Method, system and electronic device for identifying target boundary in ultrasonic image

Abstract

The purpose of the present invention is to provide a method, system and electronic device for identifying target boundaries in ultrasonic images, which relates to the field of boundary recognition technology. The method includes: obtaining an ultrasonic image to be identified; inputting the ultrasonic image to be identified into a target mask model to obtain a mask image of the target to be identified; the target mask model is obtained by training a convolutional neural network using pre-enhanced historical ultrasonic images; the pre-enhanced historical ultrasonic images are obtained by sequentially pre-processing, offline enhancement processing and online enhancement processing of the historical ultrasonic images; using a contour search algorithm to determine multiple contours in the mask area of ​​the mask image of the target to be identified; determining the contour with the largest area as the contour of the target to be identified. The present invention can improve the accuracy of target boundary recognition in ultrasonic images by constructing a target mask model and pre-processing, offline enhancement processing and online enhancement processing of the training data.

Description

Technical Field

[0001] The present invention relates to the technical field of boundary recognition, and in particular to a method, system and electronic equipment for identifying a target boundary in an ultrasonic image. Background Art

[0002] Ultrasound imaging technology has the advantages of being dynamic, non-invasive, fast, and low-cost. Computer-assisted ultrasound analysis programs can reduce the work of ultrasound medical staff in ultrasound image analysis to a certain extent. Among them, the automatic edge delineation technology of the target can assist the staff in delineating the precise boundary of the target and displaying the target position. The target's shape regularity, boundary clarity, main echoes, or calcification and other characteristic indicators can be analyzed later. However, ultrasound images have complex textures and high noise, and the targets to be identified are mostly close to fat and other tissues, and the boundaries between the two are not clear enough. The existing automatic delineation method is difficult to accurately divide the target boundary, and the recognition results still need to be further manually changed by the staff. Summary of the invention

[0003] The purpose of the present invention is to provide a method, system and electronic equipment for identifying target boundaries in ultrasonic images, which can improve the accuracy of target boundary identification in ultrasonic images.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] A method for identifying target boundaries in an ultrasonic image, comprising:

[0006] Acquiring an ultrasound image to be identified;

[0007] The ultrasonic image to be identified is input into a target mask model to obtain a target mask image to be identified; the target mask model is obtained by training a convolutional neural network using pre-enhanced historical ultrasonic images; the pre-enhanced historical ultrasonic images are obtained by sequentially performing pre-processing, offline enhancement processing, and online enhancement processing on the historical ultrasonic images;

[0008] Using a contour search algorithm, multiple contours within a mask area in a mask image of a target to be identified are determined;

[0009] Determine the contour with the largest area as the contour of the target to be identified.

[0010] Optionally, before inputting the to-be-identified ultrasonic image into the target mask model to obtain the to-be-identified target mask image, the method further includes:

[0011] The ultrasonic image to be identified is preprocessed to obtain a preprocessed ultrasonic image to be identified.

[0012] Optionally, before obtaining the ultrasound image to be identified, the following steps are also included:

[0013] Obtain multiple historical ultrasound images;

[0014] Marking the target areas in the plurality of historical ultrasound images respectively to obtain a plurality of historical ultrasound marked images; the marking results are the target marked areas;

[0015] Preprocessing, offline enhancement processing, and online enhancement processing are performed on the target marked areas in the plurality of historical ultrasound images respectively to obtain a plurality of pre-enhanced historical ultrasound images;

[0016] Inputting the pre-enhanced historical ultrasound image into the encoder to obtain a feature vector;

[0017] Inputting the feature vector into a decoder to obtain a historical target mask image;

[0018] Determining a loss function value according to the historical target prediction area in the historical ultrasonic annotated image and the historical target mask image;

[0019] When the loss function value is greater than the loss function threshold, adjust the parameters of the encoder and decoder and return to the step of "inputting the pre-enhanced historical ultrasound image into the encoder to obtain a historical target mask image" until the number of iterations reaches the iteration number threshold, and determine that the trained encoder and decoder are the target mask model.

[0020] Optionally, the preprocessing includes:

[0021] The expansion ratio is dynamically obtained, and the size of the target annotation area is adjusted according to the expansion ratio.

[0022] Optionally, the formula for offline enhancement processing is:

[0023]

[0024] Among them, Rect aug (x aug ,y aug , w aug ,h aug ) is the target annotation area after offline enhancement processing; x aug is the horizontal coordinate of the upper left corner of the enhanced target annotation area; y aug is the upper left corner ordinate of the enhanced target annotation area; w aug is the width of the enhanced target annotation area; h aug is the height of the enhanced target annotation area, x ori y is the upper left corner horizontal coordinate of the target annotation area before enhancement; ori w is the upper left corner ordinate of the target annotation area before enhancement; ori To enhance the width of the front target annotation area; h ori To enhance the height of the front target marking area; xshift To adjust the x-direction offset of the target annotation area according to the expansion ratio; shift To adjust the front and rear y-direction offset of the target annotation area according to the expansion ratio.

[0025] Optionally, the online enhancement processing includes: one or more of: brightness random change processing, contrast random change processing, sharpening processing, upside down flip processing, left side to right flip processing, translation processing, and rotation processing.

[0026] Optionally, the encoder is a cascaded residual convolution module.

[0027] A method for identifying target boundaries in an ultrasonic image, comprising:

[0028] An ultrasonic image acquisition module to be identified, used to acquire the ultrasonic image to be identified;

[0029] The target mask image determination module to be identified is used to input the ultrasonic image to be identified into the target mask model to obtain the target mask image to be identified; the target mask model is obtained by training the convolutional neural network using the pre-enhanced historical ultrasonic image; the pre-enhanced historical ultrasonic image is obtained by sequentially performing pre-processing, offline enhancement processing and online enhancement processing on the historical ultrasonic image;

[0030] A contour search module, used to determine multiple contours in the mask area of ​​the target mask image to be identified by using a contour search algorithm;

[0031] The module for determining the outline of the target to be identified is used to determine the outline with the largest area as the outline of the target to be identified.

[0032] An electronic device comprises a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute a method for identifying a target boundary in an ultrasonic image.

[0033] Optionally, the memory is a readable storage medium.

[0034] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0035] The purpose of the present invention is to provide a method, system and electronic device for identifying the boundary of a target in an ultrasonic image, to obtain an ultrasonic image to be identified; to input the ultrasonic image to be identified into a target mask model to obtain a mask image of the target to be identified; the target mask model is obtained by training a convolutional neural network using pre-enhanced historical ultrasonic images; the pre-enhanced historical ultrasonic images are obtained by sequentially pre-processing, offline enhancement processing and online enhancement processing of the historical ultrasonic images; using a contour search algorithm, multiple contours in a mask area in the mask image of the target to be identified are determined; and the contour with the largest area is determined as the contour of the target to be identified. The present invention can improve the accuracy of target boundary identification in ultrasonic images by constructing a target mask model and pre-processing, offline enhancement processing and online enhancement processing of training data. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0037] Figure 1 is a flow chart of a method for identifying target boundaries in an ultrasonic image in Embodiment 1 of the present invention;

[0038] Figure 2 A schematic diagram of the construction and use process of the method for automatically outlining the boundaries of ultrasonic breast nodules in Example 2 of the present invention;

[0039] Figure 3 This is a schematic diagram of the offline enhancement effect in Example 2 of the present invention;

[0040] Figure 4 This is a schematic diagram of the online enhancement effect in Example 2 of the present invention;

[0041] Figure 5 Schematic diagram of the structure of the cascaded residual module in Embodiment 2 of the present invention;

[0042] Figure 6 Schematic diagram of the BN-SegNet network structure in Example 2 of the present invention;

[0043] Figure 7 Schematic diagram of the effect of the automatic outlining method of ultrasonic breast nodules in Example 2 of the present invention. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] The purpose of the present invention is to provide a method, system and electronic equipment for identifying target boundaries in ultrasonic images, which can improve the accuracy of target boundary identification in ultrasonic images.

[0046] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0047] Example 1

[0048] like Figure 1 As shown, this embodiment provides a method for identifying target boundaries in an ultrasound image, comprising:

[0049] Step 101: Acquire an ultrasound image to be identified.

[0050] Step 102: Input the ultrasound image to be identified into the target mask model to obtain the target mask image to be identified; the target mask model is obtained by training the convolutional neural network using the pre-enhanced historical ultrasound image; the pre-enhanced historical ultrasound image is obtained by sequentially performing preprocessing, offline enhancement processing and online enhancement processing on the historical ultrasound image.

[0051] Step 103: using a contour search algorithm to determine multiple contours within the mask area of ​​the target mask image to be identified.

[0052] Step 104: Determine the contour with the largest area as the contour of the target to be identified, that is, the boundary of the target to be identified.

[0053] Before step 102, the method further includes:

[0054] Step 105: preprocessing the ultrasonic image to be identified to obtain a preprocessed ultrasonic image to be identified.

[0055] Before step 101, the method further includes:

[0056] Step 106: Acquire multiple historical ultrasound images.

[0057] Step 107: annotate the target areas in the plurality of historical ultrasound images respectively to obtain a plurality of historical ultrasound annotated images; the annotated results are the target annotated areas.

[0058] Step 108: Preprocessing, offline enhancement processing, and online enhancement processing are performed on the target annotated areas in the plurality of historical ultrasound images respectively to obtain a plurality of pre-enhanced historical ultrasound images.

[0059] The preprocessing includes: dynamically obtaining an expansion ratio, and adjusting the size of the target annotation area according to the expansion ratio.

[0060] Specifically, the formula for offline enhancement processing is:

[0061]

[0062] Among them, Rect aug (x aug ,y aug , w aug ,h aug ) is the target annotation area after offline enhancement processing; x aug is the horizontal coordinate of the upper left corner of the enhanced target annotation area; y aug is the upper left corner ordinate of the enhanced target annotation area; w aug is the width of the enhanced target annotation area; h aug is the height of the enhanced target annotation area, x ori y is the upper left corner horizontal coordinate of the target annotation area before enhancement; ori w is the upper left corner ordinate of the target annotation area before enhancement; ori To enhance the width of the front target annotation area; h ori To enhance the height of the front target marking area; x shift To adjust the x-direction offset of the target annotation area according to the expansion ratio; shift To adjust the front and rear y-direction offset of the target annotation area according to the expansion ratio.

[0063] Among them, the online enhancement processing includes: one or more of: brightness random change processing, contrast random change processing, sharpening processing, upside down flip processing, left-right flip processing, translation processing, and rotation processing.

[0064] Step 109: Input the pre-enhanced historical ultrasound image into the encoder to obtain a feature vector.

[0065] Step 1010: Input the feature vector into the decoder to obtain a historical target mask image.

[0066] Step 1011: Determine the loss function value according to the historical target prediction area in the historical ultrasonic annotated image and the historical target mask image.

[0067] Step 1012: When the loss function value is greater than the loss function threshold, adjust the parameters of the encoder and decoder and return to step 109 until the number of iterations reaches the iteration threshold, and determine that the trained encoder and decoder are the target mask model.

[0068] Example 2

[0069] Next, the method described in Example 1 is specifically described by taking the recognition and delineation of the boundary contour of breast cancer nodules as an example. The processing flow is as follows: Figure 2 shown.

[0070] 1. Image preprocessing and enhancement device

[0071] 1.1 Image Preprocessing

[0072] We used 2868 ultrasound images of breast nodules with nodule boundaries annotated by professional ultrasound physicians. In the ultrasound images, we took the circumscribed rectangle of the outlined boundary, that is, we located the circumscribed rectangle that satisfied:

[0073]

[0074] Among them, Rect(x,y,w,h) is the bounding rectangle of the nodule in the image, x and y are the horizontal and vertical coordinates of the upper left corner of the bounding rectangle, w and h are the width and height of the bounding rectangle, respectively. x is the set of horizontal coordinates of all outline points, P y is the set of ordinates of all outline points. After obtaining the basic bounding rectangle, the following formula is used to dynamically obtain the rectangle expansion ratio based on the size of the nodule bounding rectangle:

[0075]

[0076] Where ratio is the ratio of the enclosing rectangle to expand, and max(w,h) is the larger value of the rectangle width and height. Then the enclosing rectangle is expanded according to the ratio, that is:

[0077]

[0078] Among them, Rect new (x new ,y new , w new ,h new ) is the enclosed rectangle of the image after expansion, x new and new are the horizontal and vertical coordinates of the upper left corner of the circumscribed rectangle, w new and h new is the width and height of the circumscribed rectangle, x old and oldare the horizontal and vertical coordinates of the upper left corner of the circumscribed rectangle before expansion, w old and h old is the width and height of the bounding rectangle before expansion, and ratio is the expansion ratio calculated above. When the expansion exceeds the boundary of the image itself, the value of the image boundary is taken as the coordinate point.

[0079] 1.2 Offline Image Enhancement

[0080] In order to adapt the automatic delineation device of breast nodules to input images of various sizes and positions and increase the robustness of the automatic delineation model, in the original ultrasound image, the nodule area is dynamically cropped based on the above-mentioned expanded rectangular coordinates in the following manner. First, the offsets in the x and y directions are calculated:

[0081] x shift =[r shift_x *w ori ],r shift_x ∈{-0.2, -0.1, 0, 0.1, 0.2).

[0082] y shift =[r shift_y *h ori ],r shift_y ∈{-0.2, -0.1, 0, 0.1, 0.2).

[0083] Among them, x shift and shift Respectively represent the offset of the enhanced image in the horizontal and vertical directions of the original image, r shift_x and r shift_y Respectively represent the offset ratio of the enhanced image in the horizontal and vertical directions of the original image. Then calculate the enhanced rectangle to satisfy:

[0084]

[0085] Among them, x aug and aug are the horizontal and vertical coordinates of the upper left corner of the enhanced rectangular area, w aug and h aug are the width and height of the enhanced rectangular area, x ori and ori are the horizontal and vertical coordinates of the upper left corner of the original rectangular area before enhancement, w ori and h ori The width and height of the rectangular area before enhancement are recorded as w aug and h aug Used in the image post-processing stage. Using the enhanced Rect augThe nodule area is cut out from the original image, and then all the cut breast nodule images and the corresponding outline masks are uniformly resized to (w input *h input ). Among them, w input is the image width, here it is set to 192 pixels, h input is the image height, which is set to 128 pixels here. For nodule ultrasound images, the Lanczos interpolation algorithm is used when the image is resized. The nearest neighbor interpolation algorithm is used when the outline mask is resized. See the offline enhancement effect. Figure 3 .

[0086] 1.3 Online Image Enhancement

[0087] In order to increase the adaptability to nodule images of different brightness, size, position or quality, during the neural network training process, the following online enhancements are performed on the image before each epoch of image data is input into the network:

[0088] 1.3.1 Random brightness changes:

[0089] For a picture, calculate the intensity value I of all pixels after the change new_b :

[0090] I new_b =I old +delta(-thresh b ≤delta≤thtresh b ).

[0091] At the same time ensure

[0092] Among them, I old is the pixel intensity value before the corresponding pixel enhancement, delta is the brightness change amplitude, and thresh b Take 32, delta is at the threshold thresh b To the threshold thresh b Pick a random integer between .

[0093] 1.3.2 Random changes in contrast:

[0094] Calculate the intensity value of each pixel after the change:

[0095] I new_c =c*I old (1-thresh c ≤alpha≤1+thresh c ).

[0096] At the same time ensure

[0097] 1.3.3 Image Sharpening:

[0098] Constructing the sharpening operator Scan the image using the sharpening operator. For each 3×3 region in the image, the intensity I of its center coordinate is new (x, y) = 5*I old (x, y)-I old (x-1, y)-I old (x+1,y)-I old (x, y-1)-I old (x, y+1). new The pixel intensity of the specified coordinate point of the enhanced image, I old The pixel intensity at the specified coordinates of the image before enhancement.

[0099] 1.3.4 Randomly flip, translate or rotate the image up and down, horizontally. See the online enhancement effect Figure 4 .

[0100] 2.BN-SegNet network construction and training

[0101] like Figure 6 , the deep neural network uses an encoder-decoder structure, and the encoder part uses a cascaded residual convolution module (ResBlock) (such as Figure 5 As shown), including convolution operation (Conv), batch normalization operation (BN), and zero padding operation (ZeroPadding). The original image information is downsampled 4 times, using a 7×7 convolution and 3 maximum pooling (MaxPooling) operations for downsampling. In the last 3 downsampling operations, the input image passes through 3, 4, and 6 ResBlocks respectively. In the decoder, the feature layer after the 4 downsamplings is upsampled 4 times, and the upsampling is achieved using a two-dimensional upsampling operation (UpSampling2D). The output is restored to the original image size. Before each downsampling, the encoder passes the output to the decoder's network of the same size, and merges it with the upsampling result to achieve pixel-level position information transmission. The network structure is shown as follows Figure 5 As shown in the figure, the deep neural network includes an input image of the breast nodule region after preprocessing, with three dimensions (w, h, c), where w = 192; h = 128; c = 3. The output is the image of the nodule region mask predicted by the neural network, with two dimensions (w, h), where w = 192; h = 128.

[0102] The comprehensive loss function combining position and pixel level used in training is:

[0103] L total =L jaccard +L BCE .

[0104] Among them, L total is the comprehensive loss function, L jaccard The loss function corresponding to the jaccard coefficient is: Among them, A gt A is the actual boundary area marked by the doctor. pred is the boundary area predicted by the network, L BCE The average binary cross entropy loss for all points in the prediction region is:

[0105]

[0106] Where n is the total number of pixels in the output image, here w*h=24576, y i is the judgment value of whether there is actually a non-zero pixel at any point i, which is 0 or 1; p(y i ) is the probability that the network predicts that there is a non-zero pixel at this point, between 0 and 1.

[0107] 2. Image Post-Processing

[0108] The mask size output by the neural network is restored to w using the nearest neighbor interpolation algorithm aug *h aug Size, where w aug and h aug are the width and height of the enhanced nodule image recorded in the enhancement stage, respectively. The contour search method is used to determine the contour in the mask, and the largest contour is retained as the outline boundary of the breast nodule.

[0109] Compared with the method of manually outlining nodules by ultrasound doctors, the present invention has great advantages in execution speed and efficiency. It takes tens of seconds or even minutes for ultrasound doctors to accurately outline the edge of nodules, while the method in the present invention can complete the edge outline of breast nodules within 300 milliseconds on an ordinary office computer (model CPU i7-11700F), and complete the edge outline of breast nodules within 100 milliseconds on a computer with a GPU (model NVIDIAGTX 1660Super). It can process large-scale data in batches, greatly improve the efficiency of doctors' annotation of ultrasound images, and can be adapted to the fully automatic breast nodule detection and analysis process.

[0110] Compared with traditional boundary segmentation methods based on image edge detection, active contour search, etc., the method of the present invention has a stronger resistance to interference from surrounding tissues of breast nodules, such as breast ducts, fat, etc. For breast nodules with unclear boundaries, the segmentation method based on edge detection cannot delineate the appropriate boundaries, while the method based on deep neural network provided by the present invention can infer the appropriate boundaries based on the overall information of the nodules. Therefore, the present invention can be applied to breast nodules of different locations, different sizes, different appearances, and different categories. In general, it has higher accuracy, and the effect is as follows: Figure 7 shown.

[0111] Compared with the existing deep neural network methods based on U-Net, FPN, PSPnet and other frameworks, the present invention achieves a higher boundary segmentation accuracy by optimizing the network architecture, adding a large amount of labeled data, etc. At the same time, the image preprocessing and offline and online enhancement methods in the present invention make the present invention highly resistant to the interference of image noise and position changes.

[0112] Example 3

[0113] In order to execute the method corresponding to the above-mentioned embodiment 1 to achieve corresponding functions and technical effects, a method for identifying target boundaries in an ultrasound image is provided below, including:

[0114] The module for acquiring the ultrasonic image to be identified is used to acquire the ultrasonic image to be identified.

[0115] The target mask image determination module to be identified is used to input the ultrasound image to be identified into the target mask model to obtain the target mask image to be identified; the target mask model is obtained by training the convolutional neural network using the pre-enhanced historical ultrasound image; the pre-enhanced historical ultrasound image is obtained by preprocessing, offline enhancement processing and online enhancement processing of the historical ultrasound image in sequence.

[0116] The contour search module is used to determine multiple contours in the mask area of ​​the target mask image to be identified by using a contour search algorithm.

[0117] The module for determining the outline of the target to be identified is used to determine the outline with the largest area as the outline of the target to be identified.

[0118] Example 4

[0119] This embodiment provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute a method for identifying a target boundary in an ultrasound image described in Embodiment 1. The memory is a readable storage medium.

[0120] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0121] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for identifying target boundaries in an ultrasonic image, characterized in that: include: Acquiring an ultrasound image to be identified; Inputting the to-be-identified ultrasonic image into a target mask model to obtain a to-be-identified target mask image; The target mask model is obtained by training the convolutional neural network using the pre-enhanced historical ultrasound image; the pre-enhanced historical ultrasound image is obtained by sequentially preprocessing, offline enhancement processing and online enhancement processing of the historical ultrasound image; the convolutional neural network is an encoder-decoder structure; the encoder includes a cascaded residual convolution module; in the encoder, the original image information is downsampled 4 times, using a 7×7 convolution and 3 maximum pooling operations for downsampling respectively, and in the latter 3 downsamplings, the input image is sequentially passed through 3, 4, and 6 cascaded residual convolution modules; in the decoder, the feature layer after 4 downsamplings is upsampled 4 times, and the upsampling is achieved using a two-dimensional upsampling operation, and the output is restored to the original image size; Using a contour search algorithm, multiple contours within a mask area in a mask image of a target to be identified are determined; Determine the contour with the largest area as the contour of the target to be identified; The pre-processing comprises: Dynamically obtain the outward expansion ratio, and adjust the size of the target marked area in the historical ultrasound image according to the outward expansion ratio; The formula for dynamically obtaining the external expansion ratio is: Where ratio is the ratio of the enclosed rectangle to expand outward, and max(w,h) is the maximum value of the rectangle width w and the rectangle height h; The formula for expanding the bounding rectangle according to the ratio of the bounding rectangle is: Among them, Rect new (x new ,y new ,w new ,h new ) is the bounding rectangle obtained by adjusting the target annotation area according to the expansion ratio, x new and new are the horizontal and vertical coordinates of the upper left corner of the circumscribed rectangle after expansion, respectively. new and h new is the width and height of the enclosing rectangle after expansion, x old and old are the upper left corner horizontal coordinate and upper left corner vertical coordinate of the circumscribed rectangle before expansion, w old and h old are the width and height of the circumscribed rectangle before expansion. When the expansion exceeds the boundary of the ultrasonic image to be identified, the boundary of the ultrasonic image to be identified is taken as the circumscribed rectangle after expansion. The formula for offline enhancement processing is: Among them, Rect aug (x aug ,y aug , w aug ,h aug ) is the target annotation area after offline enhancement processing; x aug is the horizontal coordinate of the upper left corner of the enhanced target annotation area; y aug is the upper left corner ordinate of the enhanced target annotation area; w aug is the width of the enhanced target annotation area; h aug is the height of the enhanced target annotation area; x ori y is the upper left corner horizontal coordinate of the target annotation area before enhancement; ori w is the upper left corner ordinate of the target annotation area before enhancement; ori To enhance the width of the front target annotation area; h ori To enhance the height of the front target marking area; x shift To adjust the x-direction offset of the target annotation area according to the expansion ratio; shift To adjust the front and rear y-direction offset of the target annotation area according to the expansion ratio; The online enhancement processing includes: one or more of brightness random change processing, contrast random change processing, sharpening processing, upside down flip processing, left side to right flip processing, translation processing, and rotation processing.

2. A method for identifying target boundaries in ultrasonic images according to claim 1, characterized in that: Before inputting the to-be-identified ultrasonic image into the target mask model to obtain the to-be-identified target mask image, the method further includes: The ultrasonic image to be identified is preprocessed to obtain a preprocessed ultrasonic image to be identified.

3. The method for identifying target boundaries in ultrasonic images according to claim 1, characterized in that: Before obtaining the ultrasound image to be identified, it also includes: Obtain multiple historical ultrasound images; Marking the target areas in the plurality of historical ultrasound images respectively to obtain a plurality of historical ultrasound marked images; the marking results are the target marked areas; Preprocessing, offline enhancement processing, and online enhancement processing are performed on the target marked areas in the plurality of historical ultrasound images respectively to obtain a plurality of pre-enhanced historical ultrasound images; Inputting the pre-enhanced historical ultrasound image into the encoder to obtain a feature vector; Inputting the feature vector into a decoder to obtain a historical target mask image; Determining a loss function value according to the historical target prediction area in the historical ultrasonic annotated image and the historical target mask image; When the loss function value is greater than the loss function threshold, adjust the parameters of the encoder and the decoder and return to the step of "inputting the pre-enhanced historical ultrasound image into the encoder to obtain a feature vector; inputting the feature vector into the decoder to obtain a historical target mask image", until the number of iterations reaches the iteration number threshold, and determine that the trained encoder and decoder are the target mask model.

4. A target boundary recognition system in an ultrasonic image, characterized in that: The target boundary recognition system in ultrasound images is applied to the target boundary recognition method in ultrasound images according to any one of claims 1 to 3, and the target boundary recognition system in ultrasound images comprises: An ultrasonic image acquisition module to be identified, used to acquire the ultrasonic image to be identified; The target mask image determination module to be identified is used to input the ultrasonic image to be identified into the target mask model to obtain the target mask image to be identified; the target mask model is obtained by training the convolutional neural network using the pre-enhanced historical ultrasonic image; the pre-enhanced historical ultrasonic image is obtained by sequentially performing pre-processing, offline enhancement processing and online enhancement processing on the historical ultrasonic image; A contour search module, used to determine multiple contours in the mask area of ​​the target mask image to be identified by using a contour search algorithm; The module for determining the outline of the target to be identified is used to determine the outline with the largest area as the outline of the target to be identified.

5. An electronic device, characterized in that: The electronic device comprises a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute a method for identifying target boundaries in an ultrasonic image according to any one of claims 1 to 3.

6. An electronic device according to claim 5, characterized in that: The memory is a readable storage medium.

Images