An ultrasonic lesion standard section selection method and system based on deep learning

Abstract

The application provides a kind of based on deep learning's ultrasonic lesion standard section selection method and system, belong to ultrasonic technical field, including: step S1, obtain the first ultrasonic image to be handled, extract the first feature vector of ultrasonic image;Step S2, the feature vector of the average image corresponding to each standard section type is matched with the first feature vector, determine the standard section type corresponding to the average image with the highest similarity degree;Step S3, according to the standard section type, determine the position and angle corresponding to the standard section.The beneficial effect: the feature vector of the obtained ultrasonic image is matched with the feature vector of the average image corresponding to each standard section type, the standard section type corresponding to the average image with the highest similarity degree is determined, and then the position and angle corresponding to the standard section are determined according to the standard section type, which has wide applicability and practicality.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ultrasound, in particular to an ultrasound lesion standard section selection method and system based on deep learning. BACKGROUND

[0002] Clinical trials are an important means of evaluating the safety and effectiveness of new drugs, therapies or medical devices. In these trials, ultrasound technology is widely used for lesion detection and evaluation. However, in the process of scanning and acquiring ultrasound images, scanning parameters such as probe position and scanning angle are limited by the experience of the scanning operator, and it is still challenging to achieve standardized ultrasound image acquisition in clinical trials.

[0003] In the prior art, real-time ultrasound images are usually matched with the feature vectors of a plurality of preset standard section ultrasound images, and the standard section ultrasound image with the highest similarity is selected. This section selection method ignores the information redundancy between different directions, and therefore may result in underestimation or overestimation of information in some directions. SUMMARY

[0004] To solve the above technical problems, the present application provides an ultrasound lesion standard section selection method and system based on deep learning, which can be applied in clinical trials.

[0005] The technical problems solved by the present application can be implemented by the following technical solutions:

[0006] An ultrasound lesion standard section selection method based on deep learning, comprising:

[0007] Step S1: obtaining a first ultrasound image to be processed, and extracting a first feature vector of the first ultrasound image;

[0008] Step S2: matching the first feature vector with the feature vector of an average image corresponding to each standard section type, and determining the standard section type corresponding to the average image with the highest similarity to the first feature vector;

[0009] Step S3: determining the position and angle of the standard section corresponding to the standard section type.

[0010] Preferably, the step S3 further comprises:

[0011] Step S4: registering the real-time ultrasound image after adaptive adjustment of the position and angle with the standard section corresponding to the determined standard section type.

[0012] Preferably, the registration in step S4 is performed using an image registration algorithm, and the image registration algorithm uses a phase correlation algorithm, a mutual information algorithm or a deformation field algorithm.

[0013] Preferably, the step of obtaining the feature vector of the average image corresponding to each standard section type in the step S2 comprises:

[0014] Step A1, obtaining a second ultrasound image group containing lesions;

[0015] Step A2, respectively extracting features from all second ultrasound images in the second ultrasound image group to obtain a feature vector group;

[0016] Step A3, performing clustering processing on the feature vector group based on a clustering algorithm to obtain a cluster group corresponding to different lesion shapes and location types;

[0017] Step A4, calculating the average value of the feature vectors of all second ultrasound images in each cluster of the cluster group to obtain an average image of each cluster and a feature vector of the average image;

[0018] Step A5, classifying the feature vectors of all average images corresponding to the cluster group to determine the standard section type corresponding to each cluster.

[0019] Preferably, in the step A3, the clustering algorithm is a K-means clustering algorithm, or a density-based clustering algorithm, or a hierarchical clustering algorithm.

[0020] Preferably, in the step A4, the average value of the feature vectors of all second ultrasound images in each cluster of the cluster group is calculated using a feature calculation function; wherein the feature calculation function is a feature fusion function.

[0021] Preferably, it further comprises a database, wherein a standard section template corresponding to each standard section type is stored, and the standard section template at least indicates the position and angle corresponding to the standard section;

[0022] The step S3 comprises:

[0023] Step S31, selecting the corresponding standard section template from the database according to the determined standard section type;

[0024] Step S32, adaptively selecting the position and angle of the first ultrasound image according to the standard section template.

[0025] The application also provides an ultrasound lesion standard section selection system based on deep learning, which is used to implement the ultrasound lesion standard section selection method based on deep learning as described above, comprising:

[0026] A first feature extraction module is configured to obtain a first ultrasound image to be processed and extract a first feature vector of the first ultrasound image.

[0027] a feature matching module connected with the first feature extraction module, configured to match the first feature vector with a feature vector of an average image corresponding to each standard section type, and determine the standard section type corresponding to the average image with the highest similarity with the first feature vector;

[0028] a section selection module connected with the feature matching module, configured to determine the position and angle of the standard section corresponding to the standard section type.

[0029] Preferably, further comprising:

[0030] a section registration module connected with the section selection module, configured to register the real-time ultrasound image after the adaptive adjustment of the position and angle with the standard section corresponding to the determined standard section type.

[0031] Preferably, further comprising:

[0032] an image acquisition module configured to acquire a second ultrasound image group containing a lesion;

[0033] a second feature extraction module connected with the image acquisition module, configured to perform feature extraction on all second ultrasound images in the second ultrasound image group respectively to obtain a feature vector group;

[0034] a clustering module connected with the second feature extraction module, configured to perform clustering processing on the feature vector group based on a clustering algorithm to obtain a clustering cluster group corresponding to different lesion shapes and position types;

[0035] an average processing module connected with the clustering module, configured to calculate the average value of the feature vectors of all second ultrasound images in each clustering cluster of the clustering cluster group respectively to obtain an average image of each clustering cluster and a feature vector of the average image;

[0036] a classification module connected with the average processing module, configured to classify the feature vectors of all average images corresponding to the clustering cluster group to determine the standard section type corresponding to each clustering cluster.

[0037] The technical scheme of the present application has the following advantages or beneficial effects:

[0038] The present application matches the feature vector of the acquired ultrasound image with the feature vector of the average image corresponding to each standard section type, determines the standard section type corresponding to the average image with the highest similarity with the first feature vector, and then determines the position and angle of the standard section corresponding to the standard section type, which has wide applicability and practicality. Meanwhile, the comparability and accuracy of the clinical trial data can be improved, and the subjective factors of the ultrasound scanning operation can be reduced, thereby improving the scientificity and reliability of the clinical trial. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 For the preferred embodiment of the present application, the flowchart of the ultrasound lesion standard section selection method based on deep learning is shown in the figure;

[0040] Figure 2 For the preferred embodiment of the present application, the flowchart after step S3 is shown in the figure;

[0041] Figure 3 For the preferred embodiment of the present application, the flowchart of the feature vector acquisition step of the average image is shown in the figure;

[0042] Figure 4 For the preferred embodiment of the present application, the flowchart of the specific implementation of step S3 is shown in the figure;

[0043] Figure 5 For the preferred embodiment of the present application, the structure block diagram of the ultrasound lesion standard section selection system based on deep learning is shown in the figure;

[0044] Figure 6 For the preferred embodiment of the present application, the structure block diagram of the specific implementation of the ultrasound lesion standard section selection system based on deep learning is shown in the figure;

[0045] Figure 7 For the preferred embodiment of the present application, the structure block diagram of the specific implementation of the ultrasound lesion standard section selection system based on deep learning is shown in the figure. DETAILED DESCRIPTION

[0046] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0047] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0048] The present application will be further described below with reference to the drawings and specific embodiments, but not as a limitation of the present application.

[0049] Reference Figure 1 , in the preferred embodiment of the present application, in view of the above problems existing in the prior art, a method for selecting an ultrasound lesion standard section in a clinical trial based on deep learning is provided, comprising:

[0050] Step S1, acquiring a first ultrasound image to be processed, and extracting a first feature vector of the first ultrasound image;

[0051] Step S2, matching the first feature vector with the feature vector of the average image corresponding to each standard section type, determining the standard section type corresponding to the average image with the highest similarity to the first feature vector;

[0052] Step S3, determining the position and angle of the standard section according to the standard section type.

[0053] Specifically, in the present embodiment, first, a first ultrasound image to be processed is obtained, which is a currently real-time acquired image, a feature vector of the real-time acquired first ultrasound image is extracted, and is matched with the feature vector of the average image corresponding to each of a plurality of preset standard section types, the similarity is calculated, the average image most similar to the first feature vector is determined, and the standard section type corresponding to the average image is taken as the standard section type of the ultrasound image, and then the position and angle of the standard section corresponding to the standard section type are determined. In comparison, the feature vector of the average image, i.e. the average value of the ultrasound image feature vector, can better utilize the information redundancy between different directions, thereby more comprehensively describing the features of each pixel point in the ultrasound image. This method can better capture the differences between different tissues or organs, automatically select the standard section according to the position and morphology of the specific lesion, has wide applicability and practicality; at the same time, it can improve the comparability and accuracy of clinical trial data, and reduce the subjective factors of ultrasound scanning operation, thereby improving the scientificity and reliability of clinical trials.

[0054] As a preferred embodiment, wherein, as shown in Figure 2 Step S3 further comprises:

[0055] Step S4, registering the real-time ultrasound image after adaptive adjustment of the position and angle with the standard section corresponding to the determined standard section type.

[0056] Specifically, in the present embodiment, by registering the standard section with the ultrasound image, the difference between the standard section and the ultrasound image due to the deformation and noise of the ultrasound image in the prior art is solved.

[0057] As a preferred embodiment, wherein, in step S4, image registration algorithm is used for registration, and the image registration algorithm adopts phase correlation algorithm, mutual information algorithm or deformation field algorithm.

[0058] Specifically, in the present embodiment, in order to register the standard section with the ultrasound image, image registration algorithms such as phase correlation algorithm, mutual information algorithm or deformation field algorithm can be used.

[0059] Further, the step S4 further comprises verifying the accuracy of the selected standard section. In order to verify the accuracy and reliability of the embodiments of the present application, in practical application, the standard sections are evaluated and compared by some experienced doctors or ultrasonic diagnosis experts, and the sensitivity and specificity and other indexes are calculated to realize the verification of the selected standard section.

[0060] As a preferred embodiment, wherein, as shown in Figure 3 The step of obtaining the feature vector of the average image corresponding to each standard section type in step S2 comprises:

[0061] Step A1, obtaining a second ultrasonic image group containing lesions;

[0062] Wherein, a group of ultrasonic images are obtained to form the second ultrasonic image group, and the ultrasonic images in the second ultrasonic image group come from patients in clinical trials. Each second ultrasonic image contains at least one lesion. Preferably, each ultrasonic image should contain clear boundaries of the lesion and sufficient information for feature extraction.

[0063] Step A2, respectively extracting features from all second ultrasonic images in the second ultrasonic image group to obtain a feature vector group;

[0064] Wherein, features are extracted from each second ultrasonic image respectively to obtain a group of feature vectors to describe the shape and position information of the ultrasonic image. In a preferred embodiment, the feature extraction step can be realized by using existing convolutional neural network (CNN) model or HIS statistical gray distribution method, and the convolutional neural network model preferably uses network models such as VGG, ResNet or Inception. In another preferred embodiment, the CNN model can also be designed according to specific needs to realize personalized feature extraction. For each ultrasonic image, input it into the CNN model and extract the output of the last convolution layer as the feature vector of the image.

[0065] Further, before feature extraction, a data preprocessing step is further included: all ultrasonic images are preprocessed, including using anisotropic filtering and IIR filtering of front and rear frames to enhance the image, normalization and other processing to improve clustering accuracy. Then the preprocessed ultrasonic images are subjected to feature extraction.

[0066] Step A3, clustering the feature vector group based on a clustering algorithm to obtain a group of clustering clusters corresponding to different lesion shapes and positions;

[0067] The feature vectors of all images of the second ultrasound image group are clustered to obtain a group of clusters. In a preferred embodiment, the clustering operation can be implemented using common clustering algorithms such as K-Means, DBSCAN or hierarchical clustering, etc. The feature vectors are input into the clustering algorithm for clustering, and the clustering algorithm automatically divides all feature vectors into different cluster groups, each cluster group representing a lesion shape and location type. Preferably, the number of clusters and parameters can be selected according to specific requirements.

[0068] Step A4, the average value of the feature vectors of all second ultrasound images in each cluster of the cluster group is calculated respectively to obtain the average image of each cluster and the feature vector of the average image.

[0069] For each cluster, the average image of all second ultrasound images therein is calculated, and the feature vector of the average image is extracted. For each cluster, the feature vectors of all ultrasound images are averaged to obtain the average feature vector of the cluster. Then, the average feature vector is input into the CNN model, and the output of the last convolution layer is extracted as the feature vector of the average image of the cluster. In this step A4, the average value of the feature vectors in each cluster is calculated as the representative feature vector of the cluster, thereby reducing the complexity and subjectivity of feature extraction and making full use of all information.

[0070] It should be noted that, in order to improve the model reusability, the CNN model used in this step A4 can share the convolutional neural network model used in step A2. Further, the convolutional neural network model used in step A2 can also be used to implement the feature extraction operation for the first ultrasound image to be processed in the above step S1. The first ultrasound image to be processed is input into the CNN model, and the output of the last convolution layer is extracted as the feature vector of the image.

[0071] Step A5, classifying the feature vectors of all average images corresponding to the cluster group to determine the standard section type corresponding to each cluster.

[0072] The feature vectors of all average images are classified to determine the standard section type corresponding to each cluster. In a preferred embodiment, the classification operation can be implemented using common classification algorithms such as SVM, random forest or deep neural network, etc. When training the classifier, the feature vector of each average image is input as the input, and the standard section type corresponding thereto is output as the output. In practical applications, the standard section is selected as follows: for a real-time second ultrasound image, after extracting the feature vector thereof, the feature vector is compared with the representative feature vectors of all cluster groups, and the representative feature vector of the most similar cluster group is selected as the standard section.

[0073] Specifically, considering that the traditional method of calculating the feature vector of an ultrasound image is usually to obtain the feature vector by weighted averaging the gray value and direction information of each pixel point, this feature vector calculation method is simple and easy to implement, but it ignores the information redundancy between different directions, which may lead to underestimation or overestimation of information in some directions. In the present embodiment, the method of calculating the average of the feature vector of an ultrasound image can better utilize the information redundancy between different directions, thereby more comprehensively describing the features of each pixel point in the ultrasound image; at the same time, this method can better capture the differences between different tissues or organs and improve the accuracy of classification and diagnosis.

[0074] In addition, calculating the average of the feature vector of an ultrasound image can also reduce the amount of calculation, thereby improving the training efficiency of the model. Because only the average of the feature vector of each ultrasound image needs to be calculated, and the feature vector of each pixel point does not need to be calculated, the calculation time and memory requirement can be greatly reduced.

[0075] As a preferred implementation, in step A3, the clustering algorithm is a K-means clustering algorithm, or a density-based clustering algorithm, or a hierarchical clustering algorithm.

[0076] Specifically, in the present embodiment, a hierarchical clustering algorithm is used to cluster all ultrasound images, and the feature vectors are input into the hierarchical clustering algorithm for clustering. The hierarchical clustering algorithm will automatically divide all feature vectors into different clustering clusters, and each clustering cluster represents a lesion morphology and position type, so as to obtain different lesion morphologies and position types. The hierarchical clustering algorithm is a clustering method based on tree structure, and its main idea is to divide data into different hierarchical structures and gradually merge data points with similar features, finally obtaining a tree structure.

[0077] As a preferred implementation, in step A4, the average of the feature vectors of all second ultrasound images in each clustering cluster of the clustering cluster group is calculated using a feature calculation function; and the feature calculation function is a feature fusion function.

[0078] Specifically, in the present embodiment, in the above step A4, for each ultrasound image in the data set, the feature vector thereof is calculated, which can be realized by using a feature calculation function based on the scikit-learn library in Python language, such as a Feature Union function or an FMRI to Feature function, etc. These functions can merge the feature vectors of multiple ultrasound images into one feature vector. For each ultrasound image, the average of the feature vector thereof can be calculated by using the mean function in NumPy (Numerical Python).

[0079] Further, in the step A5, the classification operation is performed using a deep neural network, the deep neural network includes a convolutional layer, a pooling layer, and a fully connected layer, and different from a conventional VGG, RESNET or the like, the deep neural network adopts a large convolutional kernel of 13*13 for the first layer of the convolutional layer, and the output of the first layer of the convolutional layer is directly connected by skip connection and then added to the fully connected layer, so that the features of high and middle layers are maximally reserved and are not easily lost with the deepening of the layers, and compared with the network of the same number of layers, the sensitivity and specificity are both improved.

[0080] As a preferred embodiment, the method further comprises a database, and the database stores a standard section template corresponding to each standard section type, and the standard section template at least indicates a position and an angle corresponding to the standard section;

[0081] As shown in Figure 4 , the step S3 comprises:

[0082] In the step S31, the standard section template corresponding to the determined standard section type is selected from the database;

[0083] In the step S32, the position and the angle of the first ultrasound image are adaptively selected according to the standard section template.

[0084] Specifically, in the embodiment, the standard section is selected according to the standard section type. For each standard section type, a standard section template is determined in advance and stored in the database. In actual application, the standard section type corresponding to the real-time ultrasound image is determined by matching, and then the standard section template corresponding to the standard section type is selected from the database, and the position and the angle similar to the standard section template are adaptively selected in the real-time ultrasound image.

[0085] The application further provides an ultrasound lesion standard section selection system based on deep learning, which is used to implement the ultrasound lesion standard section selection method based on deep learning as described above, and as shown in Figure 5 , the system comprises:

[0086] The first feature extraction module 1 is used to acquire the first ultrasound image to be processed and extract a first feature vector of the first ultrasound image;

[0087] The feature matching module 2 is connected to the first feature extraction module 1 and is used to match the first feature vector with the feature vector of each average image corresponding to each standard section type, and determine the standard section type corresponding to the average image with the highest similarity with the first feature vector;

[0088] The section selection module 3 is connected to the feature matching module 2 and is used to determine the position and the angle corresponding to the standard section according to the standard section type.

[0089] As a preferred implementation, as shown in Figure 6 , it further comprises:

[0090] The section registration module 4 is connected to the section selection module 3 and is configured to register the real-time ultrasound image after the adaptive adjustment of the position and the angle with the standard section corresponding to the standard section type determined.

[0091] Specifically, in actual application, due to the deformation and noise of the ultrasound image, there will be certain differences between the standard section and the ultrasound image. In this embodiment, the standard section is registered with the ultrasound image to avoid the differences between the standard section and the ultrasound image from reducing the comparability and accuracy of the test data. Further, in order to register the standard section with the ultrasound image, an image registration algorithm such as a phase correlation algorithm, a mutual information algorithm or a deformation field algorithm can be used to implement the registration operation.

[0092] As a preferred implementation, as shown in Figure 7 , it further comprises:

[0093] The image acquisition module 5 is configured to acquire a second ultrasound image group containing a lesion;

[0094] The second feature extraction module 6 is connected to the image acquisition module 5 and is configured to perform feature extraction on all second ultrasound images in the second ultrasound image group respectively to obtain a feature vector group;

[0095] The clustering module 7 is connected to the second feature extraction module 6 and is configured to perform clustering processing on the feature vector group based on a clustering algorithm to obtain a clustering cluster group corresponding to different lesion shapes and position types;

[0096] The average processing module 8 is connected to the clustering module 7 and is configured to calculate the average value of the feature vectors of all second ultrasound images in each clustering cluster of the clustering cluster group respectively to obtain an average image of each clustering cluster and a feature vector of the average image;

[0097] The classification module 9 is connected to the average processing module 8 and is configured to classify the feature vectors of all average images corresponding to the clustering cluster group to determine the standard section type corresponding to each clustering cluster.

[0098] The advantage or beneficial effect of the above technical scheme is that the feature vector of the obtained ultrasonic image is matched with the feature vector of the average image corresponding to each standard section type, the standard section type corresponding to the average image with the highest similarity with the first feature vector is determined, and then the position and angle corresponding to the standard section type are determined according to the standard section type, so that the application has wide applicability and practicability; meanwhile, the comparability and accuracy of the clinical trial data can be improved, and the subjective factors of the ultrasonic scanning operation are reduced, so that the scientificity and reliability of the clinical trial are improved.

[0099] The above is only the preferred embodiment of the present application, and does not limit the implementation and protection scope of the present application. Those skilled in the art should realize that any equivalent replacement and obvious change made according to the present specification and drawings should be included in the protection scope of the present application.

Claims

1. A deep learning-based method for selecting a standard section of an ultrasound lesion, characterized by, The method comprises the following steps: S1, obtaining a first ultrasound image to be processed, and extracting a first feature vector of the first ultrasound image; S2, matching the first feature vector with a feature vector of an average image corresponding to each standard section type, and determining the standard section type corresponding to the average image with the highest similarity with the first feature vector; S3, determining the position and angle of a standard section according to the standard section type; The step S2 comprises the following steps of obtaining the feature vector of the average image corresponding to each standard section type: A1, obtaining a second ultrasound image group containing lesions; A2, respectively extracting features of all second ultrasound images in the second ultrasound image group to obtain a feature vector group; A3, performing clustering processing on the feature vector group based on a clustering algorithm to obtain a clustering cluster group corresponding to different lesion shapes and positions; A4, respectively calculating the average value of the feature vectors of all second ultrasound images in each clustering cluster of the clustering cluster group to obtain an average image of each clustering cluster and a feature vector of the average image; A5, classifying the feature vectors of all average images corresponding to the clustering cluster group to determine the standard section type corresponding to each clustering cluster; the step A5 uses a deep neural network for classification, and the deep neural network comprises a convolution layer, a pooling layer and a full connection layer, the convolution layer uses a 13*13 convolution kernel, and the output of the convolution layer is directly connected and then added to the full connection layer. 2.The deep learning-based ultrasound lesion standard cross-section selection method according to claim 1, characterized in that, The step S3 further comprises the following step S4: S4, registering the real-time ultrasound image after adaptive adjustment of the position and angle with the standard section corresponding to the determined standard section type. 3.The deep learning-based ultrasound lesion standard cross-section selection method according to claim 2, characterized in that, The step S4 uses an image registration algorithm for registration, and the image registration algorithm uses a phase correlation algorithm, a mutual information algorithm or a deformation field algorithm. 4.The deep learning-based ultrasound lesion standard cross-section selection method according to claim 1, characterized in that, In the step A3, the clustering algorithm is a K-means clustering algorithm, a density-based clustering algorithm or a hierarchical clustering algorithm. 5.The deep learning-based ultrasound lesion standard cross-section selection method according to claim 1, characterized in that, In the step A4, the average value of the feature vectors of all second ultrasound images in each clustering cluster of the clustering cluster group is calculated by using a feature calculation function; and the feature calculation function is a feature fusion function. 6.The deep learning-based ultrasound lesion standard cross-section selection method according to claim 1, characterized in that, The method further comprises a database, and the database stores a standard section template corresponding to each standard section type, and the standard section template at least indicates the position and angle of the standard section; The step S3 comprises the following steps: S31, selecting the standard section template corresponding to the determined standard section type from the database according to the determined standard section type; S32, adaptively selecting the position and angle of the first ultrasound image according to the standard section template. 7.A deep learning based ultrasound lesion standard section selection system, characterized in that, The method is used for implementing the deep learning-based ultrasound lesion standard section selection method according to any one of claims 1-6, and comprises: a first feature extraction module, configured to obtain a first ultrasound image to be processed, and extract a first feature vector of the first ultrasound image; The feature matching module, connected with the first feature extraction module, is configured to match the first feature vector with a feature vector of an average image corresponding to each standard section type, and determine the standard section type corresponding to the average image with the highest similarity to the first feature vector; The section selection module, connected with the feature matching module, is configured to determine a position and an angle of a standard section corresponding to the standard section type.

8. The deep learning-based ultrasound lesion standard view selection system of claim 7, wherein, Further comprising: The section registration module, connected with the section selection module, is configured to register the real-time ultrasound image after the adaptive adjustment of the position and the angle with the standard section corresponding to the determined standard section type.

9. The deep learning-based ultrasound lesion standard view selection system of claim 7, wherein, Further comprising: The image acquisition module is configured to acquire a second ultrasound image group containing a lesion; The second feature extraction module, connected with the image acquisition module, is configured to perform feature extraction on all second ultrasound images in the second ultrasound image group respectively to obtain a feature vector group; The clustering module, connected with the second feature extraction module, is configured to perform clustering processing on the feature vector group based on a clustering algorithm to obtain a clustering cluster group corresponding to different lesion shapes and position types; The average processing module, connected with the clustering module, is configured to calculate an average value of feature vectors of all second ultrasound images in each clustering cluster of the clustering cluster group respectively to obtain an average image of each clustering cluster and a feature vector of the average image; The classification module, connected with the average processing module, is configured to classify the feature vectors of all average images corresponding to the clustering cluster group to determine the standard section type corresponding to each clustering cluster.

Images