Ultrasonic image segmentation method and device

Abstract

The present invention relates to the field of computer vision, and provides an ultrasonic image segmentation method and device, the method comprising: acquiring an ultrasonic image; inputting the ultrasonic image into an image segmentation model, and obtaining an ultrasonic image segmentation result output by the image segmentation model; wherein the ultrasonic image segmentation result is obtained by training based on an ultrasonic training image and a segmentation label corresponding to the ultrasonic training image; the image segmentation model is used to extract features of the ultrasonic image, and perform spatial and channel reconstruction convolution on the extracted ultrasonic features, and use a labeled multi-layer perception mechanism to obtain an ultrasonic image segmentation result. The present invention obtains better performance by reducing redundant features, and significantly reduces complexity and computational cost, thereby improving the segmentation effect of ultrasonic images.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and in particular to an ultrasonic image segmentation method and device. Background Art

[0002] With the development and progress of China's medical level, ultrasound imaging has become the fastest and most economical ultrasound image screening technology. However, with the increasing number of patients, medical workers need to spend a lot of manpower and time on ultrasound image diagnosis, among which the most time-consuming and labor-intensive is ultrasound image segmentation.

[0003] At present, the commonly used ultrasound image segmentation mainly uses Convolutional Neural Networks (CNN) to capture the nonlinear mapping between input and output, and automatically learns local features and high-level abstract features through a multi-layer network structure to achieve ultrasound image segmentation.

[0004] However, due to the problems of large differences between training samples and small perceptual regions (ROIs) in medical images, and the inability of CNN to reasonably propagate low-level features to high-level features, the segmentation effect is poor and it is difficult to balance the precision and recall rates. Therefore, more accurate segmentation of ultrasound images is still a difficult problem that needs to be solved urgently. Summary of the invention

[0005] The present invention provides an ultrasonic image segmentation method and device to solve the defect of poor image segmentation effect of convolutional neural network in the prior art, obtain better performance by reducing redundant features, and significantly reduce complexity and calculation cost, thereby improving the segmentation effect of ultrasonic images.

[0006] The present invention provides an ultrasonic image segmentation method, comprising: acquiring an ultrasonic image; inputting the ultrasonic image into an image segmentation model to obtain an ultrasonic image segmentation result output by the image segmentation model; wherein the ultrasonic image segmentation result is obtained by training based on an ultrasonic training image and a segmentation label corresponding to the ultrasonic training image; the image segmentation model is used to extract features of the ultrasonic image, and perform spatial and channel reconstruction convolution on the extracted ultrasonic features, and utilize a labeled multi-layer perception mechanism to obtain an ultrasonic image segmentation result.

[0007] According to an ultrasonic image segmentation method provided by the present invention, the image segmentation model includes a feature extraction layer, a first plug-and-play layer, a labeled multi-layer perception layer, a second plug-and-play layer and an image segmentation layer; an ultrasonic image is input into the image segmentation model to obtain an ultrasonic image segmentation result output by the image segmentation model, including: inputting the ultrasonic image into the feature extraction layer to extract features, and obtaining ultrasonic features output by the feature extraction; inputting the ultrasonic features into the first plug-and-play layer to spatially reconstruct the ultrasonic features, and performing channel reconstruction on the first spatial refinement features obtained based on the spatial reconstruction, and obtaining the first channel refinement features output by the first plug-and-play layer; inputting the first channel refinement features into the labeled multi-layer perception layer to utilize the labeled multi-layer perception Mechanism, marking and projecting the first channel refined features to obtain the spatial information features output by the marked multi-layer perception layer; inputting the spatial information features and the first channel refined features into the second plug-and-play layer to perform feature fusion on the spatial information features and the first channel refined features, and spatially reconstructing the first features obtained based on the feature fusion, and performing channel reconstruction on the second spatial refined features obtained based on the spatial reconstruction, to obtain the second channel refined features output by the second plug-and-play layer; inputting the second channel refined features and the ultrasonic features into the image segmentation layer to perform feature fusion on the second channel refined features and the ultrasonic features, and upsampling the second features obtained based on the feature fusion, to obtain the ultrasonic image segmentation result output by the image segmentation layer.

[0008] According to an ultrasonic image segmentation method provided by the present invention, spatial reconstruction of ultrasonic features is performed, including: normalizing the ultrasonic features to obtain a first normalized feature; generating corresponding weights according to the first normalized feature and a first preset nonlinear activation function, and separating the ultrasonic features according to the generated weights to obtain a first sub-feature and a second sub-feature; transforming the first sub-feature at least three times, and transforming the second sub-feature at least three times; performing feature fusion using the first sub-feature after the second transformation and the second sub-feature after the third transformation to obtain a first fused feature; performing feature fusion using the first sub-feature after the third transformation and the second sub-feature after the second transformation to obtain a second fused feature; splicing the first fused feature and the second fused feature to obtain a first spatial refinement feature;

[0009] The first feature obtained based on feature fusion is spatially reconstructed, including: normalizing the first feature to obtain a second normalized feature; generating corresponding weights according to the second normalized feature and a second preset nonlinear activation function, and separating the first feature according to the generated weights to obtain a third sub-feature and a fourth sub-feature; transforming the third sub-feature at least three times, and transforming the fourth sub-feature at least three times; fusing the third sub-feature after the second transformation with the fourth sub-feature after the third transformation to obtain a third fused feature; fusing the third sub-feature after the third transformation with the fourth sub-feature after the second transformation to obtain a fourth fused feature; and splicing the third fused feature with the fourth fused feature to obtain a second spatially refined feature.

[0010] According to an ultrasonic image segmentation method provided by the present invention, channel reconstruction is performed on the first spatial refinement feature obtained based on spatial reconstruction to obtain the first channel refinement feature, including: segmenting the first spatial refinement feature to obtain a first segmentation feature and a second segmentation feature; performing global convolution on the first segmentation feature, and performing point-by-point convolution on the first segmentation feature, and performing feature fusion of the feature obtained by the global convolution and the feature obtained by the point-by-point convolution to obtain a first splicing feature; performing point-by-point convolution on the second segmentation feature, and splicing the feature obtained by the point-by-point convolution with the second segmentation feature to obtain a second splicing feature; respectively pooling the first splicing feature and the second splicing feature to obtain the corresponding first pooling feature and second pooling feature, and splicing the first pooling feature and the second pooling feature to obtain a third splicing feature; performing category probability prediction on the third splicing feature to obtain a probability distribution; performing weighted fusion on the first splicing feature and the probability distribution to obtain a fifth fusion feature; performing weighted fusion on the second splicing feature and the probability to obtain a sixth fusion feature; performing feature fusion on the fifth fusion feature and the sixth fusion feature to obtain the first channel refinement feature;

[0011] Channel reconstruction is performed on the second spatial refinement feature obtained based on the spatial reconstruction to obtain the second channel refinement feature, including: segmenting the second spatial refinement feature to obtain a third segmentation feature and a fourth segmentation feature; performing global convolution on the third segmentation feature, and performing point-by-point convolution on the third segmentation feature, and fusing the feature obtained by the global convolution with the feature obtained by the point-by-point convolution to obtain a fourth splicing feature; performing point-by-point convolution on the fourth segmentation feature, and splicing the feature obtained by the point-by-point convolution with the fourth segmentation feature to obtain a fifth splicing feature; pooling the fourth splicing feature and the fifth splicing feature respectively to obtain the corresponding third pooling feature and fourth pooling feature, and splicing the first pooling feature and the second pooling feature to obtain a sixth splicing feature; performing category probability prediction on the sixth splicing feature to obtain a probability distribution; weighted fusion of the fourth splicing feature and the probability distribution to obtain a seventh fusion feature; weighted fusion of the fifth splicing feature and the probability respectively to obtain an eighth fusion feature; and feature fusion of the seventh fusion feature and the eighth fusion feature to obtain the second channel refinement feature.

[0012] According to an ultrasonic image segmentation method provided by the present invention, a first channel refinement feature is marked and projected by a marked multi-layer perception mechanism to obtain a spatial information feature, including: moving the channel axis of the first channel refinement feature along the channel dimension of the first channel refinement feature, and dividing the moved first channel refinement feature into multiple marks, so as to capture the local information and global information between the segmented marks by using the multi-layer perception mechanism to obtain a first spatial feature; processing the first spatial feature element by element by using a preset activation function, and encoding the processed feature to obtain a coded feature; moving the channel axis of the coded feature along the height dimension of the coded feature, and dividing the moved coded feature into multiple marks, so as to capture the local information and global information between the segmented marks by using the multi-layer perception mechanism to obtain a second spatial feature; normalizing and re-projecting the second spatial feature to obtain a re-projection feature; and fusing the re-projection feature with the first channel refinement feature to obtain a spatial information feature.

[0013] According to an ultrasound image segmentation method provided by the present invention, before inputting the ultrasound image into an image segmentation model, the method includes: obtaining an ultrasound training image and a segmentation label corresponding to the ultrasound training image; using the ultrasound training image as input data for training, using the segmentation label corresponding to the ultrasound training image as a label for training, training the model to be trained, and obtaining an image segmentation model for segmenting ultrasound images.

[0014] The present invention also provides an ultrasonic image segmentation device, comprising: an image acquisition module, which acquires an ultrasonic image; an image segmentation module, which inputs the ultrasonic image into an image segmentation model to obtain an ultrasonic image segmentation result output by the image segmentation model; wherein the ultrasonic image segmentation result is obtained by training based on an ultrasonic training image and a segmentation label corresponding to the ultrasonic training image; the image segmentation model is used to extract features of the ultrasonic image, perform spatial and channel reconstruction convolution on the extracted ultrasonic features, and utilize a labeled multi-layer perception mechanism to obtain an ultrasonic image segmentation result.

[0015] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any one of the ultrasound image segmentation methods described above is implemented.

[0016] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the ultrasound image segmentation method described in any one of the above is implemented.

[0017] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any one of the above-mentioned ultrasound image segmentation methods.

[0018] The ultrasonic image segmentation method and device provided by the present invention extract features from the acquired ultrasonic image and perform spatial and channel reconstruction convolution on the extracted ultrasonic features, thereby reducing redundant features in the spatial dimension and redundant features in the channel dimension, thereby obtaining better performance, and utilizing a labeled multi-layer perception mechanism to capture local and global information between labels, so as to significantly reduce the number of model parameters, computational complexity and inference time while effectively processing the image segmentation task, thereby improving the segmentation performance without significantly increasing the number of parameters, thereby achieving faster and more accurate object segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 It is a flowchart of the ultrasound image segmentation method provided by the present invention;

[0021] Figure 2 It is a schematic diagram of the process of training the image segmentation model provided by the present invention;

[0022] Figure 3 It is a structural schematic diagram of the ultrasonic image segmentation device provided by the present invention;

[0023] Figure 4 It is a structural schematic diagram of the training module provided by the present invention;

[0024] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are 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.

[0026] Figure 1 A schematic flow chart of an ultrasound image segmentation method provided by the present invention is shown, and the method comprises:

[0027] S11, acquiring an ultrasound image;

[0028] S12, inputting the ultrasound image into the image segmentation model to obtain the ultrasound image segmentation result output by the image segmentation model; wherein the ultrasound image segmentation result is obtained by training based on the ultrasound training image and the segmentation labels corresponding to the ultrasound training image; the image segmentation model is used to extract the features of the ultrasound image, and perform spatial and channel reconstruction convolution on the extracted ultrasound features, and use a labeled multi-layer perception mechanism to obtain the ultrasound image segmentation result.

[0029] It should be noted that the step numbers "S1N" in this specification do not represent the order of the ultrasound image segmentation method. Figure 2 The ultrasound image segmentation method of the present invention is described.

[0030] Step S11, acquiring an ultrasound image.

[0031] Step S12, inputting the ultrasound image into the image segmentation model to obtain the ultrasound image segmentation result output by the image segmentation model; wherein the ultrasound image segmentation result is obtained based on the ultrasound training image and the segmentation label corresponding to the ultrasound training image; the image segmentation model is used to extract the features of the ultrasound image, and perform spatial and channel reconstruction convolution on the extracted ultrasound features, and use the labeled multi-layer perception mechanism to obtain the ultrasound image segmentation result.

[0032] In the present embodiment, the image segmentation model includes: a feature extraction layer, which extracts features from an input ultrasound image to obtain ultrasound features; a first plug-and-play layer, which spatially reconstructs the ultrasound features, and performs channel reconstruction on the first spatial refinement features obtained based on the spatial reconstruction to obtain first channel refinement features; a labeled multi-layer perception layer, which uses a labeled multi-layer perception mechanism to label and project the first channel refinement features to obtain spatial information features; a second plug-and-play layer, which performs feature fusion on the spatial information features and the first channel refinement features, and spatially reconstructs the first features obtained based on the feature fusion, and performs channel reconstruction on the second spatial refinement features obtained based on the spatial reconstruction to obtain second channel refinement features; an image segmentation layer, which performs feature fusion on the second channel refinement features and the ultrasound features, and upsamples the second features obtained based on the feature fusion to obtain an ultrasound image segmentation result.

[0033] Specifically, feature extraction is performed on the input ultrasound image to obtain ultrasound features, including: performing at least three convolutions on the input ultrasound image, and after each convolution, downsampling the features obtained by the convolution using batch normalization (BatchNormalization) and maximum pooling to obtain ultrasound features.

[0034] It should be added that a 3×3 convolution with a stride of 1 and a padding of 1 and a 2×2 maximum pooling with a stride of 2 can be used, which can be set according to actual design requirements and is not further limited here. In addition, at least three convolutions are set to reduce the spatial and channel redundancy between features.

[0035] In an optional embodiment, the ultrasonic feature is spatially reconstructed, including: normalizing the ultrasonic feature to obtain a first normalized feature; generating corresponding weights according to the first normalized feature and a first preset nonlinear activation function, and separating the ultrasonic feature according to the generated weights to obtain a first sub-feature and a second sub-feature; transforming the first sub-feature at least three times, and transforming the second sub-feature at least three times; fusing the first sub-feature after the second transformation with the second sub-feature after the third transformation to obtain a first fused feature; fusing the first sub-feature after the third transformation with the second sub-feature after the second transformation to obtain a second fused feature; and splicing the first fused feature and the second fused feature to obtain a first spatially refined feature.

[0036] It is worth noting that the ultrasonic features are separated by normalization processing and weights generated by the first preset nonlinear activation function to separate feature maps with large amounts of information from feature maps with small amounts of information, and then the separated first sub-features and second sub-features are transformed and reconstructed to generate features with more information and save space, thereby reducing redundant features in the spatial dimension and suppressing spatial redundancy. In addition, the first preset nonlinear activation function can adopt functions such as Sigmoid, which can be set according to actual design requirements and is not further limited here.

[0037] Correspondingly, the first feature obtained based on feature fusion is spatially reconstructed, including: normalizing the first feature to obtain a second normalized feature; generating corresponding weights according to the second normalized feature and a second preset nonlinear activation function, and separating the first feature according to the generated weights to obtain a third sub-feature and a fourth sub-feature; transforming the third sub-feature at least three times, and transforming the fourth sub-feature at least three times; fusing the third sub-feature after the second transformation with the fourth sub-feature after the third transformation to obtain a third fused feature; fusing the third sub-feature after the third transformation with the fourth sub-feature after the second transformation to obtain a fourth fused feature; and splicing the third fused feature with the fourth fused feature to obtain a second spatially refined feature.

[0038] It should be added that the spatial reconstruction of the first feature obtained based on feature fusion can specifically refer to the above-mentioned spatial reconstruction of the ultrasound feature, which will not be repeated here.

[0039] In an optional embodiment, channel reconstruction is performed on the first spatial refinement feature obtained based on spatial reconstruction to obtain a first channel refinement feature, including: segmenting the first spatial refinement feature to obtain a first segmentation feature and a second segmentation feature; performing global convolution on the first segmentation feature, and performing point-by-point convolution on the first segmentation feature, and fusing the feature obtained by the global convolution with the feature obtained by the point-by-point convolution to obtain a first splicing feature; performing point-by-point convolution on the second segmentation feature, and splicing the feature obtained by the point-by-point convolution with the second segmentation feature to obtain a second splicing feature; pooling the first splicing feature and the second splicing feature respectively to obtain the corresponding first pooling feature and second pooling feature, and splicing the first pooling feature and the second pooling feature to obtain a third splicing feature; performing category probability prediction on the third splicing feature to obtain a probability distribution; performing weighted fusion on the first splicing feature and the probability distribution to obtain a fifth fusion feature; performing weighted fusion on the second splicing feature and the probability respectively to obtain a sixth fusion feature; and performing feature fusion on the fifth fusion feature and the sixth fusion feature to obtain the first channel refinement feature.

[0040] It is worth noting that the above feature fusion can adopt element-level summation to perform summation operations on the elements at corresponding positions for the two fused features, so as to facilitate calculation and significantly improve the operation speed. In addition, the above weighted fusion can adopt element-level multiplication to perform multiplication operations on the elements at corresponding positions for the two fused features, so as to extract local subtle features and greatly improve the operation speed.

[0041] It should be noted that by segmenting the first spatial refinement features and performing different convolution operations on the segmented features, the number of channels of the two groups of features are compressed using convolution kernels, and through specific conversion operations and fusion operations, channel-refined features are obtained, thereby reducing redundancy in the channel dimension.

[0042] Correspondingly, the second spatial refined feature obtained based on the spatial reconstruction is subjected to channel reconstruction to obtain the second channel refined feature, including: segmenting the second spatial refined feature to obtain a third segmentation feature and a fourth segmentation feature; performing global convolution on the third segmentation feature, and performing point-by-point convolution on the third segmentation feature, and performing feature fusion of the feature obtained by the global convolution with the feature obtained by the point-by-point convolution to obtain a fourth splicing feature; performing point-by-point convolution on the fourth segmentation feature, and splicing the feature obtained by the point-by-point convolution with the fourth segmentation feature to obtain a fifth splicing feature; respectively pooling the fourth splicing feature and the fifth splicing feature to obtain the corresponding third pooling feature and fourth pooling feature, and splicing the first pooling feature and the second pooling feature to obtain a sixth splicing feature; performing category probability prediction on the sixth splicing feature to obtain a probability distribution; performing weighted fusion on the fourth splicing feature and the probability distribution to obtain a seventh fusion feature; performing weighted fusion on the fifth splicing feature and the probability to obtain an eighth fusion feature; and performing feature fusion on the seventh fusion feature and the eighth fusion feature to obtain the second channel refined feature.

[0043] Similarly, the channel reconstruction of the first spatial refinement feature obtained based on the spatial reconstruction may specifically refer to the channel reconstruction of the second spatial refinement feature obtained based on the spatial reconstruction, which will not be repeated here.

[0044] In an optional embodiment, the number of tokenized multi-layer perception layers is at least four, and half of them together with the feature extraction layer and the first plug-and-play layer constitute an encoder, and the other half together with the second plug-and-play layer and the image segmentation layer constitute a decoder, and in the decoding stage, a structure and jump connection symmetrical to those of the encoding stage are adopted to directly transfer the low-level features of the encoding stage to the decoding stage, and combine them with the high-level features after upsampling, so as to compensate for the loss of detail information that may be caused by multiple downsampling during the encoding process. In this way, the model can not only use high-level semantic information for decoding, but also use the detail information in the low-level features to more effectively reconstruct the details and contours of the image and achieve a more accurate image segmentation effect.

[0045] Furthermore, the first channel refinement feature is marked and projected by using a marked multi-layer perception mechanism to obtain a spatial information feature, including: moving the channel axis of the first channel refinement feature along the channel dimension of the first channel refinement feature, and dividing the moved first channel refinement feature into multiple tags, so as to use the multi-layer perception mechanism to capture the local information and global information between the segmented tags and obtain the first spatial feature; using a preset activation function to process the first spatial feature element by element, and encoding the processed feature to obtain a coded feature; moving the channel axis of the coded feature along the height dimension of the coded feature, and dividing the moved coded feature into multiple tags, so as to use the multi-layer perception mechanism to capture the local information and global information between the segmented tags and obtain the second spatial feature; normalizing and reprojecting the second spatial feature to obtain a reprojection feature; and fusing the reprojection feature with the first channel refinement feature to obtain a spatial information feature.

[0046] It should be noted that the preset activation function can use a Gaussian Error Linear Unit (GELU for short), which can be determined according to the actual design and is not further limited here. In addition, by moving the channel axis of the feature along different dimensions of the feature and segmenting the feature to obtain a label, the local and global information between the labels is captured, thereby obtaining the corresponding global feature. While maintaining good segmentation performance, the model's parameter quantity, computational complexity and reasoning time are significantly reduced, providing reasonable support for subsequent image segmentation.

[0047] In an optional embodiment, an ultrasound image is input into an image segmentation model to obtain an ultrasound image segmentation result output by the image segmentation model, including: inputting the ultrasound image into a feature extraction layer to extract features, and obtaining ultrasound features output by the feature extraction layer; inputting the ultrasound features into a first plug-and-play layer to spatially reconstruct the ultrasound features, and performing channel reconstruction on the first spatially refined features obtained based on the spatial reconstruction, and obtaining first channel-refined features output by the first plug-and-play layer; inputting the first channel-refined features into a labeled multi-layer perception layer, and utilizing a labeled multi-layer perception mechanism to refine the first channel. The features are marked and projected to obtain the spatial information features output by the marked multi-layer perception layer; the spatial information features and the first channel refinement features are input into the second plug-and-play layer for feature fusion, and the first features obtained based on the feature fusion are spatially reconstructed, and the second spatial refinement features obtained based on the spatial reconstruction are channel reconstructed to obtain the second channel refinement features output by the second plug-and-play layer; the second channel refinement features and the ultrasonic features are input into the image segmentation layer for feature fusion, and the second features obtained based on the feature fusion are upsampled to obtain the ultrasonic image segmentation result output by the image segmentation layer.

[0048] In an optional embodiment, before inputting the ultrasound image into the image segmentation model, the method includes: training the image segmentation model. Specifically, referring to Figure 2 , train the image segmentation model, including:

[0049] S21, obtaining an ultrasound training image and a segmentation label corresponding to the ultrasound training image.

[0050] In this embodiment, obtaining an ultrasound training image and a segmentation label corresponding to the ultrasound training image includes: obtaining an ultrasound training image; and marking the ultrasound training image to obtain a corresponding segmentation label.

[0051] It should be noted that the acquired ultrasound training images may come from existing channels, such as public papers, the Internet, medical systems, etc., which are not further limited here. In addition, acquiring ultrasound training images includes: using at least three different ultrasound devices to collect ultrasound images of preset body parts of the target population to obtain ultrasound training images.

[0052] Furthermore, in order to improve the accuracy and diversity of training, the target population can be selected from different age groups and people of different occupations. The age groups may be the elderly, middle-aged and young people, etc. The occupations may be remote mobilizers, office staff, grassroots workers, people with health risks, etc. The preset body parts can be set according to the actual needs for medical diagnosis, such as the medial and lateral gastrocnemius, vastus lateralis and tibialis anterior muscles, and include different movements and contraction types of corresponding body parts, which are not further limited here.

[0053] In addition, based on the collected ultrasound training images, they are divided into training sets and test sets according to a preset ratio, and the training set is used to train the training model, and after the training, the trained model is tested using the test set. The following mainly describes the process of training the training model using the training set, and the testing of the trained model using the test set can refer to the training process, which will not be repeated here.

[0054] In an optional embodiment, after obtaining the ultrasound training image and the segmentation label corresponding to the ultrasound training image, the method includes: performing data enhancement on the ultrasound training image by using a data enhancement strategy.

[0055] Further, the data enhancement strategy includes at least one of cropping, rotation, translation, flipping, sharpening, illumination and occlusion. For example, the data enhancement strategy includes cropping, rotation, translation, flipping, sharpening, illumination or occlusion. For another example, the data enhancement strategy includes at least two of cropping, rotation, translation, flipping, sharpening, illumination and occlusion. By adopting at least two data enhancement strategies, the acquired samples are respectively enhanced so as to increase the amount of training data of the model and improve the segmentation accuracy of the model.

[0056] S22, using the ultrasound training image as input data for training, using the segmentation label corresponding to the ultrasound training image as the label for training, training the training model to obtain an image segmentation model for segmenting ultrasound images.

[0057] In this embodiment, the model to be trained is trained, including: inputting ultrasound training images into the model to be trained to obtain a segmentation prediction result output by the model to be trained; constructing a loss function based on the segmentation prediction result and the segmentation label, and ending the training based on the convergence of the loss function.

[0058] It should be noted that the model to be trained can be an existing network including at least one network structure, or it can be another network specified by the user. The network to be trained usually includes a feature extraction layer, a first plug-and-play layer, a labeled multi-layer perception layer, a second plug-and-play layer, an image segmentation layer, and a loss function; according to a preset iteration rule, the above-mentioned ultrasound training image or the ultrasound training image after data enhancement is input into the model to be trained to obtain a trained image segmentation model.

[0059] In summary, the embodiments of the present invention extract features from the acquired ultrasound image and perform spatial and channel reconstruction convolution on the extracted ultrasound features, thereby reducing redundant features in the spatial dimension and redundant features in the channel dimension, thereby obtaining better performance, and utilizing a labeled multi-layer perception mechanism to capture local and global information between labels, so as to significantly reduce the number of model parameters, computational complexity, and inference time while effectively processing image segmentation tasks, thereby improving segmentation performance without significantly increasing the number of parameters, thereby achieving faster and more accurate object segmentation.

[0060] The ultrasonic image segmentation device provided by the present invention is described below. The ultrasonic image segmentation device described below and the ultrasonic image segmentation method described above can be referred to each other.

[0061] Figure 3 A schematic structural diagram of an ultrasonic image segmentation device is shown, the device comprising:

[0062] An image acquisition module 31 is used to acquire an ultrasound image;

[0063] The image segmentation module 32 inputs the ultrasound image into the image segmentation model to obtain the ultrasound image segmentation result output by the image segmentation model; wherein the ultrasound image segmentation result is obtained based on the ultrasound training image and the segmentation label corresponding to the ultrasound training image; the image segmentation model is used to extract the features of the ultrasound image, and perform spatial and channel reconstruction convolution on the extracted ultrasound features, and use the labeled multi-layer perception mechanism to obtain the ultrasound image segmentation result.

[0064] In this embodiment, the image segmentation model includes a feature extraction layer, a first plug-and-play layer, a labeled multi-layer perception layer, a second plug-and-play layer and an image segmentation layer. Accordingly, the image segmentation module 32 includes: a feature extraction unit, which inputs the ultrasound image into the feature extraction layer for feature extraction to obtain the ultrasound features output by the feature extraction layer; a first plug-and-play unit, which inputs the ultrasound features into the first plug-and-play layer to spatially reconstruct the ultrasound features, and performs channel reconstruction on the first spatial refinement features obtained based on the spatial reconstruction to obtain the first channel refinement features output by the first plug-and-play layer; a labeled multi-layer perception unit, which inputs the first channel refinement features into the labeled multi-layer perception layer to label and project the first channel refinement features using the labeled multi-layer perception mechanism to obtain to the spatial information features output by the labeled multi-layer perception layer; the second plug-and-play unit inputs the spatial information features and the first channel refinement features into the second plug-and-play layer for feature fusion, and performs spatial reconstruction on the first features obtained based on the feature fusion, and performs channel reconstruction on the second spatial refinement features obtained based on the spatial reconstruction, so as to obtain the second channel refinement features output by the second plug-and-play layer; the image segmentation unit inputs the second channel refinement features and the ultrasonic features into the image segmentation layer for feature fusion, and upsamples the second features obtained based on the feature fusion, so as to obtain the ultrasonic image segmentation result output by the image segmentation layer.

[0065] Specifically, the feature extraction unit is used to: perform at least three convolutions on the input ultrasound image, and after each convolution, use batch normalization (BatchNormalization) and maximum pooling to downsample the convolution features to obtain ultrasound features.

[0066] In an optional embodiment, the first plug-and-play unit includes: a first normalization processing subunit, which normalizes the ultrasonic feature to obtain a first normalized feature; a first separation subunit, which generates corresponding weights according to the first normalized feature and a first preset nonlinear activation function, and separates the ultrasonic feature according to the generated weights to obtain a first sub-feature and a second sub-feature; a first transformation subunit, which transforms the first sub-feature at least three times, and transforms the second sub-feature at least three times; a first reconstruction subunit, which fuses the first sub-feature after the second transformation with the second sub-feature after the third transformation to obtain a first fused feature, fuses the first sub-feature after the third transformation with the second sub-feature after the second transformation to obtain a second fused feature, and splices the first fused feature and the second fused feature to obtain a first spatial refinement feature.

[0067] In addition, the first plug-and-play unit further includes: a first segmentation subunit, which segments the first spatial refinement feature to obtain a first segmentation feature and a second segmentation feature; a first conversion subunit, which performs global convolution on the first segmentation feature, and performs point-by-point convolution on the first segmentation feature, and performs feature fusion on the feature obtained by the global convolution and the feature obtained by the point-by-point convolution to obtain a first splicing feature; a second conversion subunit, which performs point-by-point convolution on the second segmentation feature, and splices the feature obtained by the point-by-point convolution with the second segmentation feature to obtain a second splicing feature; and a second reconstruction subunit, which performs point-by-point convolution on the first splicing feature and the second splicing feature, respectively. The first and second splicing features are pooled to obtain the corresponding first pooling features and second pooling features, and the first pooling features and the second pooling features are spliced ​​to obtain the third splicing features; the first probability prediction subunit performs category probability prediction on the third splicing feature to obtain the probability distribution; the first fusion subunit performs weighted fusion on the first splicing feature and the probability distribution to obtain the fifth fusion feature; the second fusion subunit performs weighted fusion on the second splicing feature and the probability respectively to obtain the sixth fusion feature; the third fusion subunit performs feature fusion on the fifth fusion feature and the sixth fusion feature to obtain the first channel refinement feature.

[0068] In an optional embodiment, the second plug-and-play unit includes: a second normalization processing subunit, which normalizes the first feature to obtain a second normalized feature; a second separation subunit, which generates corresponding weights according to the second normalized feature and a second preset nonlinear activation function, and separates the first feature according to the generated weights to obtain a third sub-feature and a fourth sub-feature; a second transformation subunit, which transforms the third sub-feature at least three times, and transforms the fourth sub-feature at least three times; a third feature reconstruction subunit, which uses the third sub-feature after the second transformation and the fourth sub-feature after the third transformation to perform feature fusion to obtain a third fused feature, uses the third sub-feature after the third transformation and the fourth sub-feature after the second transformation to perform feature fusion to obtain a fourth fused feature, and splices the third fused feature and the fourth fused feature to obtain a second spatial refinement feature.

[0069] In addition, the second plug-and-play unit further includes: a second segmentation subunit, which segments the second spatial refinement feature to obtain a third segmentation feature and a fourth segmentation feature; a third conversion subunit, which performs global convolution on the third segmentation feature, and performs point-by-point convolution on the third segmentation feature, and performs feature fusion on the feature obtained by the global convolution and the feature obtained by the point-by-point convolution to obtain a fourth splicing feature; a fourth conversion subunit, which performs point-by-point convolution on the fourth segmentation feature, and splices the feature obtained by the point-by-point convolution with the fourth segmentation feature to obtain a fifth splicing feature; a fourth reconstruction subunit, which performs point-by-point convolution on the fourth splicing feature and the fourth splicing feature, respectively. The five splicing features are pooled to obtain the corresponding third pooling features and the fourth pooling features, and the first pooling features and the second pooling features are spliced ​​to obtain the sixth splicing features; the second probability prediction subunit performs category probability prediction on the sixth splicing feature to obtain the probability distribution; the fourth fusion subunit performs weighted fusion on the fourth splicing feature and the probability distribution to obtain the seventh fusion feature; the fifth fusion subunit performs weighted fusion on the fifth splicing feature and the probability respectively to obtain the eighth fusion feature; the sixth fusion subunit performs feature fusion on the seventh fusion feature and the eighth fusion feature to obtain the second channel refinement feature.

[0070] In an optional embodiment, the labeled multi-layer perception unit includes: a first information capture subunit, which moves the channel axis of the first channel refined feature along the channel dimension of the first channel refined feature, and divides the moved first channel refined feature into multiple tags, so as to capture the local information and global information between the segmented tags by using the multi-layer perception mechanism, and obtain the first spatial feature; a first feature processing subunit, which processes the first spatial feature element by element using a preset activation function, and encodes the processed feature to obtain the encoded feature; a second information capture subunit, which moves the channel axis of the encoded feature along the height dimension of the encoded feature, and divides the moved encoded feature into multiple tags, so as to capture the local information and global information between the segmented tags by using the multi-layer perception mechanism, and obtain the second spatial feature; a second feature processing subunit, which normalizes and reprojects the second spatial feature to obtain the reprojected feature; a seventh fusion subunit, which fuses the reprojected feature with the first channel refined feature to obtain the spatial information feature.

[0071] In an optional embodiment, the device further includes: a training module for training the image segmentation model before inputting the ultrasound image into the image segmentation model. Figure 4 , training modules, including:

[0072] A training data acquisition unit 41 is used to acquire an ultrasound training image and a segmentation label corresponding to the ultrasound training image;

[0073] The training unit 42 uses the ultrasound training image as input data for training and the segmentation label corresponding to the ultrasound training image as a label for training to train the training model to obtain an image segmentation model for segmenting the ultrasound image.

[0074] In this embodiment, the training data acquisition unit 41 includes: a training data acquisition subunit, which acquires an ultrasound training image; and a labeling subunit, which labels the ultrasound training image to obtain a corresponding segmentation label.

[0075] Furthermore, the training data acquisition subunit is used to: use at least three different ultrasound devices to collect ultrasound images of preset body parts of the target population to obtain ultrasound training images.

[0076] In an optional embodiment, the training module further includes: a data enhancement unit, which performs data enhancement on the ultrasound training image using a data enhancement strategy.

[0077] In addition, the training unit 42 includes: a segmentation prediction subunit, which inputs the ultrasound training image into the model to be trained to obtain the segmentation prediction result output by the model to be trained; a training subunit, which constructs a loss function based on the segmentation prediction result and the segmentation label, and ends the training based on the convergence of the loss function.

[0078] To summarize, the embodiment of the present invention uses an image segmentation module to perform feature extraction on the ultrasound image acquired by the image acquisition module, so as to perform spatial and channel reconstruction convolution on the extracted ultrasound features, thereby reducing redundant features in the spatial dimension and redundant features in the channel dimension, thereby obtaining better performance, and utilizing a labeled multi-layer perception mechanism to capture local and global information between labels, so as to significantly reduce the number of parameters, computational complexity, and inference time of the model while effectively processing the image segmentation task, thereby improving the segmentation performance without significantly increasing the number of parameters, thereby achieving faster and more accurate object segmentation.

[0079] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute the ultrasound image segmentation method, which includes: acquiring an ultrasound image; inputting the ultrasound image into an image segmentation model to obtain an ultrasound image segmentation result output by the image segmentation model; wherein the ultrasound image segmentation result is obtained by training based on an ultrasound training image and a segmentation label corresponding to the ultrasound training image; the image segmentation model is used to extract features of the ultrasound image, perform spatial and channel reconstruction convolution on the extracted ultrasound features, and use a labeled multi-layer perception mechanism to obtain an ultrasound image segmentation result.

[0080] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0081] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the ultrasound image segmentation method provided by the above-mentioned methods, and the method includes: acquiring an ultrasound image; inputting the ultrasound image into an image segmentation model to obtain an ultrasound image segmentation result output by the image segmentation model; wherein the ultrasound image segmentation result is obtained based on ultrasound training images and segmentation labels corresponding to the ultrasound training images; the image segmentation model is used to extract features of the ultrasound image, perform spatial and channel reconstruction convolution on the extracted ultrasound features, and use a labeled multi-layer perception mechanism to obtain an ultrasound image segmentation result.

[0082] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the ultrasound image segmentation method provided by the above-mentioned methods, the method comprising: acquiring an ultrasound image; inputting the ultrasound image into an image segmentation model to obtain an ultrasound image segmentation result output by the image segmentation model; wherein the ultrasound image segmentation result is obtained by training based on an ultrasound training image and the segmentation labels corresponding to the ultrasound training image; the image segmentation model is used to extract features of the ultrasound image, perform spatial and channel reconstruction convolution on the extracted ultrasound features, and utilize a labeled multi-layer perception mechanism to obtain an ultrasound image segmentation result.

[0083] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0084] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for segmenting an ultrasonic image, characterized in that: include: Acquiring ultrasound images; Inputting the ultrasound image into an image segmentation model to obtain an ultrasound image segmentation result output by the image segmentation model; wherein the ultrasound image segmentation result is obtained by training based on an ultrasound training image and a segmentation label corresponding to the ultrasound training image; The image segmentation model is used to extract features of the ultrasound image, perform spatial and channel reconstruction convolution on the extracted ultrasound features, and use a labeled multi-layer perception mechanism to obtain an ultrasound image segmentation result; The image segmentation model includes a feature extraction layer, a first plug-and-play layer, a tokenized multi-layer perception layer, a second plug-and-play layer and an image segmentation layer; the ultrasound image is input into the image segmentation model to obtain an ultrasound image segmentation result output by the image segmentation model, including: Inputting the ultrasonic image into the feature extraction layer to perform feature extraction, and obtaining ultrasonic features output by the feature extraction; Inputting the ultrasonic feature into the first plug-and-play layer to spatially reconstruct the ultrasonic feature, and performing channel reconstruction on the first spatially refined feature obtained based on the spatial reconstruction to obtain a first channelly refined feature output by the first plug-and-play layer; Inputting the first channel refinement feature into the labeled multi-layer perception layer, so as to label and project the first channel refinement feature by using the labeled multi-layer perception mechanism, and obtain the spatial information feature output by the labeled multi-layer perception layer; Inputting the spatial information feature and the first channel refinement feature into the second plug-and-play layer to perform feature fusion on the spatial information feature and the first channel refinement feature, and spatially reconstructing the first feature obtained based on the feature fusion, and channel reconstructing the second spatial refinement feature obtained based on the spatial reconstruction, to obtain the second channel refinement feature output by the second plug-and-play layer; The second channel refined features and the ultrasonic features are input into the image segmentation layer to perform feature fusion on the second channel refined features and the ultrasonic features, and the second features obtained based on the feature fusion are upsampled to obtain the ultrasonic image segmentation result output by the image segmentation layer.

2. The ultrasonic image segmentation method according to claim 1, characterized in that: The ultrasonic feature is spatially reconstructed, comprising: Normalizing the ultrasonic feature to obtain a first normalized feature; Generate corresponding weights according to the first normalized features and a first preset nonlinear activation function, and separate the ultrasonic features according to the generated weights to obtain a first sub-feature and a second sub-feature; Transforming the first sub-feature at least three times, and transforming the second sub-feature at least three times; The first sub-feature after the second transformation is used to fuse the second sub-feature after the third transformation to obtain a first fused feature; The first sub-feature after the third transformation and the second sub-feature after the second transformation are used to perform feature fusion to obtain a second fused feature; The first fusion feature and the second fusion feature are concatenated to obtain a first spatial refinement feature; The first feature obtained based on feature fusion is spatially reconstructed, including: Normalizing the first feature to obtain a second normalized feature; Generate corresponding weights according to the second normalized feature and the second preset nonlinear activation function, and separate the first feature according to the generated weights to obtain a third sub-feature and a fourth sub-feature; Transforming the third sub-feature at least three times, and transforming the fourth sub-feature at least three times; The third sub-feature after the second transformation is used to fuse the fourth sub-feature after the third transformation to obtain a third fused feature; The third sub-feature after the third transformation and the fourth sub-feature after the second transformation are used to perform feature fusion to obtain a fourth fused feature; The third fusion feature and the fourth fusion feature are concatenated to obtain a second spatial refinement feature.

3. The ultrasonic image segmentation method according to claim 1, characterized in that: Channel reconstruction is performed on the first spatial refinement feature obtained based on the spatial reconstruction to obtain the first channel refinement feature, including: Segmenting the first spatial refinement feature to obtain a first segmentation feature and a second segmentation feature; Performing global convolution on the first segmentation feature, and performing point-by-point convolution on the first segmentation feature, and performing feature fusion on the feature obtained by the global convolution and the feature obtained by the point-by-point convolution to obtain a first splicing feature; Performing point-by-point convolution on the second segmentation feature, and concatenating the feature obtained by the point-by-point convolution with the second segmentation feature to obtain a second concatenated feature; Pooling the first splicing feature and the second splicing feature respectively to obtain corresponding first pooling features and second pooling features, and splicing the first pooling feature and the second pooling feature to obtain a third splicing feature; Performing category probability prediction on the third concatenated feature to obtain a probability distribution; Performing weighted fusion on the first concatenated feature and the probability distribution to obtain a fifth fused feature; Performing weighted fusion on the second concatenated feature and the probability distribution to obtain a sixth fused feature; Performing feature fusion on the fifth fusion feature and the sixth fusion feature to obtain a first channel refined feature; Channel reconstruction is performed on the second spatial refinement feature obtained based on the spatial reconstruction to obtain a second channel refinement feature, including: Segmenting the second spatial refinement feature to obtain a third segmentation feature and a fourth segmentation feature; Performing global convolution on the third segmentation feature, and performing point-by-point convolution on the third segmentation feature, and performing feature fusion on the feature obtained by the global convolution and the feature obtained by the point-by-point convolution to obtain a fourth splicing feature; Performing point-by-point convolution on the fourth segmentation feature, and concatenating the feature obtained by the point-by-point convolution with the fourth segmentation feature to obtain a fifth concatenated feature; Pooling the fourth splicing feature and the fifth splicing feature respectively to obtain corresponding third pooling features and fourth pooling features, and splicing the first pooling feature and the second pooling feature to obtain a sixth splicing feature; Performing category probability prediction on the sixth concatenated feature to obtain a probability distribution; Performing weighted fusion on the fourth splicing feature and the probability distribution to obtain a seventh fusion feature; Performing weighted fusion on the fifth splicing feature and the probability distribution to obtain an eighth fusion feature; The seventh fusion feature and the eighth fusion feature are fused to obtain a second channel refinement feature.

4. The ultrasonic image segmentation method according to claim 1, characterized in that: The first channel refinement features are marked and projected using a labeled multi-layer perception mechanism to obtain spatial information features, including: Moving the channel axis of the first channel refined feature along the channel dimension of the first channel refined feature, and segmenting the moved first channel refined feature into a plurality of tags, so as to capture local information and global information between the segmented tags by using a multi-layer perception mechanism, and obtain a first spatial feature; Processing the first spatial feature element by element using a preset activation function, and encoding the processed feature to obtain an encoded feature; Moving the channel axis of the coding feature along the height dimension of the coding feature, and segmenting the moved coding feature into a plurality of tags, so as to capture local information and global information between the segmented tags by using a multi-layer perception mechanism, and obtain a second spatial feature; Normalizing and reprojecting the second spatial feature to obtain a reprojected feature; The reprojection feature and the first channel refinement feature are fused to obtain a spatial information feature.

5. The ultrasonic image segmentation method according to claim 1, characterized in that: Before inputting the ultrasound image into the image segmentation model, the method includes: Acquire an ultrasound training image and a segmentation label corresponding to the ultrasound training image; The ultrasound training image is used as input data for training, and the segmentation label corresponding to the ultrasound training image is used as a label for training. The training model is trained to obtain an image segmentation model for segmenting ultrasound images.

6. An ultrasonic image segmentation device, characterized in that: include: An image acquisition module, for acquiring ultrasound images; An image segmentation module, inputting the ultrasound image into an image segmentation model to obtain an ultrasound image segmentation result output by the image segmentation model; wherein the ultrasound image segmentation result is obtained by training based on an ultrasound training image and a segmentation label corresponding to the ultrasound training image; The image segmentation model is used to extract features of the ultrasound image, perform spatial and channel reconstruction convolution on the extracted ultrasound features, and use a labeled multi-layer perception mechanism to obtain an ultrasound image segmentation result; The image segmentation model includes a feature extraction layer, a first plug-and-play layer, a tokenized multi-layer perception layer, a second plug-and-play layer and an image segmentation layer; the image segmentation module includes: A feature extraction unit, inputting the ultrasound image into the feature extraction layer to perform feature extraction, and obtaining the ultrasound features output by the feature extraction; A first plug-and-play unit inputs the ultrasonic feature into the first plug-and-play layer to spatially reconstruct the ultrasonic feature, and performs channel reconstruction on the first spatial refinement feature obtained based on the spatial reconstruction to obtain a first channel refinement feature output by the first plug-and-play layer; A labeled multi-layer perception unit inputs the first channel refinement feature into the labeled multi-layer perception layer, so as to label and project the first channel refinement feature by using a labeled multi-layer perception mechanism, and obtain a spatial information feature output by the labeled multi-layer perception layer; The second plug-and-play unit inputs the spatial information feature and the first channel refinement feature into the second plug-and-play layer to perform feature fusion on the spatial information feature and the first channel refinement feature, and spatially reconstructs the first feature obtained based on the feature fusion, and performs channel reconstruction on the second spatial refinement feature obtained based on the spatial reconstruction, so as to obtain the second channel refinement feature output by the second plug-and-play layer; The image segmentation unit inputs the second channel refined features and the ultrasonic features into the image segmentation layer to perform feature fusion on the second channel refined features and the ultrasonic features, and upsamples the second features obtained based on the feature fusion to obtain the ultrasonic image segmentation result output by the image segmentation layer.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the ultrasound image segmentation method according to any one of claims 1 to 5 is implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the ultrasound image segmentation method according to any one of claims 1 to 5 is implemented.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the ultrasound image segmentation method according to any one of claims 1 to 5 is implemented.

Images