under different parameter settings, the accuracy, robustness and stability of the model can be evaluated. By analyzing the breast nodule segmentation index, the weaknesses and room for improvement of the model can be determined. When the index is low, the model has difficulty in accurately segmenting breast nodules in some cases, so the model structure can be adjusted, the loss function can be optimized, or the training data can be increased to improve the model performance.

[0045] Specifically, the specific process of obtaining the segmentation index of the image segmentation model is as follows: compare the image segmentation model segmentation thyroid nodule index and the image segmentation model segmentation breast nodule index with the set standard value of the image segmentation model segmentation thyroid nodule index and the standard value of the image segmentation model segmentation breast nodule index, respectively, and perform ratio operation with the allowable deviation value of the image segmentation model segmentation thyroid nodule index and the allowable deviation value of the image segmentation model segmentation breast nodule index; assign weights to the image segmentation model segmentation thyroid nodule index and the image segmentation model segmentation breast nodule index, and perform comprehensive operation processing to obtain the segmentation index of the image segmentation model.

[0046] In this embodiment, the thyroid nodule index and breast nodule index output by the image segmentation model are compared with a pre-set standard value to evaluate the accuracy of the model. The deviation between the model output and the standard value is calculated, and if the deviation is within the allowed range, the model output is considered acceptable. According to different importance, weights are assigned to the thyroid nodule index and the breast nodule index to reflect their contribution to the overall segmentation model performance. Using the assigned weights, the two indices are weighted averaged or other comprehensive operations are performed to obtain the comprehensive segmentation index of the image segmentation model. The performance of the image segmentation model in the identification of thyroid nodules and breast nodules can be more comprehensively understood, providing more reliable support for clinical treatment.

[0047] Specifically, in addition to obtaining the segmentation index of the image segmentation model by annotating and evaluating the segmentation results of the model by medical experts or experts in related fields, the segmentation index of the image segmentation model can also be calculated by the following calculation formula. The specific formula of the segmentation index of the image segmentation model is as follows:

[0048]

[0049] In the formula, ωgh represents the segmentation index of the image segmentation model, TNS represents the thyroid nodule index segmented by the image segmentation model, BGH represents the breast nodule index segmented by the image segmentation model, JKS represents the standard value of the thyroid nodule index segmented by the image segmentation model, aTR represents the standard value of the breast nodule index segmented by the image segmentation model, ΔBS represents the allowable deviation value of the thyroid nodule index segmented by the image segmentation model, ΔVA represents the allowable deviation value of the breast nodule index segmented by the image segmentation model, φ1 represents the weight factor of the thyroid nodule index segmented by the image segmentation model, and φ2 represents the weight factor of the breast nodule index segmented by the image segmentation model.

[0050] In this embodiment, the segmentation index of the image segmentation model is used to evaluate the overall performance of the model in the task of segmenting thyroid nodules and breast nodules. The image segmentation model segmentation thyroid nodule index is used to evaluate the performance of the model in the thyroid nodule segmentation task. The image segmentation model segmentation breast nodule index is used to evaluate the performance of the model in the breast nodule segmentation task. The standard value of the image segmentation model segmentation thyroid nodule index is determined and established by medical experts based on clinical experience and research data, and is used as a standard or reference for evaluating the image segmentation model segmentation thyroid nodule index. The standard value of the image segmentation model segmentation breast nodule index is determined and established by medical experts based on clinical experience and research data, and is used as a standard or reference for evaluating the image segmentation model segmentation breast nodule index. The allowable deviation value of the image segmentation model segmentation thyroid nodule index is determined by medical experts based on clinical experience and research data, and is used to determine the fault tolerance range in the thyroid nodule segmentation task. The allowable deviation value of the image segmentation model segmentation breast nodule index is determined by medical experts based on clinical experience and research data, and is used to determine the fault tolerance range in the breast nodule segmentation task. The weight factor of the thyroid nodule index segmented by the image segmentation model is determined by medical experts based on clinical experience and research data, and is used to adjust the importance of the thyroid nodule segmentation task in the overall segmentation index. The weight factor of the breast nodule index segmented by the image segmentation model is determined by medical experts based on clinical experience and research data, and is used to adjust the importance of the breast nodule segmentation task in the overall segmentation index. The segmentation index of the image segmentation model is one of the important indicators for evaluating the performance of the model in the image segmentation task. It can objectively evaluate the performance of the model in the image segmentation task. By comparing the segmentation indexes of different models, it can be determined which model is more suitable for a specific application scenario. By analyzing the components of the segmentation index, it is possible to understand the performance of the model on different tasks or different areas, thereby guiding the optimization direction of the model. If the performance on a specific task is poor, the model structure can be adjusted or the training strategy can be optimized in a targeted manner. It can be used as a basis for setting standards and goals. Setting the expected segmentation index standard value helps determine the performance level that the model should achieve and motivates the team to continuously improve the model to achieve or exceed these goals.

[0051] Specifically, the specific process of analyzing the segmentation index of the image segmentation model is as follows: based on the segmentation index of the image segmentation model, a comparative analysis is performed with the threshold of the segmentation index of the image segmentation model to obtain the segmentation compliance index of the image segmentation model; the segmentation compliance index of the image segmentation model is compared with the set segmentation compliance index range of the image segmentation model, and when the segmentation compliance index of the image segmentation model exceeds the segmentation compliance index range of the image segmentation model, the image segmentation model is corrected.

[0052] In this embodiment, by comparing the segmentation index of the image segmentation model with the threshold, the segmentation accuracy of the model can be quickly evaluated. If the segmentation index is higher than the threshold, it means that the model performs well in the segmentation task, otherwise further optimization may be required. By comparing the segmentation compliance index of the image segmentation model with the set range, it can be determined whether the segmentation performance of the model meets the expected requirements. If not, corresponding optimization strategies can be formulated based on the analysis results, such as adjusting model parameters, improving training data, etc. When the segmentation compliance index of the image segmentation model exceeds the set range, the model can be corrected in time. This real-time feedback and correction mechanism can help improve the stability and accuracy of the system, thereby improving the effect of clinical application.

[0053] Specifically, the specific process of obtaining the segmentation conformity index of the image segmentation model is as follows: the segmentation index of the image segmentation model, the image segmentation model segmentation thyroid nodule index and the image segmentation model segmentation breast nodule index are respectively compared with the threshold of the segmentation index of the image segmentation model, the threshold of the image segmentation model segmentation thyroid nodule index and the threshold of the image segmentation model segmentation breast nodule index, and comprehensive processing is performed to obtain the segmentation conformity index of the image segmentation model.

[0054] In this embodiment, in addition to obtaining the segmentation index of the image segmentation model by evaluating the conformity of the segmentation results of the model by medical experts or experts in related fields, the segmentation conformity index of the image segmentation model can also be calculated by the following calculation formula. The specific formula of the segmentation conformity index of the image segmentation model is as follows:

[0055]

[0056] In the formula, δhk represents the segmentation conformity index of the image segmentation model, ωgh represents the segmentation index of the image segmentation model, TNS represents the image segmentation model segmentation thyroid nodule index, BGH represents the image segmentation model segmentation breast nodule index, GKn represents the threshold of the segmentation index of the image segmentation model, HKv represents the threshold of the image segmentation model segmentation thyroid nodule index, and FKs represents the threshold of the image segmentation model segmentation breast nodule index. The segmentation conformity index of the image segmentation model is used to measure the segmentation quality of the model for nodules in the image. The threshold of the segmentation index of the image segmentation model is determined and established by medical experts based on clinical experience and research data, and is used as a standard or reference to determine whether the segmentation index of the image segmentation model has reached the set standard. The threshold of the segmentation index of the thyroid nodule of the image segmentation model is determined and established by medical experts based on clinical experience and research data, and is used as a standard or reference to determine whether the segmentation index of the thyroid nodule of the image segmentation model has reached the set standard. The threshold of the breast nodule index segmented by the image segmentation model is determined and established by medical experts based on clinical experience and research data, and is used as a standard or reference to determine whether the breast nodule index segmented by the image segmentation model has reached the set standard. The segmentation consistency index provides a quantitative evaluation of the performance of the image segmentation model. By comparing the consistency between the segmentation results output by the model and the true segmentation results manually annotated, the accuracy and reliability of the model can be evaluated.

[0057] The ultrasonic elastic imaging segmentation method for thyroid nodules and breast nodules comprises the following steps: S1. Acquire thyroid nodule images and breast nodule images of target patients; S2. Based on the convolutional neural network DenseNet in deep learning as an encoder, perform image feature extraction on the thyroid nodule image data and breast nodule image of the target patient, and obtain feature information in the thyroid nodule image and the breast nodule image; S3. Based on the feature information in the thyroid nodule image and the breast nodule image, design a multi-scale fusion enhancement decoder in combination with the attention mechanism and the hole convolution, model the boundary and texture of the nodule from multiple scales and ranges, and construct an image segmentation model; S4. Based on the constructed segmentation model, segment the thyroid nodule image data and breast nodule image data of the target patient, and obtain segmented image data of the thyroid nodule area and the breast nodule area; S5. Evaluate the image segmentation model, obtain the segmentation index of the image segmentation model, analyze the segmentation index of the image segmentation model, and correct the image segmentation model.

[0058] In this embodiment, a convolutional neural network (DenseNet) in deep learning is used as an encoder to extract features from thyroid nodule images and breast nodule images. This means that important feature information is extracted from the image through a neural network for use in subsequent segmentation steps. Modeling and designing a segmentation model, using the feature information extracted from thyroid nodules and breast nodules images, combined with attention mechanisms and dilated convolutions, to design a multi-scale fusion enhancement decoder. The role of this decoder is to convert the feature information obtained from the encoder into an image segmentation model so as to model the boundaries and textures of the nodules. Image segmentation uses the constructed segmentation model to segment the thyroid nodule images and breast nodule images of the target patient, thereby obtaining segmented image data of the thyroid nodule area and the breast nodule area. The image segmentation model is evaluated, the segmentation index is obtained, and it is analyzed. According to the evaluation results, the image segmentation model may need to be modified or adjusted to improve its performance and accuracy.

[0059] In summary, this application has at least the following effects:

[0060] The ultrasonic elastic imaging segmentation method and system for thyroid nodules and breast nodules can provide important support and guidance for the research and development and application of the ultrasonic elastic imaging segmentation method and system by segmenting the thyroid nodule image data and breast nodule image data of the target patient, segmenting the thyroid nodule and breast nodule image data, evaluating the image segmentation model, and analyzing and correcting it.

[0061] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0062] The present invention is described with reference to flowcharts and / or block diagrams of systems, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0063] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0064] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0065] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0066] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. An ultrasonic elastic imaging segmentation system for thyroid nodules and breast nodules, characterized in that: It includes the following modules: image acquisition module, feature extraction module, model building module, image segmentation module, and model evaluation module; The image acquisition module is used to acquire the thyroid nodule image and breast nodule image of the target patient; The feature extraction module is used to extract image features from the target patient's thyroid nodule image data and breast nodule image based on the convolutional neural network DenseNet in deep learning as an encoder, and obtain feature information in the thyroid nodule image and breast nodule image; The model building module is used to design a multi-scale fusion enhancement decoder based on the feature information in thyroid nodule images and breast nodule images, combining the attention mechanism and dilated convolution, modeling the boundaries and textures of nodules from multiple scales and ranges, and building an image segmentation model; The image segmentation module is used to segment the thyroid nodule image data and breast nodule image data of the target patient based on the image segmentation model, and obtain the segmented image data of the thyroid nodule area and the breast nodule area; The model evaluation module is used to evaluate the image segmentation model, obtain the segmentation index of the image segmentation model, analyze the segmentation index of the image segmentation model, and correct the image segmentation model; The specific process of obtaining the segmentation index of the image segmentation model is as follows: The thyroid nodule index segmented by the image segmentation model and the breast nodule index segmented by the image segmentation model are respectively compared with the set standard value of the thyroid nodule index segmented by the image segmentation model and the standard value of the breast nodule index segmented by the image segmentation model, and the ratio operation is performed with the allowable deviation value of the thyroid nodule index segmented by the image segmentation model and the allowable deviation value of the breast nodule index segmented by the image segmentation model; Assign weights to the thyroid nodule segmentation index and the breast nodule segmentation index of the image segmentation model, and perform comprehensive calculations to obtain the segmentation index of the image segmentation model; The specific formula of the segmentation index of the image segmentation model is as follows: In the formula, ωgh represents the segmentation index of the image segmentation model, TNS represents the thyroid nodule index segmented by the image segmentation model, BGH represents the breast nodule index segmented by the image segmentation model, JKS represents the standard value of the thyroid nodule index segmented by the image segmentation model, aTR represents the standard value of the breast nodule index segmented by the image segmentation model, ΔBS represents the allowable deviation value of the thyroid nodule index segmented by the image segmentation model, ΔVA represents the allowable deviation value of the breast nodule index segmented by the image segmentation model, φ1 represents the weight factor of the thyroid nodule index segmented by the image segmentation model, and φ2 represents the weight factor of the breast nodule index segmented by the image segmentation model; The specific process of analyzing the segmentation index of the image segmentation model is as follows: Based on the segmentation index of the image segmentation model, a segmentation coincidence index of the image segmentation model is obtained by comparing and analyzing the segmentation index of the image segmentation model with the threshold of the segmentation index of the image segmentation model; The segmentation conformity index of the image segmentation model is compared with the set segmentation conformity index range of the image segmentation model. When the segmentation conformity index of the image segmentation model exceeds the segmentation conformity index range of the image segmentation model, the image segmentation model is corrected.

2. The ultrasonic elastic imaging segmentation system for thyroid nodules and breast nodules according to claim 1, characterized in that: The specific process of image feature extraction for target patient thyroid nodule image data and breast nodule image is as follows: Based on the convolutional neural network DenseNet in deep learning as an encoder, the thyroid nodule images and breast nodule images of the target patients are preprocessed; The preprocessed thyroid nodule images and breast nodule images are input into DenseNet, and feature information of the thyroid nodule images and breast nodule images is extracted through the forward propagation process. The feature information includes the shape, texture, and edge of the nodules. The features extracted from DenseNet are represented in vector or matrix form, and a fully connected layer is used to map the features into a fixed-length vector space and store them in the image feature database.

3. The ultrasonic elastic imaging segmentation system for thyroid nodules and breast nodules according to claim 1, characterized in that: The specific process of building an image segmentation model is as follows: The feature information of the thyroid nodule image and breast nodule image output by the encoder is upsampled, and the weight coefficients of features of different scales are dynamically adjusted through the attention mechanism; The boundaries and textures of nodules are modeled from multiple scales and ranges, and skip connections are designed through a selective kernel network to fully integrate the intermediate features of the encoder and the output features of the decoder to build an image segmentation model.

4. The ultrasonic elastic imaging segmentation system for thyroid nodules and breast nodules according to claim 1, characterized in that: The thyroid nodule region and breast nodule region segmentation image data specifically include thyroid nodule segmentation data and breast nodule segmentation data; The thyroid nodule segmentation data includes the area of ​​the thyroid nodule segmented by the image segmentation model, the area of ​​the real area of ​​the thyroid nodule, and the area of ​​the overlapping area of ​​the thyroid nodule; The breast nodule segmentation data includes the area of ​​the breast nodule region segmented by the image segmentation model, the area of ​​the real area of ​​the breast nodule, and the area of ​​the overlapping area of ​​the breast nodule.

5. The ultrasonic elastic imaging segmentation system for thyroid nodules and breast nodules according to claim 4, characterized in that: The specific process of evaluating the image segmentation model is as follows: Based on the thyroid nodule segmentation data, the accuracy of the image segmentation model in segmenting thyroid nodules was evaluated, and the thyroid nodule segmentation index of the image segmentation model was obtained; Based on the breast nodule segmentation data, the accuracy of the image segmentation model in segmenting breast nodules is evaluated, and the breast nodule segmentation index of the image segmentation model is obtained.

6. The ultrasonic elastic imaging segmentation system for thyroid nodules and breast nodules according to claim 5, characterized in that: The specific process of obtaining the segmentation compliance index of the image segmentation model is as follows: The segmentation index of the image segmentation model, the thyroid nodule segmentation index of the image segmentation model and the breast nodule segmentation index of the image segmentation model are compared with the threshold of the segmentation index of the image segmentation model, the threshold of the thyroid nodule segmentation index of the image segmentation model and the threshold of the breast nodule segmentation index of the image segmentation model respectively, and the segmentation conformity index of the image segmentation model is obtained by comprehensive processing.

7. A method for segmenting thyroid nodules and breast nodules using ultrasonic elastic imaging, characterized in that: The following steps are involved: S1. Acquire thyroid nodule images and breast nodule images of the target patient; S2. Based on the convolutional neural network DenseNet in deep learning as an encoder, image feature extraction is performed on the thyroid nodule image data and breast nodule image of the target patient to obtain feature information in the thyroid nodule image and breast nodule image; S3. Based on the feature information in thyroid nodule images and breast nodule images, a multi-scale fusion enhancement decoder is designed by combining attention mechanism and dilated convolution to model the boundaries and textures of nodules from multiple scales and ranges, and construct an image segmentation model; S4. Based on the image segmentation model, segment the thyroid nodule image data and the breast nodule image data of the target patient to obtain segmented image data of the thyroid nodule region and the breast nodule region; S5. Evaluate the image segmentation model, obtain the segmentation index of the image segmentation model, analyze the segmentation index of the image segmentation model, and correct the image segmentation model; The specific process of obtaining the segmentation index of the image segmentation model is as follows: The thyroid nodule index segmented by the image segmentation model and the breast nodule index segmented by the image segmentation model are respectively compared with the set standard value of the thyroid nodule index segmented by the image segmentation model and the standard value of the breast nodule index segmented by the image segmentation model, and the ratio operation is performed with the allowable deviation value of the thyroid nodule index segmented by the image segmentation model and the allowable deviation value of the breast nodule index segmented by the image segmentation model; Assign weights to the thyroid nodule segmentation index and the breast nodule segmentation index of the image segmentation model, and perform comprehensive calculations to obtain the segmentation index of the image segmentation model; The specific formula of the segmentation index of the image segmentation model is as follows: In the formula, ωgh represents the segmentation index of the image segmentation model, TNS represents the thyroid nodule index segmented by the image segmentation model, BGH represents the breast nodule index segmented by the image segmentation model, JKS represents the standard value of the thyroid nodule index segmented by the image segmentation model, aTR represents the standard value of the breast nodule index segmented by the image segmentation model, ΔBS represents the allowable deviation value of the thyroid nodule index segmented by the image segmentation model, ΔVA represents the allowable deviation value of the breast nodule index segmented by the image segmentation model, φ1 represents the weight factor of the thyroid nodule index segmented by the image segmentation model, and φ2 represents the weight factor of the breast nodule index segmented by the image segmentation model; The specific process of analyzing the segmentation index of the image segmentation model is as follows: Based on the segmentation index of the image segmentation model, a segmentation coincidence index of the image segmentation model is obtained by comparing and analyzing the segmentation index of the image segmentation model with the threshold of the segmentation index of the image segmentation model; The segmentation conformity index of the image segmentation model is compared with the set segmentation conformity index range of the image segmentation model. When the segmentation conformity index of the image segmentation model exceeds the segmentation conformity index range of the image segmentation model, the image segmentation model is corrected.

Images