A segmentation model construction and contour identification method, device and computer equipment

Abstract

The application discloses a kind of segmentation model construction and contour identification method, device and computer equipment, the method includes, obtaining sample image;Sample image includes the positive sample image marked with target area and the negative sample image not marked;Negative sample image is input to the shared encoder of preset image segmentation model, carries out encoding processing, obtains encoding result;Encoding result is respectively input into segmentation branch decoder and edge detection branch decoder, obtains segmentation branch result and edge detection result;And the decoding result of segmentation branch decoder and edge detection branch decoder is input into fusion branch decoder, obtains fusion segmentation result;Based on segmentation branch result, edge detection result and fusion segmentation result, obtain the output result of preset image segmentation model;Based on positive sample image and output result, obtain the loss function of preset image segmentation model;Based on loss function, adjust the parameter of preset image segmentation model, obtain multi-branch fusion segmentation model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ultrasonic image recognition, and particularly relates to a segmentation model construction and contour recognition method and device and computer equipment. BACKGROUND

[0002] Cardiovascular disease is one of the major diseases threatening human health. The main performance is that the coronary artery develops severe atherosclerosis, which causes coronary artery stenosis, obstruction or thrombosis, leading to myocardial ischemia, hypoxia or myocardial infarction. Intravenous ultrasound (IVUS) technology is one of the most effective imaging methods for diagnosing cardiovascular diseases. Through interventional catheter technology and ultrasonic imaging technology, the internal structure of the blood vessel can be detected, and the lesion condition inside the blood vessel can be displayed. Effective analysis of the IVUS image can better help doctors determine the diagnosis result and develop a diagnosis and treatment plan.

[0003] One of the most important processes of IVUS image analysis is to delineate the lumen and the intima-media-adventitia boundary. The commonly used methods in clinic include manual delineation, virtual histology imaging technology and digital image processing technology. The IVUS image segmentation method based on digital image processing mainly includes traditional segmentation method and deep learning based method. However, due to the influence of ultrasonic speckle noise and the existence of various artifacts, lesions and surrounding structures, the traditional IVUS image segmentation algorithm is easily disturbed, and it is difficult to ensure the accuracy of automatic segmentation of the lumen and intima-media-adventitia region, and it is time-consuming. The IVUS image segmentation algorithm based on deep learning can learn through the training of neural network, realize the automatic segmentation of intima and intima-media-adventitia region in IVUS image with faster speed, and has a certain robustness to ultrasonic noise. However, in the case of limited data volume, the single task IVUS image segmentation model is difficult to learn enough image features. For IVUS images with surrounding structure interference, blood vessel bifurcation and other conditions, it is difficult to ensure the accuracy of the segmentation result, especially the positioning of the edge region pixels is not accurate enough. SUMMARY

[0004] Therefore, the technical problem to be solved by the present application is to overcome the low accuracy defect caused by various interference factors in the existing digital image processing IVUS image segmentation method, so as to provide a segmentation model construction and contour recognition method, device and computer equipment.

[0005] According to a first aspect, the present application discloses a method for constructing a multi-branch fusion segmentation model, comprising: obtaining sample images; the sample images include positive sample images marked with target regions and negative sample images without marking; inputting the negative sample images as input data into a shared encoder of a preset image segmentation model for encoding processing to obtain an encoding result; inputting the encoding result into a segmentation branch decoder and an edge detection branch decoder of the preset image segmentation model respectively to obtain a segmentation branch result and an edge detection result; inputting the decoding results of the segmentation branch decoder and the edge detection branch decoder into a fusion branch decoder of the preset image segmentation model to obtain a fusion segmentation result; obtaining an output result of the preset image segmentation model based on the segmentation branch result, the edge detection result and the fusion segmentation result; obtaining a loss function of the preset image segmentation model based on the positive sample images and the output result; adjusting parameters of the preset image segmentation model based on the loss function to obtain a multi-branch fusion segmentation model.

[0006] Optionally, the multi-branch fusion segmentation model comprises an n-level shared encoder and a multi-branch decoder, the multi-branch decoder comprises an n-level segmentation branch decoder, an n-level edge detection branch decoder and an n-1 level fusion branch decoder, and inputting the encoding result into the segmentation branch decoder and the edge detection branch decoder of the preset image segmentation model respectively to obtain the segmentation branch result and the edge detection result comprises: inputting the sample images into the first-level shared encoder to obtain first encoding output data; inputting the first encoding output data into a second-level shared encoder, a first-level segmentation branch decoder and a first-level edge detection branch decoder to obtain first segmentation data and first edge detection data; inputting output data of an m-1 level shared encoder into an m-level shared encoder to obtain output data of the m-level shared encoder, 2≤m≤n; inputting the output data of the m-level shared encoder and output data of an m-1 level segmentation branch decoder into an m-level segmentation branch decoder to obtain output data of the m-level segmentation branch decoder, 2≤m≤n; inputting the output data of the m-level shared encoder and output data of an m-1 level edge detection branch decoder into an m-level edge detection branch decoder to obtain output data of the m-level edge detection branch decoder, 2≤m≤n; obtaining the segmentation branch result based on the first segmentation data and the output data of the m-level segmentation branch decoder, and obtaining the edge detection result based on the first edge detection data and the output data of the m-level edge detection branch decoder.

[0007] Optionally, inputting the decoding results of the segmentation branch decoder and the edge detection branch decoder into the fusion branch decoder of the preset image segmentation model to obtain the fused segmentation result includes: inputting the output data of the first-level segmentation branch decoder and the output data of the first-level edge detection branch decoder into the first-level fusion branch decoder to obtain the output data of the first-level fusion branch decoder; inputting the output data of the m-th level segmentation branch decoder, the output data of the m-th level edge detection branch decoder, and the output data of the (m-1)-th level fusion branch decoder into the m-th level fusion branch decoder to obtain the output data of the m-th level fusion branch decoder; and obtaining the fused segmentation result based on the output data of the first-level fusion branch decoder and the output data of the m-th level fusion branch decoder.

[0008] Optionally, obtaining the loss function of the preset image segmentation model based on the positive sample image and the output result includes: calculating a segmentation branch loss function based on the positive sample image and the segmentation branch result; calculating an edge detection loss function based on the positive sample image and the edge detection result; calculating a fusion segmentation loss function based on the positive sample image and the fusion segmentation result; and obtaining the loss function based on the segmentation branch loss function, the edge detection loss function, and the fusion segmentation loss function.

[0009] Optionally, obtaining the sample image includes: preprocessing the sample image, labeling the preprocessed sample image, and using the labeled sample image as the positive sample image.

[0010] Optionally, acquiring the sample image further includes: performing amplification processing on the sample image to obtain an amplified sample image.

[0011] According to a second aspect, embodiments of the present invention also disclose an ultrasound imaging contour recognition method, comprising: acquiring an original image; inputting the original image into a preset multi-branch fusion segmentation model to obtain an image segmentation result of the original image; the preset multi-branch fusion segmentation model is obtained using the construction method of the multi-branch fusion segmentation model as described in the first aspect or any optional embodiment of the first aspect; and performing contour extraction based on the image segmentation result to obtain contour recognition results for each region of the original image.

[0012] Optionally, the step of extracting contours based on the image segmentation results to obtain contour recognition results for each region of the original image includes: extracting contours based on the image segmentation results to obtain a first image result; performing Gaussian blur processing on the first image result to obtain a second image result; performing binarization processing on the second image result to obtain a third image result; and performing region of interest selection processing on the third image result to obtain the final contour recognition results for each region.

[0013] Optionally, the third image result is subjected to a region of interest selection process to obtain a final region contour recognition result, including: determining a maximum connected domain containing a gray value of 255 of a center point of the third image result as a region of interest based on the third image result; obtaining a contour point set based on the intersection set of n rays emitted at n angles with equal intervals from the image center of the third image result and the region of interest; sequentially connecting the curve fitting result passing through the contour point set to obtain the final region contour recognition result.

[0014] According to a third aspect, the embodiments of the present application also disclose a construction device of a multi-branch fusion segmentation model, including: a first acquisition module configured to acquire a sample image; the sample image includes a positive sample image marked with a target region and a negative sample image without marking; an encoding module configured to input the negative sample image as input data to a shared encoder of a preset image segmentation model to perform encoding processing to obtain an encoding result; a decoding module configured to input the encoding result into a segmentation branch decoder and an edge detection branch decoder of the preset image segmentation model respectively to obtain a segmentation branch result and an edge detection result; and input the decoding results of the segmentation branch decoder and the edge detection branch decoder into a fusion branch decoder of the preset image segmentation model to obtain a fusion segmentation result; an output module configured to obtain an output result of the preset image segmentation model based on the segmentation branch result, the edge detection result and the fusion segmentation result; a loss function module configured to obtain a loss function of the preset image segmentation model based on the positive sample image and the output result; and an adjustment module configured to adjust parameters of the preset image segmentation model based on the loss function to obtain a multi-branch fusion segmentation model.

[0015] According to a fourth aspect, the embodiments of the present application also disclose an ultrasound imaging contour recognition device, including: a second acquisition module configured to acquire an original image; a segmentation module configured to input the original image into a preset multi-branch fusion segmentation model to obtain an image segmentation result of the original image; the preset multi-branch fusion segmentation model is obtained by using the construction method of the multi-branch fusion segmentation model according to the first aspect or any optional implementation manner of the first aspect; and a contour extraction module configured to perform contour extraction based on the image segmentation result to obtain a region contour recognition result of the original image.

[0016] According to a fifth aspect, the embodiments of the present application further disclose a computer device, comprising: at least one processor; and a memory connected with the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the multi-branch fusion segmentation model construction method according to the first aspect or any optional implementation manner of the first aspect or the ultrasound imaging contour identification method according to the second aspect or any optional implementation manner of the second aspect.

[0017] According to a sixth aspect, the embodiments of the present application further disclose a computer-readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of the multi-branch fusion segmentation model construction method according to the first aspect or any optional implementation manner of the first aspect or the ultrasound imaging contour identification method according to the second aspect or any optional implementation manner of the second aspect.

[0018] The technical scheme of the present application has the following advantages:

[0019] The segmentation model construction and contour identification method, device and computer device provided by the present application, the model construction method comprises: acquiring a sample image; the sample image comprises a positive sample image marked with a target region and a negative sample image without marking; inputting the negative sample image as input data to a shared encoder of a preset image segmentation model for encoding processing to obtain an encoding result; inputting the encoding result into a segmentation branch decoder and an edge detection branch decoder of the preset image segmentation model respectively to obtain a segmentation branch result and an edge detection result; and inputting the decoding results of the segmentation branch decoder and the edge detection branch decoder into a fusion branch decoder of the preset image segmentation model to obtain a fusion segmentation result; obtaining an output result of the preset image segmentation model based on the segmentation branch result, the edge detection result and the fusion segmentation result; obtaining a loss function of the preset image segmentation model based on the positive sample image and the output result; and adjusting parameters of the preset image segmentation model based on the loss function to obtain a multi-branch fusion segmentation model. After the sample image is input into the shared encoder, the segmentation branch decoder, the edge detection branch decoder and the fusion branch decoder, different segmentation results are obtained, which improves the accuracy of model identification. Based on the edge detection branch decoder, the learning ability and the generalization ability are enhanced, and through the fusion branch decoder, the image features learned by each branch task are fully utilized, which further improves the accuracy of network segmentation. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the specific embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the specific embodiments or prior art description. Obviously, the drawings described below are some embodiments of the present application, and all other drawings obtained by those of ordinary skill in the art without creative labor based on these drawings also belong to the protection scope of the present application.

[0021] Figure 1 A flow chart of a specific example of the construction method of the multi-branch fusion segmentation model in the embodiments of the present application;

[0022] Figure 2 A flow chart of a specific example of the ultrasonic imaging contour recognition method in the embodiments of the present application;

[0023] Figure 3 A flow chart of a specific example of the construction device of the multi-branch fusion segmentation model in the embodiments of the present application;

[0024] Figure 4 A principle block diagram of a specific example of the ultrasonic imaging contour recognition device in the embodiments of the present application;

[0025] Figure 5 A specific example diagram of the computer device in the embodiments of the present application;

[0026] Figure 6 A specific example diagram of the construction method of the multi-branch fusion segmentation model in the embodiments of the present application;

[0027] Figure 7 A specific example diagram of the construction method of the multi-branch fusion segmentation model in the embodiments of the present application;

[0028] Figure 8 A specific example diagram of the ultrasonic imaging contour recognition method in the embodiments of the present application;

[0029] Figure 9 Another specific example diagram of the ultrasonic imaging contour recognition method in the embodiments of the present application. DETAILED DESCRIPTION

[0030] The technical solutions of the present application will be described clearly and completely below in conjunction with the drawings. Obviously, the described embodiments are some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor also belong to the protection scope of the present application.

[0031] In the description of the present application, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer" and the like indicate the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first", "second", "third" are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0032] In the description of the present application, it should be noted that unless otherwise explicitly specified and limited, the terms "mounting", "connecting", "connecting" should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium; it can be the communication between two elements inside, it can be wireless connection, or it can be wired connection. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0033] In addition, the technical features involved in the different embodiments of the application described below can be combined with each other as long as there is no conflict.

[0034] The embodiment of the present application discloses a method for constructing a multi-branch fusion segmentation model, as shown in Figure 1 The method comprises the following steps:

[0035] Step 101: obtaining a sample image; the sample image includes a positive sample image marked with a target region and a negative sample image without marking. Illustratively, the sample image can be an IVUS image obtained by intravascular ultrasound technology, which can be used for analyzing cardiovascular diseases. The target region is the region of the intima and the middle-outer membrane of the blood vessel in the sample image, the positive sample image is the sample image after artificial annotation, and the negative sample image is the sample image without artificial annotation.

[0036] Step 102: inputting the negative sample image as input data into a preset shared encoder of an image segmentation model to perform encoding processing and obtaining an encoding result.

[0037] Illustratively, the sample image without artificial marking is input into the shared encoder of the image segmentation model as input data, and the corresponding encoding result is obtained by encoding. The structure diagram of the preset image segmentation model is as shown in Figure 7As shown, the image segmentation model comprises an input image, a shared encoder, a decoder and a segmentation structure group, the shared encoder comprises n layers of encoders, the decoder comprises a segmentation branch decoder (n layers of decoders), an edge detection branch decoder (n layers of decoders) and a fusion branch decoder (n-1 layers of decoders). The shared encoder is sequentially connected by n layers of encoders. In the embodiment of the present application, considering the size of the ultrasound image and the case of having more noise, a shared encoder with 4 layers is constructed, which can filter noise information, prevent overfitting and avoid losing too much detailed information. The shared encoder is composed of 4 layers of encoders, each of which includes a convolution block and a down-sampling block, as shown in Figure 6 As shown, the convolution block is composed of two identical sub-units, each of which includes a convolution layer, a batch normalization layer, a Dropout layer and an activation function layer connected in sequence. The down-sampling block is sequentially connected by a convolution layer with a step of 2 and an activation function layer, the convolution kernel size of the convolution layer is 3x3, and the activation function uniformly uses the LeakyRelu function. The number of layers of the shared encoder, the convolution kernel size in each encoder and the type of activation function are not limited in the embodiment of the present application, and they can be determined according to actual needs by those skilled in the art.

[0038] Step 103: respectively input the encoding results into the segmentation branch decoder and the edge detection branch decoder of the preset image segmentation model, to obtain a segmentation branch result and an edge detection result; and input the decoding results of the segmentation branch decoder and the edge detection branch decoder into the fusion branch decoder of the preset image segmentation model, to obtain a fusion segmentation result.

[0039] Exemplarily, after passing through the shared encoder, the output result of the shared encoder is decoded as the output data of the segmentation branch decoder and the edge detection branch decoder. The segmentation branch decoder and the edge detection branch decoder are the same in structure, the same in number of layers, and symmetrical with the shared encoder, and the output data of each level of the shared encoder is used as the input data in the corresponding segmentation branch decoder and edge detection branch decoder. The segmentation branch decoder and the edge detection branch decoder are sequentially connected by four decoder layers, and each decoder layer is serially connected by an upsampling block and a convolution block. The upsampling block is sequentially connected by a 2-fold upsampling layer and a convolution layer, and the convolution block is the same in structure as the convolution block in the shared encoder. The features of the same layer encoder layer and the features of the same layer decoder layer are connected through a skip-connection (residual connection), and the convolution kernel size of the convolution layer is 3x3. Compared with using a large convolution kernel alone, this series mode of multiple small convolution kernels expands the receptive field while ensuring smaller parameter quantity and less calculation quantity, and introduces more activation parameters to improve the nonlinear fitting ability of the network. The encoder uses downsampling, and the decoder uses upsampling, and the features in the downsampling and upsampling are connected through a skip-connection, so that each layer of feature map extracted by the network can be more fully utilized.

[0040] The decoding output result of the segmentation branch decoder and the edge detection branch decoder is fused and decoded as the input data of the fusion branch decoder, and the number of layers of the fusion branch decoder is one less than that of the shared encoder, the segmentation branch decoder and the edge detection branch decoder. The fusion branch decoder is sequentially connected by three decoder modules, and each fusion module is composed of a convolution block and an upsampling layer. The convolution block is sequentially connected by two same subunits. Each subunit includes a convolution layer, a batch normalization layer, a Dropout layer and an activation function layer connected in sequence, which reduces the possibility of overfitting. The size of the convolution kernel in the convolution layer is 3x3, and the activation function in the embodiment of the application is uniformly LeakyRelu function.

[0041] Step 104: obtaining the output result of the preset image segmentation model based on the segmentation branch result, the edge detection result and the fusion segmentation result. Exemplarily, after passing through the segmentation branch decoder, the edge detection branch decoder and the fusion branch decoder in the decoder, the segmentation branch result, the edge detection result and the fusion segmentation result are obtained as three segmentation results of the corresponding sample image, and the three segmentation results are used as the output result of the image segmentation model.

[0042] Step 105: obtaining the loss function of the preset image segmentation model based on the positive sample image and the output result. Illustratively, according to each segmentation result in the above step 104, the loss functions of the segmentation branch decoder, the edge detection branch decoder and the fusion branch decoder are respectively calculated according to the marked regions in the positive sample label, and the loss function of the image segmentation model is a linear combination of the loss functions of the segmentation branch decoder, the edge detection branch decoder and the fusion branch decoder.

[0043] Step 106: adjusting the parameters of the preset image segmentation model based on the loss function to obtain a multi-branch fusion segmentation model. Illustratively, the weights and biases in the image segmentation model are updated according to the loss function calculated in the above step 105, and the iteration is repeated until the last loss function no longer decreases, then the weights and biases in the last image segmentation model are retained, and the corresponding decoder with the minimum loss function is taken as the finally trained model.

[0044] The construction method of the multi-branch fusion segmentation model provided by the application comprises the following steps: obtaining a sample image; the sample image comprises a positive sample image marked with a target region and a negative sample image without marking; inputting the negative sample image as input data into a shared encoder of a preset image segmentation model to perform encoding processing and obtain an encoding result; inputting the encoding result into a segmentation branch decoder and an edge detection branch decoder of the preset image segmentation model respectively to obtain a segmentation branch result and an edge detection result; inputting the decoding results of the segmentation branch decoder and the edge detection branch decoder into a fusion branch decoder of the preset image segmentation model to obtain a fusion segmentation result; obtaining an output result of the preset image segmentation model based on the segmentation branch result, the edge detection result and the fusion segmentation result; obtaining a loss function of the preset image segmentation model based on the positive sample image and the output result; and adjusting the parameters of the preset image segmentation model based on the loss function to obtain a multi-branch fusion segmentation model. After the sample image is processed by the shared encoder, the segmentation branch decoder, the edge detection branch decoder and the fusion branch decoder, different segmentation results are obtained, which improves the accuracy of model recognition. Based on the edge detection branch decoder, the learning ability and the generalization ability are enhanced, through the fusion branch decoder, the image features learned by each branch task are fully utilized, and the accuracy of network segmentation is further improved.

[0045] As an optional embodiment of the present application, the multi-branch fusion segmentation model comprises an n-level shared encoder and a multi-branch decoder, the multi-branch decoder comprising an n-level segmentation branch decoder, an n-level edge detection branch decoder and an n-1-level fusion branch decoder, and the process of inputting the encoding result into the segmentation branch decoder and the edge detection branch decoder of the preset image segmentation model respectively to obtain the segmentation branch result and the edge detection result in step 103 mainly comprises: inputting the sample image into the first-level shared encoder to obtain first encoding output data; inputting the first encoding output data into the second-level shared encoder, the first-level segmentation branch decoder and the first-level edge detection branch decoder to obtain first segmentation data and first edge detection data; inputting the output data of the m-1-level shared encoder into the m-level shared encoder to obtain the output data of the m-level shared encoder, 2≤m≤n; inputting the output data of the m-level shared encoder and the output data of the m-1-level segmentation branch decoder into the m-level segmentation branch decoder to obtain the output data of the m-level segmentation branch decoder, 2≤m≤n; inputting the output data of the m-level shared encoder and the output data of the m-1-level edge detection branch decoder into the m-level edge detection branch decoder to obtain the output data of the m-level edge detection branch decoder, 2≤m≤n; obtaining the segmentation branch result based on the first segmentation data and the output data of the m-level segmentation branch decoder, and obtaining the edge detection result based on the first edge detection data and the output data of the m-level edge detection branch decoder.

[0046] As an optional embodiment of the present application, the process of inputting the decoding results of the segmentation branch decoder and the edge detection branch decoder into the fusion branch decoder of the preset image segmentation model to obtain the fusion segmentation result in step 103 comprises: inputting the output data of the first-level segmentation branch decoder and the output data of the first-level edge detection branch decoder into the first-level fusion branch decoder to obtain the output data of the first-level fusion branch decoder; inputting the output data of the m-level segmentation branch decoder, the output data of the m-level edge detection branch decoder and the output data of the m-1-level fusion branch decoder into the m-level fusion branch decoder to obtain the output data of the m-level fusion branch decoder; obtaining the fusion segmentation result based on the output data of the first-level fusion branch decoder and the output data of the m-level fusion branch decoder.

[0047] Exemplarily, as Figure 7The structure diagram of the network model of the 4-level shared encoder, 4-level segmentation branch decoder, 4-level edge detection branch decoder and 3-level fusion branch decoder is shown. The output of the first-level shared encoder is input into the first-level segmentation branch decoder and the first-level edge detection branch decoder, respectively. Similarly, the output of the second, third and fourth-level shared encoders is input into the second, third and fourth-level segmentation branch decoders and the second, third and fourth-level edge detection branch decoders, respectively. Except for the first level, the input data of each level of the segmentation branch decoder and the edge detection branch decoder has the output data of the corresponding upper level of the segmentation branch decoder and the edge detection branch decoder in addition to the output of the corresponding level of the shared encoder.

[0048] In the fusion branch decoder, the input of the first-level fusion branch decoder is the output data of the first-level segmentation branch decoder and the first-level edge detection branch decoder. Except for the first-level fusion branch decoder, the input of each level of the fusion branch decoder has the output data of the upper level of the fusion branch decoder in addition to the output data of the corresponding segmentation branch decoder and the edge detection branch decoder as the input data. Through the characteristics of the fusion segmentation branch decoder and the edge detection branch decoder, the learned features of the segmentation branch task and the edge detection branch task can be fully utilized.

[0049] As an optional embodiment of the present application, the step 105 comprises: calculating a segmentation branch loss function based on the positive sample image and the segmentation branch result; calculating an edge detection loss function based on the positive sample image and the edge detection result; calculating a fusion segmentation loss function based on the positive sample image and the fusion segmentation result; and obtaining the loss function based on the segmentation branch loss function, the edge detection loss function and the fusion segmentation loss function.

[0050] For example, the positive sample image is a sample image that has been accurately labeled for the intima and the media-adventitia of the intravascular ultrasound image. According to the comparison between the recognition results of each decoder and the positive sample image, the segmentation branch loss function is L Seg , the edge detection branch loss function is L Edge , and the fusion branch loss function is L Fus , and the total loss function of the segmentation model is:

[0051] L=αL Seg +βL Edge +L Fus

[0052] α and β are hyperparameters, which are set according to the training condition. The segmentation branch and the edge detection branch are auxiliary branches, and α, β∈[0,1].

[0053]

[0054] wherein N is the number of pixels, C is the number of segmentation label categories, is the probability that the nth pixel in the manually annotated image belongs to the cth label, is the probability that the nth pixel in the network prediction result belongs to the cth label.

[0055]

[0056] wherein N is the number of pixels, y (i) is the probability that the nth pixel in the manually annotated image belongs to the edge pixel, is the probability that the nth pixel in the network prediction result belongs to the edge pixel.

[0057] The loss function is back-propagated, and the weights and biases of the network model are updated. The iteration is repeated until the loss function does not decrease. The linear combination manner of the loss function in the embodiments of the present application is not limited, and those skilled in the art can determine it according to actual needs.

[0058] When the loss function reaches the minimum, a preliminary network model is obtained. The output results of the fusion branch are used as the preliminary segmentation results of the intima and the media-adventitia in the sample image. The Jaccard similarity coefficient, the Hausdorff distance and the area difference percentage are used as the segmentation result evaluation indexes. The preliminary segmentation results output by the fusion branch of each network model are evaluated. The best network model is comprehensively evaluated as the network model applied to the system.

[0059] The Jaccard similarity coefficient formula is as follows:

[0060]

[0061] R pred represents the blood vessel structure region in the prediction result, R true represents the blood vessel structure region in the manually annotated image.

[0062] The Hausdorff distance formula is as follows:

[0063]

[0064] wherein C pred represents the edge of the blood vessel structure region in the prediction result, C true represents the edge of the blood vessel structure region in the manually annotated image, and a and b are points on C pred and C true respectively. d(a, b) represents the Euclidean distance between a and b.

[0065] The area difference percentage PAD is calculated as follows:

[0066]

[0067] A represents the area of the blood vessel structure region in the prediction result, A pred A represents the area of the blood vessel structure region in the artificial marking image. true A represents the area of the blood vessel structure region in the artificial marking image.

[0068] As an optional embodiment of the present application, in step 101, the sample image is obtained by: pre-processing the sample image, marking the pre-processed sample image, and taking the marked sample image as the positive sample image.

[0069] As an example, because the sample image directly collected in the clinic is limited by the collection operation method and the collection instrument, etc., the sample image obtained may not be directly used for training of the model, and the sample image can be pre-processed at this time, where the pre-processing manner can be screening, uniform resolution, and uniform gray value operation, and the resolution size and the gray value range can be set according to specific requirements and hardware device limitations. In this embodiment, the uniform resolution used is 256x256, and the gray value range is [0, 1]. The present application does not limit the category and pre-processing manner of the sample image, and a person skilled in the art can determine it according to actual needs.

[0070] As an optional embodiment of the present application, in step 101, the sample image is obtained by: pre-processing the sample image, marking the pre-processed sample image, and taking the marked sample image as the positive sample image.

[0071] An ultrasonic imaging contour recognition method is disclosed in the present application, as shown in Figure 1 The method comprises the following steps:

[0072] Step 201: obtaining an original image. As an example, the original image is an intravascular ultrasound image obtained in the clinic, and a specific original image is shown in Figure 8

[0073] ​Step 202: inputting the original image into a preset multi-branch fusion segmentation model to obtain an image segmentation result of the original image; the preset multi-branch fusion segmentation model is obtained by using the construction method of the multi-branch fusion segmentation model in any of the above embodiments. Illustratively, the obtained original image is input into the multi-branch fusion segmentation model trained by the method embodiment to obtain a segmentation result corresponding to the original image.

[0074] Step 203: performing contour extraction based on the image segmentation result to obtain a regional contour recognition result of the original image. Illustratively, the segmentation result obtained in step 203 is actually a probability map predicted pixel by pixel, and there may be problems of non-unique intima and mid-extracardiac membrane regions and unclear edges, and there is no specific contour line, so it is necessary to clearly extract the corresponding contour according to the contour recognition result to obtain a more clear and accurate regional contour recognition result.

[0075] The ultrasonic imaging contour recognition method provided by the application comprises the following steps: obtaining an original image; inputting the original image into a preset multi-branch fusion segmentation model to obtain an image segmentation result of the original image; the preset multi-branch fusion segmentation model is obtained by using the construction method of the multi-branch fusion segmentation model in any of the above embodiments; and performing contour extraction based on the image segmentation result to obtain a regional contour recognition result of the original image. The sample image is segmented and recognized by the preset multi-branch fusion segmentation model, the accuracy of recognition is improved, contour extraction is performed again based on the recognition result, and a more clear and accurate regional contour recognition result is obtained further based on the segmentation result.

[0076] As an optional embodiment of the application, step 203 comprises the following steps: performing contour extraction based on the image segmentation result to obtain a first image result; performing Gaussian blur processing on the first image result to obtain a second image result; performing binaryzation processing on the second image result to obtain a third image result; and performing region of interest selection processing on the third image result to obtain a final regional contour recognition result.

[0077] Illustratively, the segmentation result obtained based on the multi-branch fusion segmentation model has problems of non-unique regions and unclear edges, and a part of isolated pixels and regions are removed by using Gaussian blur, and the filter kernel size can be 7x7; binaryzation means that the intima and mid-extracardiac membrane preliminary segmentation result is determined to have clear regional edges according to an automatically calculated or manually set threshold value, and the gray value is set to 0 or 255, wherein the gray value 255 represents the intima or mid-extracardiac membrane region, and the gray value 0 represents the background. The filter kernel size and the threshold value size in binaryzation are not limited in the embodiment of the application, and can be determined according to actual needs by those skilled in the art.

[0078] As an optional embodiment of the present application, in step 203, the third image result is subjected to a region of interest selection process to obtain a final region contour recognition result, comprising: determining a maximum connected domain containing a preset gray value of a center point of the third image result as a region of interest based on the third image result; obtaining a contour point set based on the intersection set of a plurality of rays at multiple angles emitted from the image center of the third image result and the region of interest; sequentially connecting the curve fitting result passing through the contour point set to obtain the final region contour recognition result.

[0079] Illustratively, after obtaining clear region edges, it is necessary to mark and divide the region edges. The corresponding division method adopts a maximum connected domain containing a gray value of 255 of a center point of an ROI (Region Of Interest, ROI for short), solves the problem of non-unique intima and media-adventitia regions, and obtains smooth intima and media-adventitia contours in line with human visual habits. The number of angles n can be set according to the size of the image resolution. In the embodiment of the present application, n = 36, as shown in a completed image segmentation result after marking. Figure 9

[0080] The present application also discloses a construction device of a multi-branch fusion segmentation model, as shown in Figure 3 The device comprises:

[0081] A first acquisition module 301 is configured to acquire a sample image. The sample image comprises a positive sample image marked with a target region and a negative sample image without marking. Illustratively, the detailed content can be found in the content of step 101 in the above method embodiment, which will not be repeated here.

[0082] An encoding module 302 is configured to input the negative sample image as input data into a shared encoder of a preset image segmentation model to perform encoding processing and obtain an encoding result. Illustratively, the detailed content can be found in the content of step 102 in the above method embodiment, which will not be repeated here.

[0083] A decoding module 303 is configured to input the encoding result into a segmentation branch decoder and an edge detection branch decoder of the preset image segmentation model respectively to obtain a segmentation branch result and an edge detection result; and input the decoding results of the segmentation branch decoder and the edge detection branch decoder into a fusion branch decoder of the preset image segmentation model to obtain a fusion segmentation result. Illustratively, the detailed content can be found in the content of step 103 in the above method embodiment, which will not be repeated here.

[0084] ​Output module 304 is used to obtain the output result of the preset image segmentation model based on the segmentation branch result, edge detection result, and fusion segmentation result. For example, details can be found in step 104 of the above method embodiment, and will not be repeated here.

[0085] The loss function module 305 is used to obtain the loss function of the preset image segmentation model based on the positive sample images and the output results. For example, details can be found in step 105 of the above method embodiment, and will not be repeated here.

[0086] The adjustment module 306 is used to adjust the parameters of the preset image segmentation model based on the loss function to obtain a multi-branch fusion segmentation model. For example, details can be found in step 106 of the above method embodiment, and will not be repeated here.

[0087] The present invention provides a device for constructing a multi-branch fusion segmentation model. The device includes: a first acquisition module 301 for acquiring sample images; the sample images include positive sample images labeled with target regions and unlabeled negative sample images; an encoding module 302 for inputting the negative sample images as input data into a shared encoder of a preset image segmentation model for encoding processing to obtain an encoding result; a decoding module 303 for inputting the encoding result into a segmentation branch decoder and an edge detection branch decoder of the preset image segmentation model respectively to obtain segmentation branch results and edge detection results; and inputting the decoding results of the segmentation branch decoder and the edge detection branch decoder into a fusion branch decoder of the preset image segmentation model to obtain a fusion segmentation result; an output module 304 for obtaining the output result of the preset image segmentation model based on the segmentation branch results, edge detection results, and fusion segmentation result; a loss function module 305 for obtaining the loss function of the preset image segmentation model based on the positive sample images and the output result; and an adjustment module 306 for adjusting the parameters of the preset image segmentation model based on the loss function to obtain a multi-branch fusion segmentation model. The sample images are processed by a shared encoder, followed by segmentation branch decoders, edge detection branch decoders, and fusion branch decoders, resulting in different segmentation outcomes and improving the model's recognition accuracy. The edge detection branch decoder enhances learning and generalization capabilities, while the fusion branch decoder fully utilizes image features learned from each branch task, further improving network segmentation accuracy.

[0088] As an optional embodiment of the present application, the multi-branch fusion segmentation model comprises an n-level shared encoder and a multi-branch decoder, the multi-branch decoder comprises an n-level segmentation branch decoder, an n-level edge detection branch decoder and an n-1-level fusion branch decoder, in the decoding module 303, the encoding results are respectively input into the segmentation branch decoder and the edge detection branch decoder of the preset image segmentation model to obtain segmentation branch results and edge detection results, comprising: inputting the sample image into the first-level shared encoder to obtain first encoding output data; inputting the first encoding output data into the second-level shared encoder, the first-level segmentation branch decoder and the first-level edge detection branch decoder to obtain first segmentation data and first edge detection data; inputting the output data of the m-1-level shared encoder into the m-level shared encoder to obtain the output data of the m-level shared encoder, 2≤m≤n; inputting the output data of the m-level shared encoder and the output data of the m-1-level segmentation branch decoder into the m-level segmentation branch decoder to obtain the output data of the m-level segmentation branch decoder, 2≤m≤n; inputting the output data of the m-level shared encoder and the output data of the m-1-level edge detection branch decoder into the m-level edge detection branch decoder to obtain the output data of the m-level edge detection branch decoder, 2≤m≤n; obtaining the segmentation branch results based on the first segmentation data and the output data of the m-level segmentation branch decoder, and obtaining the edge detection results based on the first edge detection data and the output data of the m-level edge detection branch decoder. Exemplarily, the details are shown in the content of step 103 in the above method embodiment.

[0089] As an optional embodiment of the present application, in the decoding module 303, the decoding results of the segmentation branch decoder and the edge detection branch decoder are input into the fusion branch decoder of the preset image segmentation model to obtain fusion segmentation results, comprising: inputting the output data of the first-level segmentation branch decoder and the output data of the first-level edge detection branch decoder into the first-level fusion branch decoder to obtain the output data of the first-level fusion branch decoder; inputting the output data of the m-level segmentation branch decoder, the output data of the m-level edge detection branch decoder and the output data of the m-1-level fusion branch decoder into the m-level fusion branch decoder to obtain the output data of the m-level fusion branch decoder; obtaining the fusion segmentation results based on the output data of the first-level fusion branch decoder and the output data of the m-level fusion branch decoder. Exemplarily, the details are shown in the content of step 103 in the above method embodiment.

[0090] As an optional embodiment of the present application, the loss function module 305 comprises: a first loss function sub-module, configured to calculate a segmentation branch loss function based on the positive sample image and the segmentation branch result; a second loss function sub-module, configured to calculate an edge detection loss function based on the positive sample image and the edge detection result; a third loss function sub-module, configured to calculate a fusion segmentation loss function based on the positive sample image and the fusion segmentation result; and a fourth loss function sub-module, configured to obtain the loss function based on the segmentation branch loss function, the edge detection loss function and the fusion segmentation loss function. For example, the details are as shown in the step 105 of the method embodiment.

[0091] As an optional embodiment of the present application, in the first acquisition module 301, the sample image is acquired by: a preprocessing module, configured to pre-process the sample image, and label the pre-processed sample image, and take the labeled sample image as the positive sample image. For example, the details are as shown in the step 101 of the method embodiment.

[0092] As an optional embodiment of the present application, in the first acquisition module 301, the sample image is acquired by: an expansion module, configured to perform expansion processing on the sample image to obtain an expanded sample image. For example, the details are as shown in the step 101 of the method embodiment.

[0093] The present application also discloses an ultrasonic imaging contour recognition device, as shown in the figure, the device comprises: Figure 4

[0094] The second acquisition module 401 is configured to acquire an original image. For example, the details are as shown in the step 201 of the method embodiment, which will not be repeated here.

[0095] The segmentation module 402 is configured to input the original image into a preset multi-branch fusion segmentation model to obtain an image segmentation result of the original image. The preset multi-branch fusion segmentation model is obtained by using the construction method of the multi-branch fusion segmentation model described in the above embodiment. For example, the details are as shown in the step 202 of the method embodiment, which will not be repeated here.

[0096] The contour extraction module 403 is configured to perform contour extraction based on the image segmentation result to obtain a contour recognition result of each region of the original image. For example, the details are as shown in the step 203 of the method embodiment, which will not be repeated here.

[0097] ​The application provides a multi-branch fusion segmentation model construction device, which comprises a second acquisition module 401, which is used for acquiring an original image; a segmentation module 402, which is used for inputting the original image into a preset multi-branch fusion segmentation model to obtain an image segmentation result of the original image; the preset multi-branch fusion segmentation model is obtained by using the multi-branch fusion segmentation model construction method in the above embodiment; and a contour extraction module 403, which is used for performing contour extraction based on the image segmentation result to obtain a region contour recognition result of the original image. The sample image is segmented by the preset multi-branch fusion segmentation model to identify the segmentation result, the accuracy of identification is improved, contour extraction is performed again based on the identification result, and on the basis of the segmentation result, a more clear and accurate region contour recognition result is obtained.

[0098] As an optional embodiment of the application, the contour extraction module 403 comprises: a first contour extraction module, which is used for performing contour extraction based on the image segmentation result to obtain a first image result; a Gaussian blur module, which is used for performing Gaussian blur processing on the first image result to obtain a second image result; a binaryzation module, which is used for performing binaryzation processing on the second image result to obtain a third image result; and a region selection module, which is used for performing region of interest selection processing on the third image result to obtain a final region contour recognition result. Exemplarily, the details are shown in the content of step 203 in the method embodiment, and details are not described herein again.

[0099] As an optional embodiment of the application, in the contour extraction module 403, the region selection module comprises: a determination region module, which is used for determining a maximum connected domain with a gray value of 255 of an image center point of the third image result as a region of interest based on the third image result; a contour point set module, which is used for obtaining a contour point set based on the intersection set of n rays with uniform intervals in n angles emitted from the image center of the third image result and the region of interest; and a connection module, which is used for sequentially connecting the curve fitting result through the contour point set to obtain the final region contour recognition result. Exemplarily, the details are shown in the content of step 203 in the method embodiment, and details are not described herein again.

[0100] The application further provides a computer device, as shown in the figure, which can comprise a processor 501 and a memory 502, wherein the processor 501 and the memory 502 can be connected through a bus or other means, Figure 5 for example, through a bus connection. Figure 5

[0101] ​Processor 501 can be a central processing unit (CPU). Processor 501 can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or combinations of the above types of chips.

[0102] The memory 502, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the multi-branch fusion segmentation model construction method and the ultrasound imaging contour recognition method in the embodiments of the present invention. The processor 501 executes various functional applications and data processing by running the non-transitory software programs, instructions, and modules stored in the memory 502, thereby implementing the multi-branch fusion segmentation model construction method and the ultrasound imaging contour recognition method in the above method embodiments.

[0103] Memory 502 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the processor 501, etc. Furthermore, memory 502 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory 502 may optionally include memory remotely located relative to processor 501, and these remote memories may be connected to processor 501 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0104] The one or more modules are stored in the memory 502, and when executed by the processor 501, they perform actions such as... Figure 1 or Figure 2 The embodiments shown illustrate a method for constructing a multi-branch fusion segmentation model and a method for ultrasound imaging contour recognition.

[0105] For specific details regarding the aforementioned computer equipment, please refer to the relevant documentation. Figure 1 or Figure 2 The relevant descriptions and effects in the illustrated embodiments are for understanding purposes only and will not be repeated here.

[0106] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The program can be stored in a computer readable storage medium. When the program is executed, the program can include the processes of the above-mentioned embodiment methods. The storage medium can be a magnetic disc, an optical disc, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD), etc. The storage medium can also include a combination of the above-mentioned types of memories.

[0107] Although the embodiments of the present application are described in conjunction with the drawings, various modifications and changes can be made by those skilled in the art without departing from the spirit and scope of the present application, and such modifications and changes fall within the scope defined by the appended claims.

Claims

1. A method for constructing a multi-branch fusion segmentation model, characterized in that, The method comprises the following steps: obtaining sample images; the sample images include positive sample images marked with target regions and negative sample images without marking; inputting the negative sample images as input data into a shared encoder of a preset image segmentation model for encoding processing to obtain an encoding result; inputting the encoding result into a segmentation branch decoder and an edge detection branch decoder of the preset image segmentation model respectively to obtain a segmentation branch result and an edge detection result; inputting the decoding results of the segmentation branch decoder and the edge detection branch decoder into a fusion branch decoder of the preset image segmentation model to obtain a fusion segmentation result; obtaining an output result of the preset image segmentation model based on the segmentation branch result, the edge detection result and the fusion segmentation result; obtaining a loss function of the preset image segmentation model based on the positive sample images and the output result; adjusting parameters of the preset image segmentation model based on the loss function to obtain a multi-branch fusion segmentation model; the multi-branch fusion segmentation model comprises an n-level shared encoder and a multi-branch decoder, and the multi-branch decoder comprises an n-level segmentation branch decoder, an n-level edge detection branch decoder and an n-1-level fusion branch decoder; inputting the encoding result into the segmentation branch decoder and the edge detection branch decoder of the preset image segmentation model respectively to obtain the segmentation branch result and the edge detection result, which comprises the following steps: inputting the sample images into the first-level shared encoder to obtain first encoding output data; inputting the first encoding output data into a second-level shared encoder, a first-level segmentation branch decoder and a first-level edge detection branch decoder to obtain first segmentation data and first edge detection data; inputting output data of the m-1-level shared encoder into the m-level shared encoder to obtain output data of the m-level shared encoder, 2≤m≤n; inputting output data of the m-level shared encoder and output data of the m-1-level segmentation branch decoder into the m-level segmentation branch decoder to obtain output data of the m-level segmentation branch decoder, 2≤m≤n; inputting output data of the m-level shared encoder and output data of the m-1-level edge detection branch decoder into the m-level edge detection branch decoder to obtain output data of the m-level edge detection branch decoder, 2≤m≤n; obtaining the segmentation branch result based on the first segmentation data and the output data of the m-level segmentation branch decoder, and obtaining the edge detection result based on the first edge detection data and the output data of the m-level edge detection branch decoder.

2. The method of claim 1, wherein, inputting the decoding results of the segmentation branch decoder and the edge detection branch decoder into the fusion branch decoder of the preset image segmentation model to obtain the fusion segmentation result, which comprises the following steps: inputting output data of the first-level segmentation branch decoder and output data of the first-level edge detection branch decoder into the first-level fusion branch decoder to obtain output data of the first-level fusion branch decoder; inputting the output data of the mth segmentation branch decoder, the output data of the mth edge detection branch decoder and the output data of the (m-1)th fusion branch decoder into an mth fusion branch decoder to obtain output data of the mth fusion branch decoder; obtaining the fusion segmentation result based on the output data of the first fusion branch decoder and the output data of the mth fusion branch decoder.

3. The method of claim 1, wherein, The loss function of the preset image segmentation model is obtained based on the positive sample image and the output result, and the loss function comprises: a segmentation branch loss function is calculated based on the positive sample image and the segmentation branch result; an edge detection loss function is calculated based on the positive sample image and the edge detection result; a fusion segmentation loss function is calculated based on the positive sample image and the fusion segmentation result; the loss function is obtained based on the segmentation branch loss function, the edge detection loss function and the fusion segmentation loss function.

4. The method of claim 1, wherein, The sample image is obtained, which comprises: The sample image is preprocessed, and the preprocessed sample image is labeled, and the labeled sample image is used as the positive sample image.

5. The method of claim 4, wherein, The sample image is also subjected to augmentation processing to obtain an augmented sample image.

6. An ultrasound imaging contour identification method, characterized by, It comprises: an original image is obtained; the original image is input into a preset multi-branch fusion segmentation model to obtain an image segmentation result of the original image; the preset multi-branch fusion segmentation model is obtained by using the construction method of the multi-branch fusion segmentation model according to any one of claims 1-5; contour extraction is performed based on the image segmentation result to obtain a region contour recognition result of the original image.

7. The method of claim 6, wherein, The contour extraction is performed based on the image segmentation result to obtain the region contour recognition result of the original image, which comprises: a first image result is obtained by performing contour extraction based on the image segmentation result; a second image result is obtained by performing Gaussian blur processing on the first image result; a third image result is obtained by performing binarization processing on the second image result; a final region contour recognition result is obtained by performing region of interest selection processing on the third image result.

8. The method of claim 7, wherein, The region of interest selection processing is performed on the third image result to obtain the final region contour recognition result, which comprises: a maximum connected domain containing a preset gray value of an image center point of the third image result is determined as a region of interest based on the third image result; a contour point set is obtained based on the intersection set of a plurality of rays emitted at a plurality of angles at equal intervals from the image center of the third image result and the region of interest; the final region contour recognition result is obtained by sequentially connecting the curve fitting result of the contour point set. 9.A device for constructing a multi-branch fusion segmentation model, characterized in that, It comprises: a first acquisition module is configured to acquire a sample image; the sample image comprises a positive sample image marked with a target region and a negative sample image without marking; an encoding module is configured to input the negative sample image as input data into a shared encoder of a preset image segmentation model to perform encoding processing and obtain an encoding result. The decoding module is configured to input the encoding result into a segmentation branch decoder and an edge detection branch decoder of the preset image segmentation model respectively to obtain a segmentation branch result and an edge detection result; The decoding result of the segmentation branch decoder and the edge detection branch decoder is input into a fusion branch decoder of the preset image segmentation model to obtain a fusion segmentation result; The output module is configured to obtain an output result of the preset image segmentation model based on the segmentation branch result, the edge detection result and the fusion segmentation result; The loss function module is configured to obtain a loss function of the preset image segmentation model based on the positive sample image and the output result; The adjusting module is configured to adjust parameters of the preset image segmentation model based on the loss function to obtain a multi-branch fusion segmentation model; the multi-branch fusion segmentation model comprises an n-level shared encoder and a multi-branch decoder, and the multi-branch decoder comprises an n-level segmentation branch decoder, an n-level edge detection branch decoder and an n-1-level fusion branch decoder; The decoding module is further configured to input the sample image into the first-level shared encoder to obtain first encoding output data; input the first encoding output data into a second-level shared encoder, a first-level segmentation branch decoder and a first-level edge detection branch decoder to obtain first segmentation data and first edge detection data; input output data of an m-1-level shared encoder into an m-level shared encoder to obtain output data of the m-level shared encoder, 2≤m≤n; input the output data of the m-level shared encoder and output data of an m-1-level segmentation branch decoder into an m-level segmentation branch decoder to obtain output data of the m-level segmentation branch decoder, 2≤m≤n; input the output data of the m-level shared encoder and output data of an m-1-level edge detection branch decoder into an m-level edge detection branch decoder to obtain output data of the m-level edge detection branch decoder, 2≤m≤n; The segmentation branch result is obtained based on the first segmentation data and the output data of the m-level segmentation branch decoder, and the edge detection result is obtained based on the first edge detection data and the output data of the m-level edge detection branch decoder.

10. An ultrasound imaging profile identification apparatus, characterized by, Comprise: The second acquisition module is configured to acquire an original image; The segmentation module is configured to input the original image into a preset multi-branch fusion segmentation model to obtain an image segmentation result of the original image; the preset multi-branch fusion segmentation model is obtained by using the construction method of the multi-branch fusion segmentation model in any one of claims 1-5; The contour extraction module is configured to perform contour extraction based on the image segmentation result to obtain a region contour recognition result of the original image.

11. A computer device, comprising: Comprise: At least one processor; and a memory connected with the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to cause the at least one processor to perform the steps of the method for constructing a multi-branch fusion segmentation model according to any one of claims 1-5 or the method for identifying an ultrasound imaging contour according to any one of claims 6-8.

12. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method for constructing a multi-branch fusion segmentation model according to any one of claims 1-5 or the method for identifying an ultrasound imaging contour according to any one of claims 6-8.

Images