Cross - domain cardiac ultrasound multi - sectional quality assessment method and system

Abstract

The present invention belongs to the technical field of image processing. A cross-domain cardiac ultrasound multi-section quality assessment method and system are provided. The features of the source-domain image and the target-domain image are extracted layer by layer through a multi-layer convolutional neural network, and the cross-domain similarity comparison is performed on the extracted features to obtain the locally indistinguishable features after registration. Channel weighting processing and spatial weighting processing are respectively performed on the locally indistinguishable features to obtain channel features and spatial features. The channel features, spatial features, and the locally indistinguishable features are aggregated to obtain specifically weighted features. Linear mapping is performed on the specifically weighted features, and feature embedding is performed on the result of the linear mapping according to the features of the target-domain image. The global aggregated features are obtained according to the result of the feature embedding. The quality score of the target-domain image is calculated according to the global aggregated features. The present invention enhances the accuracy and generalization performance of the quality assessment.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly to a cross - domain cardiac ultrasound multi - section quality assessment method and system. Background Technique

[0002] The statements in this part only provide background techniques related to the present invention and do not necessarily constitute prior art.

[0003] Cardiac ultrasound imaging is an important non - invasive diagnostic method and is widely used in the diagnosis and treatment of cardiovascular diseases. However, due to the complexity of the cardiac structure and the influence of the acquisition device and operation level on the ultrasound image quality, cross - domain cardiac multi - section ultrasound image quality assessment faces great challenges. Current image quality assessment techniques mainly rely on deep learning and image processing technologies. These methods often perform poorly in multi - domain applications across devices and patients, and it is difficult to adapt to the domain transfer differences between different data sources, resulting in insufficient robustness in multi - domain scenarios.

[0004] In the quality assessment of multi - domain cardiac ultrasound images, traditional deep learning algorithms are limited by factors such as noise, low resolution, and deformation of ultrasound images, and it is difficult to effectively capture general - purpose features. Especially in scenarios across devices and populations, the "domain shift" problem caused by inter - domain data differences significantly reduces the generality and accuracy of existing methods. In addition, existing algorithms often ignore the differences between local and global aggregated features in images and are difficult to accurately capture important information at different spatial scales, resulting in poor performance in image quality assessment. In addition, current algorithms usually ignore the hierarchy between local and global aggregated features in images and are difficult to comprehensively capture key information in multi - scale spaces, thus limiting their performance in image quality assessment. Summary of the Invention

[0005] To solve the deficiencies of the prior art, the present invention provides a cross - domain cardiac ultrasound multi - section quality assessment method and system. By learning domain - invariant features between different domains through the semantic transfer invariance of cross - domain features, the quality assessment network is made applicable to multi - source data from different data sources, enhancing the accuracy and generalization performance of quality assessment.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] In the first aspect, the present invention provides a cross - domain cardiac ultrasound multi - section quality assessment method.

[0008] A cross - domain cardiac ultrasound multi - section quality assessment method includes the following processes:

[0009] Obtain source - domain images and target - domain images;

[0010] Extract the features of the source domain image and the target domain image layer by layer through a multi-layer convolutional neural network, perform cross-domain similarity comparison on the extracted features, introduce local cross-domain constraints to learn the hybrid registration of the source domain image and the target domain image, and obtain the locally indistinguishable features after registration;

[0011] Perform channel weighting processing and spatial weighting processing on the locally indistinguishable features respectively to obtain channel features and spatial features, aggregate the channel features, spatial features and the locally indistinguishable features to obtain specifically weighted features, perform linear mapping on the specifically weighted features, perform feature embedding on the result of the linear mapping according to the features of the target domain image, and obtain global aggregated features according to the result of the feature embedding;

[0012] Calculate the quality score of the target domain image according to the global aggregated features.

[0013] As a further limitation of the first aspect of the present invention, the local cross-domain constraint is: , where is the source domain feature extracted from the th sample of the multi-layer convolutional neural network, is the Gaussian kernel function, is the target domain feature extracted from the th sample of the multi-layer convolutional neural network. By performing similarity registration constraints on the features extracted from each layer, local feature invariance registration is finally achieved to obtain locally indistinguishable features , The meaning of represents the source domain, The meaning of is the second-layer neural network, used to calculate the norm.

[0014] As a further limitation of the first aspect of the present invention, performing channel weighting processing on the locally indistinguishable features includes: ， where is the global pooling parameter, is the channel parameter of the first fully connected layer, is the channel parameter of the second fully connected layer, is the locally indistinguishable feature.

[0015] As a further limitation of the first aspect of the present invention, performing spatial weighting processing on the locally indistinguishable features includes: , where is the spatial parameter of the first fully connected layer, is the spatial parameter of the second fully connected layer, Parameters of the 1D convolutional layer.

[0016] As a further limitation of the first aspect of the present invention, aggregating the channel feature, the spatial feature, and the local non-differentiated feature to obtain a specific weighted feature, including: , where is the specific weighted feature, is the spatial feature, is the channel feature, is the local non-differentiated feature, is the element-wise multiplication representation.

[0017] As a further limitation of the first aspect of the present invention, obtaining a quality score of the target domain image according to the feature of the target domain image and the global aggregated feature, including:

[0018] Specific weighted feature After being linearly mapped and embedded with the feature of the target image, it is sent to the Transformer encoder to extract the global aggregated feature with structural consistency in the source domain and the target domain. The global aggregated feature is sent to the view classifier to obtain a classification result. The extracted global aggregated feature and the classification result of the view classifier are input to the Transformer decoder, and cross-domain mapping and semantic alignment are performed in combination with the feature of the target domain image. The output of the Transformer decoder is scored for image quality through a multi-layer perceptron.

[0019] In a second aspect, the present invention provides a cross-domain cardiac ultrasound multi-plane quality assessment system.

[0020] A cross-domain cardiac ultrasound multi-plane quality assessment system, characterized by including:

[0021] An image acquisition unit, configured to: acquire a source domain image and a target domain image;

[0022] A local feature extraction unit, configured to: layer by layer extract the features of the source domain image and the target domain image through a multi-layer convolutional neural network, perform cross-domain similarity comparison on the extracted features, introduce local cross-domain constraints to learn the hybrid registration of the source domain image and the target domain image, and obtain the registered local non-differentiated feature;

[0023] A global aggregated feature extraction unit, configured to: respectively perform channel weighting processing and spatial weighting processing on the local non-differentiated feature to obtain a channel feature and a spatial feature, aggregate the channel feature, the spatial feature, and the local non-differentiated feature to obtain a specific weighted feature, perform a linear mapping on the specific weighted feature, perform feature embedding on the result of the linear mapping according to the feature of the target domain image, and obtain a global aggregated feature according to the result of the feature embedding;

[0024] A quality score generation unit, configured to calculate a quality score of a target domain image based on global aggregated features.

[0025] In a third aspect, the present invention provides a computer device, including: a processor and a computer-readable storage medium;

[0026] The processor is adapted to execute a computer program;

[0027] The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, it implements the cross-domain multi-plane quality assessment method for cardiac ultrasound as described in the first aspect of the present invention.

[0028] In a fourth aspect, the present invention provides a computer-readable storage medium, which stores a computer program, and the computer program is adapted to be loaded and executed by a processor to implement the cross-domain multi-plane quality assessment method for cardiac ultrasound as described in the first aspect of the present invention.

[0029] In a fifth aspect, the present invention provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the cross-domain multi-plane quality assessment method for cardiac ultrasound as described in the first aspect of the present invention.

[0030] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0031] The present invention extracts the cross-domain semantic invariance of the anatomical structures of source domain data and target domain data, enhancing the detail robustness of local features; introduces a multi-channel hybrid Transformer to learn cross-domain hybrid semantic expressions, focuses on key features through weight allocation, and improves the specificity of evaluation; ensures efficiency and accuracy in cross-device and cross-population environments through aggregated global topological consistency registration; outputs cross-domain quality assessment results, can automatically evaluate the quality of cross-domain multi-view ultrasound images of the heart, effectively improve the diagnostic accuracy, and reduce the influence of human factors.

[0032] Advantages of additional aspects of the present invention will be partially given in the following description, partially become apparent from the following description, or be understood through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The specification drawings forming a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention.

[0034] Figure 1 It is a schematic flowchart of the cross-domain multi-plane quality assessment method for cardiac ultrasound provided in Embodiment 1 of the present invention;

[0035] Figure 2 A schematic diagram illustrating the principle of the cross-domain cardiac ultrasound multi-slice quality assessment method provided in Example 1 of the present invention;

[0036] Figure 3 Schematic diagram of the final quality scoring results provided in Example 1 of the present invention; wherein (a) is the scoring result of the first ultrasound image, (b) is the scoring result of the second ultrasound image, (c) is the scoring result of the third ultrasound image, (d) is the scoring result of the fourth ultrasound image, and (e) is the scoring result of the fifth ultrasound image;

[0037] Figure 4 A schematic diagram of a cross-domain cardiac ultrasound multi-slice quality assessment system provided in Example 2 of the present invention;

[0038] Figure 5 A schematic diagram of a computer device provided in Example 3 of the present invention. DETAILED DESCRIPTION

[0039] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0040] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0041] In the absence of conflict, the embodiments of the present invention and the features thereof may be combined with each other.

[0042] Example 1:

[0043] This implementation proposes a cross-domain cardiac ultrasound multi-section quality assessment method, such as Figure 1 As shown, the following process is included:

[0044] S1: Acquire source domain images and target domain images;

[0045] S2: A multi-layer convolutional neural network is used to extract features of the source and target domain images layer by layer. The extracted features are compared for cross-domain similarity. Local cross-domain constraints are introduced to learn the hybrid registration of the source and target domain images, and the registered local indifferent features are obtained.

[0046] S3: Perform channel weighting processing and spatial weighting processing on the local invariant features respectively to obtain channel features and spatial features, aggregate the channel features, spatial features and the local invariant features to obtain specific weighted features, perform linear mapping on the specific weighted features, perform feature embedding on the result of the linear mapping according to the features of the target domain image, and obtain global aggregated features according to the result of the feature embedding;

[0047] S4: Calculate the quality score of the target domain image according to the global aggregated features.

[0048] More specifically, as Figure 2 shown, the model architecture of the entire evaluation method includes a local invariance module, a multi-channel hybrid Transformer model, and a global transfer aggregation unit module. Through the combined action of the local invariance module, the multi-channel hybrid Transformer model, and the global transfer aggregation unit module, cross-domain multi-view cardiac ultrasound quality assessment is realized, thereby providing an effective solution for the generalizable quality assessment of multi-plane cardiac ultrasound images.

[0049] In steps S1 and S2 of this implementation manner, specifically, the source domain images in the source domain dataset and the target domain images in the target domain dataset are input into the local invariance module, and the source domain images and the target domain images are enhanced in structural consistency and local features of the images are obtained to enhance the robustness of the algorithm between different domains.

[0050] The present invention performs feature subspace mapping using the image features of the source domain and the image features of the target domain. As Figure 2 shown, a feature extractor is constructed through a multi-layer convolutional neural network to extract local features. The feature extraction by any layer of the convolutional neural network is performed using the following formula:

[0051] (1);

[0052] where is the source domain image, is the target domain image, is the feature extractor, and are the extracted source domain local features and target domain local features.

[0053] During this process, cross-domain similarity comparison is performed by extracting features layer by layer in a three-layer neural network, and local cross-domain constraints are introduced during this process Learn the hybrid registration of the source domain and the target domain, improve the alignment effect of cross-domain features by learning local structural similarity, so as to achieve the alignment of the geometric features of the local anatomical structure of the source and target domain features, and enhance the local semantic transfer invariance of the evaluation.

[0054] More specifically, taking the second-layer neural network as an example, The specific implementation process is as follows:

[0055] (2);

[0056] Among them, is the source domain feature extracted from the th sample of the multi-layer convolutional neural network, is the Gaussian kernel function, is the target domain feature extracted from the th sample of the multi-layer convolutional neural network. By performing similarity registration constraints on the features extracted from each layer, local feature invariance registration is finally achieved, and local indistinguishable features , The meaning of is the expected value, represents the source domain, The meaning of is the target domain, is the second-layer neural network, is used to calculate the norm.

[0057] In step S3 of this implementation method, specifically, it includes: inputting the registered local indistinguishable features into a multi-channel hybrid Transformer model, and optimizing the extraction and expression of cross-domain features by introducing spatial and channel weighting mechanisms.

[0058] First, selectively focus on the key regions in the image through the spatial weighting mechanism; then, filter and enhance the important image features through the channel weighting mechanism; finally, achieve the three-dimensional expression of features through weighted feature merging.

[0059] Specifically, in the channel excitation stage, the model introduces the Squeeze-and-Excitation Network as the backbone, and introduces specific channel information through the excitation operation. The specific formula is as follows:

[0060] (3);

[0061] Among them, is the global pooling parameter, is the channel parameter of the first fully connected layer, is the channel parameter of the second fully connected layer, is the local indistinguishable feature.

[0062] In the spatial excitation stage, since the undifferentiated tokenization method of the transformer assigns the same weight to each patch, which limits the expression of the correlation between important features. Therefore, the present invention improves on the basis of the Squeeze-and-Excitation Network and creates a patch-based excitation module to weight the correlation between spatial patches. The specific formula is as follows:

[0063] (4):

[0064] Wherein, is the spatial parameter of the first fully connected layer, is the spatial parameter of the second fully connected layer, is the parameter of the 1D convolutional layer.

[0065] Under the above operations, through spatial-channel collaborative weighting, multi-channel hybrid weighting is finally achieved to increase the specificity of the network:

[0066] (5);

[0067] Wherein, is the element multiplication representation, is the local undifferentiated feature, is the specific weighted feature.

[0068] By aggregating the original features and channel-spatial weighting, the specific weighted feature with specificity is finally obtained , and the model gradually optimizes the specificity of the input features during training to adapt to the complex and changing ultrasound image environment and improve the specificity and accuracy of quality assessment.

[0069] In step S4, to further improve the cross-domain generalization ability of the model for ultrasound images, a global transfer aggregation unit is proposed to achieve the deep fusion of the global semantic structure and local discriminative features. The key to the design of this module is to construct a transfer-robust aggregation feature through a dual-input Transformer decoder and a global invariance guidance mechanism. The specific process is as follows: First, taking the multi-channel hybrid Transformer specific weighted feature as the input, after linear mapping, after the position embeddings of " ", "1", "2", "3", "4", "5", "6", "7#" (i.e., taking the features of the target image as the position embeddings), it is sent into the Transformer encoder (i.e., Figure 2The encoder) is used to extract globally aggregated features with structural consistency in the source domain and the target domain. These encoded features capture high-level spatial semantic information shared across domains and form the basis for subsequent aggregation modeling. At the same time, the above encoded features are also fed into the view classifier (MLP Class) for training on the standard view recognition task (obtaining globally invariant features and performing transfer-invariant aggregation). The extracted globally aggregated features and the output of the view classifier (MLP Class) are then input into the Transformer decoder (i.e., Figure 2 the decoder in) to perform cross-domain mapping and semantic alignment in combination with the features of the target domain image. In the above process, the model learns high-level semantic representations shared between different domains, thereby constructing globally aggregated features that are robust to domain changes. Based on the local features, the obtained globally aggregated features make full use of the complementary information between local details and global structures, and finally, the image quality is scored through a multi-layer perceptron (MLP Quality) (i.e., the quality assessment result).

[0070] As Figure 3 shown, Figure 3 in (a) is the scoring result of the first ultrasound image, with a score of 5.0; Figure 3 in (b) is the scoring result of the second ultrasound image, with a score of 4.3; Figure 3 in (c) is the scoring result of the third ultrasound image, with a score of 3.1; Figure 3 in (d) is the scoring result of the fourth ultrasound image, with a score of 2.0; Figure 3 in (e) is the scoring result of the fifth ultrasound image, with a score of 1.0. It can be seen that the present invention can accurately obtain the scoring result.

[0071] Embodiment 2:

[0072] As Figure 4 shown, this implementation provides a cross-domain cardiac ultrasound multi-plane quality assessment system, including:

[0073] An image acquisition unit, configured to: acquire source domain images and target domain images;

[0074] A local feature extraction unit, configured to: layer by layer extract the features of the source domain image and the target domain image through a multi-layer convolutional neural network, perform cross-domain similarity comparison on the extracted features, introduce local cross-domain constraints to learn the hybrid registration of the source domain image and the target domain image, and obtain the registered local indistinguishable features;

[0075] The global aggregation feature extraction unit is configured to: perform channel weighting processing and spatial weighting processing on the local non-discriminative features respectively to obtain channel features and spatial features, aggregate the channel features, spatial features and the local non-discriminative features to obtain specific weighted features, perform linear mapping on the specific weighted features, perform feature embedding on the result of the linear mapping according to the features of the target domain image, and obtain global aggregation features according to the result of the feature embedding;

[0076] The quality score generation unit is configured to: calculate the quality score of the target domain image according to the global aggregation features.

[0077] The specific working processes of the above units are described in Embodiment 1 and will not be elaborated here.

[0078] It can be understood that the above units can be separately or wholly combined into one or several other units to form, or some of them can be further split into multiple smaller units with more specific functions to form, which can achieve the same operations without affecting the realization of the technical effects of the embodiments of this application. The above units are divided based on logical functions. In practical applications, the function of one unit can also be realized by multiple units, or the functions of multiple units can be realized by one unit. In other embodiments of this application, the system can also include other units. In practical applications, these functions can also be assisted by other units and can be realized by the cooperation of multiple units.

[0079] According to another embodiment of this application, the system described in this embodiment can be constructed and the method of Embodiment 1 of this application can be realized by running a computer program (including program code) capable of executing the respective steps involved in the corresponding method described in Embodiment 1 on a general computing device such as a computer including processing elements and storage elements such as a central processing unit (CPU), a random access memory (RAM), and a read-only memory (ROM). The computer program can be recorded on a computer-readable recording medium, for example, and loaded into the above computing device through the computer-readable recording medium and run therein.

[0080] Embodiment 3:

[0081] As Figure 5 shown, this implementation provides an electronic device, which includes a processor 1001, a communication interface 1002, and a computer-readable storage medium 1003. Among them, the processor 1001, the communication interface 1002, and the computer-readable storage medium 1003 can be connected through a bus or other means.

[0082] Among them, the communication interface 1002 is used to receive and send data. The computer-readable storage medium 1003 can be stored in the memory of the electronic device. The computer-readable storage medium 1003 is used to store computer programs, and the computer programs include program instructions. The processor 1001 is used to execute the program instructions stored in the computer-readable storage medium 1003.

[0083] The processor 1001 (or CPU (Central Processing Unit, central processor)) is the computing core and control core of the electronic device, and is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions to implement the corresponding method flow or corresponding function.

[0084] The processor 1001 is configured to execute the following process:

[0085] Obtain the source domain image and the target domain image;

[0086] Extract the features of the source domain image and the target domain image layer by layer through a multi-layer convolutional neural network, perform cross-domain similarity comparison on the extracted features, introduce local cross-domain constraints to learn the hybrid registration of the source domain image and the target domain image, and obtain the locally indistinguishable features after registration;

[0087] Perform channel weighting processing and spatial weighting processing on the locally indistinguishable features respectively to obtain channel features and spatial features, aggregate the channel features, spatial features and the locally indistinguishable features to obtain specifically weighted features, perform linear mapping on the specifically weighted features, perform feature embedding on the result of the linear mapping according to the features of the target domain image, and obtain the global aggregated features according to the result of the feature embedding;

[0088] Calculate the quality score of the target domain image according to the global aggregated features.

[0089] For the specific working process, see the introduction in Embodiment 1, which will not be elaborated here.

[0090] Embodiment 4:

[0091] This implementation provides a computer-readable storage medium (Memory). The computer-readable storage medium is a memory device in the electronic device and is used to store programs and data. It can be understood that the computer-readable storage medium here can include both the built-in storage medium in the electronic device and, of course, the extended storage medium supported by the electronic device. The computer-readable storage medium provides a storage space, and this storage space stores the processing system of the electronic device.

[0092] Furthermore, the storage space also stores one or more instructions suitable for being loaded and executed by the processor. These instructions may be one or more computer programs (including program code). It should be noted that the computer-readable storage medium herein may be a high-speed RAM memory or a non-volatile memory, such as at least one disk storage device; alternatively, it may be at least one computer-readable storage medium located remotely from the processor.

[0093] In one embodiment, the computer-readable storage medium stores one or more instructions; the processor loads and executes the one or more instructions stored in the computer-readable storage medium to implement the following process:

[0094] Obtain source domain images and target domain images;

[0095] The features of the source domain image and the target domain image are extracted layer by layer through a multi-layer convolutional neural network. The extracted features are compared for cross-domain similarity. Local cross-domain constraints are introduced to learn the hybrid registration of the source domain image and the target domain image, and the local indifferent features after registration are obtained.

[0096] performing channel weighting processing and spatial weighting processing on the local indifference features respectively to obtain channel features and spatial features, aggregating the channel features, spatial features and the local indifference features to obtain specific weighted features, performing linear mapping on the specific weighted features, performing feature embedding on the results of the linear mapping according to the features of the target domain image, and obtaining global aggregated features based on the result of the feature embedding;

[0097] The quality score of the target domain image is calculated based on the globally aggregated features.

[0098] Example 5:

[0099] This implementation provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the following process:

[0100] Obtain source domain images and target domain images;

[0101] The features of the source domain image and the target domain image are extracted layer by layer through a multi-layer convolutional neural network. The extracted features are compared for cross-domain similarity. Local cross-domain constraints are introduced to learn the hybrid registration of the source domain image and the target domain image, and the local indifferent features after registration are obtained.

[0102] Perform channel weighting processing and spatial weighting processing on the local indistinguishable features respectively to obtain channel features and spatial features, aggregate the channel features, spatial features and the local indistinguishable features to obtain specific weighted features, perform linear mapping on the specific weighted features, perform feature embedding on the result of the linear mapping according to the features of the target domain image, and obtain global aggregated features according to the result of the feature embedding;

[0103] Calculate the quality score of the target domain image according to the global aggregated features.

[0104] The specific working process is described in Embodiment 1 and will not be elaborated here.

[0105] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in this application can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0106] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions according to the embodiments of this application are generated in whole or in part. The computer can be a general-purpose computer, a dedicated computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center in a wired manner (for example, coaxial cable, optical fiber, digital line (DSL)) or a wireless manner (for example, infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that the computer can access or a data processing device such as a server, data center, etc. that includes one or more available media integrated. The available media can be magnetic media (for example, floppy disk, hard disk, magnetic tape), optical media (for example, DVD), or semiconductor media (for example, solid state disk (SSD)), etc.

[0107] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A cross - domain cardiac ultrasound multi - section quality assessment method, characterized in that It includes the following processes: Obtain the source domain image and the target domain image; Extract the features of the source domain image and the target domain image layer by layer through a multi-layer convolutional neural network, perform cross-domain similarity comparison on the extracted features, introduce local cross-domain constraints to learn the hybrid registration of the source domain image and the target domain image, and obtain the locally indistinguishable features after registration; Perform channel weighting processing and spatial weighting processing on the locally indistinguishable features respectively to obtain channel features and spatial features, aggregate the channel features, spatial features and the locally indistinguishable features to obtain specifically weighted features, perform linear mapping on the specifically weighted features, perform feature embedding on the result of the linear mapping according to the features of the target domain image, and obtain the global aggregated features according to the result of the feature embedding; Calculate the quality score of the target domain image according to the global aggregated features; Local cross-domain constraints is as follows: ； Among them, is the source domain feature extracted from the th sample of the multi-layer convolutional neural network, is the Gaussian kernel function, is the target domain feature extracted from the th sample of the multi-layer convolutional neural network. By performing similarity registration constraints on the features extracted from each layer, local feature invariance registration is finally achieved, and local indistinguishable features are obtained , The meaning of is the expected value, represents the source domain, The meaning of is the target domain, is the second-layer neural network, is used to calculate the norm.

2. The cross-domain cardiac ultrasound multi-plane quality assessment method according to claim 1, wherein Perform channel weighting processing on local non-differentiated features, including: ， Among them, is the global pooling parameter, is the channel parameter of the first fully connected layer, is the channel parameter of the second fully connected layer, is the local non-differentiated feature.

3. The cross-domain cardiac ultrasound multi-plane quality assessment method according to claim 1, wherein Perform spatial weighting processing on locally indistinguishable features, including: , where is the spatial parameter of the first fully connected layer, is the spatial parameter of the second fully connected layer, is the parameter of the 1D convolutional layer, is the locally indistinguishable feature.

4. The cross-domain cardiac ultrasound multi-plane quality assessment method according to claim 1, wherein Aggregate the channel feature, the spatial feature, and the local non-discriminative feature to obtain a specific weighted feature, including: , where is the specific weighted feature, is the spatial feature, is the channel feature, is the local non-discriminative feature, represents element-wise multiplication.

5. The cross-domain cardiac ultrasound multi-plane quality assessment method according to claim 4, wherein Obtain the quality score of the target domain image according to the features of the target domain image and the global aggregated features, including: Specific weighted features After linear mapping and embedding of the features of the target image, it is sent to the Transformer encoder to extract the globally aggregated features with structural consistency in the source domain and the target domain. The globally aggregated features are sent to the view classifier to obtain the classification result. The extracted globally aggregated features and the classification result of the view classifier are input into the Transformer decoder, and cross-domain mapping and semantic alignment are performed in combination with the features of the target domain image. The output of the Transformer decoder is scored for image quality through a multi-layer perceptron.

6. A cross - domain cardiac ultrasound multi - sectional quality assessment system, characterized in that, It includes: An image acquisition unit configured to: obtain the source domain image and the target domain image; A local feature extraction unit configured to: extract the features of the source domain image and the target domain image layer by layer through a multi-layer convolutional neural network, perform cross-domain similarity comparison on the extracted features, introduce local cross-domain constraints to learn the hybrid registration of the source domain image and the target domain image, and obtain the locally indistinguishable features after registration; A global aggregated feature extraction unit configured to: perform channel weighting processing and spatial weighting processing on the locally indistinguishable features respectively to obtain channel features and spatial features, aggregate the channel features, spatial features and the locally indistinguishable features to obtain specifically weighted features, perform linear mapping on the specifically weighted features, perform feature embedding on the result of the linear mapping according to the features of the target domain image, and obtain the global aggregated features according to the result of the feature embedding; A quality score generation unit configured to: calculate the quality score of the target domain image according to the global aggregated features; Local cross-domain constraints is as follows: ； Among them, is the source domain feature extracted from the th sample of the multi-layer convolutional neural network, is the Gaussian kernel function, is the target domain feature extracted from the th sample of the multi-layer convolutional neural network. By performing similarity registration constraints on the features extracted from each layer, local feature invariance registration is finally achieved, and local indistinguishable features are obtained. The meaning of is the expected value, represents the source domain, The meaning of is the target domain and is used to calculate the norm.

7. A computer device, characterized in that, It includes: A processor and a computer-readable storage medium; The processor is adapted to execute a computer program; The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, it implements the cross-domain cardiac ultrasound multi-plane quality assessment method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and the computer program is adapted to be loaded and executed by the processor to implement the cross-domain cardiac ultrasound multi-plane quality assessment method according to any one of claims 1 to 5.

9. A computer program product, characterized in that, The computer program product includes a computer program, and when the computer program is executed by the processor, it implements the cross-domain cardiac ultrasound multi-plane quality assessment method according to any one of claims 1 to 5.

Images