Breast ultrasound image segmentation method and apparatus

Abstract

This invention provides a method and device for breast ultrasound image segmentation. The method includes: acquiring an ultrasound image of a patient's breast region or a region of interest (ROI) of a breast lesion within the breast region; extracting a feature map corresponding to at least one BI-RADS feature from the ultrasound image based on a pre-trained BI-RADS feature classification model; fusing the feature map with the ultrasound image to obtain a fused feature map; and segmenting the breast lesion on the fused feature map based on a pre-trained breast lesion segmentation model to determine the breast lesion region on the ultrasound image. The BI-RADS feature classification model can accurately obtain the feature map corresponding to the BI-RADS feature. Fusing the feature map corresponding to the BI-RADS feature with the ultrasound image can highlight the breast lesion region and boundary features. Segmenting the breast lesion based on the fused feature map can improve the accuracy of breast lesion region segmentation.

Description

Technical Field

[0001] This invention relates to the field of medical ultrasound technology, specifically to a method and device for segmenting breast ultrasound images. Background Technology

[0002] Breast cancer is a malignant tumor that occurs in the epithelial tissue of the breast. According to cancer statistics, breast cancer ranks first in the incidence of malignant tumors in women, making early screening for breast cancer particularly important. Breast ultrasound can clearly show the location, shape, internal structure, and changes in adjacent tissues of each layer of breast soft tissue and lesions. It has many advantages, including being economical, convenient, non-invasive, painless, non-radioactive, and highly repeatable, and has become one of the important methods for early breast cancer screening.

[0003] The Breast Imaging Reporting and Data System (BI-RADS), proposed by the American College of Radiology (ACR), is currently a widely used and relatively authoritative grading and evaluation standard in clinical practice. BI-RADS uses standardized professional terminology to diagnose and classify all normal and abnormal imaging findings of the breast as a whole organ. However, the BI-RADS diagnostic rules are complex and numerous, making them difficult for junior or primary care physicians to memorize, thus affecting the efficiency and accuracy of clinical diagnosis. With the rapid development of artificial intelligence technology, it has become possible to apply computer-aided diagnosis to the intelligent analysis of breast ultrasound images.

[0004] Currently, when performing BI-RADS intelligent analysis of breast ultrasound images using artificial intelligence technology, the first step is to segment the breast lesion area from the breast ultrasound image, and then perform BI-RADS intelligent analysis based on the lesion area. However, due to the complexity and variability of breast lesion signs and morphology, the accuracy of existing methods for directly segmenting breast lesion areas based on breast ultrasound images is relatively low, thus affecting the accuracy of subsequent BI-RADS intelligent analysis. Summary of the Invention

[0005] This invention provides a method and device for segmenting breast ultrasound images, which addresses the problem of low accuracy in segmenting breast lesion areas using existing methods.

[0006] In a first aspect, embodiments of the present invention provide a method for segmenting breast ultrasound images, comprising:

[0007] To acquire ultrasound images of the breast region or the region of interest (ROI) of breast lesions in the breast region of the subject;

[0008] Based on a pre-trained BI-RADS feature classification model, at least one BI-RADS feature map is extracted from the ultrasound image. The BI-RADS feature classification model is obtained by fusing the feature maps of the sample ultrasound image and the sample ultrasound image output by the initial BI-RADS feature classification model, and using the fused image to iteratively train the initial BI-RADS feature classification model until the classification accuracy of the initial BI-RADS feature classification model is greater than a preset accuracy threshold.

[0009] The feature map is fused with the ultrasound image to obtain a fused feature map;

[0010] Breast lesions are segmented based on a pre-trained breast lesion segmentation model, and the lesion region is determined on the ultrasound image.

[0011] In one embodiment, extracting a feature map corresponding to at least one BI-RADS feature from an ultrasound image includes: extracting multiple feature maps corresponding to multiple BI-RADS features from the ultrasound image respectively; fusing the feature maps with the ultrasound image includes: fusing the multiple feature maps with the ultrasound image.

[0012] In one embodiment, extracting multiple feature maps corresponding to multiple BI-RADS features from an ultrasound image includes: extracting shape feature maps, edge feature maps, and posterior echo feature maps corresponding to shape features, edge features, and posterior echo features from the ultrasound image.

[0013] In one embodiment, extracting a feature map corresponding to at least one BI-RADS feature from an ultrasound image includes: extracting a comprehensive feature map corresponding to multiple BI-RADS features from the ultrasound image; fusing the feature map with the ultrasound image includes: fusing the comprehensive feature map with the ultrasound image.

[0014] In one embodiment, extracting a comprehensive feature map corresponding to multiple BI-RADS features from an ultrasound image includes: extracting a comprehensive feature map corresponding to shape features, edge features, and posterior echo features from the ultrasound image.

[0015] In one embodiment, extracting a feature map corresponding to at least one BI-RADS feature from an ultrasound image includes: extracting a shape feature map corresponding to a shape feature from the ultrasound image; fusing the feature map with the ultrasound image includes: fusing the shape feature map with the ultrasound image.

[0016] In one embodiment, extracting a shape feature map corresponding to shape features from an ultrasound image includes:

[0017] Based on a pre-trained shape feature classification model, shape feature maps corresponding to shape features are extracted from ultrasound images. The shape feature classification model is trained on sample ultrasound images labeled with the type of shape feature. During the training phase, the feature maps of the sample ultrasound images and the sample ultrasound images output by the initial shape feature classification model are fused. The fused image is then input into the shape feature classification model. Iterative training is performed with the goal of minimizing the error between the labeling result and the prediction result of the shape feature classification model until the preset conditions are met to obtain the trained shape feature classification model. The types of shape features include circles, ellipses and irregular shapes.

[0018] In one embodiment, extracting a feature map corresponding to at least one BI-RADS feature from an ultrasound image includes: extracting an edge feature map corresponding to an edge feature from the ultrasound image; fusing the feature map with the ultrasound image includes: fusing the edge feature map with the ultrasound image.

[0019] In one embodiment, extracting an edge feature map corresponding to edge features from an ultrasound image includes:

[0020] The edge feature map corresponding to the edge feature is extracted from the ultrasound image based on the pre-trained edge feature classification model. The edge feature classification model is trained on sample ultrasound images labeled with the type of edge feature. During the training phase, the feature maps of the sample ultrasound images and the sample ultrasound images output by the initial edge feature classification model are fused. The fused image is input into the edge feature classification model. Iterative training is performed with the goal of minimizing the error between the labeling result and the prediction result of the edge feature classification model until the preset conditions are met to obtain the trained edge feature classification model. The types of edge features include smoothing, blurring, angularity, differential lobes, and burrs.

[0021] In one embodiment, extracting a feature map corresponding to at least one BI-RADS feature from an ultrasound image includes: extracting a posterior echo feature map corresponding to a posterior echo feature from the ultrasound image; fusing the feature map with the ultrasound image includes: fusing the posterior echo feature map with the ultrasound image.

[0022] In one embodiment, extracting a posterior echo feature map corresponding to posterior echo features from an ultrasound image includes:

[0023] The shape feature map corresponding to the rear echo feature is extracted from the ultrasound image based on the pre-trained rear echo feature classification model. The rear echo feature classification model is trained on sample ultrasound images labeled with the type of rear echo feature. During the training phase, the feature map of the sample ultrasound image and the sample ultrasound image output by the initial rear echo feature classification model are fused. The fused image is input into the rear echo feature classification model. Iterative training is performed with the goal of minimizing the error between the labeling result and the prediction result of the rear echo feature classification model until the preset conditions are met to obtain the trained rear echo feature classification model. The types of rear echo features include enhanced, unchanged, attenuated, and mixed echoes.

[0024] In one embodiment, fusing the feature map with the ultrasound image includes: weighting the feature map and the ultrasound image according to their respective fusion weights, wherein the respective fusion weights of the feature map and the ultrasound image are obtained by fitting the contribution probabilities of the feature map and the ultrasound image to the weights using a logistic regression prediction model.

[0025] In one embodiment, breast lesion segmentation is performed on the fused feature map based on a pre-trained breast lesion segmentation model. Determining the breast lesion region on the ultrasound image includes:

[0026] The fused feature map is input into a pre-trained breast lesion segmentation model to determine the breast lesion region on the ultrasound image;

[0027] The breast lesion segmentation model is trained based on sample ultrasound images with labeled breast lesion regions. During the training phase, at least one BI-RADS feature map is extracted from the sample ultrasound image and fused with the sample ultrasound image to obtain a sample fusion feature map. The sample fusion feature map is then input into the breast lesion segmentation model. The model is iteratively trained with the goal of minimizing the error between the labeled breast lesion region and the breast lesion region predicted by the breast lesion segmentation model until the preset conditions are met to obtain the trained breast lesion segmentation model.

[0028] In one embodiment, BI-RADS features include: shape features, orientation features, edge features, internal echo features, posterior echo features, calcification features, and blood flow features.

[0029] In one embodiment, the method further includes: displaying the breast lesion region segmented from the ultrasound image on a display interface.

[0030] In one embodiment, acquiring an ultrasound image of the breast region of a subject includes: transmitting ultrasound waves to the breast region of the subject and receiving ultrasound echoes returned from the breast region to obtain ultrasound echo data, and generating an ultrasound image of the breast region of the subject in real time based on the ultrasound echo data; or, acquiring a pre-stored ultrasound image of the breast region of the subject from a storage device.

[0031] Obtaining ultrasound images of regions of interest (ROIs) of breast lesions in the breast region of a subject includes: acquiring ultrasound images of the subject's breast region; automatically identifying ROIs of breast lesions in the ultrasound images of the breast region using a target detection algorithm; and obtaining ultrasound images of ROIs of breast lesions in the subject's breast region from the ultrasound images of the breast region based on the identified ROIs.

[0032] In a second aspect, embodiments of the present invention provide an ultrasound imaging device, comprising:

[0033] Ultrasonic probe;

[0034] The transmitting circuit is used to output the corresponding transmitting sequence to the ultrasonic probe according to the set mode, so as to control the ultrasonic probe to emit the corresponding ultrasonic waves;

[0035] The receiving circuit is used to receive the ultrasonic echo signal output by the ultrasonic probe and output ultrasonic echo data.

[0036] A display is used to output visual information;

[0037] A processor for performing the breast ultrasound image segmentation method as described in any of the first aspects.

[0038] Thirdly, embodiments of the present invention provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the breast ultrasound image segmentation method as described in any of the first aspects.

[0039] The breast ultrasound image segmentation method and device provided in this invention extracts at least one feature map corresponding to a BI-RADS feature from the acquired ultrasound image based on a pre-trained BI-RADS feature classification model. The feature map is then fused with the ultrasound image to obtain a fused feature map. Finally, a pre-trained breast lesion segmentation model is used to segment breast lesions based on the fused feature map. The BI-RADS feature classification model is developed by fusing the feature maps of a sample ultrasound image and the sample ultrasound image output by an initial BI-RADS feature classification model. The fused image is used to iteratively train the initial BI-RADS feature classification model until its classification accuracy exceeds a preset accuracy threshold. The resulting trained BI-RADS feature classification model can accurately obtain the feature maps corresponding to BI-RADS features. Fusing the feature maps corresponding to BI-RADS features with the ultrasound image highlights the features of the breast lesion region and its boundaries. Segmenting breast lesions based on the fused feature map improves the accuracy of breast lesion region segmentation. Attached Figure Description

[0040] Figure 1 This is a structural block diagram of an ultrasound imaging device provided in an embodiment of the present invention;

[0041] Figure 2 A flowchart of a breast ultrasound image segmentation method provided in an embodiment of the present invention;

[0042] Figure 3 This is a schematic diagram illustrating the training process of a BI-RADS feature classification model provided in an embodiment of the present invention.

[0043] Figure 4 A schematic diagram of a fusion feature map provided in an embodiment of the present invention;

[0044] Figure 5 A flowchart of a breast ultrasound image segmentation method provided in another embodiment of the present invention;

[0045] Figure 6 This is a schematic diagram of a fusion process provided in an embodiment of the present invention;

[0046] Figure 7 A flowchart of a breast ultrasound image segmentation method provided in another embodiment of the present invention;

[0047] Figure 8 This is a schematic diagram illustrating the training process of a breast lesion segmentation model provided in an embodiment of the present invention. Detailed Implementation

[0048] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of this application. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to this application are not shown or described in the specification. This is to avoid obscuring the core parts of this application with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0049] Furthermore, the features, operations, or characteristics described in the specification can be combined in any suitable manner to form various embodiments. At the same time, the steps or actions in the method description can be rearranged or adjusted in a manner obvious to those skilled in the art. Therefore, the various orders in the specification and drawings are only for the clear description of a particular embodiment and do not imply a necessary order, unless otherwise stated that a particular order must be followed.

[0050] The serial numbers assigned to components in this document, such as "first" and "second," are used only to distinguish the described objects and have no sequential or technical meaning. The terms "connection" and "linkage" used in this application, unless otherwise specified, include both direct and indirect connections (linkages).

[0051] like Figure 1 As shown, the ultrasound imaging device provided by the present invention may include: an ultrasound probe 20, a transmitting / receiving circuit 30 (i.e., a transmitting circuit 310 and a receiving circuit 320), a beamforming module 40, an IQ demodulation module 50, a memory 60, a processor 70, and a human-computer interaction device. The processor 70 may include a control module 710 and an image processing module 720.

[0052] The ultrasound probe 20 includes a transducer (not shown) composed of multiple array elements arranged in an array. These elements can be arranged in a row to form a linear array, or in a two-dimensional matrix to form a planar array. They can also form a convex array. Each element is used to emit an ultrasonic beam according to an excitation electrical signal, or to convert a received ultrasonic beam into an electrical signal. Therefore, each element can be used to achieve the mutual conversion between electrical pulse signals and ultrasonic beams, thereby emitting ultrasonic waves to a target area of ​​human tissue (e.g., the breast area containing a breast lesion in this embodiment), and also to receive echoes of ultrasonic waves reflected back from the tissue. During ultrasound detection, the transmitting circuit 310 and the receiving circuit 320 can control which elements are used to emit ultrasonic beams and which are used to receive ultrasonic beams, or control the elements to be used in time-slotted manner to emit ultrasonic beams or receive echoes of ultrasonic beams. Elements participating in ultrasonic wave emission can be simultaneously excited by electrical signals to emit ultrasonic waves simultaneously; or elements participating in ultrasonic wave emission can be excited by several electrical signals with a certain time interval to continuously emit ultrasonic waves with a certain time interval.

[0053] In this embodiment, the user selects a suitable position and angle by moving the ultrasound probe 20 to emit ultrasound waves to the breast region 10 and receives the echo of the ultrasound waves returned from the breast region 10. The user obtains and outputs the electrical signal of the echo. The electrical signal of the echo is a channel analog electrical signal formed by the receiving array element as the channel, which carries amplitude information, frequency information and time information.

[0054] The transmitting circuit 310 generates a transmission sequence under the control of the control module 710 of the processor 70. This transmission sequence controls some or all of the multiple array elements to transmit ultrasound waves to the biological tissue. The transmission sequence parameters include the position and number of array elements, and the ultrasound beam transmission parameters (e.g., amplitude, frequency, number of transmissions, transmission interval, transmission angle, waveform, focusing position, etc.). In some cases, the transmitting circuit 310 also performs phase delay on the transmitted beam, allowing different transmitting array elements to transmit ultrasound waves at different times, so that each transmitted ultrasound beam can be focused in a predetermined region of interest. Different operating modes, such as B-image mode, C-image mode, and D-image mode (Doppler mode), may have different transmission sequence parameters. After the echo signal is received by the receiving circuit 320 and processed by subsequent modules and corresponding algorithms, a B-image reflecting the tissue anatomy, a C-image reflecting the tissue anatomy and blood flow information, and a D-image reflecting the Doppler spectrum can be generated.

[0055] The receiving circuit 320 receives the electrical signal of the ultrasonic echo from the ultrasonic probe 20 and processes it. The receiving circuit 320 may include one or more amplifiers, analog-to-digital converters (ADCs), etc. The amplifier amplifies the received ultrasonic echo signal after appropriate gain compensation, and the ADC samples the analog echo signal at predetermined time intervals, converting it into a digitized signal. The digitized echo signal still retains amplitude, frequency, and phase information. The data output from the receiving circuit 320 can be sent to the beamforming module 40 for processing, or to the memory 60 for storage.

[0056] The beamforming module 40 is signal-connected to the receiving circuit 320 and is used to perform beamforming processing such as delay and weighted summation on the signal output by the receiving circuit 320. Because the distance from the ultrasonic receiving point in the tested tissue to the receiving array element varies, the channel data of the same receiving point output by different receiving array elements has delay differences, requiring delay processing to align the phases and perform weighted summation on the different channel data of the same receiving point to obtain the beamformed ultrasonic image data. The ultrasonic image data output by the beamforming module 40 is also called radio frequency (RF) data. The beamforming module 40 outputs the RF data to the IQ demodulation module 50. In some embodiments, the beamforming module 40 can also output the RF data to the memory 60 for caching or storage, or directly output the RF data to the image processing module 720 of the processor 70 for image processing.

[0057] The beamforming module 40 can perform the above functions in hardware, firmware or software. For example, the beamforming module 40 may include a central controller circuit (CPU) capable of processing input data according to specific logic instructions, one or more microprocessor chips or any other electronic components. When the beamforming module 40 is implemented in software, it can execute instructions stored on a tangible and non-transitory computer-readable medium (e.g., memory 60) to perform beamforming calculations using any appropriate beamforming method.

[0058] The IQ demodulation module 50 removes the signal carrier through IQ demodulation, extracts the tissue structure information contained in the signal, and filters to remove noise. The signal obtained at this time is called the baseband signal (IQ data pair). The IQ demodulation module 50 outputs the IQ data pair to the image processing module 720 of the processor 70 for image processing. In some embodiments, the IQ demodulation module 50 also outputs the IQ data pair to the memory 60 for buffering or storage, so that the image processing module 720 can read the data from the memory 60 for subsequent image processing.

[0059] The processor 70 is configured to process input data according to specific logic instructions. It is a central controller circuit (CPU), one or more microprocessors, a graphics controller circuit (GPU), or any other electronic component. It can control peripheral electronic components according to input instructions or predetermined instructions, or perform data reading and / or saving on the memory 60. It can also process input data by executing programs in the memory 60. For example, it can perform one or more processing operations on the acquired ultrasound data according to one or more operating modes. The processing operations include, but are not limited to, adjusting or limiting the form of ultrasound waves emitted by the ultrasound probe 20, generating various image frames for display on the display 80 of the subsequent human-computer interaction device, or adjusting or limiting the content and form displayed on the display 80, or adjusting one or more image display settings (e.g., ultrasound images, interface components, locating regions of interest) displayed on the display 80.

[0060] The image processing module 720 processes the data output from the beamforming module 40 or the IQ demodulation module 50 to generate a grayscale image showing the changes in signal strength within the scanning range. This grayscale image reflects the internal anatomical structure of the tissue and is called a B-image. The image processing module 720 can output the B-image to the display 80 of the human-computer interaction device for display.

[0061] Human-computer interaction devices are used for human-computer interaction, that is, to receive user input and output visual information; the user input can be received by keyboard, operation buttons, mouse, trackball, etc., or by touch screen integrated with the display; the visual information output is displayed on the display 80.

[0062] The memory 60 may be a tangible and non-transitory computer-readable medium, such as a flash memory card, solid-state memory, hard disk, etc., for storing data or programs. For example, the memory 60 may be used to store acquired ultrasound data or image frames generated by the processor 70 that are not immediately displayed, or the memory 60 may store a graphical user interface, one or more default image display settings, or programming instructions for the processor, beamforming module, or IQ decoding module.

[0063] It should be noted that, Figure 1 The structure shown is for illustrative purposes only and may include structures larger than those shown. Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown. Figure 1 The components shown can be implemented in hardware and / or software. Figure 1 The ultrasound imaging device shown can be used to perform the breast ultrasound image segmentation method provided in any embodiment of the present invention.

[0064] Please refer to Figure 2The breast ultrasound image segmentation method provided in one embodiment of the present invention may include:

[0065] S201. Obtain ultrasound images of the breast region or the region of interest of breast lesions in the breast region of the subject.

[0066] In this embodiment, the acquired ultrasound images can be either ultrasound images of the breast region or ultrasound images of the region of interest within the breast lesion. Ultrasound images can be acquired in real-time using an ultrasound imaging device or read from a pre-stored storage medium.

[0067] In one optional implementation, acquiring an ultrasound image of the patient's breast region can be achieved by emitting ultrasound waves into the patient's breast region using an ultrasound probe of an ultrasound imaging device and receiving the returned ultrasound echoes to obtain ultrasound echo data. An ultrasound image of the patient's breast region can then be generated in real time based on this echo data. Specifically, a doctor can apply coupling gel to the skin surface of the patient's breast that is fully exposed, and then hold the ultrasound probe close to the patient's breast skin to perform the scan. Alternatively, pre-stored ultrasound images of the patient's breast region can be retrieved from a storage device.

[0068] In one optional implementation, obtaining an ultrasound image of the region of interest (ROI) of a breast lesion in the patient's breast region can be achieved by first acquiring an ultrasound image of the patient's breast region, then automatically identifying the ROI within the ultrasound image using a target detection algorithm, and finally obtaining an ultrasound image of the ROI from the ultrasound image of the breast region based on the identified ROI. The specific implementation of acquiring the ultrasound image of the patient's breast region can refer to the optional implementation described above. The identification of the ROI within the ultrasound image of the breast region can employ target detection algorithms based on deep learning, machine learning, or traditional image processing methods.

[0069] When using deep learning-based methods, it is necessary to first train a deep learning ROI detection model based on collected sample ultrasound images and the annotation results of regions of interest (ROIs) for breast lesions by senior physicians (the ROIs annotated by physicians can be, for example, the minimum bounding box or the boundary of the breast lesion). The ROI detection model can use, but is not limited to, RCNN, Faster RCNN, SSD, YOLO, etc. During the network training phase, the error between the detection results and the annotation results of the breast lesion ROIs during the iteration process is calculated, and the weights in the network are continuously updated with the aim of minimizing the error. This process is repeated until the detection results gradually approach the true value of the breast lesion ROI, resulting in a trained ROI detection model. This model can automatically determine the regions of interest for breast lesions in the input ultrasound image.

[0070] When using machine learning-based methods, it is necessary to first train a machine learning ROI detection model based on the collected sample ultrasound images and the annotation results of the breast lesion regions of interest by senior physicians. The ROI detection model can use, but is not limited to, machine learning models such as SVM, K-Means, and C-Means. The ROI detection model performs binary classification on the gray value or texture value of the pixels (belonging to or not belonging to the breast lesion region of interest), or divides the image into grids and performs binary classification on the small images within the grid. Then, the pixel that belongs to the breast lesion region of interest or the small image within the grid with the highest classification probability is selected as the breast lesion region of interest.

[0071] S202. Extract at least one BI-RADS feature map from the ultrasound image based on the pre-trained BI-RADS feature classification model. The BI-RADS feature classification model is obtained by fusing the feature maps of the sample ultrasound image and the sample ultrasound image output by the initial BI-RADS feature classification model, and then using the fused image to iteratively train the initial BI-RADS feature classification model until the classification accuracy of the initial BI-RADS feature classification model is greater than a preset accuracy threshold.

[0072] BI-RADS features include shape features, orientation features, edge features, internal echo features, posterior echo features, calcification features, and blood flow features. In this embodiment, feature maps corresponding to any one BI-RADS feature can be extracted from the ultrasound image. For example, a shape feature map corresponding to a shape feature can be extracted, or an orientation feature map corresponding to an orientation feature can be extracted, or an edge feature map corresponding to an edge feature can be extracted, and so on. It is also possible to extract feature maps corresponding to any number of BI-RADS features from the ultrasound image. When extracting feature maps corresponding to multiple BI-RADS features from an ultrasound image, one optional implementation is to extract multiple feature maps corresponding to multiple BI-RADS features separately from the ultrasound image. For example, a shape feature map corresponding to a shape feature, an orientation feature map corresponding to an orientation feature, and an edge feature map corresponding to an edge feature can be extracted separately. Another optional implementation is to extract a comprehensive feature map corresponding to multiple BI-RADS features from the ultrasound image, such as a comprehensive feature map corresponding to shape features, orientation features, and edge features.

[0073] In this embodiment, a feature map corresponding to at least one BI-RADS feature can be extracted from an ultrasound image based on a pre-trained BI-RADS feature classification model. The training process of the BI-RADS feature classification model can be referred to... Figure 3 .like Figure 3As shown, the BI-RADS feature classification model is obtained by fusing the sample ultrasound image with the feature map of the sample ultrasound image output by the initial BI-RADS feature classification model. The fused image is then used to iteratively train the initial BI-RADS feature classification model until its classification accuracy exceeds a preset accuracy threshold. The trained BI-RADS feature classification model has high classification accuracy, thus enabling accurate capture of corresponding feature maps from ultrasound images.

[0074] S203. The feature map is fused with the ultrasound image to obtain a fused feature map.

[0075] The feature map extracted in step S202 is fused with the ultrasound image obtained in step S201 to determine the fused feature map. For example, the feature map and the ultrasound image can be fused by concatenating or adding them. In an optional implementation, fusing the feature map and the ultrasound image may include: weighting the feature map and the ultrasound image according to their respective fusion weights, wherein the respective fusion weights of the feature map and the ultrasound image are obtained by fitting the contribution probabilities of the feature map and the ultrasound image to the weights using a logistic regression prediction model.

[0076] The contribution probabilities of the feature map and ultrasound image can be fitted separately using weighted fitting. For example, a logistic regression prediction model can be used for weighted fitting. Specifically, the sigmoid function can be used to obtain a contribution probability in the interval [0,1]. Then, these contribution probabilities are normalized to obtain the fusion weights of the feature map and ultrasound image respectively. Finally, the fused feature map is determined by the following expression:

[0077]

[0078] Where z represents the fusion feature map, x0 represents the ultrasound image, w0 represents the fusion weight of the ultrasound image, and x i Let w represent the i-th feature map. i Let represent the fusion weight of the i-th feature map, where n ≥ 1.

[0079] Please refer to Figure 4 , Figure 4 The left image in the middle section shows the acquired ultrasound image, and the right image shows the fused feature map obtained by fusing the feature map with the ultrasound image. Figure 4 It can be seen that by fusing the feature map with the original ultrasound image, the features of the breast lesion area and the boundary of the breast lesion can be highlighted, which facilitates the segmentation of the breast lesion.

[0080] S204. Based on the pre-trained breast lesion segmentation model, the fused feature map is used to segment breast lesions, and the breast lesion region is determined on the ultrasound image.

[0081] After obtaining the fused feature map, breast lesions can be segmented based on it. Breast lesion segmentation can employ semantic segmentation algorithms based on traditional machine learning, traditional image processing algorithms, or semantic segmentation algorithms based on deep learning. Semantic segmentation algorithms based on traditional machine learning typically involve splitting the image into image patches and then classifying them: first, feature extraction is performed on the split image patches, using algorithms such as PCA, LDA, or deep neural networks to build feature vectors; then, the feature vectors are classified, using algorithms such as k-NN, random forest, or SVM; finally, the classified image patches are reassembled into the whole image to obtain the segmentation result. When using traditional image processing algorithms for segmentation, region-based segmentation algorithms such as region growing, watershed algorithms, and Otsu thresholding can be used; gradient-based segmentation algorithms such as Sobel and Canny operators can also be employed. Deep learning-based semantic segmentation algorithms typically employ supervised learning strategies. They build deep neural network models by stacking convolutional layers, pooling layers, upsampling layers, deconvolutional layers, and fully connected layers. Based on these deep neural network models, a breast lesion segmentation model is constructed. The real segmentation annotations are used to create a mask image of the same size as the input image, which serves as supervision. The deep neural network model learns and outputs the segmented regions, resulting in a trained breast lesion segmentation model. Examples of deep neural network models include Mask R-CNN, U-Net, and Solo.

[0082] The breast ultrasound image segmentation method provided in this embodiment extracts at least one feature map corresponding to a BI-RADS feature from the acquired ultrasound image based on a pre-trained BI-RADS feature classification model. This feature map is then fused with the ultrasound image to obtain a fused feature map, which is used to segment breast lesions. Specifically, the BI-RADS feature classification model is developed by fusing the feature maps of the sample ultrasound image and the sample ultrasound image output by the initial BI-RADS feature classification model. The fused image is used to iteratively train the initial BI-RADS feature classification model until its classification accuracy exceeds a preset accuracy threshold. The resulting trained BI-RADS feature classification model, with its high classification accuracy, can accurately extract the corresponding feature map from the ultrasound image. By fusing the feature map corresponding to the BI-RADS feature with the ultrasound image, the characteristics of the breast lesion region and its boundary can be highlighted. Segmenting breast lesions based on the fused feature map improves the accuracy of breast lesion region segmentation, thereby enhancing the accuracy of subsequent BI-RADS intelligent analysis based on the breast lesion region.

[0083] Multiple feature maps corresponding to various BI-RADS features can be extracted from ultrasound images, and these feature maps can be fused with the ultrasound image. Then, breast lesion segmentation is performed based on the resulting fused feature map. Considering that shape features, edge features, and posterior echo features in BI-RADS features have a significant impact on boundary conditions, in an optional implementation, shape feature maps, edge feature maps, and posterior echo feature maps corresponding to the shape features, edge features, and posterior echo features can be extracted from the ultrasound image separately for fusion with the ultrasound image. Please refer to [reference needed]. Figure 5 The method provided in this embodiment may include:

[0084] S401. Obtain ultrasound images of the patient's breast region or the region of interest (ROI) of a breast lesion within the breast region.

[0085] For detailed implementation methods, please refer to S201, which will not be repeated here.

[0086] S402. Extract the shape feature map, edge feature map, and posterior echo feature map corresponding to the shape feature, edge feature, and posterior echo feature from the ultrasound image.

[0087] S403. The shape feature map, edge feature map, and posterior echo feature map are fused with the ultrasound image to obtain a fused feature map.

[0088] Please refer to Figure 6Shape feature maps can be extracted from ultrasound images using a shape feature extractor, edge feature maps can be extracted from ultrasound images using an edge feature extractor, and shape feature maps can be extracted from ultrasound images using a rear echo feature extractor. Then, the extracted shape feature maps, edge feature maps, and rear echo feature maps are fused with the ultrasound image to obtain a fused feature map. The shape feature extractor, edge feature extractor, and rear echo feature extractor can be feature extraction models trained using deep learning methods, or feature extraction operators obtained using traditional image processing methods. This embodiment does not limit the specific implementation method. For specific fusion methods, please refer to S203, which will not be elaborated here.

[0089] S404. Segmentation of breast lesions based on fusion feature maps, and determination of breast lesion regions on ultrasound images.

[0090] For detailed implementation methods, please refer to S204, which will not be repeated here.

[0091] The breast ultrasound image segmentation method provided in this embodiment fuses the ultrasound image with shape feature map, edge feature map and posterior echo feature map, which have a significant impact on the boundary conditions, so as to make the breast lesion area and the boundary of the breast lesion in the fused feature map more prominent, thereby further improving the accuracy of breast lesion area segmentation.

[0092] In addition to extracting multiple feature maps corresponding to multiple BI-RADS features separately as described above, an optional implementation can also extract a comprehensive feature map corresponding to multiple BI-RADS features from the ultrasound image, fuse the comprehensive feature map with the ultrasound image, and then segment the breast lesion based on the obtained fused feature map. This still considers the influence of multiple BI-RADS features on breast lesion segmentation, but only one comprehensive feature map needs to be fused, which helps improve fusion efficiency and thus improves the efficiency of breast lesion segmentation. For example, a comprehensive feature map corresponding to shape features, edge features, and posterior echo features can be extracted from the ultrasound image. Figure 6 The shape feature extractor, edge feature extractor, and posterior echo feature extractor in the ultrasound image are replaced with a comprehensive feature extractor, which can extract the comprehensive feature map corresponding to the shape feature, edge feature, and posterior echo feature in one step.

[0093] The above has detailed the extraction of feature maps corresponding to multiple BI-RADS features from ultrasound images for breast lesion segmentation. It should be noted that the specific combination of multiple BI-RADS features, in addition to the combinations provided in the above embodiments, can also include any combination of shape features, orientation features, edge features, internal echo features, posterior echo features, calcification features, and blood flow features. The following embodiment will illustrate the extraction of a feature map corresponding to a single BI-RADS feature from an ultrasound image for breast lesion segmentation.

[0094] Please refer to Figure 7 The breast ultrasound image segmentation method provided in this embodiment may include:

[0095] S601. Obtain ultrasound images of the patient's breast region or the region of interest (ROI) of a breast lesion within the breast region.

[0096] For detailed implementation methods, please refer to S201, which will not be repeated here.

[0097] S602. Extract shape feature maps corresponding to shape features from ultrasound images based on a pre-trained shape feature classification model.

[0098] Shape feature maps corresponding to shape features can be extracted from ultrasound images using methods based on deep learning, machine learning, or traditional image processing. In one optional implementation, the ultrasound image can be input into a pre-trained shape feature classification model. This model may include, for example, a front-end shape feature extraction network and a back-end shape feature classification network. The front-end shape feature extraction network extracts shape feature maps from the ultrasound image, while the back-end shape feature classification network outputs the type of shape feature based on the shape feature maps. The shape feature map is determined based on the output of the front-end shape feature extraction network.

[0099] The shape feature classification model is trained based on sample ultrasound images labeled with shape feature types. Its training process can be found in [reference needed]. Figure 3 The training process of the BI-RADS feature classification model, as shown, involves fusing the feature maps of the sample ultrasound images and those output by the initial shape feature classification model during the training phase. The fused image is then input into the shape feature classification model. Iterative training is performed with the goal of minimizing the error between the labeled results and the prediction results of the shape feature classification model, until a pre-defined condition is met to obtain a well-trained shape feature classification model. The shape features include circular, elliptical, and irregular shapes. The pre-defined condition can be that the classification accuracy of the shape feature classification model for the shape features is greater than a pre-defined accuracy threshold. A higher accuracy threshold can be set to accurately obtain the shape feature maps in the ultrasound images.

[0100] It should be noted that the method for extracting shape feature maps from ultrasound images based on a pre-trained shape feature classification model in this embodiment can be used for... Figure 5 In the illustrated embodiment, i.e. Figure 6 The specific implementation of the shape feature extractor can be the shape feature classification model trained in this embodiment.

[0101] S603. The shape feature map is fused with the ultrasound image to obtain a fused feature map.

[0102] For specific fusion methods, please refer to S203, which will not be elaborated here.

[0103] S604. Segmentation of breast lesions based on fusion feature maps, and determination of breast lesion regions on ultrasound images.

[0104] For detailed implementation methods, please refer to S204, which will not be repeated here.

[0105] The breast ultrasound image segmentation method provided in this embodiment accurately extracts shape feature maps that significantly affect boundary conditions from ultrasound images based on a pre-trained shape feature classification model, and fuses the shape feature maps with the ultrasound images, making the breast lesion areas and boundaries of the breast lesions more prominent in the fused feature maps. Breast lesion segmentation based on fused feature maps balances segmentation efficiency and accuracy, and can efficiently and accurately segment breast lesion areas from ultrasound images.

[0106] This invention also provides a method for segmenting breast ultrasound images, comprising:

[0107] S701. Obtain ultrasound images of the patient's breast region or the region of interest (ROI) of a breast lesion within the breast region.

[0108] For detailed implementation methods, please refer to S201, which will not be repeated here.

[0109] S702. Extract edge feature maps corresponding to edge features from ultrasound images based on a pre-trained edge feature classification model.

[0110] Edge feature maps corresponding to edge features can be extracted from ultrasound images using methods based on deep learning, machine learning, or traditional image processing. In one optional implementation, the ultrasound image can be input into a pre-trained edge feature classification model. This model may include, for example, a front-end edge feature extraction network and a back-end edge feature classification network. The front-end edge feature extraction network extracts edge feature maps from the ultrasound image, while the back-end edge feature classification network outputs the type of edge features based on the edge feature maps. The edge feature map is determined based on the output of the front-end edge feature extraction network.

[0111] The edge feature classification model is trained based on sample ultrasound images labeled with edge feature types. Its training process can be found in [reference needed]. Figure 3 The training process of the BI-RADS feature classification model, as shown, involves fusing the feature maps of the sample ultrasound images and those output by the initial edge feature classification model during the training phase. The fused image is then input into the edge feature classification model. Iterative training is performed with the goal of minimizing the error between the labeled results and the prediction results of the edge feature classification model, until a pre-defined condition is met to obtain a well-trained edge feature classification model. Edge feature types include smooth, blurred, angular, slightly lobed, and spurious. The pre-defined condition can be that the classification accuracy of the edge feature classification model for edge features is greater than a pre-defined accuracy threshold. A higher accuracy threshold can be set to accurately obtain the edge feature maps in the ultrasound images.

[0112] It should be noted that the method for extracting edge feature maps from ultrasound images based on a pre-trained edge feature classification model in this embodiment can be used for... Figure 5 In the illustrated embodiment, i.e. Figure 6 The edge feature extractor in the example can be implemented using the pre-trained front-end edge feature extraction network in this embodiment.

[0113] S703. The edge feature map is fused with the ultrasound image to obtain the fused feature map.

[0114] For specific fusion methods, please refer to S203, which will not be elaborated here.

[0115] S704. Segmentation of breast lesions based on fusion feature maps, and determination of breast lesion regions on ultrasound images.

[0116] For detailed implementation methods, please refer to S204, which will not be repeated here.

[0117] The breast ultrasound image segmentation method provided in this embodiment accurately extracts edge feature maps that significantly affect the boundary conditions from the ultrasound image based on a pre-trained edge feature classification model, and fuses the edge feature maps with the ultrasound image, making the breast lesion area and the boundary of the breast lesion more prominent in the fused feature map. Breast lesion segmentation based on the fused feature map takes into account both segmentation efficiency and accuracy, and can efficiently and accurately segment the breast lesion area from the ultrasound image.

[0118] This invention also provides a method for segmenting breast ultrasound images, comprising:

[0119] S801. Obtain ultrasound images of the patient's breast region or the region of interest (ROI) of a breast lesion within the breast region.

[0120] For detailed implementation methods, please refer to S201, which will not be repeated here.

[0121] S802. Based on a pre-trained rear echo feature classification model, extract the rear echo feature map corresponding to the rear echo feature from the ultrasound image.

[0122] Posterior echo feature maps corresponding to posterior echo features can be extracted from ultrasound images using methods based on deep learning, machine learning, or traditional image processing. In one optional implementation, the ultrasound image can be input into a pre-trained posterior echo feature classification model. This model may include, for example, a front-end posterior echo feature extraction network and a back-end posterior echo feature classification network. The front-end network extracts posterior echo feature maps from the ultrasound image, while the back-end network outputs the type of posterior echo features based on the posterior echo feature maps. The posterior echo feature map is determined based on the output of the front-end network.

[0123] The rear echo feature classification model is trained based on sample ultrasound images labeled with rear echo feature types. Its training process can be found in [reference needed]. Figure 3The training process of the BI-RADS feature classification model, as shown, involves fusing the feature maps of the sample ultrasound images and those output by the initial back echo feature classification model during the training phase. The fused image is then input into the back echo feature classification model. Iterative training is performed with the goal of minimizing the error between the labeled results and the prediction results of the back echo feature classification model, until a pre-defined condition is met to obtain a well-trained back echo feature classification model. The types of back echo features include enhanced, unchanged, attenuated, and mixed echoes. The pre-defined condition can be that the classification accuracy of the back echo feature classification model for back echo features is greater than a pre-defined accuracy threshold. A higher accuracy threshold can be set to accurately obtain the back echo feature maps in the ultrasound images.

[0124] It should be noted that the method for extracting rear echo feature maps from ultrasound images based on a pre-trained rear echo feature classification model in this embodiment can be used for... Figure 5 In the illustrated embodiment, i.e. Figure 6 The specific implementation of the rear echo feature extractor can be the rear echo feature classification model trained in this embodiment.

[0125] S803. The posterior echo feature map is fused with the ultrasound image to obtain a fused feature map.

[0126] For specific fusion methods, please refer to S203, which will not be elaborated here.

[0127] S804. Segmentation of breast lesions based on fusion feature maps, and determination of breast lesion regions on ultrasound images.

[0128] For detailed implementation methods, please refer to S204, which will not be repeated here.

[0129] The breast ultrasound image segmentation method provided in this embodiment accurately extracts the posterior echo feature map that has a significant impact on the boundary situation from the ultrasound image based on a pre-trained posterior echo feature classification model, and fuses the posterior echo feature map with the ultrasound image, making the breast lesion area and the boundary of the breast lesion more prominent in the fused feature map. Breast lesion segmentation based on the fused feature map takes into account both segmentation efficiency and accuracy, and can efficiently and accurately segment the breast lesion area from the ultrasound image.

[0130] The above sections explained how to extract shape feature maps corresponding to shape features, edge feature maps corresponding to edge features, or posterior echo feature maps corresponding to posterior echo features from ultrasound images for breast lesion segmentation. It should be noted that similar methods can be used to extract orientation feature maps corresponding to orientation features, internal echo feature maps corresponding to internal echo features, calcification feature maps corresponding to calcification features, or blood flow feature maps corresponding to blood flow features from ultrasound images for breast lesion analysis; these will not be elaborated upon here.

[0131] Based on any of the above embodiments, the breast ultrasound image segmentation method provided in this embodiment, which segments breast lesions based on fusion feature maps and determines the breast lesion region on the ultrasound image, may include: inputting the fusion feature map into a pre-trained breast lesion segmentation model so as to determine the breast lesion region on the ultrasound image.

[0132] The breast lesion segmentation model was trained based on sample ultrasound images with labeled breast lesion regions. For details of the training process, please refer to [link / reference needed]. Figure 8 .like Figure 8 As shown, during the training phase, at least one BI-RADS feature map is extracted from the sample ultrasound image and fused with the sample ultrasound image to obtain a sample fused feature map. The sample fused feature map is then input into the breast lesion segmentation model. The model is iteratively trained with the goal of minimizing the error between the labeled breast lesion region and the breast lesion region predicted by the breast lesion segmentation model until the preset conditions are met to obtain the trained breast lesion segmentation model.

[0133] To facilitate users' viewing of breast lesion areas in ultrasound images, the method provided in this embodiment, based on any of the above embodiments, may further include: displaying the breast lesion areas segmented from the ultrasound image on a display interface. For example, the boundaries of the breast lesion areas may be displayed in the ultrasound image; the breast lesion areas may be highlighted; or the segmented breast lesion areas may be displayed separately on the display interface.

[0134] This document describes various exemplary embodiments with reference to them. However, those skilled in the art will recognize that changes and modifications can be made to the exemplary embodiments without departing from the scope of this document. For example, various operational steps and components for performing operational steps can be implemented in different ways depending on the specific application or considering any number of cost functions associated with the operation of the system (e.g., one or more steps can be deleted, modified, or combined with other steps).

[0135] Furthermore, as those skilled in the art will understand, the principles herein can be reflected in a computer program product on a computer-readable storage medium pre-loaded with computer-readable program code. Any tangible, non-transitory computer-readable storage medium may be used, including magnetic storage devices (hard disks, floppy disks, etc.), optical storage devices (CD-ROMs, DVDs, Blu-ray discs, etc.), flash memory, and / or the like. These computer program instructions may be loaded onto a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to form a machine, such that instructions, which execute on the computer or other programmable data processing apparatus, can generate means for performing a specified function. These computer program instructions may also be stored in a computer-readable storage medium that can instruct the computer or other programmable data processing apparatus to operate in a particular manner, such that instructions stored in the computer-readable storage medium can form an article of manufacture, including means for implementing the specified function. The computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to perform a series of operational steps on the computer or other programmable apparatus to produce a computer-implemented process, such that instructions, which execute on the computer or other programmable apparatus, can provide steps for implementing the specified function.

[0136] While the principles herein have been illustrated in various embodiments, numerous modifications to the structure, arrangement, proportions, elements, materials, and components, particularly suited to specific environmental and operational requirements, may be used without departing from the principles and scope of this disclosure. These modifications and other alterations or alterations will be included within the scope of this document.

[0137] The foregoing specific descriptions have been described with reference to various embodiments. However, those skilled in the art will recognize that various modifications and changes can be made without departing from the scope of this disclosure. Therefore, considerations for this disclosure are to be illustrative rather than restrictive, and all such modifications are to be included within its scope. Similarly, advantages, other advantages, and solutions to problems with respect to various embodiments have been described above. However, benefits, advantages, solutions to problems, and any elements that produce these, or make them more explicit, should not be construed as critical, essential, or necessary. The term “comprising” and any other variations thereof as used herein are non-exclusive inclusion, meaning that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed or not part of the process, method, system, article, or apparatus. Furthermore, the term “coupled” and any other variations thereof as used herein refer to physical connections, electrical connections, magnetic connections, optical connections, communication connections, functional connections, and / or any other connections.

[0138] The above examples illustrate the present invention only to aid in understanding it and are not intended to limit the scope of the invention. Those skilled in the art can make various simple deductions, modifications, or substitutions based on the principles of this invention.

Claims

1. A breast ultrasound image segmentation method, characterized by, The method comprises: acquiring an ultrasound image of a breast region or a breast lesion region of interest in the breast region; extracting a feature map corresponding to at least one BI-RADS feature from the ultrasound image based on a pre-trained BI-RADS feature classification model, wherein the BI-RADS feature classification model is obtained by fusing a sample ultrasound image and a feature map of the sample ultrasound image output by an initial BI-RADS feature classification model, iteratively training the initial BI-RADS feature classification model using the fused image until the classification accuracy of the initial BI-RADS feature classification model is greater than a preset accuracy threshold, and obtaining a trained BI-RADS feature classification model; fusing the feature map and the ultrasound image to obtain a fused feature map; the fusing of the feature map and the ultrasound image comprises: weighting and fusing the feature map and the ultrasound image according to respective fusion weights of the feature map and the ultrasound image, the respective fusion weights of the feature map and the ultrasound image being obtained by fitting weights of a contribution probability of the feature map and the ultrasound image by a logistic regression prediction model; wherein the fused feature map is determined by the following expression: wherein z represents a fusion feature map, x0 represents an ultrasound image, w0 represents a fusion weight of the ultrasound image, x i represents the i-th feature map, w i represents the fusion weight of the i-th feature map, and n≥1. segmenting a breast lesion from the fused feature map based on a pre-trained breast lesion segmentation model to determine a breast lesion region on the ultrasound image; the segmentation of the breast lesion from the fused feature map based on the pre-trained breast lesion segmentation model to determine the breast lesion region on the ultrasound image comprises: inputting the fused feature map into the pre-trained breast lesion segmentation model to determine the breast lesion region on the ultrasound image; wherein the breast lesion segmentation model is trained based on sample ultrasound images with labeled breast lesion regions, in the training stage, a feature map corresponding to at least one BI-RADS feature is extracted from the sample ultrasound image, and is fused with the sample ultrasound image to obtain a sample fused feature map, the sample fused feature map is input into the breast lesion segmentation model, and the breast lesion segmentation model is iteratively trained with the minimum error between the labeled breast lesion region and the breast lesion region predicted by the breast lesion segmentation model as the target until the trained breast lesion segmentation model is obtained when a preset condition is met.

2. The method of claim 1, wherein, The extraction of the feature map corresponding to at least one BI-RADS feature from the ultrasound image comprises: extracting a plurality of feature maps corresponding to a plurality of BI-RADS features from the ultrasound image; and the fusing of the feature map and the ultrasound image comprises: fusing a plurality of the feature maps and the ultrasound image.

3. The method of claim 2, wherein, The extraction of the feature map corresponding to at least one BI-RADS feature from the ultrasound image comprises: extracting a shape feature map, an edge feature map and a back echo feature map corresponding to a shape feature, an edge feature and a back echo feature, respectively, from the ultrasound image.

4. The method of claim 1, wherein, The extracting the feature map corresponding to at least one BI-RADS feature from the ultrasound image comprises: extracting a comprehensive feature map corresponding to multiple BI-RADS features from the ultrasound image; and the fusing the feature map and the ultrasound image comprises: fusing the comprehensive feature map and the ultrasound image.

5. The method of claim 4, wherein, The extracting the comprehensive feature map corresponding to multiple BI-RADS features from the ultrasound image comprises: extracting a comprehensive feature map corresponding to a shape feature, an edge feature and a posterior echo feature from the ultrasound image.

6. The method of claim 1, wherein, The extracting the feature map corresponding to at least one BI-RADS feature from the ultrasound image comprises: extracting a shape feature map corresponding to a shape feature from the ultrasound image; and the fusing the feature map and the ultrasound image comprises: fusing the shape feature map and the ultrasound image.

7. The method of claim 3 or 6, wherein, The extracting the shape feature map corresponding to the shape feature from the ultrasound image comprises: The extracting the shape feature map corresponding to the shape feature from the ultrasound image comprises: extracting the shape feature map corresponding to the shape feature from the ultrasound image based on a pre-trained shape feature classification model, the shape feature classification model being trained based on sample ultrasound images with types of shape features being labeled, in a training stage, fusing the sample ultrasound images and feature maps of the sample ultrasound images output by an initial shape feature classification model, inputting the fused images into the shape feature classification model, and iteratively training until a pre-set condition is met to obtain the trained shape feature classification model, the types of the shape features including a circle, an ellipse and an irregular shape.

8. The method of claim 1, wherein, The extracting the feature map corresponding to at least one BI-RADS feature from the ultrasound image comprises: extracting an edge feature map corresponding to an edge feature from the ultrasound image; and the fusing the feature map and the ultrasound image comprises: fusing the edge feature map and the ultrasound image.

9. The method of claim 3 or 8, wherein, The extracting the edge feature map corresponding to the edge feature from the ultrasound image comprises: The extracting the edge feature map corresponding to the edge feature from the ultrasound image comprises: extracting the edge feature map corresponding to the edge feature from the ultrasound image based on a pre-trained edge feature classification model, the edge feature classification model being trained based on sample ultrasound images with types of edge features being labeled, in a training stage, fusing the sample ultrasound images and feature maps of the sample ultrasound images output by an initial edge feature classification model, inputting the fused images into the edge feature classification model, and iteratively training until a pre-set condition is met to obtain the trained edge feature classification model, the types of the edge features including a smooth, a blur, an angle, a differential leaf and a burr.

10. The method of claim 1, wherein, The extracting the feature map corresponding to at least one BI-RADS feature from the ultrasound image comprises: extracting a posterior echo feature map corresponding to a posterior echo feature from the ultrasound image; and the fusing the feature map and the ultrasound image comprises: fusing the posterior echo feature map and the ultrasound image.

11. The method of claim 3 or 10, wherein, extracting, from the ultrasound image, a back echo feature map corresponding to the back echo feature, including: extracting, from the ultrasound image, a shape feature map corresponding to the back echo feature based on a pre-trained back echo feature classification model, the back echo feature classification model being trained based on sample ultrasound images with types of back echo features labeled, in a training stage, fusing the sample ultrasound images and feature maps of the sample ultrasound images output by an initial back echo feature classification model, inputting the fused images into the back echo feature classification model, and iteratively training until a pre-set condition is met to obtain the trained back echo feature classification model, the types of back echo features including enhanced, unchanged, attenuated, and mixed echoes.

12. The method of claim 1, wherein, The BI-RADS features include shape features, direction features, edge features, internal echo features, back echo features, calcification features, and blood flow features.

13. The method of claim 1, wherein, The method further includes displaying the breast lesion region segmented from the ultrasound image on a display interface.

14. The method of claim 1, wherein, Obtaining an ultrasound image of a breast region of a subject includes: emitting ultrasound waves to the breast region of the subject and receiving ultrasound echoes returned by the breast region to obtain ultrasound echo data, and generating an ultrasound image of the breast region of the subject in real time based on the ultrasound echo data; or obtaining a pre-stored ultrasound image of the breast region of the subject from a storage device. Obtaining an ultrasound image of a breast lesion region of interest in a breast region of a subject includes: obtaining an ultrasound image of the breast region of the subject; automatically determining a breast lesion region of interest in the ultrasound image of the breast region by a target detection algorithm; and obtaining an ultrasound image of the breast lesion region of interest in the breast region of the subject from the ultrasound image of the breast region based on the obtained breast lesion region of interest.

15. An ultrasound imaging device, characterized by including: an ultrasound probe; a transmitting circuit configured to output corresponding transmission sequences to the ultrasound probe according to a set mode to control the ultrasound probe to emit corresponding ultrasound waves; a receiving circuit configured to receive ultrasound echo signals output by the ultrasound probe and output ultrasound echo data; a display configured to output visual information; a processor configured to perform the breast ultrasound image segmentation method according to any one of claims 1-14.

16. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed by the processor to implement the breast ultrasound image segmentation method according to any one of claims 1-14.

Images